Light Scattering by Ice Crystals - Fundamentals and Applications PDF

LIGHT SCATTERING B Y ICE CRYSTA LS
Fundamentals and Applications
This research volume outlines the scientific foundations that are central to our current under-
standing of light scattering, absorption and polarization processes involving ice crystals.
It also demonstrates how data from satellite remote sensing of cirrus clouds (comprising
various ice crystal sizes and morphologies) can be combined with radiation parameteriza-
tions in climate models to estimate the role of these clouds in temperature and precipitation
responses to climate change. Providing a balanced treatment of both the fundamentals and
applications, this book synthesizes the authors’ own work, as well as that of other leading
researchers in this area, in a coherent and logical presentation. Numerous illustrations are
included, including three-dimensional schematics, in order to provide a concise discussion
of the subject and enable easy visualization of the key concepts.
This book is intended for active researchers and advanced graduate students in atmo-
spheric science, climatology and remote sensing, as well as scholars in related fields such
as ice microphysics, electromagnetic wave propagation, geometric optics, radiative transfer
and cloud–climate interactions.
kuo-nan liou is a Distinguished Professor of Atmospheric Sciences and founding

director of the Joint Institute for Regional Earth System Science and Engineering at the
University of California, Los Angeles. Dr. Liou has received numerous awards including
the 1998 Jule G. Charney Award from the American Meteorological Society, for his pio-
neering work in the theory and application of radiative transport and its interaction with
clouds, and the 2013 Roger Revelle Medal from the American Geophysical Union for out-
standing contributions in atmospheric sciences. He is a member of the National Academy
of Engineering and of Academia Sinica.
ping yang is Professor and Head of the Department of Atmospheric Sciences, and the
David Bullock Harris Chair in Geosciences, at Texas A&M University, where his research
interests cover the areas of remote sensing and radiative transfer. He received the 2013
Ascent Award from the Atmospheric Sciences Section of the American Geophysical Union,
of which he is an elected Fellow.
L I G H T S C AT T E R I N G B Y
I C E CRY S TA L S
Fundamentals and Applications
KUO-NA N L IOU
University of California, Los Angeles
and
P I N G YA N G
Texas A&M University
with contributions by
YO S H I H I D E TA K A N O
University Printing House, Cambridge CB2 8BS, United Kingdom
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9780521889162

C Kuo-Nan Liou and Ping Yang 2016
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2016
A catalog record for this publication is available from the British Library
ISBN 978-0-521-88916-2 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Epigraph
To all the happy ice crystals in planetary atmospheres
Let there be light.

Let there be beautiful ice crystals in the air and mountain ranges.
And here come the reindeers and Santa Claus carrying Maxwell’s equations, and
light rays are shining in the wonderlands.
Let the glory of Geometric Optics for ice crystals, Newton’s optics, and sun’s
light rays rise again from the horizon.
Let ice crystals’ old friends – black carbon and dust – be not forgot for Auld
Lang Syne.
And ice crystals are carried by the ceaseless winds; and
After traveling thousands of miles up and down, the sky looks very blue.
Let there be space missions to tender ubiquitous light rays in the sky,
And all things considered, let light scattering by ice crystals in remote sensing
and climate change be a delight.
Contents
Preface page xii

1 Ice in the Earth’s atmosphere 1
1.1 Introduction to clouds 1
1.2 Geographical and temporal distributions of ice in the Earth’s atmosphere 5
1.2.1 A global perspective on clouds 5
1.2.2 An example of global cloud climatology 9
1.2.3 View from ground-based instruments 11
1.2.4 View from specific satellite instruments 14
1.3 Formation of ice crystals 17
1.3.1 The structure of ice crystals 17
1.3.2 Homogeneous and heterogeneous nucleation 19
1.3.3 Atmospheric ice nuclei 20
1.3.4 Secondary ice particle production 22
1.4 Growth of ice crystals 22
1.4.1 Growth by diffusion 22
1.4.2 Growth by accretion 29
1.5 Illustration of ice crystal morphology 30
1.5.1 Some historical perspectives 30
1.5.2 Classification of ice crystals 32
1.5.3 Computer generation of three-dimensional ice crystals 33
1.6 Ice crystal size and shape distributions in cirrus clouds 41
1.6.1 Introductory remarks 41
1.6.2 Ice crystal size and shape spectra 41
1.6.3 Correlation between ice crystal maximum dimension and width 43
1.7 Correlation between ice water content and ice crystal size 45
1.7.1 Introductory note 45
1.7.2 Correlation analysis 46
vii
viii Contents
1.8 Numerical modeling of cirrus cloud formation 50

1.8.1 A review 50
1.8.2 A two-dimensional cirrus cloud model 52
1.8.2.1 Ice microphysics 53
1.8.2.2 Radiative transfer 55
2 Fundamentals of light scattering by ice crystals 58
2.1 Introductory remarks 58
2.2 Maxwell’s equations, wave equations, and boundary conditions 60
2.2.1 Maxwell’s equations 60
2.2.2 Boundary conditions 62
2.2.2.1 Normal components 62
2.2.2.2 Tangential components 63
2.2.3 Wave equations and some analytic solutions 64
2.2.3.1 Spherical coordinates 64
2.2.3.2 Cylindrical coordinates 68
2.2.3.3 Spheroidal coordinates 72
2.3 Optical properties of ice: index of refraction 75
2.3.1 Complex index of refraction and dispersion of light 75
2.3.2 Optical properties of ice 77
2.3.3 Indices of refraction for ice 79
2.4 Definition of the single-scattering and polarization properties
of ice crystals 83
2.4.1 Representation of a simple wave 83
2.4.2 Representation of electromagnetic waves by Stokes parameters 86
2.4.3 Scattering phase matrix for an ice crystal 93
2.4.4 Extinction cross-section for a non-spherical ice crystal 95
2.4.5 Scattering phase matrix for an ensemble of ice crystals 97
2.4.6 Some examples of the single-scattering properties of ice spheres,
cylinders, and spheroids 101
2.5 Introduction to radiative transfer in ice crystal clouds: link to single
scattering 106
2.5.1 Formulation of scalar and vector radiative transfer equations 106
2.5.2 Radiative transfer in horizontally oriented ice particles 111
3 Principles of geometric optics for application to light scattering by ice crystals 115
3.1 Application of geometric optics to light scattering by ice crystals 115
3.1.1 Snell’s and Fresnel’s laws for geometric optics 118
3.1.2 Absorption effects in the context of geometric optics 120
3.1.3 Geometric ray tracing and the Monte Carlo hit-and-miss approach
to ice crystals 124
3.1.4 Illustration of ice optics in the atmosphere 127
3.1.4.1 22° and 46° halos 128
3.1.4.2 Sundogs and their geometry 129
Contents ix
3.1.4.3 Horizontally oriented plate and column crystals 131

3.1.4.4 An observed halo and arc complex pattern 134
3.2 Fraunhofer diffraction for light scattering by ice crystals 134
3.2.1 Sphere 135
3.2.2 Rectangular parallelepiped 136
3.2.3 Hexagonal cylinder 139
3.2.4 Spheroid 140
3.2.5 Some numerical results 142
3.3 Generalized geometric optics for light scattering by ice crystals 146
3.3.1 Introduction 146
3.3.2 Conventional approach 147
3.3.3 Improved geometric-optics approach using the electromagnetic
equivalence theorem 158
3.3.4 Ray-by-ray approach to extinction and absorption calculations 163
3.3.5 Contributions of surface waves to spheres: the edge effect 165
3.3.6 The geometric-optics and surface-wave approach for spheres 168
3.3.7 Application of the geometric-optics and surface-wave approach to
ice crystals 169
3.3.7.1 Fundamentals 169
3.3.7.2 Geometric optics 170
3.3.7.3 Addition of surface waves 171
3.3.7.4 Comparison with available FDTD and DDA results 173
3.3.8 Geometric-optics and surface-wave approach coupled with the
Rayleigh–Gans–Debye approximation 173
3.4 A unified theory of light scattering by ice crystals based on generalized
geometric optics 178
3.4.1 The essence of a unified theory of light scattering by ice crystals 178
3.4.2 Laboratory scattering and ice microphysics experiments with ice
crystals 185
3.4.3 Applications to snow grains contaminated by black carbon (BC)
and dust 192
4 Other useful approaches to light scattering by ice particles 197
4.1 Finite-difference time domain method 197
4.1.1 A brief review 197
4.1.2 Numerical discretization of Maxwell’s curl equations 200
4.1.3 Absorbing boundary conditions and electromagnetic fields in the
frequency domain 204
4.1.4 Near-to-far-field mapping and optical property calculations 207
4.1.5 Validation of the method and some pertinent results 209
4.1.6 Pseudo-spectral time domain method 215
4.2 T-matrix numerical method 222
4.2.1 Fundamentals of the T-matrix approach 222
4.2.2 Invariant imbedding T-matrix method 226
x Contents
4.3 Discrete dipole approximation 238

4.3.1 Fundamentals of the DDA approach 238
4.3.2 Comparison among DDA, FDTD, and PSTD 244
5 Application of light scattering by ice crystals to remote sensing 247
5.1 Introduction to remote sensing of ice clouds in the Earth’s atmosphere 248
5.1.1 Atmospheric composition and structure 248
5.1.2 Atmospheric absorption spectrum 250
5.1.3 Relative position of the sun–satellite system and radiative transfer 253
5.1.4 Bidirectional reflectance spectra of ice 254
5.1.5 The A-Train satellite constellation 257
5.2 Determination of cirrus cloud optical depth and ice crystal size 260
5.2.1 Definition of optical depth and mean effective size 260
5.2.2 Two-channel inversion technique 262
5.2.3 The importance of ice crystal phase function in the two-channel
approach 267
5.2.4 The current MODIS operation program for ice cloud studies 269
5.2.5 Global mapping of cirrus clouds 277
5.2.5.1 MODIS 277
5.2.5.2 AIRS 279
5.2.5.3 CERES-2 283
5.2.6 Detection of thin cirrus 285
5.3 Remote sensing of cirrus cloud vertical size profile 288
5.3.1 The importance of cirrus cloud vertical profiles 288
5.3.2 Development of a retrieval algorithm for vertical sizing 291
5.3.3 Validation of the vertical sizing algorithm 295
5.4 Remote sensing of cirrus clouds using reflected polarization 301
5.4.1 An historical review of polarization measurements 301
5.4.2 Reflected polarization of sunlight from ice crystal clouds 303
5.5 Principle of backscattering depolarization for differentiation of ice
crystals and water drops 306
5.5.1 Theoretical basis 307
5.5.2 Issues of orientation and theoretical results 310
5.5.3 Experimental arrangements and results 313
5.5.4 Application to lidar (laser radar) 314
5.6 Remote sensing of cirrus using reflected line spectra in
1.38 and 0.76 µm bands 318
5.6.1 A brief review of the use of line spectra for Earth
remote sensing 318
5.6.2 Reflected line spectra in the 1.38 µm band 319
5.6.3 Reflected line spectra in the 0.76 µm band 323
5.6.4 Retrieval of cirrus cloud parameters using reflected line spectra 325
Contents xi
6 Application of light scattering by ice crystals to climate studies 327

6.1 Physical basis for the spectral single-scattering parameterization
of ice crystals 328
6.1.1 Physical foundations 329
6.1.2 A generalized single-scattering parameterization for ice crystals 332
6.1.3 Absorption line formation in the scattering of light by ice crystals 333
6.1.4 Single-scattering properties of combined ice crystals and
absorbing gases 337
6.2 Radiative transfer in cirrus cloudy atmospheres 339
6.2.1 The subject of radiative transfer for application to climate studies 339
6.2.2 Radiative transfer model results versus aircraft observations for
cirrus clouds 341
6.2.3 Radiative transfer model results compared with satellite
observations 345
6.3 Radiative forcing of cirrus clouds 350
6.3.1 A discussion of the radiative forcing of cirrus clouds 350
6.3.2 Theoretical analysis for cirrus radiative forcing 351
6.4 Climatic effects of cirrus clouds 357
6.4.1 Investigation using one-dimensional climate models 357
6.4.1.1 A one-dimensional cloud–precipitation–climate model 359
6.4.1.2 A one-dimensional climate–cirrus-cloud model 362
6.4.2 Ice microphysics sensitivity investigation using GCMs 365
6.4.2.1 The effect of interactive ice crystal size on precipitation
simulations 365
6.4.2.2 The effect of ice crystal shape on cloud radiative forcing
simulations 367
6.5 Climate issues associated with cirrus clouds 370
6.5.1 Contrails and contrail cirrus 370
6.5.1.1 Early studies 371
6.5.1.2 Contemporary programs and investigations 374
6.5.2 The role of cirrus in upper-troposphere and lower-stratosphere
exchange 380
6.5.2.1 Introductory notes 380
6.5.2.2 Cirrus clouds and the mechanism of stratosphere
dehydration 382
6.5.3 Thin cirrus clouds and climate 386
6.5.3.1 Thin cirrus detection and comparison with GCM output 386
6.5.3.2 Some notes on high thin clouds in global radiative and
climate forcing 387
References 392
Index 427
Preface
The preparation of Light Scattering by Ice Crystals: Fundamentals and Applications began
about seven years ago. We thought that sufficient material should be available to compose
a high-level text reflective of the complex and intricate domain of ice crystals in the Earth’s
atmosphere and their interaction with “light” from the sun and that emitted from the Earth
and the atmosphere, with applications to remote sensing and climate studies. This text was
supposed to be a three- to four-year project; however, after sifting through the literature
for about two years, gaps emerged on various subjects, including both fundamentals and
applications. For this reason, we conducted additional research in an attempt to bridge
various gaps that are essential, from our perspective, to the unification of all subjects in a
coherent and logical manner with reference to light scattering by ice crystals. Accordingly,
we are pleased to present this text for active researchers and advanced graduate students
who are interested in general areas of atmospheric physics, atmospheric radiative transfer,
atmospheric optics, computational modeling, cloud–climate interactions, and remote sens-
ing of the atmosphere and oceans within the purview of atmospheric and climate sciences.
It is intended to complement other researchers who work in the field of light scattering by
non-spherical particles, which includes ice crystals.
“Ice in the Earth’s atmosphere,” the title of Chapter 1, plays a key role in the hydrological
cycle and precipitation processes. Furthermore, ice clouds in the upper troposphere through
their solar albedo and infrared greenhouse effects are critical elements in determining
surface and atmospheric temperature patterns within the context of greenhouse warming
and climate change induced by man-made perturbations in greenhouse gases and regional air
pollution. In Chapter 1, we introduce cloud classification, a global view of clouds in general,
and cirrus clouds in particular, followed by discussion of the formation and growth of ice
crystals. We then illustrate the complex nature of ice crystals with reference to their size and
three-dimensional morphology based on findings obtained from laboratory experiments and
aircraft observations, to establish a correlation between ice water content and ice crystal
size. This correlation is important in developing radiative transfer parameterizations in
climate models and in understanding the role of ice in climate radiative forcing. Lastly,
we present a two-dimensional cirrus cloud model to illustrate interactions of winds, ice
microphysics, and radiative transfer.
xii
Preface xiii
In Chapter 2, “Fundamentals of light scattering by ice crystals,” a number of fundamental

subjects are presented in relation to light scattering by ice crystals. We discuss the scope
and boundaries of light scattering by ice crystals and present the fundamental Maxwell’s
equations, leading to vector wave equations whose solutions require the imposition of
boundary conditions. We show that exact analytic solutions of vector wave equations exist
only for spherical, circular cylindrical, and spheroidal coordinates. The optical properties
of ice are then introduced, followed by a discussion that defines the single-scattering and
polarization properties of non-spherical ice crystals, including the meaning of the scattering
phase matrix. We subsequently discuss the link between single-scattering properties of ice
crystals, deduced from the independent scattering concept, and the transfer of radiation,
including multiple scattering and emission within ice crystal clouds.
Chapter 3, entitled “Principles of geometric optics for application to light scattering
by ice crystals,” presents the geometric-optics approach to light scattering by ice crystals,
starting with an overview of the essence of geometric optics, including diffraction and
surface waves, from several historical perspectives. We then illustrate fascinating ice optics
produced by randomly and horizontally oriented ice particles by means of Monte Carlo
geometric ray tracing. Subsequently, we demonstrate that exact solutions for diffraction
involving a number of ice crystal shapes can be analytically derived, followed by discus-
sion of conventional and improved geometric-optics approaches, and, based on a number of
postulations, introduce surface-wave contributions – the edge effect – for spheres with mod-
ification to hexagonal ice crystals. Lastly, we present a unified theory of light scattering by
ice crystals on the basis of the geometric-optics surface-wave approach. In this discussion,
theoretical phase functions are compared with those determined from controlled laboratory
light scattering and spectroscopic experiments, as well as application to light absorption
and scattering by snow grains internally contaminated by black carbon and dust particles
wherein stochastic processes are further introduced.
In Chapter 4, “Other useful approaches to light scattering by ice particles,” we confine
our presentations to three contemporary numerical approaches to light scattering by non-
spherical particles within the purview of their applications to light scattering by ice crystals,
namely the finite-difference time domain method, the T-matrix numerical method, and the
discrete dipole approximation. Single-scattering and phase matrix results determined from
these methods for applicable ice crystal size and shape ranges have been used to cross check
and calibrate those computed from a number of geometric-optics approaches. Moreover, we
develop numerical techniques to improve the first two methodologies in terms of particle
size applicability, as well as application to intricate particle shapes.
The subject of “Application of light scattering by ice crystals to remote sensing” is
presented in Chapter 5, wherein we first discuss atmospheric composition and structure,
the atmospheric absorption spectrum, sun–satellite geometry, radiative transfer, and the
contemporary A-Train satellite constellation, which are important for cirrus cloud detection
and quantification. From that base, we then present the subject of retrieving the optical depth
and ice crystal size of cirrus clouds using reflected visible and near-infrared radiation and
illustrate the importance of the phase function of ice crystals. A discussion follows on
xiv Preface
detecting thin cirrus and vertical sizing of cirrus cloud layers. We subsequently cover the
subjects of remote sensing of ice clouds using reflected polarization and the principle of
backscattering depolarization to differentiate ice crystals and water droplets. Lastly, we
present reflected line spectra in the 1.38 µm band and the oxygen A-band for inferring the
composition and optical properties of high clouds.
Chapter 6, the last chapter of the text, comprises discussions of “Application of light
scattering by ice crystals to climate studies.” Herein, we present the physical foundations
for parameterization of the extinction and absorption coefficients and phase function, for
ice crystals imbedded in gaseous absorption line formation. We then discuss delta-two-
stream and delta-four-stream approximations for efficient radiative flux transfer in non-
homogeneous plane-parallel atmospheres and compare theoretical results with aircraft
and satellite observations. On this basis, we present radiative forcing of cirrus clouds
from the viewpoint of theoretical calculations and point out the prevalence of the infrared
greenhouse effect over its solar albedo counterpart. This is followed by a presentation of
climatic effects of cirrus clouds from the perspective of one-dimensional climate models,
where we point out that cloud cover, ice water path, and ice crystal size are influenced by
temperature increases in greenhouse warming scenarios. Also discussed are examples of the
impacts of microphysics on precipitation and radiative forcings using results analyzed from
global climate model simulations. We then present a number of climatic issues associated
with cirrus clouds, including contrails and induced contrail cirrus produced by high-flying
aircraft, a man-made perturbation, their role in upper troposphere and lower stratosphere
exchanges, and the usefulness of optically thin cirrus data determined from modern satellite
instrumentation for ice cloud parameterization development in climate models.
In view of the above, the subject of light scattering by ice crystals as presented in this
text has made definitive contributions to fundamental understanding of and insight into
light scattering, absorption, and polarization processes involving ice crystals. Moreover,
light scattering by ice crystals has provided a new dimension and valuable datasets to the
development of satellite remote sensing of ubiquitous cirrus clouds comprising various ice
crystal sizes and morphologies, as well as to radiation parameterizations for these clouds
in climate models to investigate uncertainties surrounding their role in temperature and
precipitation responses to global warming and climate change.
The senior author wishes to acknowledge the National Science Foundation for its sup-
port over the last 30 years or so of his basic research on light scattering by ice crystals
and related subjects in radiative transfer at the University of Utah and the University of
California, Los Angeles. In particular, he would like to thank the Atmospheric Sciences
Section for a Creativity Award based on contributions to “Light Scattering by Ice Crystals:
Theory and Experiment” (1996) through the recommendation of the late Ronald Taylor and
subsequent support from R. R. Rogers, Bradley Smull, and Chungu Lu, Program Directors
for Dynamical and Physical Meteorology. Additionally, he thanks Roberto Peccei, a par-
ticle physicist and UCLA Emeritus Vice Chancellor for Research, and Joseph Rudnick, a
condensed matter physicist and Senior Dean of the UCLA College of Letters and Science,
for unwavering support in his pursuit of academic and research excellence and the founding
of the Joint Institute for Regional Earth System Science and Engineering at UCLA. He is
Preface xv
also exceedingly grateful to Wilfried Brutsaert and Richard Goody for their encouragement
and advice.
The co-author is grateful to Kuo-Nan Liou, Warren Wiscombe, George Kattawar, and
Jerry North for long-term mentorship. He would also like to thank Hal Maring and Lucia
Tsaoussi, NASA Program Managers, and Chungu Lu, a National Science Foundation
Program Director, for their encouragement in his pursuit of excellence in light scattering
research over the years.
We are indebted to several individuals whose efforts have directly impacted the final
production of this volume: Tara Fickle, a former graduate student in the UCLA English
Department and a part-time editor from 2008 to 2014, for her assistance with typing,
editing, and computer graphics; Janice Amar, retired Chief Operating Officer at the UCLA
Joint Institute for Regional Earth System Science and Engineering and a recalled Senior
Editor in 2015, for her assistance with final editing, permission requests, and compliance
with Cambridge University Press format requirements; Christine Cho for general support
to the authors on various aspects of the book project; Steve Ou for gathering references
and other materials for a number of sections in Chapters 5 and 6; Yu Gu for assistance
with the discussion in Section 1.8; and Lei Bi and Zhibo Zhang for contributions to
the presentation of Chapter 4 and Subsection 5.2.4, respectively. It would not have been
possible to complete the book project without the dedicated support of Yoshihide Takano,
contributing author, who has also made significant contributions to the field, in carrying
out additional calculations, checking fundamental equations, gathering references from
the Internet, as well as making revisions to three-dimensional and color presentations for
numerous figures in the text. We thank Qiang Fu, Professor of Atmospheric Sciences at
the University of Washington, who has made pioneering contributions in areas presented in
Chapter 6, for his critical review and constructive comments. Lastly, we wish to thank Bryan
Baum, Qiming Cai, Andrew Heymsfield, Yongxiang Hu, Hironobu Iwabuchi, Brian Kahn,
George Kattawar, Michael King, Patrick Minnis, Michael Mishchenko, Steven Platnick,
and Manfred Wendisch for research collaboration during the course of our work on light
scattering by ice crystals.
Finally, we acknowledge the generosity of the American Geophysical Union and John
Wiley & Sons (Figures 1.7, 1.8, 1.23, 1.24, 3.32a, 3.32c, 3.33, 5.9, 5.10, 5.11, 5.22,
5.23, 5.24, 5.25, 5.32, 6.1, 6.9, 6.12, 6.15, 6.18, 6.19, 6.20), the American Meteorological
Society (Figures 1.4, 1.5, 1.16, 1.17, 1.22, 1.25, 3.5, 4.5, 5.14, 5.26, 5.30, 5.31, 5.33, 6.3,
6.5, 6.6, 6.7, 6.13), Elsevier (Figures 1.3b, 1.13, 2.7, 2.9, 3.4, 3.11, 3.12, 3.18, 3.19, 3.20,
3.21, 3.22, 4.10, 4.13, 4.17, 4.18, 4.19, 4.20, 4.21, 4.24, 6.16), the European Geophysical
Union (Figures 5.18, 5.27), Hokkaido University (Figures 1.14, 1.15), the Institute of
Electrical and Electronics Engineers (Figures 5.17, 5.19), Nature Publishing Group (6.17),
Taylor & Francis Ltd. (Figure 1.9), the Optical Society of America (Figures 2.4, 2.8, 2.20,
3.28, 3.29, 3.30, 3.31, 4.3, 4.4, 4.6, 4.7, 4.9, 4.11, 4.12, 4.28), Oxford University Press
(Figures 1.10, 5.34), Springer-Verlag (Figure 4.8), and colleagues who have furnished
unpublished material – Les Cowley (Figure 3.9b), Edwin W. Eloranta (Figure 1.6), Marko
Riikonen (Figure 3.9a), Ulrich Schumann (Figure 1.3a), and Stefan Bauer, Janek Zimmer,
and Uwe Reiss (Figure 1.1) – for permission to use this material in this text.
1
Ice in the Earth’s atmosphere
“Let there be light,” and there was light.∗
The fact that the Earth’s sky appears blue is a consequence of the scattering of “light” by
molecules according to the theory of Rayleigh scattering. The fascinating halos and arcs
we see mixed within blue sky result from light scattering by ice crystals.
Our presentation of light scattering by ice crystals begins with an overview of clouds.
This is followed by a global view of ice distribution in the Earth’s atmosphere; formation
and growth of ice crystals; ice crystal morphology, size, and distribution; and a discussion
of cirrus cloud modeling with a linkage to some of these topics.
1.1 Introduction to clouds

I BRING fresh showers for the thirsting flowers,
From the seas and the streams;
I bear light shade for the leaves when laid
In their noonday dreams.
From my wings are shaken the dews that waken
The sweet buds every one,
When rocked to rest on their mother’s breast,
As she dances about the sun.
I wield the flail of the lashing hail,
And whiten the green plains under,
And then again I dissolve it in rain,
And laugh as I pass in thunder.
...
I am the daughter of Earth and Water,
And the nursling of the Sky;
I pass through the pores of the ocean and shores;
I change, but I cannot die.
For after the rain, when with never a stain
∗
A quotation from Genesis 1:3 to the extent to introduce the term, “Let there be light.”
1
2 Ice in the Earth’s atmosphere
The pavilion of Heaven is bare,

And the winds and sunbeams with their convex gleams
Build up the blue dome of air,
I silently laugh at my own cenotaph,
And out of the caverns of rain,
Like a child from the womb, like a ghost from the tomb,
I arise, and unbuild it again.
(Percy B. Shelley, “The Cloud” (1820))
Clouds are formed when water evaporates from oceans and other surfaces (lakes, ponds,
moist land surfaces) carried by convection, orographic, or frontal lifting and rises into the
upper, colder part of the atmosphere. Formation of clouds generally requires the interaction
of water vapor with a type of aerosol referred to as condensation nuclei or ice nuclei. A
cloud becomes visible once the water vapor has been cooled by the condition of water
or ice saturation. However, some very thin clouds cannot be seen by the human eye, and
are classified as subvisual clouds. Clouds are normally produced in a region referred to
as the troposphere (lower atmosphere), the lowest layer of the atmosphere, where weather
activities occur. Clouds are regulated by the hydrological cycle, which involves evaporation,
cloud formation, precipitation, runoff, and large-scale circulation.
In accordance with the World Meteorological Organization (WMO) definition, clouds
are conventionally classified in terms of their position and appearance (shape) in the
atmosphere. Clouds with a base height above 6 km are designated as high clouds, a category
that includes cirrus (Ci), cirrostratus (Cs), and cirrocumulus (Cc). On the basis of the U.S.
1976 Standard Atmosphere classification, 6 km corresponds to a temperature of about
249 K, which is 24 K below the freezing temperature (273 K). Thus, these clouds contain
exclusively ice particles. Cirrus clouds tend to be wispy and transparent (Figure 1.1, upper
left panel). In midlatitude, large numbers of this type of clouds are generally associated with
an approaching storm system. The upper right panel in Figure 1.1 illustrates the appearance
of Cc. Regional convective instability gives this cloud type a rolled or rippled appearance.
The upper middle panel in Figure 1.1 shows the appearance of Cs, which consists of mostly
continuous, wide sheets of clouds that cover a large area of sky and, when it is associated
with frontal systems, is a precursor to rain or snow.
The middle group of clouds, with heights between about 2 km (275 K) and 6 km,
consists of altocumulus (Ac) and altostratus (As), in which ice particles and water droplets
can coexist. The appearance of Ac, displayed in the middle left panel of Figure 1.1, is
a general indication of convective instability at the level of its formation. This cloud can
bring precipitation, usually in the form of virga, a type of precipitation trail that does not
reach the ground. Clouds of As (Figure 1.1, middle right panel) are formed when a stable
air mass is lifted to the level of condensation along a frontal system, which can produce
precipitation.
Low clouds, which are classified as having base heights below 2 km, include stratus
(St), stratocumulus (Sc), and fair-weather cumulus (Cu). These clouds contain exclusively
1.1 Introduction to clouds 3
Figure 1.1 The appearance of high (Ci, Cs, Cc), middle (Ac, As), and low (Sc, St, Cu) clouds in the
Earth’s atmosphere. These pictures are reprinted from the cloud atlas at http://www.clouds-online
.com. Copyright information: http://www.clouds-online.com/imprint.htm.
water droplets. Sc clouds have a lumpy appearance (lower left panel, Figure 1.1). They
commonly form in an unstable air mass following a cold front, and can produce light rain
or drizzle. St clouds (lower middle panel, Figure 1.1) form near the surface, usually over
coastal oceans, have a clearly defined base, and can produce drizzle. Cu clouds (lower right
panel, Figure 1.1) are the product of convective air mass instability and are often associated
with fair-weather conditions.
Other cloud types are associated with substantial vertical development. Cumulonimbus
(Cb) clouds are vertically developed cumulus produced by strong convectively unstable
conditions, principally occurring over tropical oceans and land in the summer. They appear
very dark gray with a cloud base height of about 1 km, and the cloud top can extend to
the top of the troposphere and occasionally into the lower stratosphere (15–18 km). They
generally produce thunderstorms, rain showers, and sometimes hail particles. The upper
portion of these clouds may contain pure ice or a mixture of ice and water. Nimbostratus
(Ns) is a type of cloud that tends to bring constant precipitation and low visibility. They
normally form from altostratus clouds and can thicken into lower levels during precipitation.
Figure 1.2 Cloud types defined in the vertical direction. High clouds have base heights above 6 km, middle clouds are located between about 2 km and
6 km, and low clouds have base heights below 2 km. The vertically developed cumulonimbus can grow up to 15–18 km in the tropics. Nimbostratus are
normally evolved from altostratus clouds.
1.2 Distributions of ice in the Earth’s atmosphere 5
Figure 1.2 depicts the approximate vertical positions in the troposphere of all cloud types
described above, including Cb and Ns, which produce precipitation that reaches the ground.
Cb clouds generate a significant number of anvils in the tropics. Drifting with the winds,
they can last for hours, and subsequently become high-level clouds before dissipating.
Upper-level ice crystal clouds produced by jet aircraft are known as contrails or conden-
sation trails (Appleman 1953). Contrails are visible line clouds resulting from water vapor
emissions that form behind aircraft flying in sufficiently cold air. Persistent contrails often
develop into more extensive contrail cirrus, particularly in ice-supersaturated air masses
in which ice supersaturation is generally too low to allow cirrus clouds to form naturally.
Contrails may enhance the extension of the natural cirrus cover in adjacent areas where
relative humidity is too low for the spontaneous nucleation of ice crystals. A comprehensive
analysis of jet aircraft contrails over the United States and Europe using satellite infrared
imagery has been carried out by Minnis et al. (1998a) and discussed in the IPCC report
(IPCC Report 1999) in conjunction with their climatic impact, but see Subsection 6.5.1
for further discussion. Figure 1.3a shows an unusual spiral contrail formed by a circling
aircraft observed in a NOAA-14 Advanced Very High Resolution Radiometer (AVHRR)
image west of Denmark at 1236 UT, May 22, 1998. Figure 1.3b displays a typical contrail
in the form of limited diffusion amidst a blue sky.
Finally, we would like to note that clouds have also been observed in the upper atmo-
sphere. A type of ice crystal cloud known as a polar stratospheric cloud (PSC) has often
been observed in the polar stratosphere between about 15 and 20 km. These clouds have fre-
quently been detected over the Arctic and Antarctic by limb-viewing satellite instruments
during the winter months when the ambient temperature falls below about 195 K. The
condensation of both water vapor and nitric acid (HNO3 ) results in the formation of HNO3
trihydrates, which serve as nuclei for ice crystal growth. In the stratosphere between about
20 and 30 km, a type of thin cloud usually resembling the cirrus form, and referred to as
mother-of-pearl (nacreous) clouds, has been detected. These clouds appear to be generated
by ice deposition on frozen particles of sulfuric acid. In the mesosphere between about 50
and 55 km, a cloud type known as noctilucent clouds has been observed and is well known.
These clouds are very tenuous and resemble cirrus.
1.2 Geographical and temporal distributions of ice in the Earth’s atmosphere

1.2.1 A global perspective on clouds
Clouds are global in nature, as evidenced from satellite cloud pictures. We shall confine
our discussion to the formation and dissipation of high and middle clouds based on Geo-
stationary Operational Environmental Satellite (GOES) thermal infrared (IR) images from
the perspective of large-scale weather activities, as presented in Liou (1986). Shown in
Figure 1.4a is a full disk IR picture taken at 2345 GMT, February 23, 1984. Warmer areas
are darker, while cooler areas are lighter. Temperature normally decreases with height in the
troposphere; thus, the whitest areas can safely be assumed to be high clouds. This picture
(a)
(b)
Figure 1.3 (a) An unusual spiral contrail, formed by a circling aircraft, surrounded by high clouds,
observed in a NOAA-14 Advanced Very High Resolution Radiometer (AVHRR) image west of
Denmark at 1236 UT, May 22, 1998 (Schumann 2002). (b) A typical contrail amidst a blue sky (Gao
et al. 2006 and courtesy of Randall Friedl of JPL/NASA).
(a)
(b)
Figure 1.4 (a) Full disk thermal infrared picture at 2345 GMT February 23, 1984, illustrating
globally distributed clouds in general and cirrus clouds in particular. (b) Same as (a), but for February
25, 1984 (after Liou 1986).
was taken over the Northern Hemisphere during daytime in the later part of winter, when
polar regions are generally outside the frame of the satellite camera, thus minimizing the
possibility that white areas may be associated with cold surface temperatures.
High-level cirrus clouds are globally distributed and present at all latitudes irrespective
of land, sea, or season. These clouds undergo continuous changes in area coverage, thick-
ness, texture, and position. The most striking cirriform cloud feature shown in Figure 1.4a
is the large spiral-comma-shaped pattern west of the Washington State coastline, associated
with a major surface cyclone located to the northeast of the cloud center. Lower clouds
and precipitation associated with the large-scale rising motion are present under much of
this high-cloud canopy. To the north of this cloud band and over the northernmost areas of
the Pacific, another bright cirriform area is associated with a complex of surface lows and
frontal systems. Further to the west, the leading edge of a large cirriform cloud mass is mov-
ing into the IR picture. This cloud mass represents a major storm developing off the coast of
Japan.
The most impressive area of cirriform cloudiness is seen over the Pacific Ocean between
Hawaii and Mexico, and is related to a huge, although rather weak, trough aloft linked to the
subtropical jet stream that curves southward near about (35°N, 160°W). The brightness of
these cirriform clouds coupled with light gray areas, which represent middle clouds below,
indicates an active zone of weather activity. Further east, broken, largely transverse bands of
cirriform clouds are spreading eastward into Mexico. The equatorial area is characterized
by strong, predominantly diurnal convection over western South America, with a large
production of anvil cirrus and a zone associated with the Intertropical Convergence Zone
(ITCZ) that extends across the Pacific at 10°S. In the central Pacific, a collection of
mesoscale and synoptic scale clusters of cumulonimbus, some imbedded in areas of middle
clouds, is producing the brightest (coldest) cirriform cloudiness in the picture.
The dominant cloudiness in the Southern Hemisphere is associated with cirriform clouds
produced by the strong cold front that extends north-northwestward from an occluded front
anchored in an intense low centered at (57°S, 140°W). In addition, some spiral-shaped
cirrus and middle clouds are shown in the vicinity of (31°S, 104°W); these are associated
with a low-pressure system that has been cut off from the westerlies. On the western edge
of the picture, a weak cold front with a thin cirrus band approaches New Zealand. To the
north of this front, a band of cirrus stretching north-northwestward from (30°S, 165°E) is
associated with a surface low at (19°S, 158°E) and a strong wind shear aloft.
In the ensuing 48 hours, the cirriform cloud pattern west of Washington State, as depicted
in Figure 1.4b, changes from a well-defined spiral coupled with the original surface low to a
disorganized, blotchy mass of less bright clouds in the midst of the dissipation–reformation
process, and finally to a redevelopment of brighter, more organized masses coming together
with a major storm development. A huge cirriform spiral, characteristic of extratropical
cyclones, is seen in the north Pacific. The thin line of cirrus on the poleward side of the cold
frontal band stretches from (47°N, 135°W) to (35°N, 154°W). This line occurs adjacent to
the polar jet stream that cuts across the frontal zone near the west coast of central British
Columbia. In conjunction with a huge cirriform frontal band and a spiral center at (43°N,
161°E), another major storm is also evident, which was just coming into view from the
west as shown in Figure 1.4b. In the Southern Hemisphere there has been relatively little
change in the 48-hour period between Figures 1.4a and 1.4b, due to a large-scale dynamic
blocking situation. The cirrus cloud band near (55°S, 175°W) in Figure 1.4b is associated
with a strong, new cold front beginning to sweep northeastward.
The preceding discussion makes it clear that the formation, maintenance, and dissi-
pation of high and middle clouds which contain ice particles are principally modulated
by large-scale weather features and disturbances. In the tropics, these phenomena are
related to deep-cumulus outflows associated with strong convection, characteristic of this
region. Thin and subvisual cirrus with optical depths 0.1 have not been identified in the
foregoing presentation due to the limitations of IR temperature techniques. Presented in
Subsection 1.2.2 are the clouds that were seen by specific satellite instruments and the
associated data gathered and analyzed for cloud climatology studies.
1.2.2 An example of global cloud climatology

In what follows, we present an example of global cloud climatology derived from the NOAA
High Resolution Infrared Spectrometer (HIRS) polar-orbiting satellite data from the period
1979–2001 during winter and summer (Wylie et al. 2005). The specific technique, called
the CO2 slicing method, which uses two channels in the 15 µm CO2 band and a 10 µm
window channel, has been developed to determine the statistics of cloud cover, height, and
emissivity. Many processing procedures were required to produce a global map of cloud
parameters (Wylie et al. 1994). Figure 1.5 illustrates the frequency of all clouds as well
as high clouds above 6 km. Clouds are most frequently found in the ITCZ and the midlat-
itude storm belts of the North Atlantic, North Pacific, and Antarctic Oceans, as noted in
Figures 1.4a and 1.4b. Clouds are less frequent between the subtropical high-pressure zones
over the oceans and the subtropical deserts over land.
High clouds are observed in about one-third of the HIRS observations. Their coverage
shows an annual cycle over land with the maximum occurrence during summer in each
hemisphere. The ITCZ is a region of more frequent high clouds, as are the midlatitude
storm belts in the Northern Hemisphere summer. The subtropical high-pressure systems
are evident in the region of less frequent high-cloud cover. Over the Indonesian region,
the ITCZ expands in coverage from winter to summer, whereas in the central Pacific
Ocean, it shows extension during the winter months. For the Southern Hemisphere, the
eastern Pacific Ocean off South America and the eastern Atlantic Ocean off Africa remain
relatively free of high clouds throughout the year. The North American high-cloud cover
shows little seasonal change. The high-cloud cover results derived from the CO2 slicing
method illustrate that high clouds are ubiquitous in the tropics, occurring with more than
70% frequency.
We wish to note that many past attempts have been made to classify the global dis-
tribution of clouds for climate study based on the emitted IR radiation in the 10 µm
window, plus the reflected visible radiation during daytime. The classification of cloudy
All Clouds
December, January, and February June, July, and August
High Clouds
Frequency of Clouds
0% 50% 100%
Figure 1.5 The frequency of all clouds and of high clouds above 440 hPa from 1979 to 2001, from HIRS data during boreal winter [Dec–Jan–Feb (DJF)]
and summer [Jun–Jul–Aug (JJA)] (after Wylie et al. 2005; see text for further discussion).
pixels by measured radiances is a complex decision-making process based on computer

algorithms. The best example is the International Satellite Cloud Climatology Program
(ISCCP), a WMO project developed in the 1980s (Schiffer and Rossow 1983, Rossow and
Schiffer 1999). ISCCP has also produced many useful cloud products, which have been
compared with similar results determined from the HIRS CO2 slicing approach discussed
above.
More recently, Minnis et al. (2011a) presented a cloud climatology based on NASA’s
Clouds and the Earth’s Radiant Energy System (CERES) project, analyzing data from
the Tropical Rainfall Measuring Mission (TRMM) Visible and Infrared Scanner and the
MODerate-resolution Imaging Spectroradiometer (MODIS) instruments on board the Terra
and Aqua satellites covering the period 1998 through 2007. King et al. (2013) developed a
cloud climatology from MODIS using over 12 years of continuous observations from Terra
and over 9 years from Aqua. Kahn et al. (2014) reported a cloud climatology determined
from cloud products of the Atmospheric Infrared Sounder (AIRS) and Advanced Microwave
Sounding Unit (AMSU) instrument suite. Cloud climatology products in these studies
include cloud-top temperature, ice cloud fraction, effective ice particle size, and ice cloud
optical depth. Additional discussions are presented in Subsection 5.2.5.
1.2.3 View from ground-based instruments

High-level clouds have been detected by ground-based instruments and their compositions
retrieved and quantified by lidar (laser radar) and millimeter (mm)-wave radar. In this
subsection we present two examples based on these two unique sensors.
During the 1960s, the advent of the laser as a source of energy opened up numerous pos-
sibilities for remote sensing of the atmosphere. Laser energy at optical frequencies is highly
monochromatic and coherent. With the innovation of Q-switching techniques, very short
pulses of high power could be generated, which prompted the development of backscat-
tering lidar techniques for detection of the composition and structure of clouds, aerosols,
and minor gases in the atmosphere. The lidar backscattering equation for probing the atmo-
sphere is similar to that of radar. Many innovative remote sensing techniques have been
developed, including the use of multiple-wavelength lidar to determine the composition
of minor gases and aerosol concentrations by means of differential absorption techniques,
Doppler approaches to infer the motion of particulates and molecules, and the Raman
scattering technique, which produces a weak scattering at a shifted wavelength for water
vapor measurements. In the context of cirrus cloud study, backscattering and depolarization
techniques have been used to retrieve cloud position and thickness, to differentiate ice and
water clouds, to detect ice particle orientation, and to infer particle size and optical depth
with the assistance of other sensors (see, for example, Sassen 1991). The development of a
lidar system on a space platform is discussed in the following paragraphs.
At visible wavelengths, lidar can penetrate through high clouds, which are normally opti-
cally thin with optical depths less than about 3. However, it cannot penetrate through lower
clouds, which have optical depths generally greater than 10–30. Illustrated in Figure 1.6
Top View
24 km
12 km
6 km
North
24 km
Side View
Figure 1.6 Three-dimensional observations of cirrus clouds in the vertical direction and in a hori-
zontal plane, from lidar backscattering (after E. Eloranta, personal communication 1995).
is a three-dimensional backscattering image of cirrus clouds by a scanning lidar system

across the prevailing winds. From this example, cirrus clouds appear to be highly inhomo-
geneous in the horizontal and vertical directions, and frequently have layer structure in the
vertical.
Conventional meteorological radars typically operate at centimeter (cm) wavelengths to
interpret backscattering signals from precipitating clouds that contain large raindrops and
snowflakes on the order of mm to cm sizes. Non-precipitating clouds, however, generally
consist of particles that are much smaller than mm sizes and do not produce significant
radar backscattering signals for their detection. The technical development of mm-wave
radar and its application to cloud studies is a subject of modern research. The selection of
wavelengths for a radar operated in the Earth’s atmosphere is restricted to those spectral
regions where absorption by atmospheric gases is small. The windows of H2 O and O2
absorption are located at 35 GHz (8.6 mm, Ka -band), 94 GHz (3.2 mm, W-band), 140 GHz
(2.14 mm, F-band), and 220 GHz (1.36 mm, G-band). The first mm-wave radar designed for
meteorological use was a 35 GHz system for cloud deck monitoring. It was subsequently
equipped with Doppler and polarization capabilities and employed for the observation of
(a) mm-wave radar reflectivity dBZe

-20
Height (km)
10
-30
8
-40
6 -50
18:00 Time (UTC) 17:00

(b) 1.38 μm reflectance %
2.6
36 .6ON 1.6
0.6
98oW 97.5oW 97oW
(c) 8.6 - 11 μm brightness temperature difference K
0
36.6 NO -1
-2
98oW 97.5oW 97oW
(d) P-parameter results
36.6ON Cirrus
Clear
98oW 97.5oW 97oW
Figure 1.7 A case study from the SGP-ARM site on February 11, 2001 (1730 UTC). Plotted in (a)
is the 1-hour mm-wave radar reflectivity time series from Lamont, Oklahoma. The MODIS/Terra
1.38 µm reflectance and 8.6–11 µm BTD are displayed in (b) and (c), respectively, while (d) shows
the cirrus detection P-parameter results (after Roskovensky and Liou 2003a).
clouds. A 94 GHz system with a shorter wavelength was further shown to provide effective
observations of the liquid/ice water content of cloud (in general in this text, “/” is to be
taken to mean “and”) and has been successfully deployed in a small satellite, referred to
as CloudSat, launched in April 2006. CloudSat is one of the constellation satellites in the
A-Train, discussed in Subsection 5.1.5.
Because water vapor concentration decreases rapidly with height, mm-wave radar gen-
erally operates at 35 GHz. Figure 1.7a shows an example of the 35 GHz mm-wave radar
reflectivity time series with a one-hour average observed from DOE’s Atmospheric Radia-
tion Measurement (ARM) site located at Lamont, Oklahoma, on February 11, 2001 (1730
UTC) presented in Roskovensky and Liou (2003a). The radar reflectivity depicts a distinct
two-layer distribution of cirrus clouds moving with the westerly winds. The radar time
series is given backwards, so that the earliest time matches the spatial location at the right
side of the other plots, while the latest time matches the left side. These plots are the 1.38 µm
reflectance (Figure 1.7b) and 8.6–11 µm brightness temperature difference (Figure 1.7c)
from MODIS data that can be used to detect the presence of cirrus clouds, and a thin cirrus
(optical depths between 0.1 and 0.9) mapping technique referred to as the P-parameter
method, developed in the paper (Figure 1.7d).
1.2.4 View from specific satellite instruments

A number of satellite instruments have been used to identify ice clouds and retrieve their
microphysical and optical properties in the atmosphere. However, the following exam-
ples present vertical images and profiles of ice clouds inferred from Cloud-Aerosol Lidar
and Infrared Pathfinder Satellite Observations (CALIPSO), a small satellite launched in
April 2006 specifically to study the impact of clouds and aerosols on the Earth’s radia-
tion budget and climate. It flies in formation with six other satellites in the international
A-Train constellation for coincident Earth observations (see Subsection 5.1.5). For inter-
pretation purposes, we also employ images and products determined from the Moderate-
Resolution Imaging Spectroradiometer (MODIS), referred to in Section 1.2.2, which was
designed to make 0.25–1 km resolution observations in 36 visible and infrared bands.
MODIS was launched in December 1999 on the Terra platform and in May 2002 on the
Aqua platform, which leads the A-Train.
In Figure 1.8 we show the vertical profiles of contrails and cirrus clouds derived from
the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) backscattering signal
operated at the 0.532 µm wavelength (Iwabuchi et al. 2012). The horizontal and vertical
resolutions of CALIOP data used for this presentation are 1 km and 60 m, respectively,
for altitudes between 8 and 20 km. With the help of location information determined from
MODIS data analysis, contrails were searched in the CALIOP profile. Figure 1.8b shows
CALIOP profiles corresponding to the MODIS image in Figure 1.8a.
Contrails were identified based on limited vertical extent with a remarkable backscatter-
ing, as compared with surrounding cirrus. The red vertical lines in the plots denote contrail
locations detected by MODIS image analysis. Young contrails A and B were isolated from
cirrus clouds, having confined widths within 1–2 km. The vertical extent is approximately
250 m for A and 450 m for B. A cirrus cloud with a width of about 30 km is present just
beneath contrail B, which is also shown in the MODIS image. Contrail C is not clearly
exhibited in this profile. Contrail D is imbedded in cirrus, while contrail E extends from
cirrus: both are mature, with moderate vertical and horizontal extents.
Figure 1.9a (Ou et al. 2012) illustrates another example of the vertical feature mask of ice
clouds from CALIOP polarized echoes. Between the two vertical dashed lines, which denote
the southern and northern bounds of the selected domain, clouds (light blue), extending to
5 km and mixed with dust layers (orange), have been identified using lidar depolarization
technique. These CALIOP results correspond to the MODIS/Aqua scene for March 31,
2007 at 0500 UTC over northeastern and eastern China and the East China Sea. Images for
the MODIS (RGB) overpass on March 31, 2007 at 0500 UTC over northeastern and eastern
China and the East China Sea are shown in Figure 1.9b. This case contains a significant
presence of dust layers mixed with clouds. Over northeastern China, brightness temperature
difference values between 11 and 12 µm (BTD11-12) are around −4 K, indicating a heavy
load of dust aerosols. This is a technique used to differentiate between cirrus clouds and dust
layers. Major cirrus clouds were present over northeastern China, Mongolia, and southern
Japan, where cloud temperatures are lower than 243 K. Over the remaining cloudy areas,
(a)
(b)
Figure 1.8 (a) An image of MODIS brightness temperature difference (BTD) between 11 µm and
12 µm channels over the Northeast Pacific on January 11, 2009. The cyan line shows the locations of
MODIS-CALIOP collocated pixels. The magenta line segments denote contrails detected by manual
inspection. (b) CALIOP profiles of attenuated total (lower) and perpendicular (upper) backscattering
coefficients (km−1 sr−1 ) at 532 nm wavelength corresponding to the scene in the MODIS image. The red
vertical lines show contrail locations detected by MODIS image analysis. Only cloudy cross-sectional
elements are plotted (after Iwabuchi et al. 2012).
(a) Vertical Feature Mask UTC: 2007-03-31 04:56:26.4 to 2007-03-31 05:09:55.0 Version: 3.01 Nominal
30
7
25 6
Altitude (km)
20 5
4
15 3
3 (L)
10
2
5 2 (L)
1
0
1 = clear air 2 = cloud 3 = aerosol 4 = stratospheric layer 5 = surface 6 = subsurface 7 = totally attenuated 8 = low/no confidence
(b) RGB (c) τ3 (Ocean)
Latitude (oN)
Longitude (oE)
(d) τ3 (Land, DB) (e) re
Latitude (oN)
Latitude (oN)
Longitude (oE) Longitude (oE)
Figure 1.9 (a) Result of the vertical feature mask obtained from CALIOP polarized echoes. Between
the two vertical dashed lines, which denote the southern and northern bounds of the selected domain,
clouds mixed with dust have been identified using the lidar depolarization technique. (b) Images for
the MODIS (RGB) overpass on March 31, 2007 at 0500 UTC over northeastern and eastern China
and the East China Sea. (c) Aerosol optical depths over ocean. (d) Aerosol optical depths over land.
(e) Cloud effective particle sizes. The rectangular domain contains dust over both land and water with
imbedded clouds (after Ou et al. 2012).
BTD11-12 values are around 1 K, indicating mostly water clouds. The distributions of cirrus
and water clouds are consistent with cloud-phase mask results from MODIS. In addition to
clouds, Figures 1.9c,d display aerosol optical depth maps over ocean and land, respectively,
where the results over land were obtained using the Deep Blue algorithm developed by
Hsu et al. (2004). A domain was selected within which dust aerosols, along with ice and
water clouds, are present over both ocean and land in the regions of northeastern China
and the East China Sea. Dust aerosols over ocean and land were located in the southern
and northern halves of the selected domain. Cloud particle radii are retrieved from MODIS
visible and near-IR channels and are displayed in Figure 1.9e.
1.3 Formation of ice crystals

Having presented a macroscopic view of ice clouds in the atmosphere, we shall now broach
the subject of ice cloud formation from the perspective of microphysics at the fundamental
level. The formation and growth of ice crystals are subjects that have developed and evolved
in the last few decades. A number of textbooks (Fletcher 1962, Byers 1965, Mason 1971,
Hobbs 1974, Rogers and Yau 1989, Pruppacher and Klett 1996, Wallace and Hobbs 2006,
Wang 2013) present detailed information on these subjects. However, we shall focus our
discussions on several topics to the extent that they are specifically relevant to our discussion
of light scattering and absorption by ice crystals and radiative transfer in ice clouds.
1.3.1 The structure of ice crystals

Before beginning our discussion of the initial formation of ice crystals, a few words are in
order regarding the structure of ice. Ice Ih is the hexagonal crystal form of ordinary ice,
or frozen water. Essentially all ice in the Earth’s biosphere is ice Ih. The water molecule
has three thermodynamic phases – vapor, liquid, and ice. The water vapor molecule has an
asymmetric top bending (triatomic) configuration with the oxygen atom in the middle and
one hydrogen atom on each side, with a bonding angle of 104.45°. The distance between
the oxygen and hydrogen atoms is 0.958 Å. The oxygen and hydrogen atoms have a number
of isotopic forms. Ordinary water is generally considered to be 1 H2 16 O or simply H2 O.
The water vapor molecule has three rotational constants coupled with three fundamental
vibrational modes, which produces countless pure rotational and vibrational–rotational
absorption lines covering the entire infrared and visible spectrum, critical to the discussion
of radiative transfer in planetary atmospheres.
The basic structure of ice Ih was originally proposed by Pauling (1935) and has since
been well established. The geometric arrangements of hydrogen and oxygen atoms in a
crystalline solid were later determined from analysis of the X-ray diffraction pattern. The
oxygen atoms, indicated by open circles in Figure 1.10a, are arranged in a hexagonal lattice
with each having four nearest neighbors at the corners of a rectangular tetrahedron. The
hydrogen atoms, shown as dark dots, are covalently bonded to the nearest oxygen atoms
to form H2 O molecules; at the same time, these molecules are connected to one another
by hydrogen bonds. Each molecule offers its hydrogen atoms to two other molecules and
accepts hydrogen bonds from another two. There is no long-range order in the orientation
of H2 O molecules or hydrogen bonds. In Figure 1.10a, the notation [0001] (one of the
Miller indices) is aligned with the c-axis and the picture is from a view perpendicular to
the c-axis. The unit cell of the average structure is marked by dashed lines connected by
upper-case letters in which the basic four oxygen atoms are shaded. The basic structure
Figure 1.10 (a) The crystal structure of ice Ih, with oxygen atoms denoted by white circles and
hydrogen atoms by black dots, viewed perpendicular to the c-axis. The unit cell of the average
structure is marked ABCDEFGH, while the four oxygen atoms contained within this cell are shaded.
(b) A two-dimensional view of the ice structure on the (101̄0) plane (after Petrenko and Whitworth
1999). The definition of the Miller indices is illustrated in Figure 1.13a.
then consists of a hexametric box where planes contain chair-form hexameters, similar to
sheets lying on top of one another.
The three-dimensional structure of ice Ih is complicated. Thus, a two-dimensional
layer perspective is shown in Figure 1.10b for easier visualization. There are two possible
hydrogen sites on each bond and four of these sites adjacent to each oxygen atom, such
that there are two hydrogen atoms adjacent to each oxygen atom and only one hydrogen
atom per bond. In this diagram, the Miller indices [1̄21̄0] and [101̄0] denote the directions
along one of the intersections and perpendicular to one of the prism faces, respectively (see
Figure 1.13a). By convention, a surface can be defined by four integers, which are placed in
order to complete the orientation definition; a bar over a number indicates an intersection
in a negative axial direction (e.g., see −a3 in Figure 1.13a). Crystal planes of the type
(112̄0), (101̄1), and (101̄2) would lead to rarely occurring dodecagonal, bullet pyramidal,
and pyramidal shapes, respectively (see the two lower diagrams in Figure 1.13a). For a
column or plate, the top and bottom are referred to as basal faces, which are perpendicular
to the c-axis, while the surrounding six sides are referred to as prism faces. The hexagonal
shape exhibited in the majority of ice crystals is evidently related to the basic hexagonal
symmetry within the crystal.
The crystal structure of Ih is stable down to −200°C and within atmospheric pressure.
It has a density of 0.917 g cm−3 , lower than liquid water, due to the low density of its
crystal lattice. Based on the structure of ice defined above, Subsection 1.3.2 presents the
homogeneous and heterogeneous nucleation of ice crystals.
1.3.2 Homogeneous and heterogeneous nucleation

Once a cloud extends to altitudes where the temperature is colder than 0°C, ice crystals
may form. There are two types of nucleation that can lead to the formation of ice, namely,
homogeneous and heterogeneous nucleation. Homogeneous phase transitions can lead to
the formation of ice. Homogeneous freezing of a pure liquid drop occurs when statisti-
cal fluctuations of water molecular arrangement produce a stable, ice-like structure that
can serve as an ice nucleus. Two possibilities determine the conditions for homogeneous
nucleation of freezing: the size of the stable nucleus and the probability of occurrence
of embryonic ice nuclei by random rearrangement of water molecules. These quantities
depend on the surface free energy of a crystal–liquid interface. Based on theory and exper-
iment, droplets smaller than 5 µm will freeze spontaneously at a temperature of about
−40°C. Larger droplets are predicted to freeze at slightly warmer temperatures. Some
liquid droplets exist in clouds as cold as −20°C.
Homogeneous deposition or sublimation occurs when vapor molecules form a stable ice
embryo via chance collisions. The surface free energy of a crystal–vapor interface is poorly
known; however, theory predicts that homogeneous nucleation of deposition should only
occur at extreme conditions of supersaturation. More than 20-fold supersaturation with
respect to ice is required at a temperature a few degrees below 0°C, and still higher super-
saturation at colder temperatures. Experimental confirmation of the theory of homogeneous
deposition appears to be impossible because liquid droplets will nucleate homogeneously

before the supersaturation reaches the high values required for ice. These droplets will
freeze spontaneously at temperatures colder than −40°C, making it impossible to recog-
nize ice crystals that might be formed by deposition. Homogeneous deposition cannot occur
in the atmosphere because the necessary extreme supersaturations never exist.
Ice crystals usually appear in a cloud in appreciable numbers when the temperature
decreases to below about −15°C, a range signifying heterogeneous nucleation. Water
vapor in contact with most materials freezes at these temperatures, while deposition occurs
on most surfaces at supersaturation and supercooling less than homogeneous nucleation
values. Thus, nucleation of ice in supercooled water or a supersaturated environment can
be enhanced by the presence of foreign surfaces or suspended particles.
The foreign material provides a surface onto which water molecules can impinge, stick,
bond together, and form aggregates with an ice-like structure. The larger the aggregate,
the more likely it is to be stable and to continue to exist. The probability of heterogeneous
nucleation by freezing or by deposition depends strongly on the properties of the substrate
material as well as on supercooling and supersaturation. The more tightly bound the water
molecules are to the substrate, the greater the probability of ice nucleation. In addition, if
the crystal structure of the substrate closely resembles that of an ice crystal plane, it will
increase the chances of ice nucleation. When the binding and the matching of the crystal
lattice are good, the supersaturation or supercooling required to nucleate ice on a substrate
may be much lower than that required for homogeneous ice nucleation.
Supercooled clouds in the atmosphere develop and exist in the presence of large numbers
of aerosol particles, a very small fraction of which can serve as ice nuclei at temperatures
considerably warmer than −40°C, an approximate threshold for homogeneous freezing.
There are a number of possible heterogeneous nucleation mechanisms (Rogers and Yau
1989). Ice may form directly from the vapor phase on suitable deposition nuclei, as shown
in Figure 1.11a. The lower panels show three modes of activation for freezing nuclei: some
serve first as centers for condensation, then as freezing nuclei (Figure 1.11b); some promote
freezing the instant they come into contact with a supercooled droplet (Figure 1.11c); others
cause freezing after becoming imbedded in a droplet (Figure 1.11d). A given particle (ice
nuclei, in general terms) may nucleate ice in different ways, depending on temperature and
saturation conditions, as well as being controlled by unknown probability in the course of
its history in the cloud.
1.3.3 Atmospheric ice nuclei

Observations have pointed to clay minerals, especially kaolinite, a common material found
in many soil types, as a major component of atmospheric ice nuclei. The nucleation threshold
is at a relatively warm temperature of −9°C. Snowflakes collected as they fall to the ground
are usually found to contain particles that appear to be centers for crystal growth. These
particles have been identified by electron microscopy to be kaolinite, with sizes ranging
from 0.1 µm to 4 µm. Another source of ice nuclei has been revealed by the discovery that
(a) Heterogeneous
Deposition
(b) Condensation
Followed by Freezing
(c) Contact Freezing
(d) Immersion Freezing
Figure 1.11 A schematic depiction of the processes for ice nucleation mechanisms involving atmo-
spheric ice nuclei and ice crystal formation: (a) direct heterogeneous deposition, (b) condensation in
the drop, followed by freezing, (c) contact with the drop, followed by freezing, and (d) immersion
into the drop, followed by freezing.
the bacteria in decaying plant leaf material can be effective nuclei at warm temperatures.
Research has shown that common soil particles can be active at temperatures warmer than
the threshold for kaolinite, which may explain the presence of submicron-sized nuclei of
some minor organic substance. The bacterium itself has also been found to serve as an ice
nucleus at a temperature as warm as −1.3°C, despite the fact that its nucleating ability is a
rare and changeable property. The overall significance of biogenic nuclei remains a subject
of research. Ice nuclei active at −4°C have also been observed in sea water rich in plankton.
In addition, meteoric material has been found to be an important source of atmospheric ice
nuclei. Finally, we note that the most effective material to have been used as an artificial
nucleating agent is silver iodide, discovered by Vonnegut (1947).
Measurements at coastal sites indicate more nuclei in air from sources over land than
over the ocean. The concentrations of nuclei over land also tend to decrease with altitude,
consistent with a source at the surface. Even at the South Pole, the particulates in snowflakes
are found to be clay minerals. It is clear that more than one kind of material can serve as
atmospheric ice nuclei, depending on temperature, humidity, and proximity to sources;
one may be more important than the others in a given cloud. Although the atmosphere
has an abundance of condensation nuclei, ice nuclei are scarce, regardless of their origin.
For this reason, the supercooling of cloud water to −15°C or colder appears to be rather
common.
Moreover, worldwide measurements of ice nucleus concentrations as a function of tem-
perature indicate that they tend to be higher in the Northern than the Southern Hemisphere.
On average, the number concentration N of ice nuclei per liter of air active at temperature
T tends to follow the empirical relationship ln N = a(T ∗ − T), where T ∗ is the threshold
temperature at which ice nuclei are active, typically at about −20°C (Fletcher 1962), and
a varies from 0.3 to 0.8 based on numerous studies. In urban air, the total concentration of
aerosol is on the order of 108 per liter, and only about one of them has been observed to act
as an effective nucleus at −20°C.
1.3.4 Secondary ice particle production

The occurrence of ice crystals in clouds is related to cloud type, temperature, and cloud
evolution. Observations confirm that the colder the cloud temperature, the greater the
likelihood that some ice crystals are present along with the supercooled water droplets.
Moreover, the probability of ice being present is 100% when the top temperature of the
cloud is below about −13°C. The first crystals to appear in a cloud must form on ice nuclei.
An exception to this generalization may be cirrus clouds, which form at low temperatures
(between about −20 and −60°C) in such a manner that homogeneous freezing may occur
as soon as the liquid phase appears. Additional crystals may then be produced by secondary
processes in which the primary crystals are “multiplied.”
Two mechanisms have been recognized as contributing to secondary ice particle pro-
duction: when ice crystals fracture and when freezing drops shatter or splinter. With respect
to the former, it is commonly observed that many crystals fragment in clouds. These are
produced when graupel particles overtake and collide with fragile, slower-falling dendritic
crystals. However, the process of crystal fracture appears to be difficult to quantify and
hence it is difficult to determine the rate of production of secondary particles as well as
the conditions favoring fragmentation. With respect to the latter, the mechanism of ice
multiplication is considered to be very effective under suitable conditions for production of
secondary particles when supercooled drops grow to an appropriate size and are captured
by graupel particles. At temperatures warmer than −3°C, drops tend to spread over the
ice surface instead of freezing as discrete drops. At temperatures colder than −8°C, the
freezing appears to proceed rapidly, starting with an outer shell of ice. Between these tem-
peratures, observations have revealed that small particles are ejected by drops larger than
25 µm in diameter. This mechanism of secondary particle production, sometimes called
rime-splintering, may account for high concentrations of ice crystals that are sometimes
observed in maritime cumulus clouds with temperatures warmer than −10°C.
1.4 Growth of ice crystals

1.4.1 Growth by diffusion
After ice crystals are formed in a cloud, they can undergo growth by diffusion processes,
particularly if supercooled drops coexist. On the basis of the Clausius–Clapeyron equation
derived from the conservation of the Gibbs function during phase transition, we have
des Les
= , (1.4.1a)
dT Rv T 2
(a)
12 a Evaporation (> 0oC)
b Sublimation Water
10 c Melting c
a* Evaporation (< 0oC) a
Pressure (hPa)
6 l $
Triple Point
4
Ice Water Vapor
2 a* b
0
-40 -30 -20 -10 0 10
(b)
30
25
100 x Δe sat (hPa)
20
15
10
5 Max
$
0
0 -10 -20 -30 -40 -50
Temperature (oC)
Figure 1.12 (a) Saturation vapor pressure as a function of temperature for three thermodynamic
phases of water molecules (vapor, water, ice) indicated by the regions in a two-dimensional diagram
based on the Clausius–Clapeyron equation. The triple point is denoted by a heavy dot and the solid
and dashed lines below 0°C represent ice and water, respectively. (b) Differences of saturation vapor
pressure between water and ice as a function of temperature, with the maximum at about −12°C.
where es denotes the saturation vapor pressure at a temperature T, L is the latent heat, and
Rv is the gas constant for water vapor. Integration leads to
es (T )
= exp[L(1/T0 − 1/T )/Rv ], (1.4.1b)
es0
where es0 (= 6.11 hPa) is the value at the triple point with a temperature of T0 (= 273.15 K)
at which vapor, liquid water, and ice coexist (Figure 1.12a). Because the latent heat for the
transition between vapor and ice is larger than that between vapor and water, the saturation
vapor pressure over ice is smaller than over its supercooled water counterpart. Consequently,
ice crystals will grow at the expense of supercooled droplets with a maximum growth rate
at a temperature of −12°C (Figure 1.12b). The growth of ice particles by this process is
referred to as the Wegener–Bergeron–Findeisen process (Wegener 1911, Bergeron 1935,
Findeisen 1938).
To quantify the mass increase in an ice particle, we must consider the diffusion process.
The vapor field in a cloud and ambient air can be described by the vapor density or absolute
humidity. At any point in the vapor field, the concentration of molecules, n, can be defined by
the conventional time-dependent second-order diffusion equation nt = D2 n, where
D is the molecular diffusion coefficient. Subject to the steady-state condition and imposing
boundary conditions at the positions of the ice particle and ambient air, the growth equation
for an ice crystal can be expressed in terms of vapor density in the form
dm
= 4πCD[ρv (∞) − ρv (s)], (1.4.2a)
dt
where ρ v () and ρ v (s) denote the vapor density of the ambient air and at the ice
crystal surface, respectively. To account for the non-spherical nature of ice crystals,
a parameter C has been introduced to denote the equivalent electrical capacitance (in
length units), which is a function of the size and shape of the particle. For a sphere,
C = r. For a circular disk of radius r, which can be used as an approximation for plate-type
ice crystals, C = 2rπ. Ice needles may be approximated by the formula for a prolate
spheroid of major and minor semi-axes a and b such that C = Aln [(a + A)b], where
A = (a2 − b2 )½ . For an oblate spheroid, we have C = aεarcsin ε, where the ellipticity
ε = (1 − b2 a2 )½ (Rogers and Yau 1989, Pruppacher and Klett 1996). Actual ice crystals
have more complex shapes such as commonly occurring plane dendrites and plates, which
can be reasonably approximated by a circular disk of equal area, while needles can be
approximated by long prolate spheroids.
In association with diffusion is the release of latent heat, which tends to raise the ice
crystal surface temperature above the ambient value. Analogous to mass diffusion, the
diffusion of heat away from the ice crystal is given by
dQ
= 4πCK(Ts − T ), (1.4.2b)
dt
where Ts is the temperature at the crystal surface, T is the ambient temperature, and K is the
thermal conductivity coefficient. Under stationary growth conditions, we have LI dmdt =
dQdt, where LI is the latent heat associated with the transformation between ice and water
vapor. Thus, the value of ρv (s) is determined according to this balance between the rates of
latent heating and heat transfer away from the surface, as follows:
[ρv (∞) − ρv (s)] K
= , (1.4.3)
(Ts − T ) L1 D
Using the Clausius–Clapeyron equation and the definition of saturation ratio with respect
to ice, Si = eei , where e is the vapor pressure of the ambient air and ei is the saturation
pressure over ice, an analytical expression for crystal growth can be derived and is given in
the following form:
dm 4πC(Si − 1)
= , (1.4.4)
dt (fK + fD )
where the terms in the denominator are defined by
fK = L2I /(KRv T 2 ), fD = Rv T /[Dei (T )], (1.4.5)
with ei (T) being the vapor pressure over ice at a temperature T. The growth rate varies
inversely with pressure and is a function of the temperature, with the maximum growth at
about −15°C over a wide range of pressure.
The preceding theoretical development for ice crystal growth based on diffusion of
water vapor cannot explain the different growth habits of ice or the transitions from one
habit to another. The more fundamental molecular-kinetic approach is required to explain
the formation of these habits or shapes. The surface of a growing crystal, as described in
Subsection 1.3.1, is made up of flat terraces of different heights, terminating at ledges and
separated by steps. Molecules of water that impinge onto the crystal surface are bound
more strongly at the ledges than on terraces. When unattached molecules move over the
crystal surface, they orient preferentially at ledges, leading to growth by the lateral motion
of surface steps. The growth rate is determined by the rates at which steps are generated
and advance, which in turn are controlled by temperature and supersaturation. The main
mechanism for the motion of steps is the diffusion movement of molecules across terrace
sites up to and into ledge sites. The kinetic effects determine the crystal habit, and tend to
slow the crystal growth rate, based entirely on the continuous vapor diffusion theory.
Temperature and saturation conditions determine not only growth rate but also ice crystal
habits, which are basically hexagonal structures with a range of different axis ratios. Ice
crystals have one common basic shape, a six-fold symmetric (hexagonal) prism with two
basal planes of type (0001) and six prisms of type (101̄0) in terms of Miller indices, as
shown in the two upper diagrams of Figure 1.13a. The orientation of the crystal lattice is
defined by the intersections of the planes with the hexagonal axes a1 , a2 , a3 , and c. Faces
such as (112̄0), (101̄1), and (101̄2) grow quickly to become the edges and corners of a
crystal, while faces of the type (0001) and (101̄0) grow slowly and become the bounding
faces of the crystal. Laboratory experiments have revealed that the rate of propagation of
the basal faces growing along the c-axis relative to that of the prism faces (growth along
the crystallographic direction of type [101̄0]) varies with temperature and supersaturation.
Primary ice crystal habits include column, plate, and dendrite, as shown in Figure 1.13b.
When an ice crystal grows and moves around in the cloud, its habit will be modified as a
consequence of changing temperature and saturation conditions. Sector stars can be formed
when plates develop peripheral dendritic structure, while capped columns are produced
when columns develop plates on their ends. The intricate stellar shapes have also been
frequently observed as a consequence of variations on the dendritic form (see Figures 1.14
and 1.15).
(a)
c
[0001]
a3
(0001) a2
(0001)
- -
[1210]
- - -1 0 ) -
(1010) (1010) (01 [ 0 11 0 ]
-a3
a1 -
-
[1010] [ 11 2 0 ]
--
[ 2 11 0 ]
(0001)
(1012)
-
- -
( 1 0 11 )
( 1 0 11 )
-
(1010)
- -2 0 )
(1010) ( 11
(b)
[0001]
--
[ 2 11 0 ]
--
[ 2 11 0 ]
Figure 1.13 Schematic representation of different habits of ice crystals. (a) The orientation of crystal
lattice planes in terms of the intersections of the planes with the hexagonal axes a1 , a2 , a3 , and c
is illustrated in the upper left picture. Shown at upper right is the representation of coordinates and
planes by Miller indices. The two basal planes are denoted by (0001), while (101̄0) denotes one of
the six prism planes depicted in the diagram. The numbers in square brackets denote directions; a bar
over a number stands for an intersection in a negative axial direction. Shown in the lower diagram
are representations of bullet and pyramid (after Wolff 1957). (b) Three primary ice crystal habits are
column, plate, and dendrite.
Figure 1.14 Temperature and humidity conditions for the growth of natural snow crystals. The sketches indicate the types of snow crystals (after Magono
and Lee 1966).
Figure 1.15 Meteorological classification of all types of ice and snow crystals as suggested by Magono and Lee (1966). Some of these
crystals are extremely complex.
1.4.2 Growth by accretion

After diffusion growth, ice crystals can further grow via collision and coalescence, referred
to as accretion. Ice crystal growth by collision and capture is controlled by its fall speed, u,
which has been generally determined from laboratory experiments in terms of maximum
dimension as follows (Langleben 1954):
u = kLn , (1.4.6)
where L is the maximum diameter in cm and u is in cm s−1 . For dendrites, k 160 and
n 0.3, while for columns and plates, k 234 and n 0.3. The mass (in grams) and size
of different ice crystals are usually related by empirical formulas of the form
m = aLb , (1.4.7)
where a and b are fitting coefficients for a number of ice crystal shapes that have been
determined from laboratory experiments (Mason 1971). For example, a = 1.9 × 10−2 and
b = 3 for thin hexagonal plates, and a = 9.4 × 10−4 and b = 2 for stellar crystals.
In the process of growth by accretion, there are issues concerning collection efficiency.
The first is related to the aerodynamic problem of collision efficiency, while the second is
in regards to the question of whether sticking occurs, given a collision. Collection has been
shown to be complex, with the crystal perimeter a preferred area for collisions. Because
freezing is likely to occur on contact with supercooled drops, the coalescence efficiency
is generally expected to be unity; that is, after a collision, coalescence will follow. In the
process of crystal aggregation, however, the collection efficiency is less well understood.
Open structures like dendrites are more likely to stick, given a collision, than crystals of
other shapes, and sticking has been found to be more likely at relatively warm temperatures.
Despite a number of uncertainty factors, the ice particle accretion growth may be derived
and is approximately governed by the following general form:
dm
= Ē CWC πR 2 u, (1.4.8)
dt
where CWC stands for the cloud water content (in g cm−3 ; can be ice or supercooled water),
R is an equivalent radius of the collector crystal, Ē denotes the mean collection efficiency,
and u is the difference in fall speed between the collector ice particle and small ice
crystals or supercooled water droplets.
In a mixed cloud, ice particles can increase in mass by capturing supercooled droplets,
which subsequently freeze onto them upon contact and form a coating of rime. This process,
referred to as growth by riming, leads to the formation of various rimed structures, including
rimed needle, rimed column, rimed plate, and rimed stellar. When riming proceeds beyond
a certain stage, the original shape is destroyed and the rimed particle is called graupel
(ice water). If the freezing is not immediate, denser layer structures can be formed –
centimeter-sized hail particles are extreme examples.
Aggregation is the clumping together of ice crystals to form snowflakes. When two ice
particles collide, the probability of adhesion is determined by temperature and the types
of ice particles. Complex crystals, such as dendrites, tend to adhere to one another due to
entanglement upon collision, whereas two solid plates will tend to rebound. The probability
of two colliding crystals adhering increases with increasing temperature; it is particularly
likely above temperatures of about −5°C, at which ice surfaces are said to become sticky.
Some graupel and aggregated particles will be illustrated in Subsection 1.4.1.
1.5 Illustration of ice crystal morphology

1.5.1 Some historical perspectives
In accordance with Needham and Lu (1961), the hexagonal shape (six sides) of snowflakes
was first noted by Chinese writer Han Ying in a book entitled “Han Shih Wai Chuan”
(Moral Discourses Illustrating the Han Text of the Book of Odes), written around 135 BC,
the time of the Western Han dynasty. Han Ying stated that:
Flowers of plants and trees are generally five-pointed, but those of snow, which are called “Ying,”
are always six-pointed.
In many classical Chinese writings, especially poems, six is a symbolic number, correlated
with the element water and the direction North, while five is associated with the Earth and
the Center. Hsiao Tung, a sixth-century Crown Prince of the Liang Empire (the third of the
Southern Dynasties during the Southern and Northern Dynasties period), gave an excellent
example in a poem:
The ruddy clouds flout in the four quarters of the cerulean sky
And the white snow-flakes show forth their six-petalled flowers.
There was no crystallographic explanation of the hexagonal nature of the crystals; however,
Chinese writers and poets attempted to define six-sided structure based on the principle
of the Yin-Yang and Wu-Shing (five elements) theory. It was believed that two opposing
forces operated in the entire universe, as well as among interactions between all of its
elements. Yin and Yang are said to be associated with the female (or negative aspects of
nature) and the male (or positive aspects of nature), respectively. It was noted that Yin and
Yang are also related to even and odd numbers, respectively. In principle, all elements in
the universe can be categorized as either Yin or Yang. Water is considered to possess the
spirit of Yin, a female attribute, and is associated with an even number. It happened without
explanation that the number six was assigned to the element water in ancient Chinese times.
For example, the great twelfth-century Chinese medieval philosopher of the Sung Dynasty,
Chu Hsi, wrote,
Six generated from Earth is the perfect number of water, so as snow is water condensed into crystal
flowers, these are always six-pointed.
In a text, Master Chu also attempted to further explain the formation of snow:
The reason why “flowers” or crystals of snow are six-pointed is because they are only sleet split open
by violent winds (and sleet being half-frozen rain, i.e. water) they must be six-pointed. Just so, if you
throw a lump of mud on the ground, it splashes into radiating angular petal-like form. Now six is a
Yin number, and thai-yin-hsuan-ching-shih (selenite crystal) is also six-pointed, with sharp prismatic
angular edges. Everything is due to the numbers inherent in Nature.
From the standpoint of work on ice nucleation in the twentieth century, it is interesting to
note that Master Chu connected snowflakes to the translucent hexagonal crystals of gypsum
or calcium sulfate, types of minerals that can serve as effective ice nuclei. The idea of snow
being six-pointed and related to Yin appears to have been accepted in the continuous
writings of ancient Chinese naturalists as a fact of nature or as a symbolic explanation, but
no further study on its hexagonal structure has been made. Although not within the context
of ice nucleation, Ho and Needham (1959) have discussed in some detail the observed halos
and parhelia (see Subsection 3.1.4) reported in ancient Chinese literature.
It appears that the subject of snow crystal shape was not formally documented by the
early Arabs or Classical Greeks. The first European observations are found in the writing
in 1260 of Albertus Magnus, who thought that snowflakes were star-shaped and fell only in
February and March. Almost 300 years later, Olaus Magnus, a Scandinavian bishop, wrote
in 1555 about snowflakes but appeared to overlook their essential hexagonal symmetry.
It was not until 1611 that the great German astronomer Johannes Kepler, in his book A
New Year’s Gift of Hexagonal Snow (Kepler 1611), described the hexagonal symmetry
of snowflakes and attempted to explain this structure on an atomistic basis, leading to a
modern discussion of cubical and hexagonal shapes in relation to mathematical theories of
close-packing of spheres.
After Kepler, Descartes (1637) presented sketches of snowflake crystals based on obser-
vations he had made. We also note that a light ray, which undergoes minimum deviation
from a spherical particle to explain, in part, the generation of rainbows, is referred to as a
Descartes ray. Further to this work, Erasmus Bartholinus published sketches of snowflakes
showing the branching of hexagonal stars in 1660.
With the invention of the microscope in the latter half of the seventeenth century,
advances in the study of snowflakes in Europe were rapid. Robert Hooke in 1665 published
a microscopic view of snowflakes, followed by Martens (1675), who went to the Arctic
as a ship’s barber on a whaler, and documented the correlations between the shapes of
snow crystals and meteorological conditions, and Donato Rossetti, a Canon of Livorno
in Italy, who in 1681 published a classification of snowflakes and a detailed drawing of
a hexagonal plate crystal. In 1761, Johann Wilcke of Sweden appears to have been the
first experimentalist to artificially produce snow crystals, using iodoform and camphor as
nucleation agents.
The first systematic classification of snowflake morphology was made by Scoresby
(1820), and described the shapes of the columnar and complex forms of ice crystals.
Scoresby also noted a relationship between snowflake shape and air temperature, an impor-
tant connection first suggested by Guettard (1762) and more definitively established by
the work of Fritsch (1853). In Japan, Toshitsura Doi (1832, 1839), a feudal lord of Koga,
presented numerous excellent sketches of snow crystals. The last great collection of snow
crystal pictures before the advent of modern science is attributed to Glaisher (1855). The
microphotography technique was established and used extensively in Europe to study snow
crystals and classify them into planar, columnar, and combinations of the two by G. Hell-
mann (1893). Bentley and Humphreys (1931) presented 6000 photomicrographs of snow
crystals taken by Bentley over a period of nearly 50 years in Vermont, U.S.A.
1.5.2 Classification of ice crystals

Several twentieth-century studies have presented the dependence of the crystal shape on
temperature. However, the most comprehensive investigations were undertaken by Ukichiro
Nakaya and his associates in Japan in the late 1930s, using controlled conditions to grow
ice crystals in the laboratory. The well-known Nakaya diagram illustrates that the habits of
ice crystals can be largely grouped in terms of temperature and the saturation condition of
water vapor in the cloud chamber. Since then, the field of ice microphysics has been well
developed at temperatures between about −5 and −35°C with low ice saturation values. In
the range of −10 to −20°C and above ice saturation, hollow column ice is seen to form,
and when water vapor supply increases, pristine plates, thick plates of skeleton forms, and
crystals with broad and sector-like branches are produced below water saturation, where
we also see hollow column, solid thick plate, solid bullet, and long solid needle. Above
water saturation, we see the formation of solid long needle, hollow column, sheath, and
scroll between −5 and −10°C. Between −10 and −20°C, we see a variety of plate types,
including thick plate of skeleton form, crystals with sector-like branches, fern-like, ordinary
dentritic, and stellar crystals, as well as hexagonal plate. For temperatures between −20
and −40°C, hollow column, hollow bullet, bullet rosette (combination of bullets), and
aggregate (combination of columns) are seen. When a sufficient number of small water
droplets are available, rimed needle, column, plate, and stellar crystal as well as graupel
can be observed.
All of the cloud physics texts cited previously (e.g., Pruppacher and Klett 1996) present
ice crystal morphology as a function of temperature and ice and water saturations with
a degree of similarity, as shown in Figure 1.14 (Magono and Lee 1966). Moreover, Ono
(1970) conducted an extensive investigation of ice crystal habits in atmospheric ice clouds,
which had temperatures between −2 and −32°C, based on the replicator results of more
than 10 000 ice crystals sampled from aircraft. These crystal habits were found to be in
agreement with those gathered in the laboratory (Nakaya 1954, Hallett and Mason 1958,
Kobayashi 1961) and similar to those displayed in Figure 1.14. Finally, the identification
of ice crystal morphology (or shape or habit) described above can be seen in Figure 1.15,
where a diverse variety of complex ice crystal shapes and snowflakes which occur in the
atmosphere and near the surface are displayed (Magono and Lee 1966), with nomenclature
defined in the figure.
More recently, Walden et al. (2003), using a conventional replicator technique, collected
ice crystals over the Antarctic in winter, many of which were identified as hollow and
solid columns (diamond dust), plates, pyramids, bullet clusters, and spheroids (blowing
snow). Bailey and Hallett (2004, 2009) presented a comprehensive morphology (or habit
or shape) diagram for atmospheric ice particles. The authors derived these morphology
results through both laboratory studies and a number of field observations, utilizing a new
ice particle probe called the Cloud Particle Imager (CPI), and stated that most ice crystals
are defective and irregular in shape to varying degrees and are mostly polycrystalline
at temperatures below −20°C. They proposed a new habit diagram which retained the
traditional descriptions of habits presented in Figure 1.14 for temperatures above −18°C.
The dominant habits range from plates (0 to −4°C) to columns (−4 to −8°C) to plates (−8
to −22°C). But for lower temperatures, their results revealed that the habit is dominated
by polycrystals of various forms, with two distinct habit regions, involving platelike from
−20 to −40°C and columnar from −40 to −70°C. The new diagram also emphasizes
that most individual crystals are complex, irregular, and imperfect in appearance to some
extent, including single crystals such as plates and columns that are most common at low
ice supersaturation. Also, it was pointed out that the majority of very small ice crystals
growing at low ice supersaturation are compact faceted polycrystals, not spheroids, as had
been suggested in previous publications.
For the purpose of illustration, we provide two examples of the observed crystal habit.
Lawson et al. (2006) collected numerous ice particles at the South Pole Station in summer
and sorted the CPI images of ice crystals into nine habit categories, including small plates
and faceted polycrystals (resembling spheroids), plates, rosettes, budding rosettes, long
columns, thick plates/short columns, small irregulars, blowing snow, and complex crystals
with side planes (Figure 1.16; see also Walden et al. 2003). Figure 1.17, taken from Bailey
and Hallett (2004), illustrates the common types of polycrystals of varying sizes, observed
between −20 and −40°C, collected from numerous laboratory and in situ observations
published by a number of investigators and data centers.
In Figure 1.18, we present a general depiction of ice crystal size and shape associated
with temperature in the atmosphere. Ice crystal size and shape captured by a replicator
balloon sounding system in Marshall, Colorado, on November 10, 1994 are shown as a
function of height and relative humidity. The relative humidity was measured by a cryogenic
hygrometer (dashed line) and Vaisala RS80 instruments (solid line and dots). Also shown
is temperature as a function of height. It is apparent that at the cloud top, pristine and
small columns and plates are predominant. At the cloud bottom, however, irregular bullet
rosettes and aggregates are produced, presumably due to accretion processes associated
with vertical mixing and gravitational pulling.
1.5.3 Computer generation of three-dimensional ice crystals

In order to carry out light scattering and absorption calculations and analysis, ice crys-
tal shape, size, and orientation must be defined mathematically in accordance with their
geometry. Takano and Liou (1989a, 1995) and Yang and Liou (1998a) have developed
computational techniques to generate ice crystal shapes used in light-scattering calcula-
tions involving various types of ice crystals. In the following discussion, we present a
Figure 1.16 Examples of CPI images of ice crystals sorted into habit categories during Antarctic
summer (after Lawson et al. 2006).
Figure 1.17 Laboratory and in situ examples of commonly observed polycrystals between −20 and
−40°C obtained from a number of experiments and field aircraft observations (after Bailey and Hallett
2004).
number of computer-generated 3-D ice crystal shapes introduced in previous sections,

including a solid column, a solid plate, a droxtal, two spheroids, a hollow column, a bullet,
a bullet rosette, an aggregate ice crystal, a dendrite, and a Koch snowflake. We divide the
ice crystal shapes into four groups, as shown in Figure 1.19.
14
13 -70
12
-60
11
Temperature (oC)
-50
Height (km)
10
9 -40
8 -30
7
-20
6
-10
5
4 0
0 10 20 30 40 50 60 70 80 90 100
Relative Humidity (%)
Figure 1.18 Ice crystal size and shape as a function of height and relative humidity as captured
by a replicator balloon sounding system in Marshall, Colorado on November 10, 1994. The relative
humidity was measured by a cryogenic hygrometer (dashed line) and Vaisala RS80 instruments (solid
line and dots). Also shown is temperature as a function of height (Liou 2002, courtesy of Andrew
Heymsfield).
Figure 1.19a shows three quasi-spherical ice particles: droxtal, prolate spheroid, and
oblate spheroid. For a droxtal, only the radius of a circumscribed sphere R and the two
angles θ 1 and θ 2 are required to define its geometric properties. For example, crystal lengths
a1 , a2 , L1 , L2 , and h are defined by
⎫
a1 = R sin θ1 , a2 = R sin θ2 , ⎪
⎬
L1 = R cos θ1 , L2 = R cos θ2 ,⎪ (1.5.1)
⎭
h = a1 (L1 − L2 )/(a1 − a2 ).
Other geometric properties of a droxtal, such as the position vectors of vertices and the
unit vectors normal to the individual faces, can be expressed in terms of R, θ 1 , and θ 2 .
For example, the unit vector n̂n = (nnx , nny , nnz ) associated with the six upper trapezoidal
faces of a droxtal is defined as follows:
nπ ⎫
nnx = sin α sin ,⎪⎪
3 ⎪
nπ ⎬
nny = sin α cos ,⎪ (1.5.2)
3 ⎪
⎪
⎭
nnz = cos α,
Figure 1.19 Computer-generated ice crystal shapes: (a) quasi-spherical ice particles, (b) column
types, (c) plate types, and (d) bullet rosette types. Note that 2a = D, the ice crystal width. See text
for further discussion.
where n = 1, . . . , 6. The parameter α is an angle between a unit vector to an upper

trapezoidal face and the z-axis, as shown in Figure 1.20 (far left), and is given by

2(h + L1 − L2 )
α = sin−1 . (1.5.3)
[3a 2 + 4(h + L1 − L2 )]1/2
The geometry of a spheroid is defined by
x2 y2 z2
+ + =1 (1.5.4)
a2 a2 c2
z
unit vector z z z z
a of normal
g’ q’
x x x x
f’
O O O O O
y y y y
position vector
of vertex
(a2, 0, -L2) (a) (b) (c) (d)
Figure 1.20 The figure at far left presents the definitions of a unit vector, a position vector, and the
angle α involving a droxtal. The four pictures on the right show (a) a column in Cartesian coordinates
Oxyz, which can undertake three rotations to form an aggregated ice crystal; (b) rotation by an angle
γ ʹ with reference to its c-axis; (c) tilt of the c-axis from the z-axis by an angle θʹ in the x-z plane; and
(d) rotation of the c-axis about the z-axis by an angle φʹ.
where a and c are the lengths of the semi-axes, as shown in Figure 1.19a. The semi-axis
length a is the equatorial radius of a spheroid and c is the distance from the center to the
pole along the symmetry axis. When a < c, the spheroid is referred to as prolate (center);
when a > c, it is called oblate (far right).
Figure 1.19b consists of a solid column, a hollow column, and a solid column aggregate.
Let the radius of a hexagon be a and its half-height be c (= L2, where L is the length).
The equations which define the six hexagonal surfaces (1–6) can be expressed as
nπ nπ √
3
sin x + cos y− a = 0, n = 1, 2, 3, 4, 5, 6, (1.5.5)
3 3 2
while the top and bottom surfaces (7 and 8) are defined by
cos[(n − 7)π]z − c = 0, n = 7, 8. (1.5.6)
For a hollow column with a depth of d, the basal planes expressed in Eq. (1.5.5) are replaced
by two sets of six slant hollow planes in the form

n n+1 n n+1
− sin π − sin π (±d)x + cos π − cos π (±d)y
3 3 3 3
n
n
n+1
+ cos π sin π − sin π
3 3 3 (1.5.7)
n
n
n+1
− sin π cos π − cos π
3 3 3
× a(z ∓ c ± d) = 0, n = 1, 2, 3, 4, 5, 6.
In this equation, the term (z − c + d) corresponds to the top hollow face, while the term
(z + c − d) corresponds to the bottom hollow face.
To form an aggregated ice particle (far right in Figure 1.19b), we consider a hexagonal
ice crystal element whose c-axis is parallel to the z-axis of a Cartesian coordinate system
Oxyz, as shown in Figure 1.20a. An ice crystal can be rotated to any orientation by three
n=1 n=2 n=3 n=4
Figure 1.21 Computer generation of a Koch snowflake based on the procedure described in the text.
consecutive coordinate transforms: (1) rotation by an angle γ ʹ with reference to its c-axis
as in Figure 1.20b, (2) tilting of the c-axis from the z-axis by an angle θ ʹ in the x-z plane
as in Figure 1.20c, and (3) rotation of the c-axis about the z-axis by an angle φʹ as in
Figure 1.20d. This procedure is repeated for a number of hexagonal elements attached in
an arbitrary manner to produce an aggregated crystal. The spatial position of a specific
element can be determined by the coordinate values of the particle’s center: x0 , y0 , and z0 .
In Figure 1.19c, the equation defining the geometry of a solid plate (left) is the same
as that for a solid column; see Eqs. (1.5.5) and (1.5.6). For dendrites (center), we have
attached six branches to a hexagonal plate. Let the branch height bt and branch width bb
be defined as shown in Figure 1.19c; then each branch plane (12) can be expressed in the
form

π π

n+m
(a + bt) sin n − a sin n + bb sin π
3 3 3
π
× x − (a + bt) cos n

3

π π n+m
− (a + bt) cos n − a cos n + bb cos π
3 3 3
π
× y − (a + bt) sin n = 0, n = 1, 2, 3, 4, 5, 6, m = 2 or 4. (1.5.8)
3
The shape of a snowflake (right) can be constructed using the following procedure. We start
with an equilateral triangle with sides of unit length as an initiator, then add to the middle
third of each side a -shaped peninsula with three side lengths of 13, resulting in a Star of
David (a six-pointed star). Subsequently, on each of the 12 sides of the resulting six smaller
triangles, we carry out the same additions to yield 48 sides, and continue the process n
times, which will lead to the formation of a Koch snowflake (von Koch 1904). Usually,
n = 4 is sufficient for demonstration purposes. A Koch snowflake, which has a fractal
dimension of log4log3 = 1.262, develops only on the side planes; the two basal planes are
flat. Thus, a 3-D Koch snowflake contains two flat basal planes and 3 × 4n highly irregular
side planes (with 6-sided symmetry) associated with n fractal iterations (see Figure 1.21
for n = 4).
2a
(a) (b)
L a
χ l
O t
O
d
O O χ
Side View Top View Side View Top View
(c) (d) l3
l1
L
l2
bb a1 a2
O a3
O
O O
bt
a
Side View Top View Side View Top View
Figure 1.22 Side and top views of (a) a 3-D hollow column (O denotes the center); (b) a bullet
rosette (χ is the angle between bullets); and (c) the dendrite defined in Figure 1.19. Also shown is
(d) a capped column mapped onto side and top planes (after Takano and Liou 1995).
In order to conduct light scattering exercises, we must define the required surface area
and volume.
√ Let the length of an equilateral triangle be a. Then, its 2-D area is given by
A0 = ( 34)a2 . The area generated by n iterations is then given by
An = An−1 + (3 × 4n−1 /32n )A0 . (1.5.9)
The volume of a Koch snowflake is simply Vn = An L, where L is the length defining √ the two
basal planes. Also, the radius of a Koch snowflake can be defined by a∗ = ( 33)a. For
comparison purposes, we may also define an equivalent sphere with a radius aeq such that
Vn = (4π3)aeq 3
. The Koch snowflake is a concave shape which differs from the conventional
convex ice particles such as plates and columns.
In Figure 1.19d, consider a solid bullet rosette with six branches (far left). For a single
bullet, one of the hollow depths d is replaced by −t, where t is the tip length of the bullet.
Consider a single bullet whose tip is at the origin and c-axis is along the z-axis. The bullet is
rotated by an angle in the zenith direction and by another angle in the azimuthal direction.
Then four or six rotated bullets can be combined to produce a bullet rosette. To construct a
hollow bullet rosette with six branches, we may replace the solid bullet by a hollow bullet
as shown in the far right, where the hollow depth is defined as H.
1.6 Size and shape distributions in cirrus clouds 41
To further define the shapes of a number of irregular ice crystals displayed in Figure 1.19,
we illustrate in Figure 1.22 the top and side views (a) of a hollow column, (b) a four-bullet
rosette, and (c) a dendrite. Also shown is (d) a horizontal view of a capped column, which
frequently occurs in tropical cumulonimbus. All the relevant geometric parameters and
angles are defined in this figure and preceding discussions.
1.6 Ice crystal size and shape distributions in cirrus clouds

1.6.1 Introductory remarks
A substantial number of ice crystals in the atmosphere have been observed in cirrus clouds,
due to their high location and low temperature. The formation, maintenance, and dissipation
of these clouds are principally governed by large-scale synoptic features and disturbances.
In the tropics, they are related to deep-cumulus outflows. The ice crystal shapes depend on
temperature and relative humidity, as well as whether they undergo collision and coales-
cence processes in the clouds.
Weickmann (1948) observed that at humidity levels approaching water saturation, ice
crystals in cirrus and cirrocumulus have prismatic skeleton shapes that occur in hollow and
cluster crystals (bullet rosettes). Between ice and water saturation, ice crystals in cirrus
grow in the form of prisms. In cirrostratus, where relative humidity is normally close to
ice saturation, ice crystals are primarily individual with full crystals such as columns,
prisms, and plates. In midlatitude cirrus, where substantial aircraft observations have been
conducted, ice crystal shapes are conventionally classified as a function of temperature
(Heymsfield and Platt 1984, Heymsfield and Iaquinta 2000). For most cirrus clouds, hollow
columns and hexagonal plates are the most abundant types near the cloud top. Spatial ice
crystals such as bullet rosettes are the predominant forms above about −40°C, while hollow
or solid columns prevail below about −50°C. Between these temperatures, convective cirrus
clouds contain a majority of spatial crystal forms, whereas stable cirrus clouds are primarily
composed of hollow columns. As discussed in Subsection 1.6.2, numerous field campaigns
have been designed to understand the composition and structure of cirrus clouds.
1.6.2 Ice crystal size and shape spectra

Typical ice crystal size distributions for cirrus clouds have been developed on the basis of
a number of intensive field observations from aircraft platforms. Following is a summary
of these programs and the ice crystal size and shape data that were collected.
At midlatitude, the First ISCCP Regional Experiment (FIRE) for cirrus study was
conducted in 1986 (I) over Madison, Wisconsin, and in 1991 (II) over Coffeyville, Kansas.
During FIRE I, ice crystal size distributions and habit images were gathered from 2D-C and
2D-P image probes, which measured ice crystal sizes from 50 to 100 µm and from 1000 to
3000 µm, respectively. During FIRE II, balloon-borne replicators were used to capture ice
crystal size ranging from about 10 to about 1000 µm with a resolution of 2 µm (Miloshevich
and Heymsfield 1997). The European cirrus experiment carried out in 1989 also provided
useful ice crystal size and shape data. A fourth midlatitude cirrus field experiment was
carried out in the spring of 2000 during an intensive observation period in the vicinity
of Lamont, Oklahoma under the auspices of the Department of Energy’s Atmospheric
Radiation Measurement (ARM) program. During this experiment, the University of North
Dakota Citation aircraft equipped with 2D-C and 2D-P probes was used to measure ice
crystal size and shape and other microphysical data. During the Subsonic Aircraft: Contrail
and Cloud Effect Special Study (SUCCESS) conducted in 1996, useful ice crystal and
microphysical datasets were obtained for cirrus, contrails, and contrail cirrus. Some of the
contrail ice crystals were analyzed by Liou et al. (1998).
Observations of ice crystal size and shape distributions for cirrus over the tropics were
made during the Central Equatorial Pacific Experiment (CEPEX) conducted in 1993. Air-
craft microphysical measurements were carried out by a 2-D optical probe (30–300 µm)
and video ice particle sampler (VIPS, < 30 µm). Based on available measurements in the
tropics, ice crystal sizes ranged from about 10 to 2000 µm and presented in four dominant
shapes: bullet rosettes, aggregates, hollow columns, and plates, similar to those occurring
in midlatitude. Smaller ice crystal sizes generally occur in cloud-top (colder temperature)
conditions, whereas larger ice crystal sizes are associated with warmer temperatures and/or
the developing stage of clouds associated with convection (Rolland et al. 2000). Tropical
ice crystal data were also available from the Kwajalein Experiment carried out in 1999
under the auspices of the Tropical Rainfall Measuring Mission (TRMM). The cloud tem-
peratures ranged from 0 to −70°C with optical depths between 20 and 30 associated with
deep cumulus convection. The most recent tropical cirrus experiment was conducted in
a field campaign called the Cirrus Regional Study of Tropical Anvils and Cirrus Layers
(CRYSTAL) Florida Area Cirrus Experiment (FACE) in the summer of 2002 using NASA
WB57 high-flying aircraft.
Ice crystal size data were collected via CPI and video VIPS, as well as forward scattering
spectrometer probe (FSSP) for ice crystals measuring from a few micrometers to about
30 µm in the −58 to −76°C temperature range. To characterize ice crystal size distribution
and account for habit factor, a mean effective ice crystal size has been defined as 1.5 times
the ratio of ensemble particle volume and the particle projected area, as follows (Foot 1988,
Francis et al. 1994, Wyser and Yang 1998, Yue et al. 2007):

3
De (l, k) = [
i fi,l (L)Vi (L)]nk (L)dL [
i fi,l (L)Ai (L)]nk (L)dL, (1.6.1)
2 k k
where the i and k indices stand for habit and size distributions, respectively; Vi (L) and
Ai (L) are the volume and projected area, respectively, of an ice crystal with a maximum
dimension L for the ith habit; fi,l (L) is the fraction of the ith habit in the lth habit distribution;
nk (L) dL is the number density of ice crystals in the size interval (L, L + dL) for the kth size
distribution. The factor 32 takes into account the spherical counterpart in terms of volume
and area. In this manner, there is only one unique De value for a given set of the lth habit
distribution and kth size distribution (Yue et al. 2007).
1.6 Size and shape distributions in cirrus clouds 43
Figure 1.23 depicts the ice crystal size distributions for a variety of cirrus covering a
number of temperature and cloud development conditions. Twelve size models for cirrus
were developed on the basis of the preceding data, with a large spectrum of De ranging
from 24 to 124 µm; for example, 30 µm for cirrus with a temperature of −60°C, 42 µm for
cirrostratus, 58 µm for a cirrus with a temperature of −40°C, and 124 µm for a developing
cirrus uncinus. This type of classification of ice crystal size distribution is useful for the
development of look-up tables for the purpose of retrieving ice crystal size and optical depth
from satellite radiance measurements. In this presentation, we also note that ice crystal size
in cirrus clouds spans values from a few to thousands of micrometers. Also depicted in
Figure 1.23 are four size distributions for contrails, which range from a few to about 50 µm
(Liou et al. 1998).
Finally, ice crystal shape and size data in Arctic ice and stratiform clouds associated with
frontal systems were collected during the First ISCCP Regional Arctic Cloud Experiment.
Ice crystals larger than 40 µm were collected, and inspection of their shapes shows a
combination of pristine and irregular types, including solid and hollow columns, prisms,
plates, aggregates, and branched particles (Korolev et al. 1999). In the Arctic region, ice
crystal data were also collected in the fall of 2004 during the DOE’s ARM Mixed-Phase
Arctic Cloud Experiment (MPACE) at its North Slope Alaska site. Additionally, we note
that the extensive collection of ice particles at a surface station in the Antarctic illustrates
the prevalence of long needle ice crystal types (Grenfell and Warren 1999).
1.6.3 Correlation between ice crystal maximum dimension and width

Ice particles are three-dimensional in nature and intercept light from all directions. Due
to their irregularity, the size and shape of an ice crystal have generally been defined by
its maximum dimension on the basis of two-dimensional photographs and/or images, as
displayed in Section 1.5. However, the maximum dimension is not sufficient for the purposes
of light scattering and absorption analysis: an additional dimension is needed to define the
ice crystal volume and effective geometric cross-section (Yang et al. 2000a).
Laboratory and field observations have established that the maximum dimension, L,
and width, D, are related for certain types of ice crystals (Auer and Veal 1970, Heymsfield
1972). Reanalyzing the data gathered by Auer and Veal, Mitchell and Arnott (1994) derived
the following empirical relationship for columns:
D = 2aLb , (1.6.2)
where a = 0.35 and b = 1 for L < 100 µm, and a = 3.48 and b = 0.5 for L > 100 µm. As
illustrated in the morphology section (Section 1.5), many columns exhibit hollow structure,
which must be accounted for in light-scattering calculations. We may postulate that the
cavity depth in a column varies randomly, given by d = 2ξ <d>, where ξ denotes a
random number uniformly distributed in (0, 1) and < > signifies a mean value. A value of
<d>L = 0.25 has been used by Yang et al. (2000a) in light-scattering calculations.
Figure 1.23 Representative ice crystal size distributions for cirrus clouds as a function of ice crystal maximum dimension covering a range of
temperature and cloud type. Also shown are four size distributions for contrails (after Baum et al. 2000b, Rolland et al. 2000, and Liou et al. 1998
with modifications).
1.7 Correlation between water content and crystal size 45
For plates, the maximum dimension refers to D, while L is the length along the c-axis.
Their relationship has been developed and is given by Pruppacher and Klett (1996) and
Mitchell and Arnott (1994) as follows:
L = a(D/2)b , (1.6.3)
where a = 2.4883 and b = 0.474 for D2 ranging from 5 to 1500 µm.
For bullet rosettes, the L–D relationship can also be defined by the empirical equation
for columns, but with a = 1.1552 and b = 0.63. Moreover, the pyramidal tip, t, of a bullet
element may be specified via the following relationship:
√
3
t= tan α, (1.6.4)
4D
where α is the inclination angle with respect to the major axis of the bullet elements.
According to a series of photographs taken and analyzed by Greenler (1980), α 28°.
For aggregates, the constructing procedure has been developed by Yang and Liou (1998a)
using a relative unit to specify ice crystal dimension. Furthermore, to account for surface
inhomogeneity of the preceding ice particles, we may use a two-dimensional Gaussian
probability function (Cox and Munk 1954) to represent the roughness of the ice surface in
light scattering calculations.
Lastly, we wish to note that the spatial orientation of ice crystals in cirrus is a non-
negligible factor in addressing light scattering and polarization properties. Jayaweera and
Mason (1965) have studied the behavior of freely falling cylinders in a viscous fluid and
found that if the ratio of diameter to maximum dimension is less than 1, cylinders tend
to fall with their long axes horizontal. Observations by Ono (1969) of natural ice clouds
in the atmosphere revealed that columnar and plate crystals fall with their major axes
oriented horizontally. Platt et al. (1978) demonstrated that in cirrus clouds at approxi-
mately −15°C, ice crystals are predominately plate shape, oriented horizontally, based
on lidar backscattering measurements. In particular, it was noted that the return signals
from ice clouds of a vertically pointed laser beam retain the polarization of the inci-
dent energy and that, for this to occur, the plates must be perpendicular to the laser
beam.
1.7 Correlation between ice water content and ice crystal size
1.7.1 Introductory note
In recent years, developments in cloud modeling have included prognostic equations used
to predict ice water content (IWC) for high-level clouds formed in climate and general
circulation models (GCMs). This is a milestone accomplishment from the standpoint of
incorporating a physically based cloud microphysics scheme into these models, and equally
essential from the perspective of studying cloud–radiation interactions. However, cloud par-
ticle size is also an independent parameter that affects radiation transfer. For example, for
a given IWC in clouds, smaller particles will reflect more sunlight than larger counterparts,
an effect recognized by Twomey et al. (1984) and Liou and Ou (1989) that is associated
with aerosol–cloud indirect effects. As presented in preceding sections, ice crystal size and
shape in the Earth’s atmosphere are complex and intricate. After initial homogeneous and/or
heterogeneous nucleation involving suitable aerosol particles and atmospheric conditions,
ice crystal growth is governed by diffusion processes and subsequent actions by means
of collision and coalescence. These processes are complicated by the nature of the ice
crystal’s hexagonal and irregular shape. Incorporating a fully interactive ice microphysics
scheme based on first principles into a GCM appears to be an extremely difficult compu-
tational task. Innovative parameterization based on theory and observation to determine
ice crystal size from the model-generated IWC must be developed for GCM and climate
applications.
It has been common practice in GCMs to prescribe a mean effective ice crystal size
De (e.g., Gu et al. 2003). A number of GCMs have also used temperature to determine
De (Kristjansson et al. 2005, Gu and Liou 2006). This approach is rooted in earlier
ice microphysics observations from aircraft, and attests to the fact that small and large
ice crystals are related to cold and warm temperatures in cirrus cloud layers. Ou and Liou
(1995) developed a parameterization equation relating cirrus temperature to mean effective
ice crystal size based on a large number of midlatitude cirrus microphysics data presented
by Heymsfield and Platt (1984). Ou et al. (1995) reduced large standard deviations in the
size–temperature parameterization by incorporating a dimensional analysis between IWC
and De . Using CEPEX data, McFarquhar et al. (2003) developed a De parameterization as
a function of IWC for use in a single-column model.
1.7.2 Correlation analysis

In this subsection, ice crystal size and shape distributions obtained from in situ measure-
ments conducted in numerous field campaigns have been used to understand the correlation
between ice water content (IWC) and mean effective ice crystal size (De ). In the correlation
analysis, we have divided available datasets in accordance with three geographical areas
(tropics, midlatitude, and Arctic) because of their distinct ice cloud formation processes
(Liou et al. 2008). A significant fraction of tropical cirrus are generated from towering
cumulus convection, but the majority of midlatitude cirrus clouds are primarily related to
large-scale frontal and synoptic systems and mesoscale topographical forcings. In Arctic
regions, the formation of ice clouds appears to be directly related to cold temperature,
large-scale transport of sensible and latent heat, and boundary layer turbulence. We first
introduce the ice water content, defined as

IWC = V (L)ρi n(L)dL, (1.7.1)
where n(L) is the ice crystal size distribution, V(L) is the volume of an individual ice crystal
that accounts for the shape factor, ρ i is the density of ice, and L is the ice crystal maximum
dimension. We may define a mean effective ice crystal size related to IWC in the following
general form:

De = V (L)n(L)dL A(L)n(L)dL

= IWC ρi A(L)n(L)dL = IWC/ρi Ac , (1.7.2)
where we have neglected the constant factor 32 in Eq. (1.6.1), A(L) is the cross-sectional
area for an individual ice crystal, and Ac represents the total projected area for a given
ice crystal size and habit distribution. Heymsfield and McFarquhar (1996) found that Ac
aIWCb , where a and b are empirical coefficients. The Ac –IWC relation revealed that a
direct correlation between De and IWC exists. Larger (smaller) values of IWC imply larger
(smaller) values of De , which is in agreement with ice crystal growth by means of diffusion
and accretion. The definition of this mean effective size efficiently accounts for ice crystal
size and shape distributions in light-scattering calculations (Fu and Liou 1993, Yang et al.
2000a). However, their relationship is not unique, but is constrained by Ac . The procedures
developed by Yue et al. (2007) and Yang et al. (2000a, 2005) were followed for IWC, De ,
and Ac calculations required in correlation development.
Uncertainty in the measurement of small ice crystals (<100 µm) from aircraft platforms
has been an important issue in scientific discussion. Shattering of millimeter-sized ice
particles due to collision with the probe can artificially enhance the concentration of small
ice crystals (see Heymsfield et al. 2006). In view of the uncertainty in small ice crystal
measurements, we have conducted three independent IWC–De correlation experiments:
(1) maximizing small ice crystals; (2) reducing the concentration of these smaller ice
crystals (Nsm ) by one order of magnitude; and (3) reducing Nsm by two orders of magnitude.
Experiment 2 was used as the base run, while the other two give a possible range of
parameterized De due to uncertainties in small ice crystal measurements.
A total of 40 469 in situ measurements of ice crystal size distribution in the tropics were
available for analysis, including 5460 from CRYSTAL-FACE and 35 009 from CEPEX. The
former data were collected by instruments on board NASA’s WB57 on nine different dates
during July 2002. CEPEX data included measurements from a 2-DC probe for ice crystals
larger than about 100 µm and a parameterization of Nsm based on VIPS measurements
(McFarquhar and Heymsfield 1997) obtained in March and April 1993. These instru-
ments were on board Aeromet Learjet. Both the original CEPEX and CRYSTAL-FACE
datasets were comprised of ice crystal measurements averaged over 30-second intervals.
Due to the low sampling rate of CRYSTAL-FACE data, size distributions derived from
this experiment were averaged over 5 minutes. After selecting datasets having more than
five-channel size measurements to ensure proper size average, only 261 CRYSTAL-FACE
and 11 032 CEPEX cases were used in the correlation study. From the analysis of CEPEX
data, about 34% and 66% of ice crystals are solid columns and bullet rosettes/aggregates,
respectively.
3
10
(a) Polar Ice Clouds
102
1
10
Mean Effective Diameter (De, μm)
MPACE (10/17-10/18, 2004)

Best Fit
3
10
(b) Midlatitude Cirrus
2
10
10
1 -65oC < T < -40oC
ARM (3/9/2000)
-40oC < T < -20oC
ARM (4/18/1997)
Best Fit
3
10
(c) Tropical Cirrus
2
10
1
10 CRYSTAL-FACE (7/2002)
CEPEX (3-4/1993)
Best Fit
0
10
10-6 10-5 10-4 10-3 10-2 10-1 100 101
Ice Water Content (IWC, g m–3)

Figure 1.24 Mean correlation curves with standard deviations (vertical bars) for IWC and De for (a)
the Arctic region, (b) midlatitude, and (c) the tropics, using ice crystal data collected during a number
of field campaigns listed in this figure and discussed in the text (after Liou et al. 2008).
Figure 1.24c displays 11 293 data points in a 2-D logarithmic domain. The IWC values
cover a range from 10−6 to 1 g m−3 , while De ranges from 20 to 200 µm. A χ 2 best
fit was used for these observed data to obtain the best parameterization equation in a
polynomial form, as follows:
ln(De ) = a + b ln(IWC) + c[ln (IWC)]2 , (1.7.3)
where a = 5.4199, b = 0.352 11, and c = 0.012 680. Uncertainty in Nsm can result in
deviations from the base run ranging from 5% to 40%, as IWC decreases, much larger
than the standard deviation from statistical uncertainty. We have also made attempts to
correlate De and in situ temperature measurements (−70 to −20°C); however, the 2-D
data points are extremely scattered and do not show a consistent pattern. In this case, it
appears that temperature is not a suitable variable for De determination, probably due to
the predominant convective nature of cloud systems in the tropics, where vertical velocity
is a more significant parameter in regulating De .
A total of 4033 in situ size distribution measurements were obtained for midlatitude cases
taken at the DOE/ARM Southern Great Plain site. Ice crystals larger than 100 µm were
measured by the 2D-C probe on board the UND Citation. Aircraft datasets were composed
of measurements averaged over a 5-second interval. We selected size distributions that
had measurements greater than 5-size channels, resulting in 3919 cases. IWC ranges from
10−4 to 10−1 g m−3 , while De has values from 30 to 140 µm. Although habit and shape
information from ARM data sources was not available, previous studies revealed that for
midlatitude cirrus clouds, ice crystal shape spans from bullet rosettes and aggregates (60%)
to hollow columns (20%) to plates (20%) for L > 70 µm. For L < 70 µm, shapes are 50%
bullet rosettes, 25% plates, and 25% hollow columns (Baum et al. 2000b).
Correlations between De and IWC were improved by dividing the temperature into
two groups: −40 to −20°C (warm cirrus) and −65 to −40°C (cold cirrus), as shown
in Figure 1.24b. For warm cirrus, the correlation coefficients for parameterization are:
a = 5.2375, b = 0.131 42, and c = 0. For cold cirrus, we have a = 4.3257, b =
0.265 35, and c = 0.021 864. De for warm cirrus is generally larger than that for cold
cirrus, and the range of De and IWC for midlatitude cirrus is narrower than its tropical
counterpart.
In the Arctic region, analysis was based on in situ data collected during the DOE’s
ARM MPACE experiment at the North Slope of Alaska site in the fall of 2004, as shown
in Figure 1.24a. Ice clouds were observed on two days, October 17 and 18, consisting of a
total of 1705 cases. After a data quality check, only 468 cases were selected, which were
largely from the UND Citation 2D-C measurements. The data points were averaged over
a 30-second period to ensure adequate statistical sampling. For ice particles <100 µm, a
gamma distribution was used to extrapolate SDs to 2 µm based on the empirical coefficients
derived by Boudala et al. (2002). In terms of habit, Korolev et al. (1999) gave the percentage
of pristine and irregular habits in different temperature bins for high-latitude ice clouds.
Because temperatures were much lower during MPACE observations, ice clouds were
assumed to contain relatively more pristine particles (20%), with a ratio of columns to
plates of 100/20, resulting in 3.3% plates and 16.7% columns. The remaining irregular
ice particles include 40% bullet rosettes and 40% aggregates. The correlation coefficients
for parameterization are as follows: a = 4.8510, b = 0.331 59, and c = 0.026 189. The
values of IWC and De in Arctic ice clouds range from 10−2 to 1 g m−3 and from 50 to
120 µm, respectively. Nsm sensitivity experiments show that small particles contribute 5 to
20% as IWC increases, which is smaller than in the tropical and midlatitude cases. Arctic
ice cloud IWC and De have narrower ranges than those in the other two regions. We were
unable to obtain an adequate correlation between De and temperature, which ranges from
−57 to −17°C for this dataset, possibly due to less stratification of the polar temperature
profile.
1.8 Numerical modeling of cirrus cloud formation

1.8.1 A review
Most of the ice particles in the Earth’s atmosphere reside in cirrus clouds. In Section 1.2,
we showed that cirrus cloud systems are normally associated with synoptic and mesoscale
disturbances and are partly related to deep cumulus outflows. Day-to-day experience reveals
that cirrus clouds have a fibrous appearance (see Figure 1.1). As discussed in Section 1.2,
satellite imagery suggests that large portions of the tropics are covered by extensive cirrus
systems which vary spatially and temporally. Cirrus formations associated with mesoscale
complexes may have large vertical extents, with bases near 5–6 km and tops between 12
and 16 km, which evolve during the life cycle of the complexes and are modulated by
large-scale disturbances. At various stages, the cirrus system may involve a mid-level or
a single elevated deck in the upper atmosphere (Houze et al. 1981). Tropical anvils are
relatively stable and long-lived. Unlike water clouds, these high anvils contain irregular ice
crystals at low concentrations and can be optically thin and non-black. The influence of
cirrus clouds on the atmospheric radiation field depends on their solar and thermal infrared
radiative properties, which in turn are modulated by cloud composition and position. Cloud–
radiation feedback is important in the formation, maintenance, and dissipation of tropical
anvils produced by strong diurnal convection in the equatorial region.
Outflow cirrus clouds from tropical towers appear to be maintained in a convectively
active state by radiative flux gradients within the clouds (Danielsen 1982). These clouds are
warmed and lifted into the stratosphere by net radiative heating. Danielsen has suggested that
extended anvils would become radiatively destabilized by cooling at tops and warming at
bases. The resulting increase in the lapse rate within the anvils would drive convective fluxes
by providing an upward flux of water vapor. The additional moisture would promote rapid
ice crystal growth and fallout, hence serving as a dehydration mechanism for the tropical
stratosphere. Ackerman et al. (1988) have computed heating rates in typical tropical anvils
and found that the difference between heating rates at the cloud bottom and top ranges
from 30 to 200 K d−1 , leading to convective instability in the anvils. Radiative heating
could have important consequences for upward mass transport in the tropics. Using the
radiative heating rates presented by Ackerman et al. (1988), Lilly (1988) has analyzed the
dynamic mechanism of the formation of cirrus anvils using a mixed layer model, and has
shown that destabilization of the layer could be produced by the strong radiative heating
gradient. Although these studies have shed some light on the dynamics of tropical anvils,
a comprehensive understanding of their evolution and the role of radiative heating during
their evolution requires an analysis of results produced by an appropriate cloud model.
Global cirrus cloud fields play an important role in the radiation field of the Earth–
atmosphere system and, hence, significantly affect the atmospheric thermal structure and
climate (Liou 1986, 1992). This importance has been highlighted by a number of intensive
composite field observations: the First International Satellite Cloud Climatology Project
(ISCCP) Regional Experiment (FIRE) I in October–November 1986; FIRE II in November–
December 1991; the European experiment on cirrus (ICE/EUCREX) in 1989; and Subsonic
Aircraft: Contrail and Cloud Effect Special Study (SUCCESS) in April 1996. However, it
appears that fundamental understanding of the mechanisms for the formation, maintenance,
and dissipation of cirrus in the atmosphere remains limited and requires in-depth studies
from both observational and modeling perspectives.
Several attempts have been made to develop numerical models to investigate the role
of various physical processes in the life cycle of cirrus clouds using a one-dimensional
(1-D) model, in which the vertical velocity is prescribed. The 1-D model permits the use
of a highly complex microphysics module to study the evolution of ice clouds (Heymsfield
and Sabin 1989). Inclusion of the length of semi-major axes and the aspect ratio to define
hydrometers was developed by Chen et al. (1997) in a 1-D model to study cirrus clouds.
Employing a 1-D model, Detwiler and Ramaswamy (1990) investigated the evolution of
microphysics and heating rate profiles of a quiescent cirrus deck in a motionless atmosphere
without including nucleation and collision. Jensen et al. (1994a, 1994b) developed a 1-D
model which incorporated the explicit simulation of ice crystal size distributions, a number
of microphysical processes, and three groups of species, including condensation nuclei,
activated solution drops, and ice crystals, the source of which was primarily homogeneous
nucleation of the (NH4 )2 SO4 droplet.
Cirrus clouds are a dynamic and thermodynamic system that involves the intricate
coupling of microphysics, radiation, and dynamic processes (Gultepe and Starr 1995). A
multidimensional setting is thus required for interaction and feedback studies. In their
pioneering work, Starr and Cox (1985a, 1985b) developed a two-dimensional (2-D) model
and showed that the effects of radiative processes and vertical transport are both significant
in cirrus cloud formation and maintenance. Gu and Liou (1997a, 1997b, 2000) constructed a
2-D cirrus model that includes a second-order turbulence closure and found that turbulence
can significantly modulate the supersaturation condition and hence affect the rate at which
water vapor is converted to ice. Lin (1997) and Khvorostyanov and Sassen (1998a, 1998b)
have incorporated explicit microphysics schemes in 2-D models for the simulation of ice
crystal size distributions.
The modeling of anvil cirrus clouds has normally been connected with 3-D cumulonim-
bus simulations. Tao and Moncrieff (2009) have undertaken a comprehensive review of the
development of the cloud resolving model (CRM) from 1-D, 2-D, and 3-D perspectives
over the last four decades. The 3-D CRM has been used to investigate the evolution of cir-
rus clouds with reference to understanding the interaction of dynamics, microphysics, and
radiative transfer. In Subsection 1.8.2, we present a 2-D cirrus cloud model developed by Gu
and Liou (2000) to demonstrate the linkage to cloud fields with reference to the formation
and growth of ice crystals (Sections 1.3 and 1.4), ice crystal morphology (Section 1.5), ice
crystal size and shape distributions (Section 1.6), and light scattering and absorption by
ice crystals (Chapters 3 and 6). Cloud fields are governed by dynamic and thermodynamic
processes in the atmosphere in addition to microphysics and radiation interactions.
1.8.2 A two-dimensional cirrus cloud model

When modeling stratiform cirrus clouds, we consider two-dimensional space and use the
Boussinesq approximation, which treats a fluid as incompressible; that is, air density is
considered to be constant except when it is coupled with gravity in the buoyancy term
of the vertical momentum equation. Satisfying this approximation requires that the depth
of convection be much smaller than the scale height of the basic state, which is about
8 km. The convective region of cirrus studied has a depth of around 1 km during the entire
simulation period. The Coriolis term is neglected in view of the fact that the motion in
cirrus is small-scale.
Consider a cloud-free initial state which is horizontally uniform and at rest except for
a constant vertical motion w0 . Since turbulence plays an important role in the formation
and dissipation of stratus and stratiform clouds (Moeng 1986, Liu and Krueger 1998),
a second-order turbulence closure was incorporated in the formation of the cirrus cloud
model. Also, the explicit evolution of ice crystal size distribution was accounted for in cloud
formation. Based on the preceding considerations, the mean field equations for dynamic
and thermodynamic variables in the x-z plane can be expressed in the following forms (Gu
and Liou 2000):
∂u ∂u ∂u 1 ∂p ∂ 2 ∂
= −u − (w0 + w) − − u − u w , (1.8.1)
∂t ∂x ∂z ρ ∂x ∂x ∂z
∂w ∂w ∂w 1 ∂p ∂ ∂
= −u − (w0 + w) − + gB − u w − w 2 , (1.8.2)
∂t ∂x ∂z ρ ∂z ∂x ∂z
where u and w are velocity components in the horizontal and vertical directions, respec-
tively. The unsubscripted variables represent the resolved components with respect to
a prescribed background (initial) state, which is denoted by subscript 0. Variables with
primes represent unresolved turbulent fluctuations. B is the buoyancy term, which can be
represented by the potential temperature θ , the specific humidity of water vapor q, and the
ice water mixing ratio qc in the form
θ
B= + ε q − qc , (1.8.3)
θ0
where ε = (1 − ε)/ε, and ε is the ratio of the molecular weight of water to that of dry air.
The basic equations for θ , q, and the number densities of ice crystals for the ith size
group ni can be written as
∂θ ∂θ ∂(θ0 + θ ) ∂ ∂
= −u − (w0 + w) + π−1 (QC + QR ) − u θ − w θ , (1.8.4)
∂t ∂x ∂z ∂x ∂z
∂q ∂q ∂(q0 + q) ∂ ∂
= −u − (w0 + w) −C− u q − w q , (1.8.5)
∂t ∂x ∂z ∂x ∂z

∂ni ∂q dmi ∂ ∂
=− ni − [uni ] − [(w + w0 − Uf )ni ]
∂t ∂m dt ∂x ∂z

∂ni ∂ ∂
+ − u ni − w ni , (1.8.6)
∂t nuc ∂x ∂z
where π = (p0 /pr )R/Cp , with pr = 1000 mb; Cp is the specific heat at constant pressure;
mi is the mean mass of an individual particle in the ith size group; and Uf is the terminal
velocity of an ice crystal. The first term on the right-hand side of Eq. (1.8.6) represents
the net accumulation of ice crystal number density of a particular size due to diffusional
growth. The second term is the contribution from horizontal advection, while the third
term is the vertical convergence of ice crystal number density determined by vertical air
motion and ice crystal terminal velocity Uf . The fourth and fifth terms denote the nucleation
source and eddy diffusion contributions, respectively. To obtain solutions for the evolution
of ice crystal size distributions, the ice crystal length spectrum is discretized into a finite
number of intervals (bins) of 15 µm. A partial differential equation can then be expressed
for each bin and solved numerically. It is clear that several physical factors are involved
in the governing equations: radiation, saturation ratio, air temperature, and horizontal and
vertical motion.
Radiative heating, QR , can be computed from the radiative transfer scheme discussed in
Subsection 1.8.2.2. The ice water mixing ratio, qc , the heating rate due to phase change, QC ,
and the deposition or sublimation between water vapor and ice water, C, can be connected
to the ice crystal size spectrum in the forms

N
qc = ni mi , (1.8.7)
i=1
Ls C
QC = , (1.8.8)
Cp

N

dmi ∂n1
C = Cdif + Cnuc = ni + m1 , (1.8.9)
i=1
dt ∂t nuc
where the ice crystal size spectrum has been discretized into N bins; Ls is the latent heat of
sublimation; and Cdif and Cnuc represent sources and sinks of moisture due to diffusional
growth and nucleation of ice crystals, respectively. The ice crystals initially generated by
nucleation are denoted by the mass of the smallest size group, m1 . Thus, Cnuc is the product
of m1 and the rate of change of number density of ice crystals due to nucleation in this
size group. When the nucleation process is terminated, Cnuc = 0. In the following two
subsections, we discuss ice microphysics, interactions between radiation and an individual
ice crystal, and the radiative heating rate produced in cloud fields.
1.8.2.1 Ice microphysics

The subject of ice microphysics includes ice nucleation, diffusional growth, and terminal
velocity. Ice nucleation processes can be generally classified into four modes: homogeneous
freezing nucleation, heterogeneous freezing nucleation, heterogeneous deposition nucle-
ation, and contact freezing nucleation (see Figure 1.11). Homogeneous freezing nucleation
usually occurs at or below −40°C, while heterogeneous freezing is likely to be more
effective when sufficient water is involved (Chen and Lamb 1994). The nucleation process
requires the presence of ice nuclei (IN). Fletcher (1962) developed a relationship between
the effective IN concentration and temperature as follows: NIN = N0 exp[b (TSTD − T)],
where N0 and b are constants and TSTD = 273.15 K. This expression implies that NIN is a
well-defined function of temperature. However, Meyers et al. (1992) found that Fletcher’s
equation tends to under-predict ice crystal concentrations for temperatures warmer than
−20°C, while over-predicting them for colder clouds. Gagin (1972) and Huffman (1973)
found that the number concentration of effective IN increases with increasing supersatu-
ration over ice, Si , in the form NIN = cSik , where c and k are constants determined from
Huffman (1973) and Chen and Lamb (1994). Heterogeneous deposition nucleation can be
taken into account using this relationship. The effective INs nucleate immediately, implying
that their appearance is the same as the creation of new ice particles. The nucleation process
is assumed to be irreversible; that is, dNIN
0.
The time rate of change of mass for an individual ice crystal denoted in Eq. (1.8.9) can
be obtained from the classic mass growth theory, including the conservation of water mass
and total energy at the ice crystal surface (see Subsection 1.4.1). The diffusional growth for
an individual ice crystal is governed by the following two equations:
dm
+ 4πC D f1 ρ = 0, (1.8.10)
dt
dm
Ls + HR − 4πC K f2 T = 0, (1.8.11)
dt
where m is the mass of an individual crystal; C is the equivalent radius for diffusive
processes of a non-spherical crystal; D and K are the diffusion coefficients for water
vapor and heat, respectively; f1 and f2 are the combined ventilation and surface curvature
factors, respectively; ρ and T are the density and temperature differences between the
ice crystal and the environment, respectively; and HR denotes the radiative effect on the
individual ice crystal. Analytic expressions for mass growth rate and ice crystal temperature
can be derived by solving Eqs. (1.8.10) and (1.8.11) simultaneously. Moreover, using the
Clausius–Clapeyron equation, the ideal gas law, and Kelvin’s expression for vapor pressure,
we can obtain the following equation for ice crystal growth:
dm
= AS − BH R , (1.8.12)
dt
where
2s 4πf2 K C Ls
S = (si − 1) − , A= , B= ,
ρi Rw T r B Rw T 2 B 2
L2s f2 K Rw T
B = + (1.8.13)
Rw T 2 f1 D es
and si is the ice saturation ratio, s is the surface free energy of ice, ρi is the bulk ice density,
T is the temperature in kelvin, Rw is the gas constant for water vapor, r is half of the ice
crystal width, and es is the saturation vapor pressure with respect to ice at temperature
T. The values for the unknown coefficients were taken from Laube and Holler (1988),
Ramaswamy and Detwiler (1986), Heymsfield (1972), and Auer and Veal (1970). The
mean mass of ice crystals of each size group can be calculated based on parameterization
relationships.
The terminal velocity of ice particles, Uf , defined in Eq. (1.8.6) can be expressed by the
following relationship (see Subsection 1.4.2):
Re η
Uf = , (1.8.14)
ρL
where η denotes the air viscosity; L is the ice crystal characteristic length (maximum
dimension), which can be computed from the empirical relationship given by Auer and
Veal (1970); and Re is the Reynolds number, which can be calculated from the unified
Davis number (Böhm 1989).
1.8.2.2 Radiative transfer

Following Fu and Liou (1993), the single-scattering properties of hexagonal ice crystals
can be computed from the parameterization using a mean effective size, De , to repre-
sent the ice crystal size distribution in radiative transfer calculations. Detailed ice crystal
shapes, which are a function of temperature and supersaturation, were not accounted for
in the study. Based on physical principles, the extinction coefficient normalized by IWC,
the single-scattering albedo, and the expansion coefficients for the phase function may be
expressed as third-degree polynomials in 1De . The coefficients in the polynomials that
are wavelength-dependent are determined by numerical fitting to the exact results obtained
from the light-scattering and absorption programs using 11 observed ice crystal size distri-
butions (Takano and Liou 1989a, Fu and Liou 1993). See Section 6.1 for further discussion.
The single-scattering parameterization was performed for 6 solar and 12 IR bands. The
radiation parameterization program is driven by the mean effective ice crystal size and ice
water path, which is the product of cloud thickness and IWC determined from

IWC = ρ n(mi )mi , (1.8.15)
i
where n(mi ) is computed from the model and ρ is the ice density.
For an individual ice crystal, the heating produced by the radiation effect must be
the product of the absorption cross-section and the net radiative flux density, and can be
expressed as follows:
∞
HR = σa,λ (mi )[Fλ+ + Fλ− − 2πBλ (T )]dλ ∼
= σa (mi )[F + + F − − 2σ T 4 ], (1.8.16)
0
where HR is the rate of radiative energy gained by an individual particle, which is determined
separately for each size bin, F+ and F− are the upward and downward net radiative fluxes,
and Bλ (T) is the Planck function at temperature T. To simplify the calculations, we may use
a mean absorption cross-section of an ice crystal, σa , which can be obtained from

I
σa = σa,i /I , (1.8.17)
i=1
9
10 min
20 min
40 min
8 60 min
120 min
Height (km)
5
0 2 4 6 8 10
Ice Water Content (mg m-3)
Figure 1.25 Vertical profile of the horizontally averaged IWC simulated from a 2-D cirrus cloud
formation model at five time steps (after Gu and Liou 2000).
where I is the total number of spectral bands in the parameterization and σ a,i is the
absorption cross-section of an ice crystal for a given wavelength, determined from radiation
parameterization. The radiative flux within a volume element of cloud can be obtained from
the radiation scheme. In this manner, the ice crystal size distribution is fully interactive with
radiation via the ice crystal mean effective size and IWC. The shapes of ice crystals were
assumed to be hexagonal solid columns and plates; other habits depicted in Figure 1.19
were not accounted for. Also, the 2-D cirrus cloud model did not account for collision and
coalescence events, which are critical for the formation of precipitation.
Lastly, we present in Figure 1.25 the vertical profile of IWC, the horizontally averaged
IWC, during a 120 min cirrus cloud simulation. Ice crystals initially form by deposition
nucleation associated with saturated air produced by dynamic and thermodynamic forcings.
At t = 10 min, IWC increases from 0 to about 5.7 mg m−3 . The nucleation process stops
when the supersaturation ratio begins to decrease due to deposition. Ice crystals are then
subject to diffusion growth, advection, sedimentation, and eddy diffusion. Ice crystals
become larger and fall to lower levels. At t = 20 min, IWC reaches its maximum value
of about 9 mg m−3 , and the location of its maximum shifts from 7.55 km at t = 10 min to
7.15 km at t = 20 min. At the same time, the cloud base is moving downward. After ice
crystals descend into the drier lower region, sublimation occurs and the simulated maximum
IWC decreases to about 4.2 mg m−3 at t = 40 min. After t = 60 min, a near steady state
is reached. Observations of cirrus IWC in weak synoptic forcing situations have been
reported to be between 3 and 8 mg m−3 over a horizontal path longer than 3 km and with
a temperature range of −30 to −40°C (Heymsfield 1975). The model results thus appear
to compare reasonably with the previous observations. The updrafts in the cloud layer are
larger than the downdrafts in the earlier stage of the simulation, with maximum values
of about 30 and 5 cm s−1 , respectively, at t = 20 min. Toward the end of the simulation,
updrafts and downdrafts are of the same magnitude, with a value of about 15 cm s−1 . The
simulation results are comparable to the observed vertical velocities in cirrostratus, with
typical values ranging from 5 to 20 cm s−1 and a maximum of about 60 cm s−1 (Heymsfield
1977).
In summary, in the development of a numerical model for the formation and evolution of
cirrus clouds, we have shown that ice microphysical properties in terms of ice nucleation,
diffusional growth, and terminal velocity, which are related to ice crystal shape and size,
become significant. Moreover, in order to evaluate cloud radiative heating and the heating
of individual ice particles, the basic scattering and absorption data for various ice crystal
sizes and shapes are required for accurate calculations of the transfer of solar and thermal
infrared radiation in association with model development.
2
Fundamentals of light scattering by ice crystals
In this chapter we present a number of fundamental subjects related to light scattering by

ice crystals. In Section 2.1, we discuss the scope of light scattering by ice crystals from
the perspectives of the independent scattering concept, the same light frequency before and
after scattering events, and the coupled absorption and induced polarization. Section 2.2
covers the fundamental Maxwell’s equations, leading to vector wave equations and required
boundary conditions for solutions. We show that exact analytic solutions of vector wave
equations exist only for spherical, circular cylindrical, and spheroidal coordinates. The
optical properties of ice are subsequently introduced in Section 2.3, followed by a discussion
in Section 2.4 of the definition of the single-scattering and polarization properties of non-
spherical ice crystals. Lastly, in Section 2.5 we discuss the link between the single-scattering
properties of ice crystals, deduced from the independent scattering concept, and the transfer
of radiation, which includes multiple scattering and emission in clouds containing an
ensemble of ice crystals.
2.1 Introductory remarks

In Chapter 1, we presented the formation of ice crystals in the atmosphere and the complexity
of their structure, shape, and size. We showed that the shape and size of an ice crystal are
primarily governed by temperature and supersaturation, but that they generally have a
basic hexagonal structure. In the atmosphere, if ice crystal growth involves collision and
coalescence, the crystal’s shape can be extremely complex. Observations based on aircraft
optical probes and replicator techniques for widespread midlatitude, tropical, Arctic, and
contrail cirrus show that these clouds are composed largely of ice crystals in the shape of
bullet rosettes, solid and hollow columns, plates, and aggregates, various forms of dendrites,
and ice crystals with irregular surfaces. The size of ice particles normally ranges from a few
micrometers to thousands of micrometers. In the presence of sunlight, some of the defined
ice crystals produce fascinating optical phenomena, including 22◦ and 46◦ halos, sundogs,
and numerous arcs and bright spots.
The scattering of light is a physical process by which a particle in the path of an
electromagnetic wave continuously abstracts energy from the incident wave and reradiates
58
2.1 Introductory remarks 59
that energy in all directions. Therefore, a particle may be thought of as a point source of
scattered energy. Scattering is a fundamental physical process associated with light and its
interaction with matter. In the presentation of light scattering by ice crystals, the following
three limitations apply.
First, in atmospheric scattering, it has normally been assumed that the light scattered
by molecules and particulates has the same frequency (or wavelength or wavenumber) as
the incident light. Shift frequencies produced by high-energy laser light such as Raman
scattering will not be within the purview of the present discussion.
Second, as illustrated in Chapter 1, the number density of ice crystals in clouds is
relatively small and ice crystals are separated by distances which are much larger than their
sizes, such that each would scatter light in exactly the same way if all other particles did not
exist. This is referred to as independent scattering. The independent scattering assumption
greatly simplifies the problem of light scattering involving a spectrum of particles within
the context of a “sufficiently small” volume so that the single-scattering properties of ice
crystals (as well as water droplets and aerosols) can be defined for application to radiative
transfer calculations. The concept of independent scattering allows for the use of energy
quantities instead of electric fields in the analysis of the propagation of electromagnetic
waves in planetary atmospheres.
Third, in a scattering volume which contains many particles, each particle is exposed to,
and also scatters, the light which has already been scattered by other particles. Scattering
more than once is referred to as multiple scattering. In principle, in the presence of particles,
light can undergo an infinite number of scattering events. In connection with the concept
of independent scattering discussed above, we will cover multiple scattering in ice crystal
clouds to the extent of introducing random and horizontal orientations of ice crystals. The
link between light scattering (single scattering events) and multiple scattering (radiative
transfer) will be discussed.
Scattering is frequently accompanied by absorption. Grass looks green because it scatters
green light while it absorbs red and blue light. The absorbed energy is converted into some
other form, and it is no longer present as red or blue light. Both scattering and absorption
remove energy from a beam of light traversing a medium. The beam of light is said to be
attenuated, and this attenuation is referred to as extinction. Thus, extinction is a result of
scattering plus absorption. In a non-absorbing medium, scattering is the sole process of
extinction. Absorption of sunlight by ice crystals is generally negligible at ultraviolet (UV)
and visible wavelengths. However, substantial absorption occurs at infrared and microwave
wavelengths.
All scattering processes lead to polarization in association with the propagation of
light beams. A light beam is characterized by certain polarization configurations that are
defined by the vibration of the electric vector and by the phase difference between the two
components of this vector. The polarization of a light beam cannot be visualized even in
the visible spectrum and requires a specific instrument for its detection. In Section 2.2, we
shall introduce Maxwell’s equations, wave equations, and boundary conditions, basic to the
discussion of the propagation of electromagnetic waves and their interaction with matter.
60 Fundamentals of light scattering by ice crystals
2.2 Maxwell’s equations, wave equations, and boundary conditions

2.2.1 Maxwell’s equations
The state of excitation established in space by the presence of electric charges constitutes
an electromagnetic field, which is governed by the fundamental Maxwell’s equations. The
field is represented by the electric vector E and the magnetic induction vector B. Addi-
tionally, a second set of vectors involving the electric current density j, the electric
displacement D, and the magnetic vector H must be introduced in order to describe
the effect of the electromagnetic field on material objects. Subject to the condition that
the physical properties of the medium are continuous at every point in its neighborhood, the
space and time derivatives of these five vectors are related by the following four equations
(Maxwell 1865):
1 ∂D 4π
∇×H= + j, (2.2.1a)
c ∂t c
1 ∂B
∇×E=− , (2.2.1b)
c ∂t
∇ · D = 4πρ, (2.2.1c)
∇ · B = 0, (2.2.1d)
where t denotes time and c is the velocity of light in vacuum, which is approximately
equal to 3 × 1010 cm s−1 . Equation (2.2.1c) can be considered as a defining equation for
the electric charge density ρ, while Eq. (2.2.1d) implies that no free magnetic poles exist.
Following Born and Wolf (1975), we use the Gaussian system of units in this presentation.
Since div curl (∇ · ∇×) of any vector is equal to 0, e.g., ∇ · ∇ × H = 0, a div (dot
product) operation on Eq. (2.2.1a) leads to
1 ∂D
∇·j=− ∇· . (2.2.2)
4π ∂t
Differentiating Eq. (2.2.1c) with respect to t leads to
∂ρ
+ ∇ · j = 0. (2.2.3)
∂t
This equation represents the equation of continuity in an electromagnetic field where charge
is conserved in the surroundings of any point.
In order to uniquely determine the field vectors from a given distribution of currents
and charges, the preceding equations must be augmented by relationships describing the
behavior of substances under the influence of the field. The electron conduction response
to the electric field and the two constitutive relations describing the aggregate response of
bond charges to the electric field and corresponding magnetic induction can be defined by
the following three equations:
j = σ E, D = εE, B = μH, (2.2.4)
2.2 Maxwell’s equations and boundary conditions 61
where σ is referred to as specific (or electrical) conductivity, ε is the permittivity, and

µ is the magnetic permeability. Substances for which σ
= 0 are called conductors. For
example, metals are excellent conductors in which conductivity decreases with increasing
temperature. However, in some classes of materials, known as semiconductors, conductivity
over a wide range increases with temperature. Substances for which σ is negligibly small are
referred to as dielectrics. In this case, their electric and magnetic properties are completely
determined by ε and µ. For most substances (air, for example), the magnetic permeability
µ is practically unity.
Maxwell’s equations connect the field vectors through simultaneous differential equa-
tions. To obtain individual differential equations for the vectors, we shall confine our
attention to a part of the field where there are no charges (ρ = 0) or currents (j = 0), and to
a medium which is homogeneous, so that ε and µ are constants. Subject to these conditions,
we obtain the so-called source-free Maxwell equations:
ε ∂E
∇×H= , (2.2.5a)
c ∂t
μ ∂H
∇×E=− , (2.2.5b)
c ∂t
∇ · E = 0, (2.2.5c)
∇ · H = 0. (2.2.5d)
In the following discussion, we will use Eqs. (2.2.5a)–(2.2.5d) to derive the electromagnetic
wave equation. Note that Eqs. (2.2.5c) and (2.2.5d) can be directly obtained by performing
a dot product operation on Eqs. (2.2.5a) and (2.2.5b).
Consider now a plane electromagnetic wave in a periodic field defined by a circular
frequency ω such that
E → Eeiωt , H → Heiωt , (2.2.6)
√
where i = −1 and we have used exp(iωt) in the derivation in order to have a negative
value for the imaginary part mi of the refractive index (see Subsection 2.3.1). Using these
transformations, Eqs. (2.2.5a) and (2.2.5b) can then be written as
∇ × H = ikm2 E, (2.2.7a)
∇ × E = −ikH, (2.2.7b)
where k = 2π/λ(= ω/c) is the wavenumber denoting the propagation constant in vacuum,
√
ω is the circular frequency, λ is the wavelength in vacuum, m = ε is the complex
refractive index of the medium at the frequency ω, and μ(∼ 1) is the permeability of air
noted previously.
By performing the curl operation on Eq. (2.2.7b), we obtain
∇ × ∇ × E = −ik∇ × H. (2.2.8)
(a) n2 n (b)
A t2
e2 n
Δh t
Δh ΔA
nA S
e1 t1
e2 S Δs
e1 n1
Figure 2.1 Illustration of the derivation of (a) normal boundary conditions for the magnetic induction
vector B and the electrical displacement vector D and (b) tangential boundary conditions for the
electric vector E and the magnetic vector H. All notations are defined in the text.
Since ∇ × ∇ × E = ∇(∇ · E) − ∇ 2 E and ∇ · E = 0, we can substitute ∇ × H from

Eq. (2.2.7a) to obtain the following equation:
∇ 2 E = −k 2 m2 E. (2.2.9)
Likewise, we can derive the following relationship from Eqs. (2.2.7a) and (2.2.5d) for the
magnetic vector H:
∇ 2 H = −k 2 m2 H. (2.2.10)
The preceding two equations illustrate that the electric vector and the magnetic induction
in a homogeneous medium satisfy the following vector wave equation:
∇ 2 A + k 2 m2 A = 0, (2.2.11)
where the vector A can be either E or H. In Subsection 2.2.2, we shall introduce the
boundary conditions that are required for solving the vector wave equation.
2.2.2 Boundary conditions

Across any surface separating two media, shape changes may occur in the parameters
characterizing the media so that the field vectors are expected to undergo corresponding
changes. Under the conditions of no charges (ρ = 0), no currents (σ = 0), and magnetic
permeability µ 1, the two media are then defined only by the permittivity, ε1 and ε2 , or
by the complex indices of refraction, m1 and m2 .
2.2.2.1 Normal components

Solution of the vector wave equations requires suitable boundary conditions (Born and
Wolf 1975). The boundary surface separating the two media is denoted by the symbol S
and the associated normal vector n, as shown in Figure 2.1a. Consider an infinitesimally
thin transition layer such that the parameters change rapidly but continuously from their
values near S in (1) to their counterparts in (2). Within this transition layer as well as in (1)
and (2), the field vectors and their first derivatives are assumed to be continuous bounded
functions of position and time. We may construct a small cylinder of height h and cross-
sectional area A, and let the directions of the surface normal unit vectors, n1 and n2 , at the
bottom and the top of the cylinder, be as indicated in Figure 2.1a. Integrating the divergence
equations (2.2.1c) and (2.2.1d) over the volume V = h A of the cylinder and using
Gauss’ divergence theorem, we have

∇ · B dV = B · dA = 0,
V
(2.2.12)
∇ · D dV = D · dA = 0,
V
where the integrals over A are over the closed cylinder surface. To evaluate the surface
integrals, we assume that the cylinder height h → 0 so that the contribution of the walls
may be neglected. In this manner, we may write

B · dA = B1 · n1 dA + B2 · n2 dA = 0,
A A
(2.2.13)
D · dA = D1 · n1 dA + D2 · n2 dA = 0.
A A
Since n2 = −n1 = n as shown in Figure 2.1a, it follows that

(B2 − B1 ) · ndA = 0, (D2 − D1 ) · ndA = 0. (2.2.14)
A A
Assuming that the field vectors are constant within an infinitesimal area element A,
we can then obtain the boundary conditions for the normal components of B and D as
follows:
(B2 − B1 ) · n = 0, (D2 − D1 ) · n = 0. (2.2.15)

These equations state that the normal components of B and D across any discontinuous
surface are continuous.
2.2.2.2 Tangential components

To express the behavior of the tangential components of B and D at the surface S, we shall
consider a closed surface A and the associated normal vector nA , as depicted in Figure 2.1b.
The tangential unit vectors along the lower and upper boundaries of A are denoted by t1
and t2 . We can perform the integration of Eqs. (2.2.1a) and (2.2.1b) (j = 0) and use the
Stokes integral theorem to obtain the following two equations:

1 ∂B
∇ × E · nA dA = E · ds = − · nA dA,
A A c ∂t
(2.2.16)
1 ∂D
∇ × H · nA dA = H · ds = · nA dA,
A A c ∂t
where the first surface integrals have been replaced by closed line integrals.
The line integrals can be decomposed into two parts in each equation. In the limit
h → 0, contributions from the corresponding sides vanish. Thus, we have

1 ∂B
E1 · t1 ds + E2 · t2 ds = − · nA dA,
s s A c ∂t
(2.2.17)
1 ∂D
H1 · t1 ds + H2 · t2 ds = · nA dA.
s s A c ∂t
For a small surface A, we have t2 = −t1 = nA × n, so that

1 ∂B
(E2 − E1 ) · nA × nds = − · nA dA,
s A c ∂t
(2.2.18)
1 ∂D
(H2 − H1 ) · nA × nds = · nA dA.
s A c ∂t
For small values of A, all integrands can be considered to be constant. It follows that
1 ∂B
(E2 − E1 ) · nA × n = − · nA h,
c ∂t
(2.2.19)
1 ∂D
(H2 − H1 ) · nA × n = · nA h.
c ∂t
The right-hand sides of these equations vanish in the limit of h → 0, because the field
vectors and their derivatives with respect to time remain bounded. Finally, we obtain the
results
n × (E2 − E1 ) = 0, n × (H2 − H1 ) = 0, (2.2.20)
which state that the tangential components of the field vectors are continuous across a
discontinuous surface S. Equations (2.2.15) and (2.2.20) are necessary for the solution of
the differential vector wave equations denoted in Eqs. (2.2.9) and (2.2.10).
2.2.3 Wave equations and some analytic solutions

In light scattering problems, the solution of the vector wave equation denoted in Eq. (2.2.1)
subject to appropriate boundary conditions can be obtained only if a suitable coordinate
system can be defined to conform the particle geometry. In the following subsections, we
present solutions to the vector wave equation for the cases of a sphere, a cylinder, and
a spheroid, in which spherical, cylindrical, and spheroidal coordinates can be imposed,
respectively.
2.2.3.1 Spherical coordinates

We fully realize that the detailed solution for light scattering by a sphere has been presented
in historical papers by Lorenz (1890), Mie (1908), and Debye (1909), and is now referred
to as the Lorenz–Mie or simply the Mie theory. However, we wish to capture some new
ar
z
af
r P(r, q, f)
z
θ
y
φ aq
x
y
x
Figure 2.2 Definitions of the parameters in spherical coordinates, where x = r sin θ cos φ, y =
r sin θ sin φ, z = r cos θ ; and ar , aθ , and aφ are orthogonal vectors with P a point on the sphere.
developments to introduce the manner in which the exact solution of the vector wave
equation can be derived from spherical coordinates. We shall largely follow the formulation
developed in Liou (1977, 2002). Referring to Eq. (2.2.11), we may write the scalar wave
equation in the form
∇ 2 ψ + k 2 m2 ψ = 0, (2.2.21)
where ψ is a scalar which satisfies the scalar wave equation. Consider the vectors Mψ and
Nψ in spherical coordinates (r, θ , φ) (Figure 2.2), defined by the following two equations:

∂ 1 ∂ 1 ∂
Mψ = ∇ × [ar (rψ)] = ar + aθ + aφ × [ar (rψ)]
∂r r ∂θ r sin θ ∂φ
(2.2.22)
1 ∂(rψ) 1 ∂(rψ)
= aφ − aφ ,
r sin θ ∂φ r ∂θ
mkNψ = ∇ × Mψ

2 (2.2.23)
∂ (rψ) 1 ∂ 2 (rψ) 1 ∂ 2 (rψ)
= 2
+ m 2 2
k (rψ) ar + aθ + aφ .
∂r r ∂r∂θ r sin θ ∂r∂φ
These equations satisfy the vector wave equation defined in Eq. (2.2.11) subject to
Eq. (2.2.21). The terms ar , aθ , and aφ are unit vectors in spherical coordinates.
Let u and v be two independent solutions of the scalar wave equation defined in
Eq. (2.2.21). The electric and magnetic field vectors can then be expressed as
E = Mv + iNu , (2.2.24)
H = m( −Mu + iNv ). (2.2.25)
These two equations will satisfy Eqs. (2.2.9) and (2.2.10). Using Eqs. (2.2.22) and (2.2.23),
E and H can then be written explicitly in spherical coordinates as follows:

i ∂ 2 (ru) 1 ∂(rv) i ∂ 2 (ru)
E= + m 2 2
k (ru) ar + + aθ
mk ∂r 2 r sin θ ∂φ mkr ∂r∂θ

(2.2.26)
1 ∂(rv) 1 ∂ 2 (ru)
+ − + aφ ,
r ∂θ mkr sin θ ∂r∂φ

i ∂ 2 (rv) m ∂(ru) i ∂ 2 (rv)
H= + m k (rv) ar + −
2 2
+ aθ
k ∂r 2 r sin θ ∂φ kr ∂r∂θ

(2.2.27)
m ∂(ru) i ∂ 2 (rv)
+ + aφ .
r ∂θ kr sin θ ∂r∂φ
We now return to the scalar wave equation defined in Eq. (2.2.21), which can be expressed
in spherical coordinates as follows:

1 ∂ 2 ∂ψ 1 ∂ ∂ψ 1 ∂ 2ψ
r + sin θ + + k 2 m2 ψ = 0. (2.2.28)
r 2 ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin θ ∂φ 2
We now define the solution of this equation in terms of three independent variables in
the form
ψ(r, θ, φ) = R(r)(θ )(φ). (2.2.29)
Upon substituting Eq. (2.2.29) into Eq. (2.2.28) and dividing the entire equation by
ψ(r, θ , φ), we obtain

1 1 ∂ 2 ∂R 1 1 ∂ ∂ 1 1 ∂ 2
r + sin θ + + k 2 m2 = 0.
r 2 R ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂φ 2
(2.2.30)
Subsequently, we may multiply Eq. (2.2.30) by r2 sin2 θ to obtain

2 1 ∂ 2 ∂R 1 ∂ ∂ 1 ∂ 2
sin θ r + sin θ sin θ + k m r sin θ +
2 2 2 2
= 0.
R ∂r ∂r ∂θ ∂θ ∂φ 2
(2.2.31)
Since the first three terms in this equation consist of only the variables r and θ , but not φ,
Eq. (2.2.31) can be valid only under the condition that
1 d 2
= const = −2 , (2.2.32)
dφ 2
where, for mathematical convenience, the constant has been chosen to be −2 , and denotes
an integer. Combining Eqs. (2.2.31) and (2.2.32), we then have

2 1 ∂ 2 ∂R 1 ∂
sin θ r + sin θ sin θ + k 2 m2 r 2 sin2 θ − 2 = 0. (2.2.33)
R ∂r ∂r ∂θ
Upon dividing Eq. (2.2.33) by sin2 θ , we obtain the following equation:

1 ∂ ∂R 1 1 ∂ ∂ 2
r2 + k 2 m2 r 2 + sin θ − = 0. (2.2.34)
R ∂r ∂r sin θ ∂θ ∂θ sin2 θ
This equation consists of two independent variables, namely R and . Thus, we must have

1 d dR
r2 + k 2 m2 r 2 = const = n(n + 1), (2.2.35)
R dr dr

1 1 d d 2
sin θ − = const = −n(n + 1) (2.2.36)
sin θ dθ dθ sin2 θ
to satisfy Eq. (2.2.34), where n is an integer. Again, the selection of the constant
−n(n + 1) is for mathematical convenience. After rearranging Eqs. (2.2.32), (2.2.35),
and (2.2.36), we obtain the following three equations:

d 2 (rR) n(n + 1)
+ k m −
2 2
(rR) = 0, (2.2.37)
dr 2 r2

1 d d 2
sin θ + n(n + 1) − = 0, (2.2.38)
sin θ dθ dθ sin2 θ
d 2
+ 2 = 0. (2.2.39)
dφ 2
The single-value solution of Eq. (2.2.39), a second-order differential equation, is given by
= a cos φ + b sin φ, (2.2.40)
where a and b are arbitrary constants. Furthermore, Eq. (2.2.38) is the well-known
equation for spherical harmonics. For mathematical convenience, we may introduce a new
variable µ = cos θ so that Eq. (2.2.38) can be rewritten in the form

d 2 d 2
(1 − μ ) + n(n + 1) − = 0. (2.2.41)
dμ dμ 1 − μ2
The solutions to Eq. (2.2.41) can be expressed using the associated Legendre polynomials,
which are the spherical harmonics of the first kind, in the form
= pn (μ) = pn ( cos θ ). (2.2.42)
Finally, in order to solve the remaining Eq. (2.2.37), we set

√
kmr = ρ, R = (1/ ρ)Z(ρ), (2.2.43)
to obtain the following equation:

d 2Z 1 dZ (n + 1/2)2
+ + 1− Z = 0. (2.2.44)
dρ 2 ρ dρ ρ2
The solution of Eq. (2.2.44) can be expressed as the general cylindrical function of order
n + ½ in the form
Z = Zn+1/2 (ρ). (2.2.45)
It follows that the solution of Eq. (2.2.37) can be expressed as
√
R (kmr) = Zn+1/2 (kmr)/ kmr. (2.2.46)
Combining Eqs. (2.2.40), (2.2.42), and (2.2.46), the elementary wave functions at all points
on the surface of a sphere can now be expressed as
√
ψ(r, θ, φ) = Zn+1/2 (kmr)pn ( cos θ )(a cos φ + b sin φ)/ kmr. (2.2.47)
Each cylindrical function denoted in Eq. (2.2.46) can be expressed as a linear combina-
tion of two cylindrical functions of standard type in terms of the Bessel functions Jn+½ (ρ)
and Nn+½ (ρ). We may define two standard functions as follows:

ψn (ρ) = πρ/2Jn+1/2 (ρ), χn (ρ) = − πρ/2Nn+1/2 (ρ). (2.2.48)
The functions ψn are regular in every finite domain of the ρ plane, including the origin,
whereas the functions χn cannot represent waves inside the sphere. On the basis of the
definitions in Eq. (2.2.48), Eq. (2.2.46) can be rewritten in the form
rR (kmr) = cn ψn (kmr) + dn χn (kmr), (2.2.49)
where cn and dn are arbitrary constants. Equation (2.2.49) represents the general solution
of Eq. (2.2.37). It follows that the general solution of the scalar wave equation (2.2.21) can
be expressed as
∞
n
rψ(r, θ, φ) = pn (cos θ )[cn ψn (kmr) + dn χn (kmr)](a cos φ + b sin φ).
n=0 =−n
(2.2.50)
Subsequently, the electric and magnetic field vectors of electromagnetic waves can be
derived from Eqs. (2.2.26) and (2.2.27). In the case when cn = 1 and dn = i, we have

ψn (ρ) + iχn (ρ) = πρ/2H (2)1 (ρ) = ξn (ρ), (2.2.51)
n+ /2
(2)
where Hn+ 1/2 is the half-integral-order Hankel function of the second kind. It has the
property of vanishing at infinity in the complex plane, suitable for the representation of
scattered waves.
The complete solution for light scattering of a plane wave by a homogeneous sphere
requires the coordinate transformation of electric vectors, matching the incident, scattered,
and transmitted waves at spherical boundaries, and mapping of the solution to the far field,
leading to the well-known Lorenz–Mie theory.
2.2.3.2 Cylindrical coordinates
The exact solution for the scattering by an infinite homogeneous dielectric circular cylinder
for normal incidence – that is, a light beam perpendicular to the axis of the cylinder – was
az
z
af
P(r, f, z)
ar
z
y
x φ r
y
x
Figure 2.3 Definition of the parameters in cylindrical coordinates, where x = r cos φ, y = r sin φ,
z = z; and ar , aφ , and az are orthogonal vectors with P a point on the cylinder.
derived by Rayleigh (1918). The solution for arbitrary oblique incidence was solved by
Wait (1955). Subsequent numerical investigations concerning the scattering of light beams
by infinite circular cylinders have been carried out by van de Hulst (1957), Greenberg et al.
(1967), Kerker (1969), and Liou (1972a). Liou (1972b) used circular cylinders to model
light scattering by randomly oriented ice particles in the atmosphere.
In the cylindrical coordinate system shown in Figure 2.3, Mψ defined in Eq. (2.2.22)
can be written in the form
1 ∂ψ ∂ψ
Mψ = ∇ × (az ψ) = ar − aφ . (2.2.52)
r ∂φ ∂r
Moreover, using Eq. (2.2.23), we have
mkNψ = ∇ × Mψ

∂ ∂ψ ∂ 1 ∂ψ
=− − ar + aφ (2.2.53)
∂z ∂r ∂z r ∂φ

1 ∂ ∂ψ 1 ∂ 1 ∂ψ
+ r − − az .
r ∂r ∂r r ∂φ r ∂φ
It follows that

1 ∂ 2ψ 1 ∂ 2ψ 1 ∂ ∂ψ 1 ∂ 2ψ
Nψ = ar + aφ − r 2 + 2 az . (2.2.54)
mk ∂z∂r r ∂z∂φ r ∂r ∂ r r ∂φ 2
On the basis of Eqs. (2.2.24), (2.2.25), (2.2.52), and (2.2.54), the electric and magnetic
vectors can be expressed in the forms

1 ∂v i ∂ 2u
E= + ar
r ∂φ mk ∂z∂r

∂v i ∂ 2u i 1 ∂ ∂u 1 ∂ 2u (2.2.55)
+ − + aφ − r + 2 2 az ,
∂r mkr ∂z∂φ mk r ∂r ∂r r ∂φ

m ∂u i ∂ 2v
H= − + ar
r ∂φ k ∂z∂r

∂u i ∂ 2v i 1 ∂ ∂v 1 ∂ 2v
+ m + aφ − r + 2 2 az . (2.2.56)
∂r kr ∂z∂φ k r ∂r ∂r r ∂φ
The scalar wave equation in cylindrical coordinates is given by

1 ∂ ∂ψ 1 ∂ 2ψ ∂ 2ψ
r + 2 2
+ + m2 k 2 ψ = 0. (2.2.57)
r ∂r ∂r r ∂φ ∂z2
This equation is separable by letting
ψ(r, φ, z) = R(r)(φ)Z(z). (2.2.58)
On substituting Eq. (2.2.58) into Eq. (2.2.57) and dividing the entire equation by ψ(r, φ,
z), we obtain

1 d dR(r) 1 d 2 (φ) 1 d 2 Z(z)
r + 2 + + m2 k 2 = 0. (2.2.59)
rR(r) dr dr r (φ) dφ 2 Z(z) dz2
Since the first two terms in Eq. (2.2.59) contain variables r and φ, the only possible solution
of this equation is
1 d 2 Z(z)
+ m2 k 2 = c1 , (2.2.60a)
Z(z) dz2
where c1 is a certain constant. Rearranging this equation, we have
d 2 Z(z)
+ (m2 k 2 − c1 )Z(z) = 0, (2.2.60b)
dz2
where the term m2 k2 − c1 > 0. The solution of Eq. (2.2.60b) is an exponential function:
Z(z) = e−ihz , (2.2.61)
where we have set the arbitrary constant h = (m2 k2 − c1 )1/2 and used a negative sign in the
exponential function.
In view of Eqs. (2.2.59) and (2.2.60a), we must also have the following equation:

1 d dR(r) 1 d 2 (φ)
r + 2 + c1 = 0. (2.2.62)
rR(r) dr dr r (φ) dφ 2
On multiplying by r2 throughout this equation, we get

r d dR(r) 1 d 2 (φ)
r + c1 r 2 + = 0. (2.2.63)
R(r) dr dr (φ) dφ 2
This equation contains two independent terms, so that we must have
1 d 2 (φ)
= −c2 , (2.2.64)
(φ) dφ 2
where c2 is a certain constant. The solution of this differential equation can be expressed
in terms of an exponential function in the form
√
(φ) = e±i c2 φ
. (2.2.65)
√
We may set c2 equal to an integer n such that the solution is a periodic function in φ with
a period of 2π in the form
(φ) = einφ , n = 0, ±1, ±2, . . . (2.2.66)
On the basis of Eqs. (2.2.63) and (2.2.64) and the definition of c2 = n2 , the first term in
Eq. (2.2.63) can be written as follows:

r d dR(r)
r + c1 r 2 − n2 = 0. (2.2.67a)
R(r) dr dr
After rearrangement, we have

d 2 R(r) 1 dR(r) n2
+ + c1 − 2 R(r) = 0. (2.2.67b)
dr 2 r dr r
Considering a positive c1 so that c1 = j2 and setting jr = x, Eq. (2.2.67b) can be written in
the form

d 2 R(x) 1 dR(x) n2
+ + 1 − R(x) = 0. (2.2.67c)
dx 2 x dx x2
The solution of Eq. (2.2.67c) is any Bessel function of order n, denoted by Zn (x), as
follows:
R(r) = Zn (j r), (2.2.68)
where j = (m2 k2 − h2 )12 . By combining Eqs. (2.2.61), (2.2.66), and (2.2.68), the general
solution of the scalar wave equation in cylindrical coordinates can be written as follows:
ψ(r, φ, z) = Zn (j r)einφ e−ihz . (2.2.69)
To find the complete solution for light scattering of a plane wave by an infinite circular
cylinder, we may follow the procedures outlined for the spherical case, including coordinate
transformation of the electric vectors; matching of the incident, scattered, and transmitted
waves at the cylindrical boundaries; and mapping of the solution to the far field. Separa-
tion of the TE and TM modes in the internal and scattered waves corresponding to each
η=
η=0
os π_
3
- co 3
ξ > ξ0
=c
s
π_
η
η (I)
= π_
-c ξ0 s 6
os _π co
6 η=
(II) ξ < ξ0
y z
z
F O F η =1
η = -1
1
se de
Ca Mo
TE
(i) H
(i) E
2
se ode
(i) H Ca M OF = OF =
TM
(i) E
Figure 2.4 Coordinate system (η, ξ, φ) for the scattering by a prolate spheroid with a semifocal
distance . The z-axis is chosen as the axis of revolution. The incident plane contains the incident
direction and the z-axis. The x-axis is along the incident plane. For the transverse magnetic (TM)
mode, the E vector is at the incident plane, while for the transverse electric (TE) mode, the H vector is
along the incident plane. The incident angle ζ is the angle in the incident plane between the incident
direction and the z-axis (after Asano and Yamamoto 1975).
polarization mode of the incident wave is required to determine unknown coefficients in

the wave functions (see e.g., Liou 1972a).
2.2.3.3 Spheroidal coordinates

An exact scattering solution, similar to the Lorenz–Mie theory of light scattering by spheres
and the Rayleigh–Wait theory of light scattering by cylinders, can be derived for particles
with spheroidal shapes (Asano and Yamamoto 1975, Asano and Sato 1980). Prolate and
oblate spheroids may be used to approximate ice cylinders and plates, respectively. The
spheroidal coordinates can be obtained by the rotation of an ellipse about an axis of sym-
metry. Using this rotation, the spheroidal coordinate systems are defined by the following
parameters: η, the angular coordinate; ξ , the radial distance; and φ, the azimuthal angle.
Shown in Figure 2.4 are prolate spheroidal coordinates, which can be related to Cartesian
coordinates by the following transformation:
x = (1 − η2 )1/2 (ξ 2 − 1)1/2 cos φ,

y = (1 − η2 )1/2 (ξ 2 − 1)1/2 sin φ, (2.2.70)
z = ηξ,
where is the semifocal distance, −1 ≤ η ≤ 1, 1 ≤ ξ ≤ ∞, and 0 ≤ φ ≤ 2π. For the

oblate system, the ξ term in x and y should be replaced by (ξ 2 + 1)1/2 , with 0 ≤ ξ ≤ ∞.
The size and shape of an ellipse are defined by the semifocal distance and the eccentricity e.
1/2
For the prolate and oblate systems, e = 1/ξ0 and e = 1/(ξ02 + 1) , respectively, where ξ 0
is the value of ξ at the surface. If = 0, the spheroidal coordinates reduce to the spherical
coordinates. In the far field, ξ → , η → cos θ , and ξ → r.
If ψ satisfies the scalar wave equation defined in Eq. (2.2.21), vectors associated with
electric and magnetic fields can be expressed in the spheroidal coordinate system such that
they satisfy the vector wave equations denoted by ψmn , and the solutions to the vector wave
equation can be obtained from the following two equations:
Mmn = ∇ × (a · ψmn ), (2.2.71)
mkNmn = ∇ × Mmn , (2.2.72)
where m (not the subscript index) is the complex refractive index and k = 2πλ, where
λ is the wavelength. The solution of the scalar wave equation can be expressed in terms
of scalar spheroidal wave functions. The vector a is an arbitrary constant unit vector
(a = e) or the position vector (a = r). The vectors defined in Eqs. (2.2.71) and (2.2.72)
are solenoidal, such that the field vectors E and H can be expressed in terms of these two
vectors in spheroidal coordinates.
The scalar wave equation (2.2.21) is separable in the spheroidal coordinate system in
terms of three second-order ordinary linear differential equations based on three variables:
η, ξ , and φ. The solutions of the equation in φ are eimφ , cos mφ, and sin mφ, where m =
0, 1, 2, . . .
The variable η satisfies the following equation:

d 2 dSmn (η) m2
(1 − η ) + λmn ∓ c η −
2 2
Smn (η) = 0, (2.2.73)
dη dη 1 − η2
where c = · mk, m is the complex refractive index, k is the wavenumber, Smn is the
spheroidal angular function of order m and degree n of the first kind, which can be defined
in the range −1 ≤ η ≤ 1, and λmn is a separation constant, which is a function of c.
The eigenvalues λmn (c) and λmn (−ic), and the associated eigenfunctions Smn (c, η) and
Smn (−ic, η), correspond to the prolate and oblate coordinate systems, respectively. The
eigenfunctions can be expressed in terms of the associated Legendre functions of the first
kind as follows:
∞

mn
Smn (±c, η) = dr (±c)Pm+rm
(η), (2.2.74)
r=0,1
where we have defined c = +c and −c = −ic, and drmn (±c) are the expansion coefficients
related to the prolate (+c) and oblate (−c) systems. The prime on the summary symbol
signifies that the summation is over only even (odd) values when n − m is even (odd). It
follows that the spheroidal angle functions are dependent not only on the angular component
but also on the properties of the medium c. Note that, for spherical geometry, the angular
functions reduce to Pnm (η), independent of c [see Eq. (2.2.42)]. Based on the theory of
Sturm–Liouville differential equations, the spheroidal angle function has the following
orthogonal properties:
1
0 (n
= n ),
Smn (η)Smn (η)dη = (2.2.75)
−1 mn (n = n ),
where the normalization constant

∞

2(r + 2m)! mn 2
mn = d . (2.2.76)
r=0,1
(2r + 2m + 1)r! r
The radical functions Rmn (ξ ) satisfy the following differential equation:

d dRmn (ξ ) m2
(ξ 2 ∓ 1) − λmn − c2 ξ 2 ± 2 Rmn (ξ ) = 0. (2.2.77)
dξ dξ (ξ ± 1)
(j )
This equation is normalized such that, for cξ → ∞, four asymptotic forms for Rmn (j = 1,
2, 3, 4) can be obtained.
The appropriate expansions in terms of spherical Bessel functions for the prolate (+c)
and oblate (−ic) systems can be expressed in the form

(r + 2m)!

Rmn (±c, ξ ) = 1
(j ) mn
dr (±c)
r=0,1
r!
m/2
ξ2 ∓ 1 (r+n−m) (r + 2m)! mn (j )
× i dr (±c) Zm+r (±cξ ) , (2.2.78)
ξ2 r=0,1
r!
where we have again replaced c by +c (prolate) and −ic (oblate) by −c, and ξ = iξ for
(j )
oblate, for simplicity of presentation. The notation Zn (cξ ) is the nth-order spherical Bessel
function (i = 1), Newmann (j = 2) and Hankel functions of the first kind (j = 3) and the
second kind (j = 4). Because of the properties of the spherical Bessel functions, Rmn (1)
is
used to represent the wave inside the spheroid and the incident wave. At large distances
(3)
from the spheroid, Rmn is suitable for representing the scattered wave.
On the basis of the preceding solutions for the three independent variables representing
spheroidal coordinates, complete solutions for the scalar wave can now be written as
follows. For the prolate system, we have

e cos mφ
(j )
ψmn (c, η, ξ, φ) = Smn (c, η)Rmn (c, ξ )
(j )
. (2.2.79)
o sin mφ
For the oblate system, the solutions can be obtained by replacing c by −ic and ξ by iξ .
The notations e and o refer to even and odd dependence on φ; i.e., cos mφ and sin mφ,
respectively. The superscript j will take the values 1 or 3 for the incident and internal waves
and for the scattered wave, respectively.
Finally, the field vectors E and H can now be expressed in terms of the vector spheroidal
wave functions based on Eqs. (2.2.71) and (2.2.72). These functions can be written as

r e r e
Mmn and Nmn , (2.2.80)
o o
which can be obtained from the scalar functions in Eq. (2.2.79) for the two types of spheroids
(prolate and oblate) by setting a = r as the position vector. The components of spheroidal
vector wave functions can be found in Flammer (1957). The unknown coefficients in the
wave functions must be determined from boundary conditions in which it is necessary to
separate TE from TM modes in the internal and scattered waves corresponding to each
polarization mode of the incident wave (Asano and Yamamoto 1975).
Examples of the single-scattering properties of ice spheres, ice cylinders, and ice
spheroids will be presented after discussion of ice optical properties and numerous defini-
tions given in reference to the scattering phase matrix, extinction efficiency, single-scattering
albedo, and asymmetry factor.
2.3 Optical properties of ice: index of refraction

2.3.1 Complex index of refraction and dispersion of light
Electric dipoles are generated when positive and negative charges are impelled to move in
opposite directions by an applied electric field in a dielectric. The product of the charges
and the separation distance between the positive and negative charges is referred to as the
dipole moment, which when divided by unit volume is denoted as the polarization P. The
displacement vector D (charge per area) in Maxwell’s equations in a dielectric is defined
by
D = εE = E + 4πP. (2.3.1)
The permittivity of the medium is then given by
ε = 1 + 4πP · E/E 2 . (2.3.2)
√
In terms of ε and the permeability μ, the velocity of light can be written as c = 1/με.
The permeability μ of air or water is nearly equal to the permeability μ0 (= 1) of vacuum,
i.e., μ ≈ μ0 . Furthermore, the index of refraction is defined as the ratio of the velocity of
light c0 in vacuum to that in the medium:
√
m = c0 /c ≈ ε = (1 + 4πP · E/E 2 )1/2 . (2.3.3)
The polarization vector for N dipoles can be written in terms of the polarizability α as
follows:
P = N αE. (2.3.4)
Substituting Eq. (2.3.4) into Eq. (2.3.3) leads to
m2 = 1 + 4πN α. (2.3.5)
The polarization vector defined in Eq. (2.3.4) can also be defined in terms of the charge
e of the electron and a vector distance r in the form
P = N er. (2.3.6)
In view of Eqs. (2.3.4) and (2.3.6), we have the following relationship:
αE = er. (2.3.7)
Furthermore, the force generated by electric and magnetic fields is given by the following
Lorentz force equation:
F = e[E + (μ/c)v × H], (2.3.8)
where v denotes the velocity of an electron, which is very small compared to the velocity
of light. Hence, the force produced by the magnetic field may be neglected. The force on a
vibrating system in terms of displacement D = εE is produced by three actions: electron
acceleration; the damping force which carries away energy when vibrating electrons emit
electromagnetic waves proportional to electron velocity; and the restoring force of vibra-
tions, which is proportional to the distance r. From Newton’s second law, we can express
balance forces as follows:
F eE d 2r dr
= = +γ + ξ r, (2.3.9)
m∗e m∗e dt 2 dt
where m∗e is the mass of the electron and γ and ξ are the damping and restoring coefficients,
respectively. In scalar form, the balance equation can be expressed as
d 2r dr eE
2
+γ + ξr = ∗ . (2.3.10)
dt dt me
The homogeneous solution of this second-order differential equation is given by
r = r0 eiωt = r0 ei2πν̃t , (2.3.11)
where r0 denotes the value when r → 0, ω is the circular frequency, and ν̃ is the frequency.
Substituting Eq. (2.3.11) into Eq. (2.3.10), we obtain
[(ξ − 4π2 ν̃ 2 ) + i2πν̃γ ]r = eE/m∗e . (2.3.12)
√
The natural (or resonant) frequency is defined as ν̃0 = ξ /2π. Thus, we have
er e2 1
α= = ∗ 2 2
E me 4π ν̃0 − ν̃ 2 + i2πγ ν̃
(2.3.13)
e2 ν̃02 − ν̃ 2 i γ ν̃
= ∗ − .
me 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2 2π 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2
0 0
Let the real and imaginary parts of the index of refraction (or refractive index) be mr and
mi , respectively, so that the complex index of refraction is defined in the form
m = mr − imi , (2.3.14)
where mr and mi are positive real numbers. From Eqs. (2.3.5) and (2.3.13), we can then
obtain the following equations:
4πN e2 ν̃02 − ν̃ 2
m2r − m2i = 1 + ∗ 2 , (2.3.15)
me 4π2 ν̃ − ν̃ 2 2 + γ 2 ν̃ 2
0
2N e2 γ ν̃
2mr mi = . (2.3.16)
m∗e 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2
0
For air, the real part mr ≈ 1 and the imaginary part mi (mr − 1). Additionally, in the
neighborhood of the resonant frequency, ν̃ ∼ = ν̃0 , so that (ν̃ 2 − ν̃02 ) = (ν̃ + ν̃0 ) · (ν̃ − ν̃0 ) ∼
=
2ν̃0 (ν̃ − ν̃0 ). The half-width of the natural broadening is dependent on the damping and is
given in the form αN = γ /4π, while the line strength S = πN e2 /(m∗e c). It follows that the
real part can be derived and is given in the form
N e2 ν̃ − ν̃0
mr − 1 = − ∗
. (2.3.17)
4πme ν̃0 (ν̃ − ν̃0 )2 + αN2
The absorption coefficient kν̃ is directly related to the imaginary part mi and can be written
as follows:
4πν̃0 mi S αN
kν̃ = = . (2.3.18)
c π (ν̃ − ν̃0 )2 + αN2
Figure 2.5 illustrates (mr − 1) and kν̃ as functions of frequency. The former increases
as the frequency increases when (ν̃0 − αn ) > ν̃. This mode is referred to as normal disper-
sion, under which light is dispersed by a prism into its component colors. For the region
(ν̃0 + αN ) > ν̃ > (ν̃0 − αN ), (mr − 1) decreases with increasing frequency, referred to as
anomalous dispersion. For the range ν̃ > (ν̃0 + αN ), normal dispersion takes place again;
however, (mr − 1) is smaller than unity. The absorption coefficient with the half-width αN
is referred to as the normal or Lorentz profile (Lorentz 1906).
2.3.2 Optical properties of ice

Theoretical computations of the radiative properties of ice clouds require knowledge of
laboratory measurements of the complex refractive index m(ν) = mr (ν) − imi (ν), where ν
is the wavenumber in a vacuum (ν = ν̃0 /c). In terms of wavelength, we have λ = 1/ν. The
real part of the refractive index determines the phase speed of the electromagnetic wave,
while the imaginary part of the refractive index is related to the absorption coefficient kν
defined in Eq. (2.3.18) in the wavenumber domain.
As shown in Eqs. (2.3.2) and (2.3.3), permittivity ε is related to polarization P. In
general, polarization refers to the variation of the electric vector. For example, we refer to
linear polarization to imply that the electric vector vibrates in a line defined in space (see
Subsection 2.4.1). In a related issue, we shall define the velocity of light in a medium. The
velocity of light in an isotropic material is the same in all directions. However, crystals are
mr -1
aN kν
-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
(ν - ν0)/aN
Figure 2.5 Real and imaginary parts of the complex index of refraction as a function of the nor-
malized frequency, where mr is the real refractive index, α N is the half-width, kν̃ is the absorption
coefficient, and (ν̃ − ν̃0 ) is the frequency deviation from the reference frequency ν̃0 .
generally optically anisotropic and a light beam is broken up into two waves which travel
with different velocities through the crystal. One of the two waves, called the ordinary
wave, travels with the same velocity in all directions in the crystal; therefore, its vector
surface is a sphere.
The velocity of the other wave, called the extraordinary wave, varies with the direction
of propagation through the crystal and its vector surface is an ellipsoid of revolution. For
optically uniaxial crystals, the spherical vector surface coincides with the ellipsoidal vector
surface at only two points, and these are at the ends of the axis of revolution of the ellipsoid,
as shown in Figure 2.6. In this case the velocity of the ordinary wave is equal to that of the
extraordinary wave when light travels along the axis of revolution of the ellipsoid, which
is the optical axis of the crystal. If the extraordinary wave is slow relative to the ordinary
wave, the crystal is said to be optically positive (Figure 2.6a). If the reverse is true, the
crystal is said to be optically negative (Figure 2.6b). The ordinary and extraordinary waves
are polarized at right angles to each other. The electric vector of the ordinary wave always
vibrates perpendicular to the optical axis, while that of the extraordinary wave is in the plane
defined by the propagation direction and the optical axis. This phenomenon is referred to
as birefringence.
Ice is a uniaxial, doubly refracting, and optically positive crystal. Its optical axis coincides
with the crystal’s c-axis. Two principal indices of refraction are defined for uniaxial crystals.
One is based on the velocity of the extraordinary wave in a direction normal to the wave
(a) Optic axis (b) Optic axis
Ordinary
wave
Extraordinary
wave
Figure 2.6 Spherical ellipsoidal vector surfaces for an optically uniaxial crystal; the difference
between the surfaces is exaggerated in the two diagrams: (a) denotes the positive crystal, while (b)
denotes the negative crystal.
front when the wave travels perpendicular to the optical axis, and the other is the refractive
index for the ordinary wave. Thus, for a positive uniaxial crystal, the principal index of
refraction me for the extraordinary wave is defined as
me = c/v1 , (2.3.19)
where v1 is the minimum normal velocity of the extraordinary wave. The index of refraction
mo for the ordinary wave is given by
mo = c/v2 , (2.3.20)
where v2 is the velocity of the ordinary wave. The indices of refraction me and mo of ice
differ only in the third decimal place at a temperature of −3°C for visible wavelengths
(Merwin 1930, Hobbs 1974).
According to Born and Wolf (1975), the effective index of refraction meff for extraordi-
nary waves can be obtained from
1 cos2 κ sin2 κ
2
= 2
+ , (2.3.21)
meff mo m2e
where κ is the angle between the c-axis of the crystal and the propagation direction of
the waves. As shown in Takano and Liou (1989a), the effect of ice birefringence on the
single-scattering properties of hexagonal ice crystals is negligible. Thus, for all practical
purposes it suffices to use the average refractive index in atmospheric scattering, absorption,
and polarization calculations.
2.3.3 Indices of refraction for ice

The determination of mr and mi for ice begins with transmission measurements through
thin ice films in a laboratory setting. On the basis of Beer’s law, the transmission of light
through a medium for a given wavenumber ν, neglecting multiple scattering, can be written
as follows:
Iν (s)
Tν = = exp(−kν · ), (2.3.22)
Iν (0)
where Iν (0) is the incident intensity, Iν (s) is the emergent intensity at a position s, is
the thickness of a thin ice film, and kν , the absorption coefficient in units of (length)−1 , is
equal to 4πνmi , as noted previously. Experiments to measure the absorption coefficient,
related to the imaginary refractive index, employed geometry with normal incidence, i.e.,
perpendicular to the thin film. In Eq. (2.3.22), we have neglected the reflection component,
which is small in normal cases. The absorption coefficient so defined includes the effects of
both the true absorption of energy and the scattering of energy. Complications caused by film
thickness, the required absorbing substrate, and other pertinent parameters have been noted
by Toon et al. (1994). Spectral measurements are conducted to cover a range of wavenumber.
An initial estimate of optical constraints is made to determine film thicknesses, and the best
value of imaginary index is obtained based on measured transmissions by minimizing the
mean square error between measured and calculated transmission for a number of films of
varying thickness.
The real refractive index can be obtained from the Kramers–Kronig relationship
(Kramers 1927, Kronig 1926) in the forms
∞
2 ν Im[ε(ν )]
Re[ε(ν)] = 1 + P dν , (2.3.23a)
π 0 ν2 − ν2
∞
2ν Re[ε(ν )] − 1
Im[ε(ν)] = − P dν , (2.3.23b)
π 0 ν2 − ν2
where P implies that the principal value of the integral is taken; Re and Im represent the
real and imaginary parts, respectively; and the integration covers all wavenumbers.
The real and imaginary parts of the refractive index for ice have been comprehensively
reviewed and tabulated by Irvine and Pollack (1968), based on the available laboratory mea-
surements of reflection and transmission. Bertie et al. (1969) have measured the absorptance
and reflectance of a film of ice in the range of 8000 to 30 cm−1 at a temperature of −173°C
and have derived real and imaginary indices of refraction. Schaaf and Williams (1973)
have performed similar measurements for ice at −7°C in the wavenumber range of 5000
to 300 cm−1 . Seki et al. (1981) have measured the reflectance of a single crystal of hexag-
onal ice at a temperature of 80 K in the 0.044 to 0.207 µm ultraviolet wavelength range.
A review of the ice refractive index has been given by Warren (1984). Toon et al. (1994)
determined the infrared optical constants of H2 O based on the transmission of infrared light
through thin films of varying thickness over a frequency range from 7000 to 500 cm−1 at
temperatures below 200 K. Additional measurements were reported by Kou et al. (1993),
Clapp et al. (1995), Gosse et al. (1995), Grundy and Schmitt (1998), Zhang et al. (2001),
Rajaram et al. (2001), Curtis et al. (2005), and Mätzler (2006). Several of these authors
performed measurements at temperatures lower than 210 K. Warren and Brandt (2008) have
updated the values reported in Warren (1984). A summary of these measurements in terms
of spectral and temperature ranges is depicted in Figure 2.7.
300
250
Temperature (K)
200
150
100
10-2 10-1 100 101 102 103 104 105 106 107
Wavelength (μm)
Figure 2.7 Available datasets for the imaginary part of the refractive index of ice as a function
of spectral coverage and temperature range, obtained by Warren and Brandt (2008), Gosse et al.
(1995), Toon et al. (1994), Grundy and Schmitt (1998), Zhang et al. (2001), Rajaram et al. (2001),
Clapp et al. (1995), Curtis et al. (2005), and Mätzler (2006) (after Iwabuchi and Yang 2011). Several
measurements were made at temperatures much lower than the melting temperature.
Warren & Brandt (2008)

Warren (1984)
10 0
10 −2
10 −4
10 −6
10 −8
0 1 2 3 4 5
Figure 2.8 Right panel: Real and imaginary indices of refraction for ice as a function of wavelength
from 0.2 to 100 µm. The insert is the imaginary refractive index from 0.2 to 5 µm covering the solar
spectrum. These datasets were compiled by Warren (1984) and updated by Warren and Brandt (2008).
Left panel: all data points from these two sources, presented in the domain of real and imaginary
indices of refraction.
Real Index of Refraction (a)
10-1 100 101 102 103 104 105 106

Imaginary Index of Refraction
(b)
10-1 100 101 102 103 104 105 106

Wavelength (μm)
105 104 103 102 101 100 10-1 10-2

Wavenumber (cm-1)
Figure 2.9 (a) Real and (b) imaginary parts of the refractive index of ice for a temperature range
from 160 to 270 K as a function of wavelength to 106 µm (after Iwabuchi and Yang 2011). Also shown
is the corresponding wavenumber.
Spectral ice indices of refraction are fundamental parameters that determine the relative
scattering and absorption properties of ice particles due to a single-scattering event for
a given wavelength and ice crystal size. If there is no absorption, real-part values are
responsible for the scattering processes. When absorption is involved, the amount of energy
scattered and absorbed depends on both real and imaginary parts, as well as on wavelength
2.4 Definition of scattering properties for ice crystals 83
and on particle size, shape, and composition. In order to compute solar and thermal infrared
radiative fluxes that are absorbed, reflected, or transmitted through ice clouds, accurate
values of the real and imaginary indices of refraction are required. They are equally
important in the development of remote sensing techniques to infer the size and shape
of water and ice particles and cloud optical depth.
Figure 2.8 illustrates ice refractive index datasets as a function of wavelength from 0.2
to 100 µm, as compiled by Warren (1984) and Warren and Brandt (2008). These datasets
have been widely used for calculations of the scattering and absorption properties of
ice crystals. A logarithmic scale is used for the lower right panel such that differences
between the two compilations are not clearly shown, except in the ultraviolet and visible
regions. The left panel displays the variation of the imaginary part versus the real part on a
uniform scale. It is clear that the two compilations show large differences in the microwave
region.
Figure 2.9 illustrates the real and imaginary parts of the refractive index of ice for
a range of temperature from 160 to 270 K in the wavelength domain from 0.05 µm to
100 cm. Also shown is the equivalent wavenumber domain. The dependence of the real
part on temperature appears to be small in all wavelength ranges. However, a significant
dependence of the imaginary part on temperature is shown for wavelengths longer than
about 20 µm, particularly in microwave regions.
2.4 Definition of the single-scattering and polarization properties of ice crystals

2.4.1 Representation of a simple wave
By means of scattering processes, ice crystals in the path of electromagnetic waves con-
tinuously abstract energy from the incident waves and reradiate that same energy in all
directions. Thus, we shall first describe the representation of electromagnetic waves. An
electromagnetic wave is characterized by electric and magnetic vectors E and H, respec-
tively, which form an orthogonal set with the direction of propagation of the wave. In any
medium, E and H are related, and it is customary to use E in scattering discussions. We
say light is polarized in a certain direction when the vibration of the electric vector E
concentrates in that direction. It follows that the direction of polarization is defined as the
direction of the electric vector.
The flow of energy and the direction of wave propagation are represented by the Poynting
vector S, as depicted in Figure 2.10. This vector is given by (Born and Wolf 1975, in
Gaussian units)
c
S= E × H, (2.4.1)
4π
where |S| has units of flux density. The electric field vector E can be decomposed into
two components, El and Er , representing the electric vectors parallel (l) and perpendicular
(r) to a plane through the direction of propagation. The plane so defined is called the
E S
El
E
Er H
Figure 2.10 Propagation of an electromagnetic wave in terms of electric and magnetic vectors and
the definition of the Poynting vector S. An electric vector E in space can be arbitrarily decomposed
into two orthogonal components, referred to as parallel (l) and perpendicular (r) to a plane through
the direction of propagation.
plane of reference, and its selection is generally arbitrary. In practice, we select the plane
containing the incident and scattered beams as the common plane of reference for the two
beams.
Consider an electromagnetic wave that propagates in the z direction with a propagation
constant k (= 2πλ), where λ is the wavelength, and a circular frequency ω (= kc),
and with positive amplitudes and phases (al , ar ) and (δl , δr ) for the electric field of an
electromagnetic wave in the l(El ) and r(Er ) directions, respectively. Thus, we have
El = al e−iδl e−ikz+iωt , Er = ar e−iδr e−ikz+iωt , (2.4.2)
where El and Er are complex, oscillating functions. We may set ζ = kz − ωt and take the
cosine representation for the case when the plane wave is time harmonic, so that
El = al cos(ζ + δl ), Er = ar cos(ζ + δr ). (2.4.3)
It follows that
El /al = cos ζ cos δl − sin ζ sin δl ,

(2.4.4)
Er /ar = cos ζ cos δr − sin ζ sin δr .
Multiplying the first and second equations by (sin δr , cos δr ) and (sin δl , cos δl ), respec-
tively, and subtracting one from the other, we obtain
(El /al ) sin δr − (Er /ar ) sin δl = cos ζ sin(δr − δl ),

(2.4.5)
(El /al ) cos δr − (Er /ar ) cos δl = sin ζ sin(δr − δl ).
r
(al cosδ,ar)
(al,ar cosδ )
2ar l
(-al,-ar cosδ)
(-al cosδ ,-ar)

2al
Figure 2.11 Geometric representation of an elliptically polarized electromagnetic wave. The term δ
is the phase difference and al and ar are the amplitudes of the electric field in the l and r directions,
respectively. Intercepts with the rectangle are indicated by crosses.
By squaring and then adding the above two equations, we obtain

(El /al )2 + (Er /ar )2 − 2(El /al )(Er /ar ) cos δ = sin2 δ, (2.4.6)
where the phase difference δ = δr − δl .
Equation (2.4.6) represents the equation of a conic, the associated determinant of which
is given by

1/al2 −cos δ/(al ar ) sin2 δ

= 2 2 ≥ 0. (2.4.7)
−cos δ/(al ar ) 1/ar2 al ar
Thus, the conic equation in (2.4.6) represents an ellipse. For this reason, an electromagnetic
wave is referred to as an elliptically polarized wave, as illustrated in Figure 2.11. The
ellipse is inscribed in a rectangle whose sides are parallel to the coordinate axes and have
lengths of 2al and 2ar . The ellipse touches the sides at the points (±al , ±ar cos δ) and
(±al cos δ, ±ar ).
Two special cases are of particular importance. If δ = mπ (m = 0, ±1, ± 2, . . . ), then
Eq. (2.4.6) becomes

El Er 2 El Er
± = 0, i.e., =∓ . (2.4.8)
al ar al ar
This equation describes two perpendicular lines. The wave for this case is said to be linearly
polarized. On the other hand, if δ = mπ/2 (m = ±1, ±3, . . .) and al = ar = a, we have
El2 + Er2 = a 2 . (2.4.9)
This equation describes a circle, and we call the wave in this case circularly polarized.
The polarization is called right-handed when sin δ > 0, and left-handed when sin δ < 0.
Right-handed and left-handed refer to the direction of rotation (direction of fingers) when
δ=0 sin δ < 0
δ=π sin δ > 0
Figure 2.12 Representation of linear (left) and circular (right) polarization on a plane, where δ is
the phase difference between El and Er , and left- and right-hand circular polarization are defined
according to the sign of sin δ.
the thumb is pointed in the direction of propagation. Geometrical representations of linear

and circular polarization are shown in Figure 2.12.
In Subsection 2.4.2, we introduce the Stokes vector to define the complete polarization
properties of an electromagnetic wave.
2.4.2 Representation of electromagnetic waves by Stokes parameters

To define the complete polarization properties of electromagnetic waves, four param-
eters are required: El , Er , δ l , and δ r . In reference to intensity, a set of four quanti-
ties referred to as the Stokes parameters, first introduced by Stokes (1852), has been
defined. Because intensity is proportional to the absolute square of the electric field, we
may define the following four parameters, neglecting a constant of proportionality, as
follows:
I = El El∗ + Er Er∗ , (2.4.10a)
Q = El El∗ − Er Er∗ , (2.4.10b)
U = El Er∗ + Er El∗ , (2.4.10c)
V = −i(El Er∗ − Er El∗ ), (2.4.10d)

where an asterisk denotes the complex conjugate value. The parameters I, Q, U, and V are
real quantities that satisfy I 2 = Q2 + U 2 + V 2 . Furthermore, based on the two electrical
y
x
χ
b
O
l
Figure 2.13 Geometric representation of elliptical polarization of a light beam in which the direction
of propagation is into the paper, a and b are the lengths of the semi-major and semi-minor axes,
respectively; χ is the orientation angle between the Ol and Ox axes; and β is the ellipticity angle
whose tangent is the ratio of the ellipse traced by the endpoint of the electric vector, i.e., tan β = ± ba,
where + and − stand for right- and left-handed polarization, respectively. See also Figure 2.11.
fields defined in Eq. (2.4.2), we have

I = al2 + ar2 , (2.4.11a)
Q = al2 − ar2 , (2.4.11b)
U = 2al ar cos δ, (2.4.11c)
V = 2al ar sin δ. (2.4.11d)
The Stokes parameters can be expressed in terms of the geometry defining an ellipse.
Let β denote an angle whose tangent is the ratio of the axes of the ellipse traced by the
endpoint of the electric vector, as displayed in Figure 2.13. If the semi-major and semi-
minor axes of the ellipse are given by a and b, respectively, then tan β = ±b/a. Also, let χ
be the orientation angle between the major axis of the ellipse and the l direction. When the
plane waves are time harmonics, we may express the electric field vectors along the l and
r directions in terms of amplitude and phase using the cosine representation, as shown in
Eq. (2.4.3).
Let x and y denote the directions along the major and minor axes, respectively, Then the
electric fields in the x-y plane may be written

Ex cos χ sin χ El
= , (2.4.12a)
Ey −sin χ cos χ Er
where Ex and Ey may also be expressed in terms of amplitudes (a, b) and an arbitrary
phase δ0 using cosine and sine representations such that they satisfy the elliptical equation
in the forms
Ex = a cos(ξ + δ0 ), (2.4.12b)
Ey = ±b sin(ξ + δ0 ). (2.4.12c)
We may define an angle β, referred to as the ellipticity angle, as follows:
a = a0 cos β, ∓b = a0 sin β. (2.4.13a)
Equations (2.4.12b) and (2.4.12c) can then be written in the forms
Ex = a0 cos β cos(ξ + δ0 ), (2.4.13b)
Ey = −a0 sin β sin(ξ + δ0 ), (2.4.13c)
where a02 is equivalent to I. Using Eqs. (2.4.12a), (2.4.13b), and (2.4.13c), we obtain

El cos χ −sin χ a0 cos β cos(ξ + δ0 )
=
Er sin χ cos χ −a0 sin β sin(ξ + δ0 )

(2.4.14)
cos β cos χ cos(ξ + δ0 ) + sin β sin χ sin(ξ + δ0 )
= a0 .
cos β sin χ cos(ξ + δ0 ) − sin β cos χ sin(ξ + δ0 )
The two components in Eq. (2.4.14) can be expressed as
El = a0 [cos β cos χ (cos ξ cos δ0 − sin ξ sin δ0 )
+ sin β sin χ (sin ξ cos δ0 + cos ξ sin δ0 )]
(2.4.15a)
= a0 [(cos β cos χ cos δ0 + sin β sin χ sin δ0 ) cos ξ
+ (−cos β cos χ sin δ0 + sin β sin χ cos δ0 ) sin ξ ],
Er = a0 [cos β sin χ (cos ξ cos δ0 − sin ξ sin δ0 )
− sin β cos χ (sin ξ cos δ0 + cos ξ sin δ0 )]
(2.4.15b)
= a0 [(cos β sin χ cos δ0 − sin β cos χ sin δ0 ) cos ξ
− (cos β sin χ sin δ0 + sin β cos χ cos δ0 ) sin ξ ].
From Eq. (2.4.3), we have
El = al (cos ξ cos δl − sin ξ sin δl ), (2.4.16a)
Er = ar (cos ξ cos δr − sin ξ sin δr ). (2.4.16b)
Comparing Eqs. (2.4.15a) and (2.4.15b) with Eqs. (2.4.16a) and (2.4.16b), the following
relationships emerge:
a0 (cos β cos χ cos δ0 + sin β sin χ sin δ0 ) = al cos δl , (2.4.17a)
a0 (−cos β cos χ sin δ0 + sin β sin χ cos δ0 ) = −al sin δl , (2.4.17b)
a0 (cos β sin χ cos δ0 − sin β cos χ sin δ0 ) = ar cos δr , (2.4.17c)
−a0 (cos β sin χ sin δ0 + sin β cos χ cos δ0 ) = −ar sin δr . (2.4.17d)
Eliminating δl by performing square operations leads to
al2 = a02 (cos2 β cos2 χ + sin2 β sin2 χ ). (2.4.18a)
Similarly, eliminating δr yields
ar2 = a02 (cos2 β sin2 χ + sin2 β cos2 χ ). (2.4.18b)
It follows from Eqs. (2.4.18a) and (2.4.18b) that
I = a02 = al2 + ar2 = Il + Ir , (2.4.19a)
Q = Il − Ir = a02 [cos2 β(cos2 χ − sin2 χ ) + sin2 β(sin2 χ − cos2 χ )] = I cos 2β cos 2χ .

(2.4.19b)
Moreover, substituting Eqs. (2.4.17a)–(2.4.17d) into the expression for U = 2al ar cos δ,
we derive the following equations:
U = 2al ar (cos δl cos δr + sin δl sin δr )

= 2a02 [(cos β cos χ cos δ0 + sin β sin χ sin δ0 )(cos β sin χ cos δ0
− sin β cos χ sin δ0 ) + (cos β cos χ sin δ0 − sin β sin χ cos δ0 )
× (cos β sin χ sin δ0 + sin β cos χ cos δ0 )].
= 2a02 [cos2 β sin χ cos χ cos2 δ0 − sin β cos β cos2 χ sin δ0 cos δ0
+ sin β cos β sin2 χ sin δ0 cos δ0 − sin2 β sin χ cos χ sin2 δ0
+ cos2 β sin χ cos χ sin2 δ0 + sin β cos β cos2 χ sin δ0 cos δ0 (2.4.20)
− sin β cos β sin χ sin δ0 cos δ0 − sin β sin χ cos χ cos δ0 ]
2 2 2
= 2a02 [cos2 β sin χ cos χ − sin2 β sin χ cos χ

+ sin β cos β(−cos2 χ sin δ0 cos δ0 + sin2 χ sin δ0 cos δ0
+ cos2 χ sin δ0 cos δ0 − sin2 χ sin δ0 sin δ0 )]
= 2a02 sin χ cos χ (cos2 β − sin2 β) = I sin 2χ cos 2β.
In a similar manner, substituting Eqs. (2.4.17a)–(2.4.17d) into the expression for the term
V = 2al ar sin δ yields
V = I sin 2β. (2.4.21)
In summary, the Stokes parameters can be expressed in terms of total intensity, ellipticity,
and orientation angles in the forms
I = Il + Ir , (2.4.22a)
Q = Il − Ir = I cos 2β cos 2χ , (2.4.22b)
U = I cos 2β sin 2χ , (2.4.22c)
V = I sin 2β. (2.4.22d)

Note that I and V are independent of the orientation angle χ .
V
I
2β U
Y
Q 2χ
Figure 2.14 Representation of Stokes parameters (I, Q, U, V), denoted by heavier lines, on a Poincaré
sphere. The angles β and χ are defined in Figure 2.13. When β is positive (negative), the polarization
is said to be right- (left-)handed.
Equations (2.4.22a)–(2.4.22d) may be represented in Cartesian coordinates on a sphere

referred to as the Poincaré sphere (Poincaré 1892), as illustrated in Figure 2.14. The radius
of the sphere is given by I, while the zenithal and azimuthal angles are defined by π/2 − 2β
and 2χ , respectively. Thus, Q, U, and V denote the lengths in the x, y, and z directions,
respectively. On this sphere, the northern and southern hemispheres represent right-handed
and left-handed elliptical polarization, respectively. The north and south poles denote right-
handed and left-handed circular polarization, respectively, and points in the equatorial plane
represent linear polarization. For a simple wave, we have I 2 = Q2 + U 2 + V 2 . Moreover,
from Eqs. (2.4.22a)–(2.4.22d), the plane of polarization and the ellipticity angle can be
expressed in terms of Stokes parameters as follows:
tan 2χ = U/Q, (2.4.23a)

1/2
sin 2β = V /(Q2 + U 2 + V 2 ) . (2.4.23b)
In the preceding discussion, we have assumed a constant amplitude and phase in repre-
senting the wave vibration using Eqs. (2.4.22a)–(2.4.22d). However, the actual light beam
consists of many simple waves in very rapid succession. Within a very short duration (e.g.,
seconds), millions of simple waves are collected by a detector. Consequently, measurable
intensities are associated with the superposition of many millions of simple waves with
independent phases. If the operator denotes the time average over a time interval (t1 , t2 ),
then the Stokes parameters of the entire beam of light for this time interval may be expressed
as

I = al2 + ar2 = Il + Ir ,

Q = al2 − ar2 = Il − Ir ,
(2.4.24)
U = 2al ar cos δ,
V = 2al ar sin δ.
In this case, we can prove that I 2 ≥ Q2 + U 2 + V 2 . A light beam is generally charac-

terized by partial elliptical polarization. If the phase differences between the two electric
components are 0° or an integer multiple of 180° (i.e., β = 0), the light beam is linearly
polarized (V = 0). If, on the other hand, the amplitudes of the two electric components are
the same and their phase differences are an odd-integer multiple of 90°, then the light beam
is circularly polarized (Q = U = 0). When the ellipticity angle β is positive or negative,
the circular polarization is said to be right- or left-handed. These cases have been discussed
previously using electric vectors.
Additionally, we may represent the Stokes parameters in terms of detectable variables.
Consider a retardation ε in the r direction with respect to the l direction and let the component
of the electric field vector be in a direction making an angle with the positive l direction.
For a simple wave at time t, the electric field can then be represented in the form
E(t; ; ε) = El cos + Er e−iε sin

(2.4.25)
= al cos e−iξ + ar e−i(δ+ε)−iξ sin .
The expressions for El and Er have been defined in Eq. (2.4.2). For a time interval (t1 , t2 ),
the time average intensity is then given by
I (, ε) = E(t; ; ε)E ∗ (t; ; ε)

= al2 cos2 + ar2 sin2 + 12 2al ar cos δ sin 2 cos ε (2.4.26a)
− 1
2
2al ar sin δ sin 2 sin ε,
where ∗ denotes the complex conjugate value; see Eq. (2.4.10a). Using Eq. (2.4.24) and
noting that Il cos2 + Ir sin2 = (I + Q cos 2)/2, we obtain
I (, ε) = 12 [I + Q cos 2 + (U cos ε − V sin ε) sin 2]. (2.4.26b)
On the basis of Eq. (2.4.26b), the Stokes parameter can now be expressed in terms of
retardation and polarization angle in the form (Liou 1975a)
I = I (0, 0) + I (π/2, 0),

Q = I (0, 0) − I (π/2, 0),
(2.4.27a)
U = I (π/4, 0) − I (3π/4, 0)
V = − [I (π/4, π/2) − I (3π/4, π/2)] .
Detector
Q
Light
Source
Θ
I0 P M S
Figure 2.15 A schematic diagram of the optical setup in a laboratory setting for the measurement
of F() in which the symbols P, M, Q, and A denote polarizer, modulator, quarter-wave plate, and
analyzer, respectively; and S denotes the center of scattering.
Consequently, the Stokes parameters of a light beam can be represented by a combination

of several polarizers oriented at the positions 0, π4, π2, and 3π/4 and a compensator
(e.g., a quarter-wave (π/2) plate), which can detect circular polarization defined in the
element V.
In conjunction with the experimental setup, several optical components can be used,
including linear polarizers, quarter-wave plates, and an electro-optic modulator, which all
affect the state of polarization of a light beam as it crosses the component. In mathematical
terms, this can be described as obtaining the Stokes vector of the transmitted beam by
multiplying the Stokes vector of the incident beam by a 4 × 4 transformation matrix, referred
to as the Mueller matrix (Shurcliff 1962). The Mueller matrix of an optical component is
defined in terms of the component’s orientation relative to a reference plane which, in turn,
is defined by an angle between the scattering plane (the horizontal plane in laboratory
experiments) and the principal axis of the component, and can be written as follows:
⎡ ⎤ ⎡ ⎤⎡ ⎤
I M11 M12 M13 M14 I0
⎢ Q ⎥ ⎢ M21 M22 M23 M24 ⎥ ⎢ Q0 ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ (2.4.27b)
⎣ U ⎦ ⎣ M31 M32 M33 M34 ⎦ ⎣ U0 ⎦ ,
V M41 M42 M43 M44 V0
where the terms Mij (i, j = 1–4) denote the 4 × 4 Mueller matrix elements. Referring to
Figure 2.15, the scattered Stokes vector can be expressed in terms of the incident Stokes
vector I0 and a number of optical component arrangements as follows (Coulson 1988, Kuik
et al. 1991):
I() = A Q F()M P I0 , (2.4.27c)
where A, Q, M, and P represent the Mueller matrices of the analyzer, quarter-wave plate,
modulator, and polarizer, respectively, which have orientation angles with respect to the
incoming light beam, F() is the transformation matrix defined in Eq. (2.4.32) below, and
is the scattering angle. The analyzer is a polarizer inserted in front of the detector, while
the modulator uses the linear electro-optic effect such that the voltage over a cell can be
varied sinusoidally in time.
To quantify the polarization of a light beam, it is customary to define the following
parameter, referred to as the degree of polarization:
1/2
PO = (Q2 + U 2 + V 2 ) /I . (2.4.28a)
If the ellipticity is neglected, we may define the degree of linear polarization in the form
LP = −Q/I = −(Il − Ir )/(Il + Ir ). (2.4.28b)
Let there be light: Light occurring in nature is called natural light. It is light whose
intensity remains unchanged and is unaffected by the retardation of one orthogonal com-
ponent relative to another when resolved in any direction in the transverse plane. This
means that natural light is defined from Eq. (2.4.26b) as, I (, ε) = I /2. Its intensity is
hence independent of and ε. Thus, the necessary and sufficient condition that light be
natural is Q = U = V = 0. Under this condition, the degree of polarization defined in Eq.
(2.4.28a) for natural light is zero. As a consequence, natural light is also referred to as
unpolarized light. Light emitted from the sun is unpolarized. However, after interacting
with molecules and particles through scattering events, the unpolarized sunlight generally
becomes partially polarized. Natural light characterized by Q = U = V = 0 can be shown
to be equivalent to a mixture of any two independent oppositely polarized streams of half
the intensity.
In the atmosphere, light is generally partially polarized and its Stokes parameters (I, Q,
U, V) can be decomposed into two independent groups characterized as natural light and
elliptically polarized light, as follows:
⎡ ⎤ ⎡ ⎤ ⎡ 2 ⎤
I I − (Q2 + U 2 + V 2 )1/2 (Q + U 2 + V 2 )1/2
⎢Q⎥ ⎢ 0 ⎥ ⎢ Q ⎥
⎢ ⎥=⎢ ⎥+⎢ ⎥ . (2.4.29)
⎣U ⎦ ⎣ 0 ⎦ ⎣ U ⎦
V 0 V
Having presented the essence of the physical and mathematical definitions of a light
beam, we shall now introduce the single-scattering (light scattered just once) properties of
ice crystals.
2.4.3 Scattering phase matrix for an ice crystal

Consider an ice particle of arbitrary shape and size. The scattered electric field at a distance
r from the particle must be related to the two components of the incident electric field
(Eli , Eri ). In the far field, the 2 × 2 amplitude matrix that transforms the incident electric
vector into the scattered electric vector may be written as

s

i

i
El exp(−ikr + ikz) S2 S3 El A2 A3 El
= = , (2.4.30)
Ers ikr S4 S1 Eri A4 A1 Eri
where z is the vertical direction in Cartesian coordinates, Sj (j = 1, 2, 3, 4) are the amplitude

functions, and the matrix comprising Aj (j = 1, 2, 3, 4) is referred to as the transformation
matrix. Because of the symmetry of spherical particles with respect to the incident beam,
S3 = S4 = 0.
Using the Stokes parameters defined above, the incident and scattered electric vectors
can be expressed in terms of their intensity components as follows:
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = F ⎢ Q0 ⎥ , (2.4.31)
⎣ U ⎦ k 2 r 2 ⎣ U0 ⎦
V V0
where the subscript 0 denotes the incident beam and the transformation matrix F is given
by
⎡ ⎤
F11 F12 F13 F14
⎢ F21 F22 F23 F24 ⎥
F=⎢ ⎣ F31 F32 F33 F34 ⎦ ,
⎥ (2.4.32)
F41 F42 F43 F44
where
F11 = 12 (M2 + M3 + M4 + M1 ); F12 = 12 (M2 − M3 + M4 − M1 ), (2.4.32a)

F13 = S23 + S41 ; F14 = −D23 − D41 , (2.4.32b)
F21 = 1
2
(M2 + M3 − M4 − M1 ); F22 = 1
2
(M2 − M3 − M4 + M1 ), (2.4.32c)
F23 = S23 − S41 ; F24 = −D23 + D41 , (2.4.32d)
F31 = S24 + S31 ; F32 = S24 − S31 , (2.4.32e)
F33 = S21 + S34 ; F34 = −D21 + D34 , (2.4.32f)
F41 = D24 + D31 ; F42 = D24 − D31 , (2.4.32g)
F43 = D21 + D34 ; F44 = S21 − S34 . (2.4.32h)
Expressions for the matrix elements in terms of the electric fields can be derived from
the definition of the Stokes parameters and are given by
Mk = |Sk |2 , (2.4.33a)
Skj = Sj k = (Sj Sk∗ + Sk Sj∗ )/2, (2.4.33b)
−Dkj = Dj k = (Sj Sk∗ − Sk Sj∗ )i/2, j, k = 1, 2, 3, 4. (2.4.33c)
The preceding matrix elements are real numbers. In light scattering and radiative transfer,
it is conventional to define the scattering phase matrix, P, such that its first element is
normalized to unity as follows:

2π π
P11 ()
sin ddφ = 1, (2.4.34)
0 0 4π
where and φ denote the scattering and azimuthal angles, respectively.
To complete the definition of the phase matrix, we need to introduce the scattering
cross-section, which represents the amount of incident flux that is removed from the
original direction as a result of a single-scattering event. This flux is distributed isotropically
throughout the area of a sphere whose radius is R and whose center is the scatterer. The
scattering cross-section is related to the first element of the transformation matrix defined
in Eq. (2.4.32) in the form
% 4 &
1 2π π 1
σs = 2 Mk sin d dφ. (2.4.35)
k 0 0 2 k=1
Using Eq. (2.4.35), the scattering phase matrix can be defined in terms of the transfor-
mation matrix as follows:
P 1
= F. (2.4.36)
4π σs k 2
Subsequently, the scattered Stokes parameters can be expressed in the form
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = eff P ⎢ Q0 ⎥ , (2.4.37)
⎣U ⎦ ⎣
4π U0 ⎦
V V0
where the effective solid angle associated with scattering is given by eff = σs /r 2 . If no
assumption is made about the shape and position of the scatterer, the scattering phase matrix
consists of 16 non-zero elements as follows:
⎡ ⎤
P11 P12 P13 P14
⎢ P21 P22 P23 P24 ⎥
P(general) = ⎢
⎣ P31 P32 P33 P34 ⎦ .
⎥ (2.4.38)
P41 P42 P43 P44
2.4.4 Extinction cross-section for a non-spherical ice crystal

The extinction cross-section for a spherical particle can be derived from the solution for the
Lorenz–Mie theory discussed in Subsection 2.2.3.1. In this subsection, we shall formulate
the extinction cross-section for a non-spherical ice particle with arbitrary orientation.
Consider the conceptual diagram shown in Figure 2.16. A polarized incident wave can
be resolved into two components, the TE and TM modes for which the electric vector
vibrates perpendicular and parallel to the incident plane, respectively. For the TE mode
Far field
Θ = 0O
i
E i s
E +E
2 2
| E i |2 - | E i + E s | = se | E i |
Figure 2.16 A conceptual diagram for the evaluation of extinction cross-section in an arbitrarily
oriented non-spherical ice particle. All notations are defined in the text.
(case 1) the scattered electric field in the forward direction from Eq. (2.4.30) is given by

s

El e−ikr+ikz S3 (0) e−ikr+ikz
E ( = 0) =
s
s = Eri = S(0)Eri , (2.4.39)
Er ikr S1 (0) ikr
where S (0) = [S3 (0), S1 (0)]T and the symbol T denotes the transpose of a matrix.
Further, consider a point (x, y, z) in the forward direction, i.e., = 0. In the far field,
since x z and y z, we have
1 x2 + y2
r = (x 2 + y 2 + z2 ) /2 ≈ z + . (2.4.40)
2z
Superimposing incident and scattered electric fields in the forward direction yields

S(0) −k(x 2 +y 2 )/2z
Er + E = Er I +
i s i
e , (2.4.41)
ikz
where Eir = [Eri , Eri ]T and the unit matrix I = [1, 1]T . The far-field combined flux density
in the forward direction is then proportional to

2 2 2 S (0) −k(x 2 +y 2 )/2z
IEir + Es ≈ Eri I + Re e , (2.4.42)
kz i
where Re[ ] represents the real part of the argument. We may integrate the combined flux
density over the cross-sectional area A of a scattering particle to obtain the total power of
the combined image as follows:

I
2 |Eir + Es |dxdy = A − σ e , (2.4.43)
E i
r A
where A = [A, A] and σ e = [σ e,11 , σ e,12 ]T . We may set the integration limits to infinity
T
for mathematical convenience to obtain

∞ ∞
2πz
e−k(x +y )/2z dxdy =
2 2
. (2.4.44)
−∞ −∞ ik
It follows that the extinction cross-section for the TE mode (case 1) is given by
σe,1 = σe,11 + σe,12 = (4π/k 2 )Re[S1 (0)] + (4π/k 2 )Re[S3 (0)]
= (4π/k 2 )Re[S1 (0) + S3 (0)]. (2.4.45a)
Following the same procedure, the extinction cross-section for the TM mode (case 2) can
be derived as follows:
σe,2 = (4π/k 2 )Re[S2 (0) + S4 (0)]. (2.4.45b)
The preceding two equations are the generalized extinction cross-sections for a non-
spherical particle. For incident natural light, the extinction cross-section is the average of
the two linear polarization components:
σe = (σe,1 + σe,2 )/2. (2.4.45c)
For a group of randomly oriented non-spherical ice particles, to be discussed in the next
subsection, because of the sign cancelation of S3 (0) and S4 (0), Eq. (2.4.45c) becomes
approximately (Yang and Liou 1997, Liou et al. 2014)
2π
σe ∼
= 2 Re[S1 (0) + S2 (0)]. (2.4.45d)
k
2.4.5 Scattering phase matrix for an ensemble of ice crystals

The preceding discussion was concerned with the scattering of light by an ice crystal.
Scattering by a group of ice crystals is determined by the orientation and size of individual
crystals with respect to the incident light beam. For continuity in the present discussion, we
shall consider a sample of non-spherical ice crystals of the same size, randomly oriented in
space. The scattering phase matrix can then be expressed as
2π π/2
1
P() = P (α , γ )σs (α , γ ) sin α dα dγ , (2.4.46)
2πσs 0 0
where α and γ are the orientation angles of a non-spherical particle with respect to the
incident light beam, and P denotes the scattering phase matrix for a single particle. In
this case, the scattering phase matrix is solely a function of the scattering angle. Further,
the scattering cross-section defined in Eq. (2.4.35) for randomly oriented non-spherical ice
crystals can be written in the form

1
σs = σs (α , γ ) sin α dα dγ , (2.4.47)
2π
where σs denotes the scattering cross-section for a single crystal. For a symmetric hexagonal
crystal, we may define two orientation angles with reference to the incident beam. Let α be
the complement of the zenith angle, and β be an angle that is mapped on √ the2x-y plane with
reference to the incident beam. The surface area of the basal plane is 3 3D /8, while that
of the prism plane
√ is DL. The cross-sectional area for the basal plane mapped along the
light beam is 3 3 (D 2 /8) sin α. The cross-sectional area for the prism plane also depends
on the position of the light beam with reference to six prime planes. For example, if the
light beam is at an angle of β = π/6, the mapped cross-section along the beam would be
at maximum. Thus, we should have DL cos α cos(π/6 − β). The geometric cross-section
for the entire hexagonal crystal is the sum of the preceding mapped cross-sections:
√
3 3 2
G(α, β) = D sin α + DL cos α cos(π/6 − β). (2.4.48)
8
Defining the average cross-section for randomly oriented ice crystals and carrying out the
integrations over α and β, we obtain

6 π/6 π/2
3 √ 2
Ḡ = G(α, β) cos αdαdβ =( 3D /4 + DL). (2.4.49)
π 0 0 4
√
Note that the surface area of a hexagonal crystal is given by S = 3( 3D 2 /4 + DL). It
follows that in the context of the optical extinction theorem, the extinction cross-section,
which is twice the geometric cross-section, is related to the surface area by σe = 2Ḡ = S/2.
Thus, the average geometric cross-section for randomly oriented plate/column crystals Ḡ
is one-fourth of their surface areas (Vouk 1948).
If ice crystals are randomly oriented in space, the law of reciprocity can be applied
(Perrin 1942, van de Hulst 1957). In this case, the directions of the incident and scattered
polarized beams can be reversed to achieve the same result. It follows that the amplitude
functions (S3 , S4 ) in Eq. (2.4.30) must be equivalent to (−S4 , −S3 ), so that

S2 −S4
S∗ = . (2.4.50)
−S3 S1
The transformation matrix corresponding to this 2 × 2 amplitude matrix can then be written
in the form [see Eq. (2.4.32)]
⎡ ∗ ∗ ∗ ∗ ⎤
F11 F12 F13 F14
⎢ ∗ ∗ ∗ ∗ ⎥
⎢ F21 F22 F23 F24 ⎥
∗ ⎢
F =⎢ ∗ ⎥, (2.4.51)
∗ ∗ ∗ ⎥
⎣ F31 F32 F33 F34 ⎦
∗ ∗ ∗ ∗
F41 F42 F43 F44
where the matrix elements are defined by
∗ ∗
F11 = 12 (M2 + M4 + M3 + M1 ); F12 = 12 (M2 − M4 + M3 − M1 ), (2.4.51a)
∗ ∗
F13 = −S24 − S31 ; F14 = D24 + D31 , (2.4.51b)
∗ ∗
F21 = 1
2
(M2 + M4 − M3 − M1 ); F22 = 12 (M2 − M4 − M3 + M1 ), (2.4.51c)
∗ ∗
F23 = −S24 + S31 ; F24 = D24 − D31 , (2.4.51d)
∗ ∗
F31 = −S23 − S41 ; F32 = −S23 + S41 , (2.4.51e)
∗ ∗
F33 = S21 + S34 ; F34 = −D21 − D34 , (2.4.51f)
∗ ∗
F41 = −D23 − D41 ; F42 = −D23 + D41 , (2.4.51g)
∗ ∗
F43 = D21 − D34 ; F44 = S21 − S34 . (2.4.51h)
The scattered and incident Stokes parameters in this case are related via the scattering phase
matrix in the form [see Eq. (2.4.37)]
⎡ ⎤ ⎡ ⎤⎡ ⎤
I P11 P12 P13 P14 I0
⎢ Q ⎥ eff ⎢ P21 P22 P23 P24 ⎥ ⎢ Q0 ⎥
⎢ ⎥= ⎢ ⎥⎢ ⎥. (2.4.52)
⎣U ⎦ 4π ⎣ P31 P32 P33 P34 ⎦ ⎣ U0 ⎦
V P41 P42 P43 P44 V0
Consider an arbitrarily small volume containing a sample of non-spherical particles

whose scattering phase matrix is P, and let there be the same number of reciprocal particles
such that its scattering phase matrix is P∗ [corresponding to F∗ ; see Eq. (2.4.36)]. Then,
the average scattering phase matrix P corresponding to this volume can be expressed as
(P + P∗ )/2 and is proportional to the following transformation matrix:
⎡ ⎤
P11 P12 P13 P14
⎢ P21 P22 P23 P24 ⎥
P = ⎢ ⎥
⎣ P31 P32 P33 P34 ⎦ , (2.4.53)
P41 P42 P43 P44
where
P11 = 12 (M2 + M4 + M3 + M1 ); P12 = 12 (M2 − M1 ), (2.4.53a)

P13 = 1
(S
2 23
+ S41 − S24 − S31 ); P14 = 1
2
(−D23 − D41 + D24 + D31 ), (2.4.53b)
P21 = 12 (M2 − M1 ); P22 = 12 (M2 − M3 − M4 + M1 ), (2.4.53c)
P23 = 12 (S23 − S41 − S24 + S31 ); P24 = 12 (−D23 + D41 + D24 − D31 ), (2.4.53d)
P31 = 12 (S24 + S31 − S23 − S41 ); P32 = 12 (S24 − S31 − S23 + S41 ), (2.4.53e)
P33 = S21 + S34 ; P34 = −D21 , (2.4.53f)
P41 = 12 (D24 + D31 − D23 − D41 ); P42 = 12 (D24 − D31 − D23 + D41 ), (2.4.53g)
P43 = D21 ; P44 = S21 − S34 . (2.4.53h)
In terms of the elements expressed as Pij , we have
P12 = P21 , P13 = −P31 , P14 = P41 , P23 = −P32,

P24 = P42 , P34 = −P43 . (2.4.54)
Using these six relationships, the scattering phase matrix can then be written as
⎡ ⎤
P11 P12 P13 P14
⎢ P P22 P23 P24 ⎥
⎢ 12 ⎥
P = ⎢ ⎥. (2.4.55)
⎣ −P13 −P23 P33 P34 ⎦
P14 P24 −P34 P44
In this case, the scattering phase matrix contains ten independent elements. We shall
now consider a sample of randomly oriented ice particles, which have a plane of symmetry
and let the incident Stokes parameters be changed to (I0 , Q0 , −U0 , −V0 ). The scattered
Stokes parameters must also be changed to (I, Q, −U, −V ). Thus, we may replace U0 and
V0 by −U0 and −V0 to obtain
I = P11 I0 + P12 Q0 + P13 U0 + P14 V0
= P11 I0 + P12 Q0 − P13 U0 − P14 V0 . (2.4.56a)
For this equation to be valid, we must have P13 = P14 = 0. Likewise, we have, for the
Q, U, and V components,
Q = P21 I0 + P22 Q0 + P23 U0 + P24 V0
= P21 I0 + P22 Q0 − P23 U0 − P24 V0 , (2.4.56b)
U = P31 I0 + P32 Q0 + P33 U0 + P34 V0
= P31 I0 + P32 Q0 − P33 U0 − P34 V0 , (2.4.56c)
V = P41 I0 + P42 Q0 + P43 U0 + P44 V0
= P41 I0 + P42 Q0 − P43 U0 − P44 V0 . (2.4.56d)
Consequently, we must have P23 = P24 = 0, P31 = P32 = 0, and P41 =

P42 = 0 to satisfy Eqs. (2.4.56b)–(2.4.56d). It follows that the scattering phase matrix in
this case is given by
⎡ ⎤
P11 P12 0 0
⎢ P21 P22 0 0 ⎥
⎢ ⎥. (2.4.57)
⎣ 0 0 P33 P34 ⎦
0 0 P43 P44
Moreover, from Eq. (2.4.54), we note that P21 = P12 and P43 = −P34 . Thus, the
scattering phase matrix for randomly oriented non-spherical ice particles, which obey the
law of reciprocity, can be expressed as follows (the symbol is neglected for presentation
convenience):
⎡ ⎤
P11 P12 0 0
⎢ P12 P22 0 0 ⎥
P(ice crystals) = ⎢
⎣ 0
⎥. (2.4.58a)
0 P33 P34 ⎦
0 0 −P34 P44
It follows that the scattering phase matrix reduces to six independent elements. In the case
of spherical particles, S3 = S4 = 0, so that P22 = P11 and P44 = P33 . Therefore, there are
only four independent scattering phase matrix elements, as follows:
⎡ ⎤
P11 P12 0 0
⎢ P12 P11 0 0 ⎥
P(spheres) = ⎢
⎣ 0
⎥. (2.4.58b)
0 P33 P34 ⎦
0 0 −P34 P33
Consider now a spectrum of ice particles randomly oriented in space with a size dis-
tribution given by n(L), where L is the major axis of an ice particle. A parameter referred
to as the aspect ratio LD, where D denotes the width, is used to define the particle size.
We may employ a mean effective size in association with light-scattering calculations. The
scattering phase matrix associated with a suitable volume, which contains a spectrum of
ice particles defined by a size distribution n(L), can be obtained from
L2 L2
P() = P(, L)σs (L)n(L)dL σs (L)n(L)dL, (2.4.59a)
L1 L1
where L1 and L2 are the lower and upper limits of ice particle length. Let the total number
of ice crystals in the cloud be N. Then the scattering (and extinction) cross-section for a
spectrum of ice crystals of different sizes and shapes can be obtained from the following:
L2
σs,e = σs,e (L)n(L)dL/N . (2.4.59b)
L1
As stated previously, the condition under which the scattering phase matrix is composed
of six independent elements is that non-spherical particles are randomly oriented in space
in such a manner that every one of them has a plane of symmetry based on which the
law of reciprocity can be applied: that the incident and scattered beams are reversible. In
Subsection 1.5.2, we gave examples of typical ice crystal sizes and shapes. Assuming that
they are randomly oriented in space, it would be almost impossible to prove that every
one of these ice crystals has a plane of symmetry that would obey the law of reciprocity.
Nevertheless, numerical calculations for light scattering by ice crystals carried out by
Takano and Liou (1989a, 1995) show that the scattering phase matrix elements P13 , P14 ,
P23 , P24 , P31 , P32 , P41 , and P42 are practically zero and that P12 = P21 and P34 = −P43 .
2.4.6 Some examples of the single-scattering properties of ice spheres, cylinders,

and spheroids
In this subsection, we present a number of pertinent single-scattering results for ice spheres,
ice circular cylinders, and ice spheroids. The extinction efficiency Qext is defined as
σ e G, where G is the geometric cross-section with respect to the incident direction. For a
spherical particle with a radius of r, it is simply πr2 . Moreover, the single-scattering albedo,
representing the fraction of light that is scattered out of the incident light beam, is defined
as follows:
= σs (scattering cross section)/σe (extinction cross section). (2.4.60a)
Thus, (1 – ϖ) represents the fraction of light that is absorbed. From the total scattered
intensity, we define a parameter referred to as the asymmetry factor in the form

1 1
g = < cos > = P11 (cos ) cos d cos , (2.4.60b)
2 −1
Extinction Efficiency Single-Scattering Albedo Asymmetry Factor

4 1.0 1.0
0.8 0.8
3
r = 10 μm
0.6 0.6
2
0.4 0.4
1
0.2 0.2
0 0.0 0.0
4 1.0 1.0
0.8 0.8
3
r = 100 μm
0.6 0.6
2
0.4 0.4
1
0.2 0.2
0.0 0.0
0
10−1 100 101 102 10−1 100 101 102 10−1 100 101 102
Wavelength (μm)
Figure 2.17 Extinction efficiency, single-scattering albedo, and asymmetry factor as functions of
wavelength from 0.2 to 100 µm for an ice sphere with a radius of 10 µm (top panels) and 100 µm
(bottom panels). These results are computed from the exact Lorenz–Mie solution.
where P11 () denotes the normalized phase function, and is directly related to the total
scattered intensity, as a function of the scattering angle . The g-factor is the first moment
of the phase function and represents the strength of forward scattering. The extinction
efficiency, single-scattering albedo, and asymmetry factor are the three basic parameters
for input to a radiative transfer program for radiative flux calculations. For remote sensing
purposes, however, we need the phase function for bidirectional reflectance analysis; and
if polarization is involved, information from the full scattering phase matrix is required.
Figure 2.17 depicts the extinction efficiency Qext , the single-scattering albedo ϖ, and
the asymmetry factor g for an ice sphere with radii of 10 and 100 µm as functions of 0.2
to 100 µm wavelength. The results were calculated from Lorenz–Mie theory, employing
the complex (real and imaginary) indices of refraction depicted in Figure 2.8. Qext for r =
10 µm can be explained in terms of the size parameter 2πrλ. For r = 100 µm, Qext is close
to 2 as a result of the limit of large size parameters when the wavelength becomes smaller.
Minima for ϖ at λ = 2.85, 10.75, and 42.76 µm when r = 10 µm result from the Christiansen
effect (Christiansen, 1884, 1885), which occurs when the real part of the refractive index
approaches 1, while the corresponding imaginary counterpart is substantially larger, leading
to the domination of absorption. For r = 100 µm at λ 3 µm, ϖ has a value of 0.5 due
to strong absorption. The maxima of g-factor for r = 10 µm occur in concurrence with
the minima of ϖ, and vice versa, at λ 23 µm, contributed from diffracted and externally
reflected rays, which are more dominant when ϖ is smaller. At λ 3 µm, the g-factor for
Extinction Efficiency Single-Scattering Albedo Asymmetry Factor

4 1.0 1.0
0.8 0.8
3
r = 10 μm
0.6 0.6
2
0.4 0.4
1
0.2 0.2
0 0.0 0.0
4 1.0 1.0
0.8 0.8
3
r = 100 μm
0.6 0.6
2
0.4 0.4
1
0.2 0.2
0.0 0.0
0
10−1 100 101 102 10−1 100 101 102
10−1 100 101 102
Wavelength (μm)
Figure 2.18 Extinction efficiency, single-scattering albedo, and asymmetry factor for an ice circular
cylinder with a radius of 10 µm (top panels) and 100 µm (bottom panels) at normal incidence.
r = 100 µm can be interpreted in terms of geometric optics as in the case of r = 10 µm. At

λ 3 µm, the g-factor becomes relatively flat due to strong absorption.
Figure 2.18 shows Qext , ϖ, and g for a circular cylinder at normal incidence computed
from a Lorenz–Mie-like program (Liou 1972a). Qext is similar to the values presented in
Figure 2.17 due to geometric similarity, except for the enhanced amplitude of sinusoidal
oscillations around 2 at λ 3 µm. ϖ for r = 10 µm is also similar to that of the sphere at
λ 30 µm, while ϖ for r = 100 µm is similar to that of the sphere over all wavelengths.
The g-factor for r = 10 µm is larger than that for the sphere, but for r = 100 µm it is larger
than the spherical case at λ 3 µm, although the results are comparable to the g-factor for
a sphere at λ 3 µm.
Figure 2.19a illustrates the phase function for polydisperse ice spheres having a gamma
size distribution with an effective radius re of 20 µm, corresponding to cirrostratus clouds.
We have selected four remote sensing wavelengths for presentation. At visible wavelengths,
noticeable features are a diffraction peak at the scattering angle = 0°, the primary rainbow
feature at = 136°, and a glory feature at = 180°. At λ = 1.38 µm, the rainbow peak
reduces and spreads due to a small size parameter. At λ = 3.7 µm, the rainbow shifts to
= 154°. Rainbow and glory intensities are reduced as a result of absorption. At
λ = 10 µm, the phase function consists of diffracted and externally reflected rays (see
Subsection 3.1.2) caused by strong absorption, rendering it featureless.
Figure 2.19b shows the corresponding degree of linear polarization (LP). At λ =
0.63 µm, a negative polarization at 70° results from transmitted rays without internal
0.8
(a) (b)
1x106
Cs (re = 20 μm) 0.6
Degree of Linear Polarization

1x104
0.4
Phase Function
1x10 2 1 =0.63 m 0.2
1
1.38 m
0
1x100
3.7 m
=0.63 µm
−0.2 1.38 µm
1x10−2 10 m 3.7 µm
10 µm
−0.4
0 60 120 180 0 60 120 180
Scattering Angle (deg.)
Figure 2.19 (a) Phase function and (b) degree of linear polarization for a group of spheres governed
by a gamma size distribution with an effective radius of 20 µm, corresponding to cirrostratus. In the
phase function diagram, the vertical scale applies to the lowest curve, while the upper curves are
shifted upward by a factor of 10.
reflection, while maxima at = 96°, 136°, and 144° are produced, respectively, by exter-
nally reflected rays, rainbows, and supernumerary bows. Intense variation of LP around
= 180° is produced by the glory feature. At λ = 1.38 and 3.7 µm, maxima produced
by external reflection at = 100° and 112°, respectively, are reduced due to smaller size
parameters. At λ = 3.7 µm, the maximum produced by rainbows shifts to = 154°. At λ =
10 µm, a single maximum occurs, produced by external reflection in conjunction with strong
absorption.
In Figure 2.20, some characteristics of the normalized phase function P11 for randomly
oriented spheroids are observed in backscattering regions; that is, a rather flat angular
distribution with a small increase at the backscattering. Features such as rainbows and
glories produced by scattering by large spheres are greatly reduced for spheroids of large
ab. The element −P12 P11 (= LP) gives the degree of linear polarization for the single
scattering of unpolarized incident light. It has positive values at scattering angles 120°
in contrast to negative values in the case of spheres.
The element P22 P11 represents the ratio of the intensity component depolarized or
cross-polarized to the total scattered intensity. The depolarization ratio for total intensity is
defined as = 1 − P22 P11 , which is a measure of non-sphericity because P22 = P11 and
= 0 for homogeneous spheres. For small angles 25°, is very small. However, from
102 1.0 102 1.0
Prolate Oblate
m = 1.33 0.5 m = 1.33 0.5
1
101 a = 15 10 a = 15
a/b = 5 a/b = 5
0 0
0 0
10 10
-0.5 P43/P11 -0.5 P43/P11
10-1 (c) 10-1 (c)

P11 -1.0 P11 -1.0
1.0 1.0
(a) (a)
10-2 10-2
1.0 0.5 1.0 0.5
P22 /P11 P22 /P11
P44/P11 P44/P11
0.5 0 0.5 0
0 -0.5 P33/P11 0 -0.5 P33/P11

-P12/P11 -P12/P11
(b) (d) (b) (d)
-0.5 -1.0 -0.5 -1.0
0 60 120 180 0 60 120 180 0 60 120 180 0 60 120 180
Figure 2.20 Angular distribution of the elements of normalized scattering matrix for randomly oriented prolate (left panels) and oblate (right panels)
spheroids with m = 1.33, α = 15, and ab = 5: (a) the normalized phase function P11 ; (b) the element P22 P11 and the degree of linear polarization
−P12 P11 ; (c) the element P43 P11 ; and (d) the elements P33 P11 and P44 P11 . The scattering matrix elements for area equivalent spheres are shown by
dotted lines (after Asano and Sato 1980).
side to backscattering regions, values are relatively large in the case of spheroids. The
element P43 P11 for randomly oriented spheroids is close to that for equivalent spheres.
Also, good agreement extends to larger scattering angles than in the case of LP. For large
scattering angles, however, P43 P11 for spheroids is larger than that for equivalent spheres.
The element P44 P11 for randomly oriented spheroids is generally larger than P33 P11 . The
case of thin oblate spheroids is an exception, however. We find that differences between
P44 P11 and P33 P11 vary as a function of angle in a manner similar to P22 P11 , and that
≥ |P44 − P33 |/P11 with the equality valid at = 0 and 180°. At these angles, P33 =
±P22 , as predicted from the symmetry relations discussed by van de Hulst (1957).
2.5 Introduction to radiative transfer in ice crystal clouds: link to single scattering
The preceding discussion applies to scattering by an ice crystal just once, including cou-
pled absorption and polarization processes and scattering by an ensemble of ice crystals,
assuming that the independent scattering concept is applicable such that each crystal scat-
ters the light beam independently (single scattering). In the latter case, we may consider
a sufficiently small volume in an ice cloud so that its single-scattering properties can be
defined on the basis of Maxwell’s fundamental electric and magnetic vector equations and
Stokes’ vector intensity (I, Q, U, V). We shall present the link between these two funda-
mentals and the transfer of radiation in planetary atmospheres, in which multiple scattering
(and absorption or emission) takes place. The term “light scattering” thus includes single
scattering and multiple scattering, the latter also referred to as “radiative transfer.”
The subject of radiative transfer covers a variety of fields, including astrophysics, applied
physics and optics, planetary science, atmospheric science, and meteorology, as well as
various engineering disciplines. Astrophysicists pioneered the development of this field
at the beginning of the twentieth century (Schuster 1905, Schwarzschild 1906, Eddington
1916, Menzel 1966). Prior to 1950, the subject of radiative transfer was also an important
research area in applied physics and nuclear engineering, associated with neutron transport.
In his landmark book, Chandrasekhar (1950) presented the subject of radiative transfer in
plane-parallel atmospheres as a branch of mathematical physics and developed numerous
solution methods and techniques, including the consideration of polarization. The principle
of radiative transfer has also been extensively employed by planetary scientists, particularly
in association with the remote sensing of planetary atmospheres by means of spectroscopy
and polarimetry (Chamberlain and Hunten 1987), as well as by atmospheric scientists
to study the transfer of solar and thermal infrared radiation in the Earth’s atmosphere
containing clouds and aerosols, for climate, weather, and remote sensing applications
(Goody and Yung 1989; Liou 1992, 2002; Mishchenko et al. 2000a, 2000b).
2.5.1 Formulation of scalar and vector radiative transfer equations

In accordance with the independent scattering concept introduced in Section 2.1, we may
define the intensity, Iλ , in units of energy per second per square meter per wavelength
per solid angle (e.g., W m−2 λ−1 sr−1 ), where W (watt) = joule per second, m = meter,
λ = wavelength in units of µm, and sr = steradian. See also Eq. (2.4.10a). For simplicity
2.5 Radiative transfer in ice crystal clouds 107
(a) Z (b)
Z
(m, f)
z
s i2
i1 (m ', f ')
P2
Θ
q q P1
q'
O Y O
f
f'
f−
Figure 2.21 (a) Definitions of the zenith angle θ and the azimuthal angle φ with reference to the
Cartesian coordinate system (x, y, z), where s represents a position vector in space. (b) Scattering
plane OP1 P2 with respect to the meridian planes OP1 Z and OP2 Z. All angles in the figure are defined
in the text.
of presentation, we set I = Iλ . It is understood that I represents monochromatic intensity

confined within a solid angle. Thus, we say that intensity acts like a pencil of radiation,
which is a function of position, defined by (x, y, z) in Cartesian coordinates, and direction,
defined by a solid angle in terms of zenith and azimuthal angles (Figure 2.21a). We shall
first consider I as a scalar and follow the formulation developed by Chandrasekhar (1950)
based on the energy conservation principle.
A pencil of radiation traversing an ice cloud will be weakened by its interaction with ice
crystals. If the intensity I becomes I + dI after traversing a thickness ds in the direction of
its propagation, then we have
dI = −kρI ds, (2.5.1)
where ρ is the density of the cloud and k denotes the mass extinction (scattering +
absorption) cross-section (in units of area per mass) for radiation of a given wavelength.
At the same time, the intensity can be strengthened by multiple scattering from all other
directions into the pencil plus self-emission from ice crystals under the Kirchhoff–Planck
thermal equilibrium condition, all at the same wavelength. We define the source function
coefficient j such that the enhancement in intensity due to multiple scattering and emission
is given by
dI = jρds, (2.5.2)
where the source function coefficient j has the same physical meaning as the mass extinction
cross-section. Combining the two radiation sources defined in Eqs. (2.5.1) and (2.5.2), we
obtain
dI = −kρI ds + jρds. (2.5.3)

For presentation convenience, we may define the source function J = jk to obtain
dI (s)
− = −I + J. (2.5.4)
kρds
This represents the general radiative transfer equation without imposition of a coordinate
system.
For atmospheric applications, we may consider that the atmosphere in localized portions
is plane-parallel such that intensity and atmospheric parameters (e.g., temperature and
gaseous profiles) are horizontally homogeneous, while variation only occurs in the vertical
direction. In this case, it is convenient to measure linear distances normal to the plane of
stratification (Figure 2.21a). If z denotes this component, then the general equation for
radiative transfer defined in Eq. (2.5.4) becomes
dI (z; θ, φ)
cos θ = −I (z; θ, φ) + J (z; θ, φ), (2.5.5)
kρdz
where θ denotes the zenith angle in reference to the upward normal and φ the azimuthal
angle in reference to the x-axis.
For convenience of analysis, we introduce the non-dimensional normal optical depth (or
simply optical depth) τ defined by the following integration:
∞
τ= k(z )ρ(z )dz , (2.5.6)
z
so that dτ = −kρdz. In terms of the optical depth and by measuring intensity downward
from the outer boundary (top of the atmosphere), we have
dI (τ ; μ, φ)
μ = I (τ ; μ, φ) − J (τ ; μ, φ), (2.5.7)
dτ
where µ = cos θ . This is the basic equation for radiative transfer in plane-parallel atmo-
spheres. The source function representing intensity enhancement during the transfer of
radiation includes three terms: (1) multiple scattering, (2) direct radiation from the sun
attenuated to the level τ , and (3) emission from ice crystals subject to the Kirchhoff–Planck
condition. All three of these components with the same wavelength are traversing in the
direction of intensity defined by (µ, φ).
Thus, the basic scalar equation for the transfer of diffuse intensity (i.e., scattering more
than once) as a function of optical depth τ and direction denoted by (µ, φ) in plane-parallel
atmospheres can be written in the form

dI (τ ; μ, φ)
μ = I (τ ; μ, φ) − I (τ, μ , φ )P (μ, φ; μ , φ )dμ dφ
dτ 4π 4π (2.5.8)

− F P (μ, φ; −μ0 , φ0 )e−τ /μ0 − (1 − ' ω)B[T (τ )],
4π
where the second term on the right-hand side represents multiple scattering contribu-
tions from all directions (4π solid angle), where the single-scattering albedo has been
defined in Eq. (2.4.60a); the phase function P (μ, φ; μ , φ ) represents the normalized 3-D
distribution of scattered intensity, which redirects the incoming intensity defined by (μ , φ )
to the outgoing intensity defined by (μ, φ); the third term defines the solar flux F at the
top of the atmosphere, attenuated to the level τ and undergoing a single scattering event in
the direction (−μ0 , φ0 ) via the phase function P (μ, φ; −μ0 , φ0 ); and the last term is the
emission contribution under the local thermodynamic equilibrium condition such that emis-
sion is equal to absorption, where B[T(τ )] is the Planck function defined by the temperature
at the position τ .
In order to describe the radiation field completely at each point in space, the scalar
intensity I must be replaced by the vector intensity I = (I, Q, U, V ), in terms of the
four Stokes parameters defined by the intensity, the degree of polarization, the plane of
polarization, and the ellipticity of light beams as a function of incoming and outgoing
directions, as shown in Eqs. (2.4.22a)–(2.4.22d). Thus, the radiative transfer equation can
be written in vector form as follows:
dI(τ ; μ, φ)
μ = I(τ ; μ, φ) − J(τ ; μ, φ), (2.5.9)
dτ
where the source function J(τ ; μ, φ) is a vector consisting of four elements, which can
be obtained as follows. Consider a differential increment dJ(τ ; μ, φ; μ , φ ), produced by
multiple scattering involving a pencil of radiation of solid angle d in the direction (μ , φ ).
The diffuse intensity vector I(τ ; μ , φ ), which generates the source term, is in reference
to the meridian plane OP1 Z, defined in Figure 2.21b. However, the scattering phase matrix
P() derived from the scattering theory is in reference to the plane of scattering OP1 P2 that
contains the incident and scattered beams. We must then transform I(τ ; μ , φ ) to the plane
of scattering in order to obtain the proper source function by first redirecting I(τ ; μ , φ )
to the plane of scattering by applying the transformation matrix L(−i1 ), such that the
angle −i1 is defined between the meridian plane OP1 Z and the plane of scattering OP1 P2 ,
and the minus sign signifies that the rotation of the plane is counterclockwise. Thus, the
contribution to the source function with reference to the plane of scattering at the point
P2 is given by the term denoted by the symbol (∗ ) = P()L(−i1 )I(τ ; μ , φ )d /4π,
where the approximation of a constant single-scattering albedo for a cloud is made. The
transformation matrix for the Stokes vector (I, Q, U, V ) defined on the x-y plane relative
to that on the l-r plane can be obtained on the basis of Eq. (2.4.12a), and it is given by
⎡ ⎤
1 0 0 0
⎢ 0 cos 2χ sin 2χ 0 ⎥
L(χ ) = ⎢
⎣ 0 −sin 2χ cos 2χ 0 ⎦ .
⎥ (2.5.10)
0 0 0 1
The Stokes parameters I and V are invariant in the transformation process. Based on matrix
manipulations, we also have L(χ1 )L(χ2 ) = L(χ1 + χ2 ) and the inverse matrix L−1 (χ ) =
L(−χ ).
To transform this vector, denoted by (∗ ) above, to the scattering direction (μ, φ) repre-
senting the meridian plane OP2 Z, we must again apply the transformation matrix L(π − i2 )
through the angle (π − i2 ) clockwise, where i2 denotes the angle between the meridian plane
OP2 Z and the plane of scattering OP1 P2 . Thus, the desired differential source function due
to the diffuse component can be expressed as
dJ(τ ; μ, φ; μ , φ ) = Z(μ, φ; μ , φ )I(τ ; μ , φ )d /4π, (2.5.11)
where the phase matrix is defined by
Z(μ, φ; μ , φ ) = L(π − i2 )P()L(−i1 ). (2.5.12)
The source function vector for multiple scattering can be obtained by performing the
integration over all incoming directions signified by (μ , φ ) as follows:
2π 1
1
J(τ ; μ, φ; μ , φ ) = Z(μ, φ; μ , φ )I(τ ; μ , φ )dμ dφ . (2.5.13)
4π 0 −1
We have used Z and P to represent the phase matrix and the scattering phase matrix,
respectively, to differentiate between the two. From spherical trigonometry, the angles i1
and i2 can be derived and are given by

sin(φ − φ ) 1 − μ2 sin(φ − φ ) 1 − μ 2
sin i1 = , sin i2 = . (2.5.14)
sin sin
Following the same procedure, the direct component of the source function vector associated
with the point source I (−μ, φ) = δ(μ − μ0 )δ(φ − φ0 )F , where δ is a delta function and
F is the downward solar flux vector, is given by
1
J(τ ; μ, φ) = Z(μ, φ; −μ0 , φ0 )F e−τ /μ0 . (2.5.15)
4π
Combining Eqs. (2.5.15) and (2.5.13), the transfer of sunlight, including full polarization,
can be written as follows:
2π 1
dI(τ ; μ, φ) 1
μ = I(τ ; μ, φ) − Z(μ, φ; μ , φ )I(τ ; μ , φ )dμ dφ
dτ 4π 0 −1 (2.5.16)
1
− Z(μ, φ; −μ0 , φ0 )F e−τ /μ0 + (1 − )B(T )Ie ,
4π
where B(T) is the Planck intensity at temperature T; and Ie = (I, Qe , 0, 0), with −Qe the
linear polarization component associated with emission. In the preceding formulation, we
have assumed that an optical depth can be defined for the medium. This applies to spherical
particles that are randomly located with the scattering phase matrix P given by Eq. (2.4.58b),
and to non-spherical particles that are randomly oriented with the scattering phase matrix
P given by Eq. (2.4.58a). In these cases, we may replace the phase function depicted in Eq.
(2.5.8) with the 4 × 4 phase matrix Z to account for the full polarization effect. The phase
matrix Z has a number of unique properties associated with the symmetry principle of light
beams. For the preceding cases, a light beam can reverse its direction with the final results
being the same, so that Z(−μ, −φ; −μ , −φ ) = Z(μ, φ; μ , φ ).
θ
φ - φ' Out
θ'
Θ
In π-
O
α
α'
γ - γ'
Y
φ' Z' φ
Figure 2.22 Geometry of single scattering by a non-spherical ice particle. The coordinate system
(x , y , z ) refers to the crystal axis, while (x, y, z) is fixed in space. The angles (θ, θ ) are zenith
angles associated with the incident and scattered beams with respect to the (x, y, z) coordinates, while
φ = (φ − φ ) are corresponding azimuthal angles. The angles (α, α ) are zenith angles with respect
to the orientation of the particle, while (γ , γ ) correspond to azimuthal angles. These angles can be
transformed to (x , y , z ) coordinates in terms of a set of angles (, φ) and (α , γ ), for analysis of
the light scattering by horizontally oriented ice crystals.
2.5.2 Radiative transfer in horizontally oriented ice particles

The scattering of a light beam by a non-spherical ice particle is dependent on the directions
of the incoming and outgoing radiation, and on the particle’s orientation with respect to the
incoming beam. In order to formulate the transfer of solar radiation in a medium comprised
of horizontally oriented ice particles, we may begin by assuming that the intensity varies
only in the z direction, referred to as the plane-parallel condition.
Referring to Figure 2.22, we may select a fixed coordinate system (x, y, z) such that the
z-axis is along the zenith direction. Additionally, we let (x , y , z ) be a coordinate system
referred to the crystal axis, with z along the c-axis. Angles (φ , φ, γ , γ ) are azimuthal
angles corresponding to (θ , θ, α , α) defined in the figure, and is the scattering angle.
The scattering parameters for a non-spherical ice particle, including the phase function
and the extinction and scattering cross-sections, may be expressed with respect to either
of these two coordinate systems. Thus, we may express them symbolically as follows:
P (α, γ ; μ , φ ; μ, φ) = P (α , γ ; cos , φ); σe,s (α, γ ; μ , φ ) = σe,s (α , γ ), (2.5.17)
where φ = φ − φ . It is clear that the phase function is dependent on the directions of

the incident and scattered beams as well as the orientation of the non-spherical ice particle.
The extinction and scattering cross-sections, however, depend only on the direction of the
incident beam and the ice particle orientation.
For a sample of non-spherical ice particles randomly oriented in space, the average
single-scattering properties may be expressed in the forms
2π π/2
1
P (cos , φ) = P (α , γ ; cos , φ)σs (α , γ ) sin α dα dγ ,
2πσs 0 0
2π π/2 (2.5.18)
1
σe,s = σe,s (α , γ ) sin α dα dγ .
2π 0 0
The extinction and scattering cross-sections for randomly oriented non-spherical particles
are independent of direction. If all non-spherical ice particles have rotational symmetry
(e.g., circular cylinders), then the phase function is independent of the azimuthal angle φ.
Consequently, multiple scattering of the diffuse intensity in randomly oriented, symmetrical
non-spherical ice particles can be formulated using six independent scattering phase matrix
elements.
The spatial orientation of hexagonal and irregular ice crystals in cirrus clouds is a
significant factor in the transfer of radiation in the atmosphere. The fact that numerous
halos and arcs have been observed demonstrates that specific orientation of ice particles
must exist in some cirrus. Based on laboratory experiments, cylinders with a diameter-to-
length ratio of less than 1 tend to fall with their long axes horizontally oriented. Observations
of columnar and plate crystals in cirrus clouds have shown that these particles fall with
their major axes parallel to the ground. The orientation of ice particles in cirrus clouds
has been observed by numerous lidar measurements based on the depolarization technique
in the backscattering direction. The depolarization ratio of the backscattered return from
horizontally oriented plates is close to zero (see Section 5.5), but this ratio increases
significantly as the lidar scans a few degrees from the vertical. Specific orientation occurs
when the ice particles have relatively large sizes and defined shapes, such as columns and
plates do. However, if the ice crystals are irregular, as aggregates are, a preferred orientation
is unlikely to occur. Furthermore, smaller ice crystals in cirrus clouds where substantial
turbulence occurs tend to orient in three-dimensional space. Finally, it has been noted that
ice particle orientation and alignment are closely modulated by the electric field in clouds.
In the case of horizontally oriented ice crystals, their single-scattering parameters are
dependent on the direction of the incident light beam. Thus, the conventional formulation
for the multiple-scattering problem requires modification. Takano and Liou (1989b) used
realistic scattering parameters and the Stokes vector for horizontally oriented ice crystals
in association with the adding method for radiative transfer. Takano and Liou (1993)
further presented the theoretical formulation and numerical calculations for the transfer of
polarized thermal infrared radiation in optically anisotropic media with specific application
to horizontally oriented ice particles. In the following paragraphs, we present a unified
theoretical formulation that is applicable to both solar and thermal infrared radiative transfer,
including polarization, for horizontally oriented ice crystals.
When ice particles are randomly oriented in a horizontal plane, we have α = π/2 from
Eq. (2.5.17). Thus, the phase function and cross-sections are dependent only on the incident
angle and may be symbolically written in the forms

2 π/2
P (μ , φ ; μ, φ) = P (π/2, γ ; μ , φ ; μ, φ)dγ , (2.5.19a)
π 0
2π π/2
1
σe,s (μ) = 2 σe,s (π/2, γ ; μ, φ)dγ dφ. (2.5.19b)
π 0 0
With the preceding understanding of the incident direction of a light beam with respect to ice
particle geometry, we may define a differential normal optical depth such that d τ̃ = −β̃e dz,
where the vertical extinction coefficient β̃e = βe (μ = 1) and z is the distance. The general
equation governing the transfer of the Stokes vector may be expressed in the form
dI(τ̃ ; μ, φ)
μ = k(μ)I(τ̃ ; μ, φ) − J(τ̃ ; μ, φ), (2.5.20)
d τ̃
where the actual extinction coefficient normalized by its vertical counterpart is defined by
k(μ) = βe (μ)/β̃e , in Eq. (2.5.16) is replaced by the single-scattering albedo defined by
(μ) = βs (μ)/βe (μ), and βs is the scattering coefficient matrix, which has a form similar
to the 4 × 4 extinction coefficient matrix.
For horizontally oriented particles, the extinction coefficient is dependent on both the
energy characteristics of the incident beam and its state of polarization, referred to as the
dichroism of the scattering medium. This generally occurs when the light beam passes
through a cloud of aligned non-spherical particles associated with electric and/or magnetic
fields, as noted previously. Because of dichroism, the extinction coefficients corresponding
to the Stokes vector are represented by a 4 × 4 extinction matrix (Martin 1974, Mishchenko
1991, Liou and Takano 2002). If ice particles are randomly oriented in space such that
each one of them has a plane of symmetry and the law of reciprocity may be applied,
the scattering phase matrix P consists of only six independent elements, as shown in
Eq. (2.4.58a). In this case, k(μ) = 1 and βs , βe , and are independent of µ. For all
practical purposes, we may use the extinction coefficient βe and the single-scattering albedo
ω̃ in scalar forms in the calculations for horizontally oriented column and plate crystals.
We may approach the radiative transfer problem for ice particles randomly oriented in
a horizontal plane using the single-scattering and optically thin approximations, and omit
the emission term. The phase function and single-scattering parameters are now dependent
on the direction of the incident beam. In the case of the transfer of solar radiation, we must
use the normal optical depth τ̃ and the cosine of the solar-zenith angle-dependent single-
scattering albedo, (μ0 ). We must also distinguish between the reflection and transmission
of radiation from above and below, since the phase functions for horizontally oriented ice
particles differ in these two configurations (Takano and Liou 1989b, Liou and Takano
2002).
Figure 2.23 depicts an example of a non-dimensional reflected light beam that has been
scattered for horizontally oriented (2-D) and randomly oriented (3-D) columns using a
c d
e
3D 2D
0 0 0
0
0
0
le
5
5
le
10
10
g Zen 20 40
15
15
Zen 20 40 60 60 g
20
20
An An
25
25
30
30
35
al
35
ith A ith A al
40
40
45
45
120
uthΔf
50
120
50
uth
55
60
55
ngle 60 ngle
60
60
im Δf
65
65
70
70
,q i m ,q 80 180
75
80 180
75
80
z
80
Az
85
85
Figure 2.23 Non-dimensional reflected light at a wavelength of 0.63 µm for ice columns randomly
oriented in space (3-D) and oriented parallel to the ground (2-D columns) in the plane of zenith and
azimuthal angles. The cosine of the solar zenith angle and the optical depth used in the calculations
are 0.5 and 1, respectively. In the 3-D case, two features are seen: (a) the antisolar peak and (b) limb
brightening. For the 2-D column case, three peak features are: (c) the broad peak around the antisolar
point, (d) the broad peak around the subsun, and (e) the lower tangent arc. See also Subsection 3.1.4.
representative cirrostratus ice crystal size distribution with an optical depth of 1 in the
plane defined by the zenith (θ ) and relative azimuthal (φ − φ0 ) angles. The wavelength and
the cosine of the solar zenith angle used are 0.63 µm and 0.5, respectively. Note that the
domain for the reflectances is from 60° to 180° scattering angles, which does not cover
the commonly observed halos located at 22° and 46°. In the 3-D case, the maximum, at
θ ≈ 80◦ and close to the principal plane φ − φ0 = 0◦ , is related to limb brightening. A
small maximum associated with the antisolar peak is seen at θ ≈ 40◦ and φ − φ0 = 120◦ .
Otherwise, the reflectance variations are relatively small in the linear scale. In the case
of 2-D columns, we see numerous reflection maxima. The chief ones are: (1) the subsun
located at θ ≈ 60◦ in the principal plane produced by external reflections; (2) the lower
tangent arc located at θ ≈ 80◦ in the principal plane generated by two refractions; and (3)
the antisolar peak located at θ ≈ 60◦ and φ − φ0 = 180◦ , caused by internal reflections.
Much larger anisotropy occurs in this case than in the 3-D case. In realistic cirrus clouds,
we would anticipate that some of the large and defined ice particles (plates and columns)
are horizontally oriented (see also discussions in Subsections 1.6.2 and 3.1.4).
3
Principles of geometric optics for application to light
scattering by ice crystals
In Chapter 1, we showed that ice crystals are complex and intricate in terms of their
morphology and size – essentially no single ice crystal that has been observed in the Earth’s
atmosphere and laboratory setting exhibits the exact same size, shape, and composition
(internally mixed with ice nuclei). In Chapter 2, we followed by presenting the available
solutions for light scattering by sphere, circular cylinder, and spheroid upon which the
coordinate systems can be imposed to define these particle shapes. Thus, analytic solutions
can be derived from conventional wave equations and suitable boundary conditions to match
the incident, internal, and scattered electric and magnetic vectors in three-dimensional
space defined by coordinate systems. Computations can then be performed using solution
equations referred to as the “exact” method, to obtain the single-scattering properties of
ice spheres, cylinders, and spheroids. For instance, the Lorenz–Mie theory is the exact
“solution” for light scattering by spherical particles. Light scattering by non-spherical ice
crystals, however, cannot be resolved by the conventional approach to solutions in view
of their complex structure. This chapter presents the geometric-optics approach to light
scattering by ice crystals. We first review the essence of geometric optics from several
historical perspectives, including diffraction and surface waves.
3.1 Application of geometric optics to light scattering by ice crystals

In his pioneering optics experiments, Newton (1704) demonstrated that the refraction
of light through a prism leads to its dispersion into component colors, as depicted in
Figure 3.1a. Light seen by the retina of the human eye, referred to as white light, covers fre-
quencies between 4.3 × 1014 and 7.5 × 1014 s−1 (wavelength 0.4–0.7 µm). When incident
white light enters a prism and undergoes two consecutive refractions, the outgoing light dis-
plays seven color spectra – red bends the least, violet bends the most, and indigo, blue, green,
yellow, and orange colors bend in between in a sequential manner. In Subsection 2.3.1,
we discussed the physical basis for light dispersion, leading to the definition of complex
refractive index. Ice absorption is negligibly small at visible wavelengths (see Figure 2.8),
so the optical properties of light can be defined by the real refractive index. In all optical
phenomena discussed in the following paragraphs, water drops, ice crystals, and aerosols in
115
116 Geometric optics for light scattering by ice crystals
(a) Refraction (b) Refraction/Reflection

Sunlight
White light
Prism dispersion of light

(Newton 1704) Rainbow (Airy 1838)
(c) Surface wave (Edge effect) (d) Diffraction

Sunlight
Sunlight
Glory (van de Hulst 1947) Corona (Babinet 1837)
(e) Refraction (f) Refraction
Sunlight
Sunlight

22o halo (Huygens 1662) 46o halo (Mariotte 1681)
Figure 3.1 (a) The dispersion of white light into component colors by two refractions through a
prism. (b) The production of the primary rainbow from light rays undergoing two refractions and one
internal reflection in a raindrop. (c) The generation of glories based on the concept of surface waves,
referred to as the edge effect. (d) The formation of coronas by means of diffraction. The production of
(e) a 22° halo and (f) a 46° halo by two refractions through prism angles of 60° and 90°, respectively.
the Earth’s atmosphere act like a prism and produce visible color sequences. Indeed, geo-
metric optics in terms of tracing light rays is essential to understanding physical processes
associated with light scattering by ice crystals.
The best known atmospheric optical phenomenon is the rainbow, an inspiration for
art and mythology in human history. Rainbows in the sky are usually formed on summer
3.1 Application of geometric optics to light scattering 117
afternoons after a rain shower. They are produced by two refractions and one internal
reflection of the sun’s light rays within raindrops, as illustrated in Figure 3.1b. The most
frequently observed rainbow (first rainbow) is due to the sun’s light rays undergoing
minimum deviation, referred to as the Descartes ray, resulting in maximum intensity at an
observed angle of 137° between the sun (see the thick white line with arrows) and raindrops,
causing the white light to disperse into component colors (red outside and blue inside).
Thus, in order for a rainbow to be observed, the sun must be behind the observer. When the
raindrops are sufficiently large, it is possible to observe a secondary rainbow at an angle of
130° (blue outside and red inside). A supernumerary bow, which often appears inside the first
rainbow, is produced by wave interference that cannot be explained by geometric reflection
and refraction. The physical explanation for the formation of rainbows and supernumerary
bows is attributed to Airy (1838), and known as the Airy theory of rainbows.
When water clouds are present, an optical feature known as a glory may be observed
on mountain tops as well as from aircraft. This optical feature requires 180° backscattering
geometry. Geometric reflection and refraction cannot explain the occurrence of a glory
because water’s refractive index of 1.33 cannot produce minimum-deviation rays. van
de Hulst (1947) suggested that glories are caused by grazing-incident rays that undergo
one internal reflection, as shown in Figure 3.1c. He postulated that the 14.4° gap between
the scattering angles of 165.6° and 180°, associated with the 1.33 refractive index, could
be bridged by surface waves, which are produced by interaction of the incident waves at
grazing angles near the edges of a sphere and continuation of the wave motion along its
surface into the shadow region, referred to as the edge effect. Nussenzveig (2002) added
further insight into the edge effect. The color sequence (red outside and blue inside) is
produced by wave interference.
The formation of the optical feature known as a corona, frequently seen around the sun
or moon, can be explained by light diffraction associated with water or aerosol particles, as
illustrated in Figure 3.1d. According to Babinet’s principle (Babinet 1837), diffraction by
a cloud particle is equivalent to that produced by an aperture of the same size and shape.
Colorful rings appear around the sun when particle size is uniform; however, the color
sequence differs from the 22° halo discussed below. After a huge volcanic eruption in 1883,
color rings (red outside and blue inside) called Bishop’s rings (Bishop 1884) appeared
around the sun. These rings were produced by the diffraction of sunlight associated with
volcanic aerosols having about the same size through the optical interference of diffracted
waves.
Light scattering by ice crystals produces numerous fascinating optical features associated
with their hexagonal shapes and orientation properties, as illustrated in Subsection 3.1.4.
We have identified the two most noticeable features: 22° and 46° halos (Figure 3.1e,f),
which have color rings around the sun with red inside and blue outside (Greenler 1980).
It appears that Mariotte (1681) was the first to attribute the formation of 46° halos to
equilateral triangular prismatic ice crystals. Huygens (1662) appears to have been the first
to present a quantitative explanation for the formation of 22° halos, the most commonly
observed optical feature generated by ice crystals in cirrus clouds.
It is quite clear that the principles of geometric optics, including diffraction and surface
waves, have been followed to identify essentially all optical features that have been observed
in the Earth’s atmosphere and to provide physical and mathematical explanations as to how
these phenomena are produced. In Chapter 1 we used geometric-optics fundamentals to
develop numerous approximations for light scattering by ice crystals covering all possible
sizes from 0.1 to 1000 µm (see Figure 1.23) and shapes (see Figure 1.19). Results
derived from these approximations have been verified with respect to those computed
from the “exact” Lorenz–Mie theory and other numerical and computational methods
presented in Chapter 4. Additionally, we have assessed geometric-optics results vis-à-
vis the values determined from light-scattering and ice-microphysics experiments in a
controlled laboratory setting. This chapter presents a systematic and logical development
and application of the theories of geometric optics, diffraction, and surface waves to light
scattering by ice crystals. Snell’s and Fresnel’s laws are first introduced.
3.1.1 Snell’s and Fresnel’s laws for geometric optics

We shall consider a plane wave arriving at a boundary which separates two homogeneous
media with distinct optical properties. This wave splits into two waves: a transmitted wave
entering the second medium and a reflected wave propagating back into the first medium.
From the part of the wave that touches the surface plane, we may isolate a narrow light
beam referred to as a ray that is much smaller than the surface. Let v1 and v2 be the
velocities of propagation in the two media such that v1 > v2 , and let θi and θt be the angles
corresponding to incident and refracted waves. On the basis of the geometry depicted in
Figure 3.2, we have
sin θi /sin θt = v1 /v2 = m, (3.1.1)
where m is the refractive index for the second medium relative to the first medium. This
equation represents the essence of Snell’s law, which relates incident and refracted angles
to the indices of refraction. As background, this law was named after the Dutch astronomer
Willebrord Snellius (1580–1626); however, according to Born and Wolf (1975), Snel-
lius presented this law in an unpublished manuscript. Apparently, it was first accurately
described in 984 by Ibn Sahl, a scientist at the Baghdad court, in the manuscript “On the
Burning Instruments” (Rashed 1993).
In order to derive the Fresnel equations, we consider the electric vector of the incident
field Ei , with incidence denoted by the superscript i. The components of the incident
electric field vector perpendicular (r) and parallel (l) to the plane comprised of incident and
refracted rays mapped in rectangular coordinates, as shown in Figure 3.2, are given by
Exi = −Eli cos θi , Eyi = Eri , Ezi = Eli sin θi . (3.1.2)
From the Maxwell equations, the relationship between electric and magnetic vectors
√ √
can be shown to be H = εa × E, or E = − 1/εa × H, where a is a unit vector in
Incident Reflected
Err
Eri r
i
El El
t θi
v1
θr 1
X
2
v 2t
θt
t
El
Ert
Transmitted (or Refracted)
Figure 3.2 Graphical representation of the reflection and refraction of a plane wave traversing from
medium 1 to medium 2. The choice of positive directions for the parallel components (l) of the electric
vectors is depicted in the figure. The perpendicular components are at right angles into the plane of
reference. The terms v1 t and v2 t denote the distance in media 1 and 2, respectively. The incident
angle is denoted by θi , while the refracted angle is denoted θt . The reflected angle θr = π − θi .
the direction of propagation. Thus, the components of the magnetic vector are as follows
√
(μ ≈ 1, m = ε):
Hxi = −Eri cos θi m1 , Hyi = −Eli m1 , Hzi = Eri sin θi m1 , (3.1.3)
where m1 is the refractive index of the first medium with respect to vacuum.
In a similar manner, if Et and Er denote the transmitted (t) and reflected (r) electric
vectors, respectively, then the following relationships can be derived:
Ext = −Elt cos θt , Eyt = Ert , Ezt = Elt sin θt ,
(3.1.4)
Hxt = −Ert cos θt m2 , Hyt = −Elt m2 , Hzt = Ert sin θt m2 ,
Exr = −Elr cos θr , Eyr = Err , Ezr = Elr sin θr ,
(3.1.5)
Hxr = −Err cos θr m1 , Hyr = −Elr m1 , Hzr = Err sin θr m1 ,
where m2 is the refractive index of the second medium with respect to vacuum and θr =
180◦ − θi , as displayed in Figure 3.2.
The boundary conditions require that the tangential components of E and H be contin-
uous [see Eq. (2.2.20)]. Hence, we must have
Exi + Exr = Ext , Eyi + Eyr = Eyt ,
(3.1.6)
Hxi + Hxr = Hxt , Hyi + Hyr = Hyt .
Upon substituting all the electric and magnetic components into Eq. (3.1.6), we obtain the
following four relationships:

cos θi Eli − Elr = cos θt Elt , Eri + Err = Ert , (3.1.7a)
i
m1 cos θi Er − Err = m2 cos θt Ert , m1 Eli + Elr = m2 Elt . (3.1.7b)
Using these four equations, solutions for the electric components of the reflected and
transmitted waves in terms of the incident wave are given by
Elr = Rl Eli , Elt = Tl Eli , Err = Rr Eri , Ert = Tr Eri , (3.1.8)
where the amplitude coefficients can be obtained as follows:

cos θi − m cos θt m cos θi − cos θt
Rr = , Rl = , (3.1.9a)
cos θi + m cos θt m cos θi + cos θt
2 cos θi 2 cos θi
Tr = , Tl = , (3.1.9b)
cos θi + m cos θt m cos θi + cos θt
with m = m2 /m1 , the refractive index of the second medium relative to the first medium.
Equations (3.1.9a) and (3.1.9b) represent the well-known Fresnel formulas (Fresnel 1819).
The Poynting vector for electromagnetic waves [Eq. (2.4.1)] is expressed as S =
(c/4π)E × H. Because H can be expressed in terms of E, the flux density |S| =
√
(c/4π) ε|E|2 (μ = 1). Thus, the amounts of energy incident, reflected, and transmitted
per unit area of the boundary per unit time (e.g., W m−2 ) are given by
F i = |Si | cos θi = (c/4π)m1 |Ei |2 cos θi , (3.1.10a)

F = |S | cos θr = (c/4π)m1 |E | cos θr ,
r r r 2
(3.1.10b)
F = |S | cos θt = (c/4π)m2 |E | cos θt ,
t t t 2
(3.1.10c)
where θr = 180◦ − θi . The reflected and transmitted portions of energy in two polar-
2
ization components with respect to the incident energy are proportional to Rr,l and
Tr,l m cos θt / cos θi , respectively. We can prove that Rr,l + Tr,l m cos θt / cos θi = 1, based
2 2 2
on the principle of energy conservation. The transmitted (or refracted) parts of the energy
can then be expressed as (1 − Rr,l 2
).
3.1.2 Absorption effects in the context of geometric optics

The geometric-optics approach postulates that within a particle the effect of absorption on
the propagation direction of a ray can be neglected, so the refracted angle and ray path length
can be computed from Snell’s law and the particle’s geometry. This approach is correct if
absorption is weak, as in the case of ice and water at most solar wavelengths. For cases of
strong absorption, rays refracted into the particle are almost totally absorbed, so that the
geometric-optics method can also be used to compute external reflection. The preceding
argument is physically correct in the limits of weak and strong absorption; however, we
normal of interface
incident wave reflected wave
er
ei
qi qr
Air
Ice
qt
planes of constant planes of constant

amplitude phase
et
ea refracted wave
Figure 3.3 Illustration of a light beam traveling from air to ice, a medium with absorption. The plane
of constant amplitude of the refracted wave is parallel to the interface, which is referred to as the
equi-amplitude plane (Born and Wolf 1975). At the same time, the direction of phase propagation for
the inhomogeneous wave inside the medium is determined by Snell’s law. The notations ei , et , and
eα represent the directions of the incident wave, the refracted wave, and the normal to the interface,
respectively. The angles θ i , θ r , and θ t are defined in the figure.
shall consider the general absorption effect in the context of geometric optics based on
fundamental electromagnetic wave theory. The effect of the complex refractive index on
geometric optics has been formulated only for the Fresnel coefficients, as shown by Stratton
(1941) and Born and Wolf (1975). Following Yang and Liou (1995) and Liou (2002), we
present Snell’s and Fresnel’s laws under absorption conditions in terms of modified real
and imaginary indices of refraction. See also Dupertuis et al. (1994) and Chang et al.
(2005) for discussions of inhomogeneous wave properties in the context of geometric
ray-tracing.
Referring to Figure 3.3, consider the propagation of an incident wave from air into
ice. The wave vectors associated with the incident and reflected waves are real quantities
because these waves, which are outside the ice medium, must have the same properties.
However, the wave vector of the wave refracted into the ice is complex, and the result is
referred to as the inhomogeneity effect. The three wave vectors can be defined by
ki = kei , kr = ker , kt = kt et + ikα eα , (3.1.11)
where ei , er , et , and eα are unit vectors defined in Figure 3.3; k = 2π/λ, where λ is the
wavelength in air; and kt and kα are two real parameters which determine the complex
wave vector of the refracted wave. For non-absorptive cases, kα is zero. The corresponding
electric vectors can be expressed as
Ei (r) = Ai exp(ikr · ei ), (3.1.12a)

Er (r) = Ar exp(ikr · e ), r
(3.1.12b)
Et (r) = At exp(ikr · e + ikα r · e ),
t α
(3.1.12c)
where Ai , Ar , and At denote amplitude vectors. Furthermore, we may define the following
normalized parameters:
Nr = kt /k, Ñi = kα /k. (3.1.13)
At the interface of the two media, where the position vector is denoted rs , the phases
must be the same for incident, reflected, and refracted waves. Thus, from Eqs. (3.1.11) and
(3.1.13), we obtain
ei · rs = er · rs = Nr (et · rs ) + i Ñi (eα · rs ). (3.1.14)
Because the wave vectors for the incident and reflected waves are real, we must have
ei · rs = er · rs = Nr (et · rs ), eα · rs = 0. (3.1.15)
On the basis of the geometry defined by Eq. (3.1.15), a generalized form of Snell’s law
can be derived and is given by
sin θi
sin θi = sin θr , sin θt = , (3.1.16)
Nr
where θi , θr , and θt denote the incident, reflected, and refracted angles, respectively
(Figure 3.3). The vector eα in Eq. (3.1.15) is normal to the interface of the two media.
It follows that the planes of constant amplitude of the refracted wave are parallel to the
interface. To determine Nr and Ñi , we use the electric field of the refracted wave, which
must satisfy the wave equation in the form [see Eq. (2.2.9)]
∇ 2 Et (r) + k 2 (mr − imi )2 Et (r) = 0, (3.1.17)
where mr and mi are real numbers representing the real and imaginary parts of the refractive
index, respectively. Substituting Eq. (3.1.12c) into Eq. (3.1.17) and using Eq. (3.1.13) leads
to the following relationships:
Nr2 − Ñi2 = m2r − m2i , Nr Ñi cos θt = mr mi . (3.1.18)
If we let Ni = Ñi cos θt , then we obtain from Eqs. (3.1.16) and (3.1.18) the following
expressions:
√
2( 2 ) 2 *1/2 +1/2 mr mi
Nr = mr − m2i + sin2 θi + m2r − m2i − sin2 θi + 4m2r m2i , Ni = .
2 Nr
(3.1.19)
These two parameters are referred to as the adjusted real and imaginary indices of refraction.
(a) (b)
110o
130o
Figure 3.4 Solid lines are reflection/refraction determined from effective indices of refraction, while
dotted lines are based on Snell’s law, given by sin θt,1 = sin θi,1 /mr and sin θt,j = mr sin θi,j , j =
2, 3, 4 . . . The ice crystal’s complex index of refraction is (a) 1.3857 − i 0.422; (b)1.0925 − i 0.248
(after Yang and Liou 2009).
The refracted wave given in Eq. (3.1.12c) can now be rewritten in the form
Et (r) = At exp(−kNi la + ikNr et · r), (3.1.20)
where la = (eα · r)/ cos θt is the distance of propagation of the refracted wave along the
direction et . The direction of the phase propagation for the inhomogeneous wave inside the
medium is determined by Nr via Snell’s law, whereas the attenuation of wave amplitude
during propagation is determined by Ni . Consequently, the refracted wave can be precisely
traced. The Fresnel reflection and refraction coefficients defined in Eqs. (3.1.9a, b) in terms
of the adjusted real and imaginary indices of refraction can then be written as follows:
Nr cos θi − cos θt 2 cos θi
Rl = , Tl = , (3.1.21a)
Nr cos θi + cos θt Nr cos θi + cos θt
cos θi − Nr cos θt 2 cos θi
Rr = , Tr = . (3.1.21b)
cos θi + Nr cos θt cos θi + Nr cos θt
In these equations, the subscripts l and r denote the parallel and perpendicular polarized
components, respectively.
Yang and Liou (2009) further developed recurrence formulas for the effective indices of
refraction associated with high-order reflection/refraction events when the ray transmission
is from an ice crystal to air. Figure 3.4a depicts a 2-D case for propagation of an incident ray
into an ice hexagon using a complex index of refraction of 1.3857 − i 0.422, corresponding
to a 12 µm wavelength, while Figure 3.4b is for a complex index of refraction of 1.0925 −
i 0.248, corresponding to an 11 µm wavelength. The incident angles for the first-order
reflection/refraction event in the computation were set at 65° and 55° (i.e., θi,1 + θr,1 =
130◦ and 110°), as shown in Figure 3.4a and Figure 3.4b, respectively. The solid lines
show the results computed using the effective indices of refraction for corresponding
orders of reflection/refraction events. The dotted lines depict the results computed from
Table 3.1 The real and imaginary parts of the effective index of refraction for
Figure 3.4a (case 1) and Figure 3.4b (case 2).
Case 1 Case 2
mr = 1.3857 and mi = 0.422 mr = 1.0925 and mi = 0.248
θi,1 = 65◦ θi,1 = 55◦
Order of reflection/
refraction event, j Nr,j Ni,j Nr,j Ni,j
1 1.423 4 .410 83 1.121 3 .241 63

2 1.394 2 .419 44 1.094 0 .247 66
3 1.394 2 .419 44 1.094 0 .247 66
4 1.420 2 .411 75 1.094 0 .247 66
5 1.395 3 .419 10 1.118 3 .242 27
the simplified Snell’s law given by θi,j = θr,j , j = 1, 2, 3 . . .; sin θt,1 = sin θi,1 /mr ; and
sin θt,j = mr sin θi,j , j = 2, 3, 4 . . . under the condition that total reflection does not occur.
The effect of wave inhomogeneity on the propagation of a localized ray is substantial for
higher-order reflection/refraction events. However, its effect on the scattering properties of
the particle is relatively small because it is significant only when an ice particle’s absorption
is substantial. For a strongly absorbing particle, the energy carried by a higher-order ray
is much lower than that carried by a lower-order ray, in view of the fact that a particle’s
absorption coupled with external reflection dominate the scattered field.
In Table 3.1 the real and imaginary parts of the effective index of refraction are listed for
cases 1 and 2, corresponding to Figures 3.4a and 3.4b, respectively. The terms Nr,j and Ni,j
are larger and smaller than mr and mi , respectively. The dependence of the effective index
of refraction on ray history is also evident from the results listed in the table. There is no
systematic change in Nr,j and Ni,j as a function of the order of reflection/refraction events.
By matching these events for the results in this table with the geometries displayed in
Figure 3.4, the effective indices of refraction for two sequential internal reflections are the
same if two hexagonal faces for these reflections are parallel. In summary, the effective
indices of refraction are dependent on both ray history and particle shape.
3.1.3 Geometric ray tracing and the Monte Carlo hit-and-miss approach
to ice crystals
Use of the Monte Carlo method in connection with geometric ray-tracing was first developed
by Wendling et al. (1979) for hexagonal ice columns and plates. See also Marchuk and
Mikhailov (1967). Takano and Liou (1995) and Yang and Liou (1998a) further applied
the hit-and-miss Monte Carlo method to trace photons in complex ice crystals, including
contributions from absorption and polarization.
(a) (b)
(c) (d)
Figure 3.5 Positions of incident photons on each crystal, determined using random numbers:
(a) hollow column, (b) bullet rosette, (c) dendrite, and (d) capped column (after Takano and Liou
1995). See Subsection 1.5.3 for the computer-generated ice crystal sizes and shapes.
We may consider a bundle of parallel rays, representing a flow of photons, which

strike an ice crystal from a direction defined by a set of two angles (α, β), where α is the
complementary angle between the incident direction and the crystal’s c-axis and β is the
azimuthal angle with respect to this axis. Consider also a plane normal to this bundle of
incident rays and the geometric shadow of an ice crystal projected onto this plane. We may
let a rectangle (defined by dimensions X and Y) enclose this geometric shadow, such that
the center of this rectangle coincides with the center of the crystal. One of the sides, X,
is parallel to the geometric shadow of the crystal’s principal axis. A point (xi , yi ) can be
subsequently selected inside this rectangle using random numbers, RN (0–1), such that

xi = X RN − 12 , yi = Y RN − 12 . (3.1.22)
In this manner, xi varies from −X2 to X2, while yi varies from −Y2 to Y2. If the point
is inside the geometric shadow of the crystal, it is regarded as an incident point on the
crystal; otherwise, it is disregarded. If there are more than two crystal planes for a photon,
the point closer to the light source is regarded as the incident point.
Figures 3.5a–3.5d show the positions of incident photons for hollow column, bullet
rosette, dendrite, and capped column. The coordinates of an incident point (xi , yi ) in these
figures can be transformed to the coordinates (x, y, z) with respect to the body-framed
coordinate system (Takano and Asano 1983) and can be expressed in terms of xi , yi , α, β,
and one unknown, zi , in the form
⎡ ⎤ ⎡ ⎤⎡ ⎤
x sin α cos β −sin β cos α cos β xi
⎣ y ⎦ = ⎣ sin α sin β cos β cos α sin β ⎦ ⎣ yi ⎦ . (3.1.23)
z −cos α 0 sin α zi
A crystal surface can generally be defined by the following equation:
a0 x + b0 y + c0 z − d0 = 0, (3.1.24a)
where the coefficients a0 , b0 , c0 , and d0 are determined by the coordinates of three

points on the crystal surface. In particular, a plane containing the points P1 (x1 , y1 , z1 ),
P2 (x2 , y2 , z2 ), and P3 (x3 , y3 , z3 ) can be expressed as

x − x1 y − y1 z − z1

x1 − x2 y1 − y2 z1 − z2 = 0. (3.1.24b)

x − x y − y z − z
2 3 2 3 2 3
For example, one of the side planes of a dendritic crystal branch √ can be defined by an
equation obtained
√ from three of the four points: [−(a + bb)/2, 3(a − bb)/2, ±L/2] and
[−(a + bt)/2, 3(a + bt)/2, ±L/2], where +L/2 denotes the upper point and −L/2
denotes the lower point. See Figure 1.19 for the graphic definitions of L, a, bt, bb, t, d,
H, and other notation defining the four ice crystal shapes depicted in Figure 1.22. On
substituting (x, y, z) in Eq. (3.1.23) for (x, y, z) in Eq. (3.1.24a), the unknown, zi , can be
evaluated. Using this zi and (xi , yi ), the coordinates of the incident point (x, y, z) can be
calculated using Eq. (3.1.23).
Once the incident coordinates have been determined, the photons are traced using the
hit-and-miss Monte Carlo method. The Fresnel reflection coefficients, Rl and Rr , are first
calculated and compared with a random number, RN. If (|Rl |2 + |Rr |2 )/2 is greater than
RN, the photon is reflected; otherwise, it is transmitted. When a photon traverses a crystal, it
can be absorbed. We can account for absorption by means of a stochastic procedure. When
a photon enters a crystal, an absorption length, la , is generated from a random number such
that
ln(RN)
RN = exp(−2kNi la ), i.e., la = − . (3.1.25)
2kNi
The random number represents the probability of transmission of the photon. The absorption
path length, la , represents the distance traversed by the photon in the crystal before it is
absorbed. An actual path length, l, between an incident point and the next internal incident
point can then be calculated on the basis of Snell’s law and the specific ice-crystal geometry.
The transmission is given by T = exp(−2kNi l). If T ≤ RN ≤ 1, then the photon associated
with this RN is absorbed. Similarly, if l is greater than la , then the photon is absorbed;
otherwise, it is transmitted without absorption. This procedure is repeated as long as photons
travel inside the crystal.
After a photon is transmitted out of the crystal or reflected externally, it can re-enter
the crystal depending on the crystal’s shape. In this case, a new incident direction can be
calculated using the direction cosine of the scattered beam. The new incident coordinates
can also be determined from the new incident direction and the coordinates of an emergent
point of the photon on the crystal surface. The foregoing procedure is repeated until the
photon escapes from the crystal. When a photon re-enters the crystal, the scattering angle
and the scattering matrix are computed with respect to the original incident direction. In the
conventional method, the number of scattered photons per unit solid angle, 2π sin θ θ, is
counted as the phase function. The single-scattering albedo is obtained from the ratio of the
number of scattered photons to the number of incident photons. The Monte Carlo method
allows us to treat complicated ice crystals effectively and can be employed in connection
with the improved geometric ray-tracing approach.
The surfaces of ice crystals may not be exactly smooth, particularly if they undergo
collision and riming processes. Also, a careful electron-microscopic examination of some
polycrystalline ice crystals reveals rough structures on the surfaces. Halo and arc patterns
that are absent from some cirrus clouds could be caused by deviations of the ice crystal
surfaces from defined hexagonal structures. Some aspects of ice crystal surface roughness
can be incorporated into geometric ray-tracing by following an idea developed by Cox and
Munk (1954) for wavy sea surfaces. A rough surface may be thought of as comprised of
a number of small facets that are locally planar and randomly tilted flat surfaces. We may
use a two-dimensional Gaussian probability function to define the surface tilt as follows:
% &
1 zx2 + zy2
p(zx , zy ) = exp − , (3.1.26a)
πσ 2 σ2
where σ is a parameter determining the degree of roughness and the parameters zx and zy
are the slopes defined for a facet of rough surface along two orthogonal directions, given
by
∂z ∂z
zx = = tan θ cos φ, zy = = tan θ sin φ, (3.1.26b)
∂x ∂y
where θ and φ are the local polar angles defining the position of the tilt of the surface facet.
In general, surface roughness of ice particles has the effect of smoothing out the scattering
maxima that occur in the phase function (see Subsection 3.4.2).
3.1.4 Illustration of ice optics in the atmosphere

The geometric-optics approach to ray tracing provides a means of identifying all the fasci-
nating cloud optics we see in the atmosphere on a daily basis, as introduced in Section 3.1.
Rainbow and glory features produced by scattering by water drops can be explained using
geometric ray tracing. Ice crystals are much more complex in view of their non-spherical
and irregular shapes, as displayed in Section 1.5. Ice crystal non-sphericity also leads to
orientation issues, which cannot be directly observed using in situ sampling techniques.
(a) (b) A Incident ray

θ'

θi θ'i
22o Halo θt θ't

Scattered ray
46o Halo
Figure 3.6 (a) The production of 22° and 46° halos from two refractions through a plate crystal
with prism angles A of 60° and 90°, respectively. (b) Definition of the minimum deviation, which
generates the maximum intensity; see text for the definition of the angles in this diagram.
Moreover, it should be pointed out that geometric optics is the only approach in light scatter-
ing that will provide physical explanations of the features that occur in the phase function
computed from “exact” solutions (e.g., the Lorenz–Mie theory) or numerical methods
(e.g., finite-difference time domain method). The fascinating optical phenomena produced
by defined ice crystals are numerous, and are presented in the following subsections.
3.1.4.1 22° and 46° halos

First, we discuss the production of 22° and 46° halos, which occur when ice crystal clouds
are present in the atmosphere. Referring to Figure 3.6, if a light beam from the sun (or
moon) passes through a prism of angle A in a plane at right angles to the refracting edge,
the deviation angle is
θ = (θi − θt ) + (θi − θt ) = 2θi − A. (3.1.27a)
The minimum deviation (maximum intensity) occurs when

dθ dθ i
=0=1+ . (3.1.27b)
dθi dθi
Since A = θt + θt , we also have 1 + dθt /dθt = 0. An obvious solution of the last two
equations is when θi = θi and θt = θt . Thus, from Eq. (3.1.27a), the incident angle at which
minimum deviation occurs is θi = (θ + A)/2. Also, the angle of refraction at minimum
deviation is given by θt = A/2. Applying Snell’s law as denoted in Eq. (3.1.1) yields
sin[(θ + A)/2] = m sin A/2. (3.1.28)
This equation represents the basic relationship for finding the minimum deviation from
a known index of refraction and prism angle. As illustrated in Figure 3.6, the possible prism
angles are 60°, 90°, and 120°. However, owing to the property of the sine function and the
fact that the refractive index of ice is about 1.31 in visible wavelengths, a prism angle of
120° cannot produce minimum deviation. The common halo has an angular radius of 22°,
produced by two refractions in a hexagonal prism. The halo of 46° is generated by two
refractions in a rectangular prism. Since the index of refraction varies with wavelength,
white light is dispersed into its component colors, with red refracted least and blue refracted
most. Halos are seen in the form of circles because of the random orientations of hexagonal
prisms in the atmosphere.
3.1.4.2 Sundogs and their geometry

Second, microphysics observations from laboratory ice clouds have revealed that hexagonal
plates and long columns fall through the air with their major axes parallel to the ground and
would most likely have a random orientation in the horizontal plane. A light beam which
reaches a sheet of ice columns or plates randomly oriented in the horizontal plane will then
produce bright spots along conical circles at the angle of minimum deviation around the
sun or the moon. When the sun is close to the horizon and thin cirrus clouds are present,
colored streaks may sometimes be observed at greater than 22°, at the same elevation as the
sun. This optical phenomenon is referred to as the parhelia of 22° or, commonly, sundogs
or mock suns. Sundogs consist of only two bright spots, since other light from the sun
deviated by clouds higher above the horizon cannot be observed. To evaluate their position,
we shall now refer to Figure 3.7, which illustrates the relevant geometry of a plate crystal
with respect to the sun’s position.
Let the lines AO and A O in Figure 3.7a be the incident (skew ray) and refracted
directions, respectively, and consider the enclosure AOCB. We can determine the following
relationships:
sin εi sin εt
sin θi = , sin θt = . (3.1.29a)
sin sin
The elevation angle εi denotes the position of the sun, and other angles are defined in the
figure. Substituting Eq. (3.1.29a) into mr sin θt = sin θi [Eq. (3.1.1), Snell’s law], we obtain
mr sin εt = sin εi . (3.1.29b)
On the basis of the geometry of the incident (skew ray) and refracted rays defined in
Figure 3.7a, we can further show that
cos θi = cos εi cos φi , cos θt = cos εt cos φt . (3.1.30)
Eliminating θ i and θ t from Eq. (3.1.30) and the Snell equation, we obtain
mr (cos εt /cos εi ) sin φt = sin φi . (3.1.31)
Set mr (cos εt /cos εi ) = mr on the left-hand side of Eq. (3.1.31). Then, from Eq. (3.1.29b),
this term can be rewritten as follows:

m2r − sin2 εi
mr = . (3.1.32a)
cos εi
(a) A

Skew ray
εi A'
θi Δ
Q θt

Z εt B
Sundog B'
O φi φt C
(> 22o halo)
OQBC XYZ
Y
X A'B', AB OQBC
(b) A
Emergent ray
Incident ray leaving prism

θ'/2 B
εi D
θ'h /2
O
Figure 3.7 (a) The geometry of a skew ray that produces a sundog in a 3-D plate crystal, and
the definition of the sun’s elevation angle εi . (b) The definition of incident and emergent rays with
reference to the elevation angle and the minimum deviation, and their mapping onto a horizontal
plane.
When mapping on a horizontal plane, Eq. (3.1.31) can also be expressed in the form
mr sin φt = sin φi . (3.1.32b)
Referring to Eq. (3.1.27b), the 22° parhelion occurs if the following conditions are met:
(φi = φi ; φt = φt ). If A is the angle of the prism, we have the following relationship:
φt + φt = A. (3.1.33)
Using Eq. (3.1.33), the minimum deviation angle projected onto the plate’s basal plane
(QOCB) is given by
θh = φi − φt + φi − φt = 2φi − A. (3.1.34)
Thus, we obtain
φi = (θh + A)/2. (3.1.35)

From Eq. (3.1.33), we have φt = A/2. On substituting this and Eq. (3.1.35) into
Eqs. (3.1.31) and (3.1.32b), we obtain

1 A
sin (θh + A) = mr (cos εt /cos εi ) sin , (3.1.36a)
2 2

1 A
sin (θh + A) = mr sin . (3.1.36b)
2 2
Moreover, based on the geometry depicted in Figure 3.7b, we can show that
cos θ = sin2 εi + cos2 εi cos θh . (3.1.37a)
This equation can be rewritten as follows:

θ θ
sin = cos εi sin h . (3.1.37b)
2 2
It follows that the angles θh and θ can be computed from Eqs. (3.1.29b), (3.1.36a), and
(3.1.37b). For the prism angle A = 60° and the sun’s elevation angle εi = 30°, the angular
and azimuthal distances θh and θ for a sundog are 24.54° and 28.41°, respectively, for
the 0.656 µm red wavelength and 25.40° and 29.41°, respectively, for the 0.405 µm violet
wavelength.
3.1.4.3 Horizontally oriented plate and column crystals

Third, in the case of horizontally oriented (2-D) plates, the scattered sunlight is confined
to four latitude belts due to specific geometry. Based on geometric ray-tracing, the zenith
angle θ ∗ can be computed from the incoming solar zenith angle θi in the forms (Takano
and Liou 1989b)
1/2 1/2
∗
π/2 − sin−1 m2r − sin2 θi , for θi > sin−1 m2r − 1 58◦
θ = 1/2 1/2 (3.1.38)
sin−1 m2r − cos2 θi , for θi < cos−1 m2r − 1 32◦ .
For example, if θi = 75◦ , the latitude belts correspond to zenith angles of
±75◦ and ± 27◦ with negative values representing mirror images. If the incident angle
is 27°, the four latitude belts are ±27◦ and ±75◦ . Due to the symmetrical property of 2-D
plates (randomly oriented with their c-axes vertical) with respect to the incoming light
beam, all scattered light is also confined to the four latitude belts. At the 75° emergent
zenith angle, the subsun, subsundog, 120° subparhelion, and antisolar peak optical features
can be observed. In addition to these optical phenomena, the anthelion (AN) located at the
180° azimuthal angle can also be observed, due to the scattering and coupling of the subsun
and antisolar peaks. At the 27° zenith angle, the Kern’s arc (KA) can be seen. The subsun
feature and sun pillars are produced by external reflections from the flat horizontal faces of
ice plates, and appear as streaks of white light above and/or below the sun.
The subpeak at θi = 82◦ in the transmitted light is caused by the lower sunvex Parry
arc of the subsun. For Parry columns (the c-axis and a pair of prisms facing horizontally),
Halos and arcs related to 22o

22o Halo, Upper & Lower 22o Halo, Sundog Subsundog
Tangent Arcs, Parry Arcs

Halos and arcs related to 46o

46o Halo 46o Halo 46o Halo 46o Halo
Circumzenithal Arc Circumhorizontal Arc Supralateral Arc Infralateral Arc

Other halos and arcs

Subsun Parhelic Circle Subparhelic Circle

120o Parhelion Helic Arc Subhelic Arc Wegener’s & Hastings’

Anthelic Arc

Figure 3.8 A summary of the origins of frequently observed halos and complex arcs produced by
horizontally oriented columns and plates, in terms of reflection and refraction. Dots denote points of
incoming, refracted, and outgoing light rays, while dashed lines denote ray paths inside a plate or
column crystal.
several optical features are observable. These include the subsun (SS) and lower sunvex
Parry arc (LSVP) in the reflected light, and the circumzenith arc (CZA) and upper suncave
(USCP) and sunvex (USVP) Parry arcs in the transmitted light. For 2-D columns (rotational
orientation about the c-axis in a horizontal plane), the lower tangent arc (LTA) is noticeable
in the reflected light. Aside from these features, the reflected and transmitted light of 2-D
columns is similar to that for randomly oriented columns.
The physical causes of the preceding optical features generated by “smooth” 2-D plate
and column crystals and other crystal types are displayed in Figure 3.8 in terms of external
reflections, two refractions, and internal reflections, based on the principles of geometric
optics. These features are summarized in three major groups: 22° halos and associated arcs,
46° halos and associated arcs, and other complex halos and arcs.
(a) Observation
Simulation
(b) q
p
m
n
o
l
k
f
g
e
i
b
d
h
j
c
r
a
Figure 3.9 (a) Halo and arc complex observed in Antarctica by Riikonen (1999). (b) A geomet-
ric ray-tracing halo simulation program developed by Les Cowley and Michael Schroeder (2009,
HaloSim computer ray tracing, Les Cowley – http://www.atoptics.co.uk) was employed for simula-
tion purposes. We have identified (a) 22° halo, (b) 46° halo, (c) 22° parhelion, (d) upper tangent arc,
(e) Parry arc, (f) circumzenithal arc, (g) supralateral arc, (h) infralateral arc, (i) Parry supralateral arc,
(j) Parry infralateral arc, (k) helic arc, (l) subhelic arc, (m) parhelic circle, (n) Wegener arc, (o) 120°
parhelion, (p) anthelion, (q) diffuse arc, and (r) a diamond shape, which is a Fraunhofer diffracted
pattern overlapping a weak sun pillar (courtesy of Yoshi Takano). See Figure 3.8 for the causes of
these fascinating halos and arcs.
3.1.4.4 An observed halo and arc complex pattern

Fourth, we present halo and arc complex patterns observed in Antarctica by Riikonen
(1999), as depicted in Figure 3.9a. Most of these patterns are rare and do not appear in
daily observations; however, they cover essentially all possible optical features produced by
ice crystals in the Earth’s atmosphere. Geometric ray-tracing computer simulation results
are displayed in Figure 3.9b, where the names of these fascinating optical phenomena
are: (a) 22° halo, (b) 46° halo, (c) 22° parhelion, (d) upper tangent arc, (e) Parry arc,
(f) circumzenithal arc, (g) supralateral arc, (h) infralateral arc, (i) Parry supralateral arc,
(j) Parry infralateral arc, (k) helic arc, (l) subhelic arc, (m) parhelic circle, (n) Wegener arc,
(o) 120° parhelion, (p) anthelion, (q) diffuse arc, and (r) a diamond shape, which is a
diffracted pattern overlapping a weak sun pillar. Physical causes of these features in terms
of reflection and/or refraction are illustrated in Figure 3.8.
3.2 Fraunhofer diffraction for light scattering by ice crystals

The theoretical development of diffraction begins with Babinet’s principle, which states
that the diffraction pattern in the far field, i.e., Fraunhofer diffraction (Fraunhofer 1821),
from a circular aperture is the same as that from an opaque disk or a sphere of the same
radius. Babinet (1837) stated the diffraction principle by citing the corona produced by
water droplets in the atmosphere. Babinet’s principle can take scalar or vector form.
The vector form accounts for the sign of electric and magnetic fields. However, when
the product of wavenumber and diffracting particle size is larger than about 1, the scalar
Babinet principle is sufficient for analysis (Jackson 1975). Moreover, the conditions for,
and limitations on, Fraunhofer diffraction have been discussed by Born and Wolf (1975).
The diffracted images for circular and rectangular apertures were presented in Born and
Wolf (1975). Diffracted images for other polygonal apertures were studied by Komrska
(1972, 1973) and Smith and Marsh (1974). Cai and Liou (1982) performed the computation
of diffracted intensity for randomly oriented hexagonal cylinders, while Takano and Asano
(1983) studied diffraction phenomena produced by randomly or horizontally oriented ice
crystals.
Liou et al. (1983b) extended their numerical computations on diffraction by hexagonal
cylinders to account for cubes and parallelepipeds. Takano and Liou (1989a) illustrated the
manner in which diffracted intensity can be added to reflected and/or refracted intensity
when the phase function for ice crystals produces a delta-function transmission at the 0°
forward scattering direction. Parviainen et al. (1994) discussed the connection of vertically
elliptical coronas with horizontally oriented pollens. While carrying out computation of the
Fraunhofer integral, Tränkle and Mielke (1994) simulated corona patterns using elliptical
shapes and produced strong brightening for birch and pine pollens. Additionally, Bi et al.
(2011a) showed good agreement between results determined from Fraunhofer diffraction
and a surface-integral method for an absorbing cube. Also, Hesse et al. (2012) presented
polar scattering patterns of diffracted and externally reflected intensities for a strongly
absorbing cube.
3.2 Fraunhofer diffraction by ice crystals 135
x
y
Plane waves
O' r
P
φ
r0
θ
z
O
r0 - r
A
u0 e-ikz
Figure 3.10 Diffraction of plane waves, denoted by u0 e−ikz , by a circular aperture with geometrical
area A, evaluated at the position P. The shaded area is perpendicular to the aperture defined at the x-y
plane. See text for notation definition.
In Subsection 3.2.1, we shall first review Fraunhofer diffraction from a circular

aperture.
3.2.1 Sphere
Let the z-axis be in the direction of propagation of the incident light, and let the wave
disturbance occur at a point P to the right of a geometric aperture of area A (Figure 3.10).
The distances from point P to point O (x, y) on the aperture area and to the origin O are
denoted r and r0 , respectively. In the far field, the light-wave disturbance at P can be derived
from Fraunhofer diffraction theory following Born and Wolf (1975) as follows:

up = −uc e−ikr dxdy, (3.2.1)
A
where uc = iu0 /rλ, u0 represents the disturbance in the original wave at point O on the
plane wave front, k = 2π/λ, and λ is the wavelength. The phase difference at P between
waves passing through points O and O can be expressed in the form
δ = k(r − r0 ) = k(x cos φ + y sin φ) sin θ, (3.2.2)
where θ is the angle between lines OP and the z-axis, and φ is the azimuthal angle on the
aperture. Substituting Eq. (3.2.2) into Eq. (3.2.1), we obtain

−ikr0
up (sphere) = uc e exp[−ik(x cos φ + y sin φ) sin θ ]dxdy. (3.2.3)
A
For a circular aperture, we can transfer rectangular coordinates (x, y) to polar coordinates
(ρ, ψ) such that x = ρ cos ψ and y = ρ sin ψ to obtain
a 2π
up = −uc e−ikr0 exp[−ikρ cos(ψ − φ) sin θ ]ρdρdψ. (3.2.4)
0 0
Using the zero-order Bessel function defined by

2π
1
J0 (y) = eiy cos α dα, (3.2.5)
2π 0
Eq. (3.2.4) can be expressed in the form
a
up = −uc e−ikr0 2π J0 (kρ sin θ )ρdρ. (3.2.6)
0
Moreover, Bessel functions have the following well-known recurrence relationship:

d
[yJ1 (y)] = yJ0 (y), (3.2.7)
dy
where J1 (y) is the first-order Bessel function. Upon integration, we have
y
y J0 (y )dy = yJ1 (y). (3.2.8)
0
Thus, from Eqs. (3.2.6) and (3.2.8), we obtain

up (sphere) = −2uc e−ik0 AJ1 (y)/y, (3.2.9)
where the geometric shadow area A = πa 2 , y = x sin θ, and the size parameter x = ka.
The scattered intensity in terms of the incident intensity I0 = |u0 |2 is then given by
ip
Id = |up |2 = I0 , (3.2.10)
k2r 2
where the diffraction angular intensity function for a single sphere is defined by
x4
ip = [2J1 (y)/y]2 . (3.2.11)
4
Diffraction depends only on the particle size parameter and is independent of the index
of refraction. If we display [2J1 (y)/y]2 = D 2 versus y, a principal maximum of 1 at y = 0
(i.e., θ = 0) will show and oscillates with gradually diminishing amplitude as y increases.
When J1 (y) = 0 and D 2 = 0, a number of minima diffraction patterns are produced. The
positions of maxima are defined by the values of y that satisfy
d 2 d
D = [2J1 (y)/y]2 = 0. (3.2.12)
dy dy
The maxima in the diffraction patterns have been used to explain the position of the corona
produced by the diffraction of uniform sized spherical aerosols.
The following analytical solutions and analyses are based on the presentation given in
Takano et al. (2012).
3.2.2 Rectangular parallelepiped

Consider a plane wave incident on a rectangular parallelepiped from a direction with zenith
angle (π2 − α) and azimuthal angle β measured with respect to the coordinate system
(a) (b) y’
y’
5 y2 6
x’=a1y’-b1 x’=a2y’+b2
4 y2 8

x’ = a1y’ - b1 x’ = a2y’ + b2 y1
1
4
2a
1 y1 x’=a3y’+b 3
1’ 5
3 2
2c z’
z’
O
O x’ ψ x’
2b
7
7’
3
7
10
2 2c 2a
6
9 8
y’
(c) x
(d)
v

a z’ y
ζ
b
x’ p
v cosζ q

c z O η c x’
ζ O z
ξa
y’ a
x
Figure 3.11 These four figures depict the geometric shadow of (a) a rectangular parallelepiped
normal to the incident direction; (b) a hexagonal cylinder normal to the incident direction (after
Takano and Asano 1983); (c) a spheroid along the semi-major axis with the semi-major and semi-
minor axes denoted by c and a, respectively, where c is along the rotational axis defined by z and the
angle ζ is defined in the figure; and (d) a three-axis ellipsoid projected onto a plane normal to the
incident direction (after Takano et al. 2012).
fixed to the parallelepiped, as shown in Figure 3.11a, where the x-, y-, and z-axes are
directed, respectively, through the centers of three sets of rectangular planes. The origin
O of the coordinate system is taken to be the center of the parallelepiped. The geometric
shadow of the parallelepiped as projected onto a plane normal to the incident direction can
be expressed in terms of the (x , y ) coordinates of its marginal vertices in a new coordinate
system where the z -axis is taken along the incident direction.
The transformation of the body-framed coordinate system X(x, y, z) into the new coor-
dinate system X (x , y , z ) can be written in the form
X = CDX, (3.2.13)
where the transformation matrices are given by

⎡ ⎤
sin α 0 −cos α
C=⎣ 0 1 0 ⎦, (3.2.14)
cos α 0 sin α
⎡ ⎤
cos β sin β 0
D = ⎣ −sin β cos β 0 ⎦ . (3.2.15)
0 0 1
Thus, if the size and shape of the parallelepiped and the propagation direction (α, β)
of the incident wave are known, the coefficients a1 , a2 , b1 , b2 , y1 , and y2 defined in
Figure 3.11a can be determined by Eq. (3.2.13). The shape of a parallelepiped is defined by
lengths a, b, and c, which are half-lengths of a rectangular parallelepiped along the three
axes. The size parameter is given by ka. Having defined a number of geometric variables,
we may rewrite the diffraction integral equation of Eq. (3.2.3) in the form

−ikr0
up (parallelepiped) = uc e exp[−i(Xx + Y y)] dxdy, (3.2.16)
Ap
where
X = k sin θ cos φ, Y = k sin θ sin φ. (3.2.17)
The integration of Eq. (3.2.16) over the geometric shadow area bounded by (2-3-4-
8-5-6-2), defined in Figure 3.11a, can be carried out by dividing the entire integration
domain into three sub-domains, including two trapezoids [(1 -4-8-5-1 ) and (3-7 -6-2-3)]
and one parallelogram (1 -5-7 -3-1 ). In this manner, we can follow the integration method
presented in Smith and Marsh (1974) to carry out these integrations, after which the
diffraction amplitude up /(uc e−ikr0 ) of a parallelepiped defined by the orientation angles
(α, β) at an arbitrary point P(θ, φ) can be obtained in the form
up (θ, φ; α, β; ka, b/a, c/a)/(uc e−ikr0 )
4 ) * ( +
= sin 12 (y2 − y1 )(a1 X + Y ) sin 12 [−a1 (y1 − y2 ) + 2b1 ] X − 12 (y1 + y2 )
X(a1 X + Y )
4 ) * ( +
+ sin 12 (y2 − y1 )(a2 X + Y ) sin 12 [a2 (y1 + y2 ) + 2b2 ] X + 12 (y1 + y2 )
X(a2 X + Y )
4
+ sin [y1 (a1 X + Y )] sin(b1 X) (3.2.18)
X(a1 X + Y )
where a1 , a2 , b1 , b2 , y1 , and y2 are explicit functions of α, β, a, b, and c as defined in the
following equations:
a1 = sin α tan β, a2 = − sin α/tan β, (3.2.19a)
b1 = sin α tan β(a sin β − b cos β) + a sin α cos β + b sin α sin β + c cos α, (3.2.19b)
b2 = −(sin α/tan β)(a sin β − b cos β) + a sin α cos β + b sin α sin β + c cos α,
(3.2.19c)
y1 = −a sin β + b cos β, y2 = a sin β + b cos β. (3.2.19d)
Using these relationships, the intensity function for a given orientation of a rectangular
parallelepiped, defined by angles α and β, can then be expressed as
2
k2
ip (θ, φ; α, β; ka, b/a, c/a) = |up /uc e−ikr0 |2 . (3.2.20)
2π
In the case of a rectangular aperture, we may define the geometric shadow area as 2aL,
where 2a and L are the lengths of the rectangle. Referring to Eq. (3.2.3), we may set the
integrating boundaries x and y to be from −a to a and from −L2 to L2, respectively. A
separation of the two integrations leads to
a L/2
up (rectangle) = uc e−kr0 exp(−ikx cos φ sin θ )dx exp(−iky sin φ sin θ )dy
−a −L/2
sin(ka cos φ sin θ ) sin [k(L/2) sin φ sin θ ]

= uc e−kr0 4a(L/2) .
ka cos φ sin θ k(L/2) sin φ sin θ
(3.2.21)
The solution of diffraction due to the light-wave disturbance associated with a rectangle
in the far field is given by the product of sine functions. It follows that the intensity function
for diffraction can be expressed as
4k 4 a 2 (L/2)2 sin2 (ka cos φ sin θ ) sin2 [k(L/2) sin φ sin θ ]

ip [θ, φ; ka, k(L/2)] = .
π2 (ka cos φ sin θ )2 [k(L/2) sin φ sin θ ]2
(3.2.22)
Alternatively, if we let α = 0, β = 0, a = b, and c = L2 in Eq. (3.2.18), we may also

derive Eq. (3.2.22) directly.
3.2.3 Hexagonal cylinder

Consider a plane wave incident on a hexagonal cylinder from a direction with the zenith
angle (π/2 − α) and the azimuthal angle β measured in reference to the coordinate system
fixed to the cylinder, where the x- and z-axes are directed, respectively, along the a- and
c-axes of the hexagonal cylinder. The origin point O of the coordinate system is fixed at
the center of the cylinder. The geometric shadow of the cylinder projected onto a plane
normal to the incident direction is expressed in terms of the (x , y ) coordinates of its
marginal vertices in a new coordinate system such that the z -axis is taken to be along the
incident direction (see Figure 3.11b). The transformation of the body-framed coordinate
system X(x, y, z) into the new coordinate system X (x , y , z ) can then be written in the
form
X = BCDX, (3.2.23)
where the transformation matrixes C and D have been defined in Eqs. (3.2.14) and (3.2.15),
and the B matrix is given by

⎡ ⎤
cos ψ sin ψ 0
B = ⎣ −sin ψ cos ψ 0⎦. (3.2.24)
0 0 1
The relationship among the three angles α, β, and ψ is defined by the following equation:
π ,
tan ψ = − tan β + sin α, (0 ≤ ψ ≤ π). (3.2.25)
3
Thus, if the size and shape of a hexagonal cylinder and the propagation direction (α, β)
of the incident wave are known, the coefficients a1 , a2 , a3 , b1 , b2 , b3 , y1 , and y2 defined
in Figure 3.11b can be determined by Eq. (3.2.23). The shape of a hexagonal cylinder is
defined by the length-to-radius ratio ca, where a and c are the semi-minor and semi-major
axes, respectively. The size parameter is given by ka.
Having defined a number of geometric variables, we may rewrite the integral equation
of Eq. (3.2.3) in the form

up (hexagon) = uc e−ikr0 exp[−ik(Xx + Y y )] dx dy . (3.2.26)
A
The geometric shadow area bounded by (1-6-5-4-10-9-8-7-1), shown in Figure 3.11b,

can be divided into three sub-domains: two trapezoids (1-6-5-4-1) and (7-10-9-8-7) and
one parallelogram (1-4-10-7-1). The operation of the matrix B in Eq. (3.2.23) can then be
applied so that the bases of the trapezoids are parallel to the x -axis. In this manner, we can
follow the method of Smith and Marsh (1974) to carry out these integrations. It follows
that the diffraction amplitude up of a hexagonal cylinder defined by the orientation position
(α, β) at an arbitrary point P(θ, φ) can be obtained in the form:
up (θ, φ; α, β; ka, c/a)/uc e−ikr0
4 ) * ( +
= sin 12 (y2 − y1 )(a1 X + Y ) sin 12 [−a1 (y1 − y2 ) + 2b1 ] X − 12 (y1 + y2 )
X(a1 X + Y )
4 ) * ( +
+ sin 12 (y2 − y1 )(a2 X + Y ) sin 12 [a2 (y1 + y2 ) + 2b2 ] X + 12 (y1 + y2 )
X(a2 X + Y )
4
+ sin [y1 (a3 X + Y )] sin(b3 X). (3.2.27)
X(a3 X + Y )
Subsequently, the diffracted intensity for a given orientation of a hexagonal cylinder defined
by angles α and β is given by an expression similar to Eq. (3.2.20).
3.2.4 Spheroid
The geometric shadow of a spheroid projected onto a plane perpendicular to the incident
direction is generally an ellipse. We may specify the orientation of the spheroid by an angle
ζ between the incident direction and its rotation axis. An ellipse is a circle of radius a
multiplied by a factor ξ in the direction Ox (see Figure 3.11c). Thus, the elliptical aperture
Ae is equal to a circular aperture Ac multiplied by ξ in the direction along Ox. Referring to

Eq. (3.2.3) and changing variables such that x = x/ξ and y = y, we obtain

−ikr0
up (spheroid) = ξ uc e exp[−ik(x ξ cos φ + y sin φ) sin θ]dx dy . (3.2.28)
Ac
We may transform the rectangular coordinates (x , y ) to polar coordinates (ρ, ψ) such
that x = ρ cos ψ and y = ρ sin ψ. In this manner, Eq. (3.2.28) can be rewritten as follows:
a 2π
up = ξ uc e−ikr0 exp[−kρl cos(ψ − σ ) sin θ]ρdρdψ, (3.2.29)
0 0
where σ = tan−1 (tan φ/ξ ) and l = (ξ 2 cos2 φ + sin2 φ)1/2 . In terms of the zero-order Bessel
function defined in Eq. (3.2.5), we have
a
up = ξ uc e−ikr0 2π J0 (kρl sin θ )ρdρ. (3.2.30)
0
Referring to Eqs. (3.2.7) and (3.2.8) regarding the relationship between the zero- and
first-order Bessel functions and to Eq. (3.2.30), we can show that
2J1 (xl sin θ )
up = uc e−ikr0 Ae , (3.2.31)
xl sin θ
where the geometric shadow area Ae = ξ πa2 and the size parameter x = ka. Hence, the
scattered intensity in terms of the incident intensity I0 = |u0 |2 can be expressed as

ξ 2 x 4 2J1 (lx sin θ ) 2
ip (θ, φ; ζ ; ka, c/a) = . (3.2.32)
4 lx sin θ
To determine ξ , we note that the geometric shadow of a spheroid onto a plane perpendic-
ular to an incident direction is an ellipse. We may specify the orientation of the spheroid by
an angle ζ between the incident direction and its rotation axis, defined by the semi-minor
axis a and the semi-major axis along the rotation axis c, as shown in Figure 3.11c. Thus,
the spheroid and the incident light beam tangent to it are defined by the following two
equations:
x2 z2
2
+ 2 = 1, x = tan ζ z + v. (3.2.33a)
a c
Upon combining these two equations, we obtain

1 tan2 ζ 2v tan ζ v2
+ z 2
+ z + − 1 = 0. (3.2.33b)
c2 a2 a2 a2
Moreover, in order to have the incident beam tangent to the spheroid, the determinant D
must be equal to 0 such that
2
v tan ζ 2 1 tan2 ζ v
D = − 2+ − 1 = 0, (3.2.33c)
a2 c a2 a2
which leads to the solution in the form v = ±(c2 tan2 ζ + a 2 )1/2 . Additionally, we find the
following relationship:
ξ a = |v| cos ζ = a[(c/a)2 sin2 ζ + cos2 ζ ]1/2 . (3.2.34a)
Thus, we obtain
ξ = [(c/a)2 sin2 ζ + cos2 ζ ]1/2 . (3.2.34b)
Once ξ is known, the scattered intensity due to diffraction by a spheroid can be evaluated
from Eq. (3.2.32).
The general case of a spheroid shape is a three-axis ellipsoid, which is governed by the
following geometric relationship:
x2 y2 z2
2
+ 2 + 2 = 1. (3.2.35)
a b c
When a three-axis ellipsoid is viewed from an arbitrary direction expressed by angles
α and β, its shadow is an ellipse, as shown in Figure 3.11d. The ellipse can be obtained
as follows. On solving X = (CD)T X , the coordinate system (x, y, z) can be expressed as
(x , y , z ). Substituting (x, y, z) into Eq. (3.2.35), we obtain a quadratic equation for the
coordinate system (x , y , z ). If the determinant D for the quadratic equation with respect
to z is 0, another quadratic equation in x and y can be obtained to represent a shadow
ellipse. Specifically, we obtain
Ax 2 + Bx y + Cy 2 = D. (3.2.36)
Upon rotation of the x and y coordinate axes by an angle η (= tan−1 [B/(A − C)]),
Eq. (3.2.36) can be simplified in the form
x /p2 + y /q 2 = 1. (3.2.37)
It follows that the semi-axis lengths p and q, and the angle η between the p-axis and x -axis
as well as its diffracted intensity can then be computed.
Finally, for randomly oriented non-spherical particles, the intensity function ip can be
obtained by performing integrations over the angles α, β, and φ in the form
2π 2π π
1
īp (θ ; ka, [b/a] , c/a) = ip (θ, φ; ka, [b/a] , c/a; α, β) cos αdαdβdφ.
4π2 0 0 0
(3.2.38)
The intensity function is now a function of the scattering angle and the particle’s geometric
parameters in reference to the incident light beam characterized by a certain wavelength.
3.2.5 Some numerical results

In Figures 3.12a–i, the intensity function is shown in logarithmic contours for a diffracting
body with the shape and orientation indicated in the inset. The wavelength used for the
calculation is 0.55 µm. Any diffracted intensity would have a maximum at the scattering
(a) 90
(b) 90
(c) 90
120 60
120 60 φ 120 60 φ φ
4 3
3 4
150 30 150 4 30 150 30

5 4
4 5
6
6 4 3 3
180 3 4 4 45 6 0 0 180 0 0 180 4 4 5 0 0
2 2 2
θ θ
4 4 4
210 210
6θ
330 330 210 330
6 6
8 8 8
240 300 240 300 240 300
270 270 270
(d) 90
(e) 90
(f) 90
120 60 φ 120 60 φ 120 60 φ
150 30 150 30 150 30

5
4
4
5
4 5 6 4
180 55 67 0 0 180 0 0 180 4 4 5 6 0 0
5
4
3 2 2 2
4 θ 3 3
4 4 4θ
θ 4
210 330 210 330 210 330
6 6 3 6
8 8 4 8
240 300 240 300 240 300
270 270 270
(g) (h) 90
(i) 90
90
60 120 60 φ 120 60 φ
120 φ
4
3
150 30 150 5 30 150 30
4
4 4 5 6
5 5 6
180 0 0 180 0 0 180 4 0 0
4 5
4 6 3
2 4 5 2 2
4
3
4 4θ 4 4
210 3 210 210
6θ
330
6θ
330 330
6
4
8 8 8
240 300 240 300 240 300
270 270 270
Figure 3.12 Diffracted intensity function contours on a logarithmic scale for an object with the shape and orientation
indicated by the inset, using an incident wavelength of 0.55 µm. The radial direction covers the scattering angle θ from
0° to 10°. The objects are (a) a circular aperture (a = 4.24 µm), (b) an elliptical aperture (a = 3 µm and c = 6 µm), (c) a
rectangular aperture (a = 3 µm and c = 6 µm), (d) a hexagonal aperture (a = 10 µm), (e) a cube (α = 45°, β = 45°, and
a = 3.8 µm), (f) a hexagonal cylinder (α = 60°, β = 30°, a = 4 µm, and c = 8 µm), (g) a hexagonal cylinder (α = 60°,
β = 20°, a = 4 µm, and c = 8 µm), (h) a rectangular parallelepiped (α = 30°, β = 30°, a = 2 µm, b = 7 µm, and c =
4 µm), and (i) a three-axis ellipsoid (α = 60°, β = 45°, a = 3 µm, b = 6 µm, c = 9 µm) (after Takano et al. 2012).
angle θ 1st (max) of 0°. The blackened parts in the figures associated with the dense contour
lines depict weak diffracted intensities. The diffraction pattern of a circular aperture is
concentric (Figure 3.12a) with the first and second minima located at θ 1st (min) = 4.5° and
θ 2nd (min) = 8.3°, respectively. These dark bands can be explained by the fact that the
first-order Bessel function J1 (y) has zeros at y = 3.832 and 7.016.
Figure 3.12b shows the diffraction pattern of an elliptical aperture whose area is equal
to that of the circle. The long axis of the elliptical contour is along the short axis of its
aperture, such that the aspect ratio is the same as that of the aperture. Figure 3.12c is for a
rectangular aperture circumscribing the ellipse in Figure 3.12b. The first maximum pattern
around θ 1st (max) = 0° has a rectangular shape with an aspect ratio equal to that of the
aperture but with a 90° rotation. Next to the first maximum pattern, four surrounding fans
of rectangular shapes are produced.
The regular hexagonal aperture produces a hexagonal pattern around θ 1st (max) = 0°,
surrounded by six fans (Figure 3.12d). However, there is a dark circle inside the hexagonal
pattern, which was also shown in the work of Smith and Marsh (1974). The scattering angle
θ 1st (min) of 2.11° at which the diffracted intensity becomes zero is identified by inspecting
the position at which the diffracted amplitude changes its sign from positive to negative.
This dark circle coincides with the first minimum of circular aperture of the same area
with a radius of 9.09 µm, corresponding to y = 3.832. The same dark circle also occurs
for a regular octagonal aperture. Figure 3.12e shows the diffracted pattern of a cube for an
oblique incident light beam. In view of the shape defined in the inset, its pattern resembles
that of a hexagon but its apex has a 30° rotation. For example, there is a dark circle-like
shape inside the hexagonal pattern containing the first maximum. Its approximate scattering
angle θ is 3.4°, close to the first minimum of a circular aperture of the same area, which
occurs at θ 1st (min) = 3.43°.
Figure 3.12f illustrates the diffraction pattern of a hexagonal cylinder for an oblique
incident light beam. In view of the projected shadow in the inset, its diffraction pat-
tern has the characteristics of both rectangular and hexagonal apertures. Figure 3.12g
shows another example of a hexagonal cylinder in which the incident direction in
Figure 3.12f is shifted by changing the rotational angle β in the horizontal direction from
30° to 20°. The diffraction pattern is also rotated by 10° and is not symmetric in reference
to the line defined by φ = 90° and φ = 270°, which differs from the patterns depicted in
Figures 3.12a–f. Figure 3.12h is for a rectangular parallelepiped such that the shape of
the projected shadow resembles a parallelogram. As a result, a parallelogram-like pattern
occurs around the incident direction and is surrounded by four fans. Figure 3.12i depicts
the diffraction pattern generated by a three-axis ellipsoid whose projected shadow is an
ellipse with a semi p-axis length of 6.6 µm, semi q-axis length of 4.0 µm, and an angle η
between the p-axis and xʹ-axis of 29.0°. In Figure 3.12b, we show the diffraction pattern
for an ellipse with its long axis along the q-axis.
Figure 3.13 depicts the normalized diffracted intensity as a function of the scattering
angle from 0 to 10° for randomly oriented spheroids, hexagonal cylinders, and rectangular
parallelepipeds with aspect ratios of 4 and 14 such that their cross-sectional areas are
(a) (b)
108 108
c/a = 4 c/a = 1/4
c/a = 4 c/a = 1/4

107 107
c/a = 4, a = b c/a = 1/4, a = b
Diffracted Intensity
106 106
105 105
104 104
103 103
0 2 4 6 8 10 0 2 4 6 8 10
Figure 3.13 Diffracted intensity functions for randomly oriented spheroids, hexagonal cylinders,
and regular parallelepipeds: (a) aspect ratio = 4; (b) aspect ratio = 14. The cross-sectional area for
both cases is that of a sphere with a size parameter 2πaλ of 100, where λ is the wavelength and a is
the radius.
equal to that of a sphere with a size parameter of ka = 100. It is evident that the two
groups (4 and 14) display similar patterns but with numerous differentiable fluctuations.
In Figure 3.12, we demonstrate that for a non-spherical particle with a specific orientation
relative to the incident light beam, the scattering light produced by diffraction displays
significant fluctuations in 3-D space. However, when an ensemble of non-spherical particles
are randomly oriented in 3-D space, the diffracted light is symmetrical with respect to the
azimuthal angle and is a function of the scattering angle only. Intensity fluctuations in this
case are much smaller than for a particular orientation. Results for the diffraction of a
group of spheres having the same sizes show substantial max–min fluctuations (see also
Figure 3.13a). These fluctuations can be largely damped out by applying a size distribution
for spherical particles.
For application to ice crystals, we may use a size distribution to smooth out small
fluctuations. The semi-exact solution for the diffraction of irregular ice particles (bullet
rosette, aggregate, snowflake, and rough-surfaced crystal) may be obtained by performing
tedious numerical calculations once the shape is defined. As a first approximation, however,
we may use diffraction patterns for columns/plates with equivalent cross-sectional areas
(Cai and Liou 1982, Takano and Liou 1989a) for application to light scattering by ice
crystals.
3.3 Generalized geometric optics for light scattering by ice crystals

3.3.1 Introduction
The scattering of light by spheres, cylinders, and spheroids can be solved by the exact theo-
ries presented in Subsection 2.2.3, and computations can be performed for size parameters
that are practical for atmospheric applications. However, an exact solution for the scattering
of light by non-spherical ice crystals covering all sizes and shapes that occur in planetary
atmospheres does not exist in practical terms. It is unlikely that one specific method can
be employed to resolve all of the scattering problems associated with non-spherical and
inhomogeneous ice crystals.
The geometric-optics approach, as discussed in Section 3.1, is the asymptotic approx-
imation of fundamental electromagnetic theory, valid for light scattering computations
involving a target whose dimension is much larger than the incident wavelength. As also
noted in that section, the geometric ray-tracing method provides a means of identifying
numerous fascinating ice optical phenomena observed in the Earth’s atmosphere. The part
of reflection/refraction within the context of geometric optics is based on the principle that
light rays can be identified at a particle surface; however, the part of diffraction is not
necessarily governed by this localization principle.
The geometric-optics method, including geometric reflection/refraction and diffraction
contributions, has been extensively used to investigate the single-scattering properties of
non-spherical ice crystals (Wendling et al. 1979, Liou and Coleman 1980, Coleman and
Liou 1981, Cai and Liou 1982, Takano and Jayaweera 1985, Takano and Liou 1989a
and 1989b, Macke 1993, Hess and Wiegner 1994, Macke et al. 1996). Application of the
ray-tracing technique to light scattering by spheres can be traced back to studies by Liou
and Hansen (1971). The authors applied this technique to light scattering by spheres and
compared the phase function and polarization patterns computed from the geometric ray-
tracing method with the “exact” Lorenz–Mie theory for a size distribution. The patterns
computed from the ray-tracing technique are reasonably accurate when the modal size
parameter is larger than 100 for a polydisperse system of spheres.
The application of geometric ray-tracing to light scattering by hexagonal ice prisms was
first carried out by Jacobowitz (1971). In his study, a sufficiently large number of equally
spaced, parallel rays were traced through a hexagonal ice crystal. The external reflection and
two refractions coupled with various orders of internal reflection were summed to determine
the scattering pattern in the far field. Diffraction contributions due to the scattering of
incident light in the forward direction were computed by Kirchhoff’s formula. Hexagonal
prisms were assumed to be infinitely long along the major axis. To circumvent the unrealistic
infinitely-long assumption, Wendling et al. (1979) combined the Monte Carlo method and
the geometric ray-tracing technique to compute the phase function of finite hexagonal
columns. In these early studies, the polarization effect and phase interferences associated
with the incident, internal, and scattered rays were not accounted for in the calculation.
The paper by Cai and Liou (1982) appears to be the first study to include a 4 × 4 scattering
phase matrix and polarization configuration as well as phase interferences in the context
3.3 Generalized geometric optics for light scattering 147
of geometric ray tracing to compute the single-scattering properties of hexagonal columns

and plates.
The theoretical foundation for “conventional” geometric ray tracing and the associated
computational algorithm developed by Cai and Liou (1982) were later improved and refined
by a number of researchers. Takano and Liou (1989a, 1989b) considered the effects of the
ice crystal’s birefringence property, horizontal orientation, and size spectrum in associa-
tion with light-scattering calculations. The single-scattering properties of ice crystals with
horizontal orientation have also been investigated by Rockwitz (1989) and Noel et al.
(2001).
Furthermore, the ray-tracing method has been applied to various complex ice crystal
shapes by Takano and Liou (1995), Macke (1993), Macke et al. (1996), Iaquinta et al.
(1995), Muinonen et al. (1997), Peltoniemi et al. (1989), and Yang and Liou (1998a), as
well as to ice crystals with internal inclusions (e.g., air bubbles and soot) by Macke et al.
(1996) and Macke (2000). Application of the ray-tracing technique with implementation of
the Monte Carlo method to complex geometries has also been reported by Nousiainen et al.
(2003) and Grynko and Shkuratov (2003). Additionally, Borovoi et al. (2002) investigated
backscattering characteristics of arbitrarily oriented hexagonal ice crystals using the ray-
tracing method. Borovoi and Grishin (2003) discussed an effective ray-tracing algorithm for
computing phase matrix elements for ice particles. Alternate approaches to the conventional
ray-tracing method have been developed by Muinonen (1989) and Yang and Liou (1995,
1996a, 1997) in which the principles of geometric optics are applied to the computation
of the near field either on the surface of, or inside, the scattering particle. The near field
obtained from the ray-tracing technique is then mapped to the far field on the basis of either
a surface-integral- or volume-integral-based electromagnetic relationship.
In Subsections 3.3.2–3.3.4, we present conventional and improved approaches to light
scattering by ice crystals, the methods dealing with absorption in the context of geometric
optics, and the numerical implementation of the Monte Carlo method in geometric ray
tracing.
3.3.2 Conventional approach

When the size of an ice crystal is much larger than the incident wavelength, we may consider
a light beam as consisting of a bundle of separate parallel rays that undergo reflection and
refraction outside and inside the ice crystal with propagation directions determined by
Snell’s law at the surface. The total field is assumed to consist of diffracted and reflected
and refracted rays. Diffracted rays pass around the ice crystal while rays impinging on it
undergo local reflection and refraction, referred to as Fresnelian interaction. The energy
carried by the diffracted and Fresnelian rays is assumed to be the same as the energy
intercepted by the ice crystal cross-section projected along the incident direction.
The intensity of the scattered light within the small scattering-angle interval θ in the
scattering direction θ in the far field can be computed from the summation of the intensity
contributed by each individual ray emerging at between θ + θ /2 and θ − θ /2. It is
(conventional)
diffracted ray
Incident wave front reflected ray
e1r
incident ray
Q1 equivalent surface E&M theory
l l
Q0
far field
e1i =ei0 l
E&M currents
(improved)
Q2
Q3
l
e2t
transmitted ray
diffracted ray
transmitted ray
e3t
Figure 3.14 A graphical depiction of the principle of the geometric ray-tracing technique, including
diffraction, for computing the single-scattering properties of an ice crystal (conventional). The sym-
bols Q1 , Q2 , and Q3 denote the entrance points of light rays; and er1 , et2 , and et3 are the corresponding
directional vectors. Also shown is a conceptual diagram of the geometric-optics integral-equation
method, including geometric ray tracing within an ice crystal and the mapping of resulting currents
to the far field by means of fundamental electromagnetic theory (improved; see Subsection 3.3.3).
usually assumed that the interference is smoothed out when ice crystals are randomly
oriented. In this case, the extinction efficiency (the ratio of the extinction cross-section
to the average projected area of the particle) of the ice crystal is 2. Based on Babinet’s
principle, diffraction by an ice crystal may be regarded as the same as diffraction by an
opening, on an opaque screen perpendicular to the incident light, which has the same
geometric shape as the projected cross-section of an ice crystal. The Fraunhofer diffraction
approach can then be employed to compute the diffraction component for column and plate
crystals (Cai and Liou 1982, Takano and Liou 1989a).
To trace the reflected and refracted rays, let np (p = 1, 2, 3 . . . ) be the unit vectors locally
normal to the particle surfaces at the incident points Qp (p = 1, 2, 3 . . . ) facing the incoming
rays, as shown in Figure 3.14. For external reflection at point Q1 , the incident direction ei1
and the incident angle θ1i are given, respectively, by the following two expressions:
ei1 = ei0 , (3.3.1a)

θ1i = cos−1 − n1 · ei1 , (3.3.1b)
where ei0 denotes the initial incident direction (Figure 3.14). Following Snell’s law,
the directions of the externally reflected ray and the corresponding refracted ray are
defined by
er1 = ei1 + 2 cos θ1i n1 , (3.3.2a)

-
et1 = ei1 /m + cos θ1i m − cos θ1t n1 , (3.3.2b)
where m is the refractive index of the scattering particle, and θ1t is the refractive angle given
by Snell’s law as follows:
-
θ1t = sin−1 sin θ1i m . (3.3.2c)
When the refractive index m is a complex number, simultaneous absorption and scattering
occur and the refracted wave within the particle is an inhomogeneous wave. In this case,
an adjusted refractive index must be used to trace the refracted rays (see Subsection 3.1.2).
For practical computations at visible and near-infrared wavelengths, the real part of the
refractive index may be used as an excellent approximation for the adjusted refractive
index to trace ray directions based on Eq. (3.1.19).
For internal reflection with orders of p = 2, 3, 4 . . . , the incident directions can be
defined in a similar manner and are given by the directions of either the first-order refracted
rays or internally reflected rays, as follows:
ei2 = et1 , (3.3.3a)
eip = erp−1 , p = 3, 4, 5 . . . (3.3.3b)
With some vector algebraic manipulations on the basis of Snell’s law, it can be shown that
the propagating directions of pth-order reflected and refracted rays are given by
erp = eip + 2 cos θpi np , (3.3.4a)

etp = meip + m cos θpi − cos θpt np , (3.3.4b)
where the incident and refraction angles, θpi and θpt , are defined by

θpi = cos−1 − np · eip , (3.3.5a)

θpt = sin−1 m sin θpi . (3.3.5b)
Total reflection occurs if the term m sin θpi in Eq. (3.3.5b) is larger than 1. In this case, a
refracted ray should not be expected so that the ray-tracing computation should be continued
only for the ray associated with total reflection. Equations (3.3.1a)–(3.3.5b) constitute a
closed set of equations for tracing the directions of all the reflected and refracted rays
associated with a given incident ray.
A localized plane electromagnetic wave is a transverse vector wave. Thus, the vector
property or the polarization configuration of electric fields associated with localized rays in
the ray-tracing computation must be accounted for, which requires the definition of various
auxiliary unit vectors. For the incident direction of an initial ray specified by a unit vector
ei0 , we define two unit vectors u0 and v0 , as shown in Figure 3.15, which are normal to the
first order ray (p=1)
u1i t
f1 fp ept up
n1 v1
e1r e1i vp
u1r θ1i
v1 θ pt
Air Air
upr
Ice Ice
epr
θ pi vp
np
θ1t v1
epi upi
vp
u1t
e1t p-th order rays (p>1)
Figure 3.15 Schematic diagrams for the direction of the incident, reflected, and refracted rays for
the first-order ray (p = 1) and pth-order rays (p > 1). The unit vectors e, u, and v are defined
with reference to polarization configuration: θ p and φ p (p = 1, 2, 3, . . . ) are zenith and azimuthal
angles, respectively: and the superscripts i and t denote incident and refracted (transmitted) directions,
respectively.
incident direction and satisfy the following relations:

u0 · v0 = 0 and v0 × u0 = ei0 . (3.3.6)
The unit vectors v0 , u0 , and ei0 defined in this equation constitute a right-handed coor-
dinate system. To define the initial rays in practice, we may specify the unit vectors v0 , u0 ,
and ei0 in the directions along the x-, y-, and z-axes of the incident coordinate system.
Because the unit vectors ei0 , u0 , and v0 are orthogonal to each other, the incident polariza-
tion configuration can be specified with respect to u0 and v0 such that the incident electric
field Ei0 can be written as follows:
Ei0 = E0u
i
u0 + E0v
i
v0 . (3.3.7)
Similarly, we may define three pairs of unit vectors (uip , vp ), (urp , vp ), and (utp , vp ) for
the ray directions along eip , erp , and etp (p = 1, 2, 3 . . . ), respectively. The unit vectors uip ,
urp , and utp are on the incident plane, as shown in Figure 3.15, while the unit vectors vp
(p = 1, 2, 3 . . . ) are defined as pointing out of the incident plane. These vectors can be
specified by the following expressions:

vp = eip × np /sin θpi , p = 1, 2, 3 . . . , (3.3.8a)
ui,r,t
p = ei,r,t
p × vp , p = 1, 2, 3 . . . (3.3.8b)
The unit vector vp is normal to the incident plane, a plane containing the incident
direction and the direction locally normal to the particle surface at the incident point, for
the pth-order reflection and refraction. The unit vectors uip , urp , and utp are parallel to
the pth-order incident plane. The vector vp in Eq. (3.3.8a) cannot be uniquely specified if
sin θpi = 0. In this case, we select vp = vp−1 . With the aforementioned unit vectors defined,
the electric fields associated with the pth-order incident, reflected, and refracted rays can
be expressed as follows:
Ei,r,t
p = Epu up + Epv
i,r,t i,r,t i,r,t
vp . (3.3.9)
Consider now the external reflection and the first-order refraction. In order to apply
the Fresnel formulas, the electric field associated with the incident ray impinging on the
point Q1 must be specified with respect to ui1 and v1 . Also, the electric field associated
with the incident ray specified in Eq. (3.3.7) can be expressed in an alternative form as
follows:
Ei0 = E1u
i
ui1 + E1v
i
v1 . (3.3.10)
Equations (3.3.7) and (3.3.10) for the incident electric vector can be expressed in terms
of a rotational matrix in the form

i i
E1u E0u
i
= 1 i
, (3.3.11a)
E1v E0v
where the rotational matrix 1 (p = 1) is given by

i

u1 · u0 ui1 · v0 cos φ1 sin φ1
1 = = . (3.3.11b)
v1 · u0 v1 · v0 −sin φ1 cos φ1
Because the field components on the left-hand side of Eq. (3.3.11a) are specified with
respect to the incident plane, the Fresnel formulas can now be applied. The externally
reflected field, Er1 , is given by
Er1 = E1u
r r
u1 + E1v
r
v1 , (3.3.12a)
where

r i i
E1u E1u E0u
r
= R1 i
= R1 1 i
. (3.3.12b)
E1v E1v E0v
In Eq. (3.3.12b), R1 is the reflection matrix for external reflection given by

R1u 0
R1 = . (3.3.12c)
0 R1v
The elements of the reflection matrix in Eq. (3.3.12c) are given by the Fresnel coefficients
defined in Subsection 3.1.1 as follows:
m cos θ1i − cos θ1t cos θ1i − m cos θ1t
R1u = , R1v = . (3.3.13)
m cos θ1i + cos θ1t cos θ1i + m cos θ1t
Likewise, the electric field associated with the first-order refracted ray is given by
Et1 = E1u
t
ut1 + E1v
t
v1 , (3.3.14a)

t i i
E1u E1u E0u
t
= T1 i
= T1 1 i
, (3.3.14b)
E1v E1v E0v
where the refraction matrix T1 is defined in the form

2 1/2
T1u 0 1 − R1u 0
T1 = = . (3.3.14c)
2 1/2
0 T1v 0 1 − R1v
In Eq. (3.3.14c), conservation of the energy for the ray due to changes in the refractive
index and ray cross-section in two media is accounted for in the refraction matrix (Cai and
Liou 1982).
For external reflection, the direction along the reflected ray is the scattering direction.
Thus, the scattering angle (note that has been used throughout most of the text) can be
expressed as

θ1s = cos−1 ei0 · er1 . (3.3.15a)
The direction perpendicular to the scattering plane can be subsequently specified by
-
vs1 = ei0 × er1 sin θ1s . (3.3.15b)
If sin θ1s = 0 in Eq. (3.3.15b), implying forward (θ1s = 0◦ ) and backward scattering (θ1s =
180◦ ), then the vector vs1 cannot be defined. In this case, we select vs1 = v0 . Having the unit
vector vs1 defined, the direction parallel to the scattering plane is given by
us1 = er1 × vs1 . (3.3.15c)
We may express the electric field associated with the externally reflected ray with respect
to two directions parallel and perpendicular to the scattering plane in the form
Er1 = E1u
s s
u1 + E1v
s s
v1 . (3.3.16)
From Eqs. (3.3.12a, b) and (3.3.16), we can show that

s i
E1u E0u
s
= 1 R1 1
s
i
, (3.3.17a)
E1v E0v
where s1 is a rational matrix given by

us1 · ur1 us1 · v1
s1 = . (3.3.17b)
vs1 · ur1 vs1 · v1
To obtain the scattering matrix, the incident field must be defined with respect to the
directions parallel and perpendicular to the scattering plane in the form
i
Ei0 = E1su
i
e0 × vs1 + E1sv
i
vs1 . (3.3.18)
Note that the unit vector ei0 × vs1 in Eq. (3.3.18) is parallel to the scattering plane. The
expression in Eq. (3.3.18) for the incident field is related to that in Eq. (3.3.7) as follows:

i i
E0u E1su
i
= 1 i
i
, (3.3.19a)
E0v E1sv
where i1 is a 2-D rotational matrix given by

u0 · ei0 × vs1 u0 · vs1
1 =
i
. (3.3.19b)
v0 · ei0 × vs1 v0 · vs1
Thus, we can express the scattered field in Eq. (3.3.17a) as follows:

s i
E1u E1su
s
= 1 R1 1 1
s i
i
. (3.3.20a)
E1v E1sv
Likewise, for the refracted rays with p = 2, we have

s i
E2u E2su
s
= 2 T2 2 T1 1 2
s i
i
. (3.3.20b)
E2v E2sv
Furthermore, for the orders p = 3, 4, 5 . . . , we have

s i
Epu Epsu
s
= p Tp p . . . R2 2 T1 1 p
s i
i
. (3.3.20c)
Epv Epsv
The definitions of sp , Tp , p , Rp , and ip are similar to the case for p = 1. In this

manner, both the incident and scattered electric field vectors are expressed with respect to
the scattering plane as shown in Eqs. (3.3.20a–3.3.20c). Thus, contributions of the scattered
rays to the transformation matrix defined in Eq. (2.4.30) can be expressed as follows. First,
for externally reflected rays (p = 1), we have
(1)
A2 A(1)
3
A =
(1)
= s1 R1 1 i1 . (3.3.21a)
A(1)
4 A (1)
1
For transmitted rays (p = 2), the transformation matrix can be expressed as

(2)
A2 A(2)
3
A =
(2)
= s2 T2 2 T1 1 i2 . (3.3.21b)
A(2)
4 A(2)
1
For p > 2, we can define the following transformation matrix as follows:

(p) ⎛ ⎞
A A
(p) 02
= sp Tp p ⎝ Rk k ⎠ T1 1 ip ,
2 3
A(p) = (p) (p)
(3.3.21c)
A4 A1 k=p−1
where 1 has been defined in Eq. (3.3.11b) and p has the same expression except that
index 1 is replaced by p. We may perform the transpose of Eq. (3.3.21c), resulting in the
following equation:
%p−1 &
) (p) *T (p)
A2 A4
(p)
) i *T 0 ) *T ) *T
A = (p) (p)
= p [ 1 ] T1
T
[ k ] Rk p Tp sp , (3.3.21d)
T
A3 A1 k=2
where the transpose of p (p = 1, 2, 3, . . . ) is given by

cos φp −sin φp
[ p ] =
T
. (3.3.21e)
sin φp cos φp
If the sign of the rotation angle φp is changed, then the signs of the non-diagonal elements
in Eq. (3.3.21d) are also changed, since Tp and Rp are diagonal matrices. As a result, we
have the following transformation matrix:
%p−1 &
A2
(p)
−A4
(p)
0
(p)∗
A = (p) (p)
= ip 1 T1 k Rk p Tp sp . (3.3.21f)
−A3 A1 k=2
The right-hand side of Eq. (3.3.21f) represents a reverse order of terms in reference to
Eq. (3.3.21c) such that every ice particle position has a reciprocal position with respect
to the initial position. Consequently, incident and scattered light beams can be reversed.
This is referred to as the law of reciprocity, as discussed in Subsection 2.4.5. For external
reflected and diffracted rays, non-diagonal elements of the transformation matrix are zero
and so the addition of these two components to geometric refraction and reflection would
not alter the consideration of reciprocity.
In the preceding discussion, the phase change associated with ray optical paths has not
been accounted for. However, light rays that impinge on an ice crystal at different locations
must experience phase changes due to their different paths. For this reason, Cai and Liou
(1982) considered the phase interference of emerging rays in ray-tracing computations.
The follow-on studies reported in the literature essentially ignored the phase shifts asso-
ciated with ray paths. Takano and Jayaweera (1985) showed that phase interference can
be smoothed out when ice crystals are randomly oriented. For practical applications, we
generally assume that ice crystals are randomly oriented in radiative transfer computations.
If ice crystals are horizontally oriented, however, the single-scattering properties depend
not only on the scattering angle but also on the azimuth of the scattering plane. In this
case, the radiative transfer calculation can be quite involved (see Subsection 2.5.2). In addi-
tion to random orientation, integration over the size spectrum will smooth out fluctuations
produced by phase interferences in the scattering pattern for one ice crystal size.
For randomly oriented ice particles, the corresponding phase matrix has only six inde-
pendent elements, as shown in Subsection 2.4.3. Thus, for the pth-order emerging ray, its
contribution to the phase matrix is given by the following expression (Takano and Jayaweera
1985):
⎡ (p) (p) ⎤
f11 f12 0 0
⎢ (p) ⎥
⎢ f12 f22 (p)
0 0 ⎥
F =⎢
(p) ⎢ ⎥, (3.3.22)
(p) (p) ⎥
⎣ 0 0 f33 f34 ⎦
(p) (p)
0 0 −f34 f44
where the matrix F(p) transforms the incident Stokes parameters to the scattered Stokes
parameters associated with the pth-order outgoing localized wave, and its elements are
defined as follows:
(p) (p) (p) (p) (p) -
f11 = M1 + M2 + M3 + M4 2, (3.3.22a)
(p) (p) (p) (p) -
f12 = f21 = − M1 + M2 2, (3.3.22b)
(p) (p) (p) (p) (p) -
f22 = M1 + M2 − M3 − M4 2, (3.3.22c)
(p) (p) (p)
f33 = S12 + S34 , (3.3.22d)
(p) (p) (p)
f34 = −f43 = −D12 , (3.3.22e)
(p) (p) (p)
f44 = S12 − S34 , (3.3.22f)
and the terms in these equations are given by
(p) 2
Mi = Ai ,
(p)
(3.3.23a)
(p) (p) (p) (p)∗ (p)∗ (p) -
Sij = Sj i = Ai Aj + Ai Aj 2, (3.3.23b)
(p) (p) √ (p) (p)∗ (p)∗ (p) -
Dij = −Dj i = −1 Ai Aj − Ai Aj 2, (3.3.23c)
where the subscripts i and j range from 1 to 4, and the asterisk indicates the complex con-
jugate. Thus, the scattering matrix associated with the various orders of external reflections
and transmissions for all the incident rays can be expressed in the following form:
N ∞ N
(p)
Fray = σj Fj σj , (3.3.24)
j =1 p=1 j =1
where j denotes the external reflection, while the various orders of transmission are asso-
ciated with the jth initial ray, N is the total number of incident rays, and σj is the
cross-section of the jth initial ray. To speed up the computation, the preceding ray-tracing
algorithm can be augmented with the Monte Carlo hit-and-miss approach described in
Subsection 3.1.3.
In addition to contributions from reflected and refracted rays, diffraction also contributes
to the scattering of the incident wave. According to Babinet’s principle (see Section 3.2),
the diffraction pattern associated with an object is the same as that for an aperture with a
shape identical to the projection of the object on a plane normal to the incident direction.
The diffraction matrix obtained by the scalar Fraunhofer diffraction theory for a scattering
particle (see e.g., Jackson 1975) has been used extensively in geometric ray-tracing studies.
It can shown that the transformation matrix associated with diffraction can be written in
the form

dif

A2 0 k2 (1 + cos θ ) 1 0
A =
dif
= exp(−ikr · ξ)d ξ ·
2
,
0 Adif
1 2π 2 0 1
projected area
(3.3.25)
in line with the A matrix defined for geometric reflection and refraction [see Eq. (3.3.21c)].
From Eq. (3.3.25), the contribution of diffraction to the scattering phase matrix, denoted
by Fdif , can be evaluated. To sum the contributions due to diffraction and Fresnel rays,
proper weighting factors must be accounted for, particularly in cases where an ice particle
is absorptive.
If an ice crystal is absorptive with respect to its incident wavelength, i.e., the imagi-
nary refractive index is non-zero, total absorption can be accounted for by considering the
absorption of individual rays. In general, the absorption cross-section of a particle depends
on the polarization configuration of the incident light. However, for randomly oriented
particles, the absorption cross-section is the average of absorption cross-sections corre-
sponding to two orthogonal polarization cases. Consider a case where polarization of the
incident light is specified as follows: (Ei0u , Ei0v ) = (1, 0). It follows that the intensity of the
first-order refracted field can be obtained from Eq. (3.3.14b) in the form
t ∗,T
E0u , Et0v Et0u , Et0v i i = (1, 0)(T1 1 )T (T1 1 )∗ (1, 0)T , (3.3.26)
(E ,E )=(1,0)
0u 0v
where the superscript T denotes the transpose of a matrix, and ∗ indicates the complex
conjugate. The intensity given by Eq. (3.3.26) is the amplitude of the Poynting vector in
which the refractive index and change in the ray cross-section due to refraction have been
implicitly accounted for in the refractive matrix given by Eq. (3.3.14c). A similar expression
can be derived for the case when the polarization of a field is given by (Ei0u , Ei0v ) = (0, 1).
Therefore, the contribution of the first-order refracted rays to the absorption cross-section
is given by

N
σabs,1 = 2−1 σj [1 − exp(−4πmi dj 1 /λ)]
j =1
) ∗,T ∗,T *
× Etpu , Etpv Etpu , Etpv i i + Etpu , Etpv Etpu , Etpv i i
(E 0u ,E0v )=(1,0) (E
0u ,E0v )=(0,1)

N
= 2−1 σj [1 − exp(−4πmi dj 1 /λ)]
j =1
) *
× (1, 0)(T1 1 )T (T1 1 )∗ (1, 0)T + (0, 1)(T1 1 )T (T1 1 )∗ (0, 1)T (3.3.27a)
where the subscript j denotes the jth initial ray, dj 1 is the distance between the first incident
point (i.e., Q1 in Figure 3.14) and the second incident point (i.e., Q2 in Figure 3.14), σj
is the cross-section of the jth initial ray, mi [denoted as Ni in Eq. (3.1.19)], is the adjusted
imaginary refractive index, and λ is the incident wavelength in a vacuum. Likewise, the
contribution of the pth-order reflected rays is

% &

N
p−1
−1 −1
σabs,p = 2 σj [1 − exp(−4πmi djp /λ)] exp −4πmi λ dj L
j =1 L=1
× [(1, 0)(Tp p . . . R2 2 T1 1 )T (Tp p . . . R2 2 T1 1 )∗ (1, 0)T

+ (0, 1)(Tp p . . . R2 2 T1 1 )T (Tp p . . . R2 2 T1 1 )∗ (0, 1)T ]. (3.3.27b)
Thus, the absorption cross-section of an ice crystal can be expressed as follows:

∞

σabs = σabs,p . (3.3.28)
p=1
In practice, the summation in Eq. (3.3.28) can be truncated for the terms with p > 10,
because the amount of energy carried by higher-order rays is relatively insignificant.
Equations (3.3.27a) and (3.3.28) provide explicit formulations for absorption cross-section
within the framework of the ray-tracing technique in which the polarization configuration
is fully accounted for. In the conventional ray-tracing method under the condition of equal
energy partition of diffraction and geometric reflection/refraction, the extinction cross-
section is twice the projected area of the scattering particle; that is, σext = 2σp , where σp is
the particle’s projected area on a plane normal to the incident direction.
One of the shortcomings of the conventional ray-tracing method is the production of
delta-transmission associated with the refraction of rays through two parallel faces of
pristine ice crystals. The delta-transmission phenomenon has been identified and discussed
in detail by Takano and Liou (1989a) and Mishchenko and Macke (1998). Let the portion of
the scattering cross-section associated with the delta-transmission be σδ . Then the scattering
cross-section can be separated into three terms as follows:
σsca = (σp − σabs − σδ ) + σδ + σp . (3.3.29a)
The first term corresponds to contributions from externally reflected rays and various
transmitted rays excluding delta-transmitted rays, the second term denotes contributions
from delta-transmitted rays, and the third term is associated with diffraction.
Let fδ be the ratio of delta-transmitted energy to the total scattered energy, defined by
fδ = σδ /σsca = σδ /(2σp − σabs ) = σδ /(σext − σabs ). (3.3.29b)
The scattered Stokes vector can be expressed as follows [see Eq. (2.4.37)]:
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = σs P ⎢ Q0 ⎥ , (3.3.30a)
⎣ U ⎦ k 2 r 2 ⎣ U0 ⎦
V V0
where (I0 , Q0 , U0 , V0 ) and (I, Q, U, V ) represent incident and scattered Stokes parameters,
respectively, and P is the normalized scattering phase matrix. Based on Fray in Eq. (3.3.24),
Fdif defined on the basis of Adif in Eq. (3.3.25), the expressions in Eqs. (3.3.29a, b), and
the associated physical meanings of these quantities and expressions, the normalized phase
function is given by (Yang and Liou 2006)
[(2σp − σabs )(1 − fδ ) − σp ] a Fray + 2(2σp − σabs )fδ δ(cos − 1)Ī¯ + σp b Fdif
P() =
2σp − σabs

¯ 1 1
= 2fδ δ(cos − 1)Ī + (1 − fδ ) − a Fray + b Fdif , (3.3.30b)
2 2
where = σsca /σext is the single-scattering albedo. In Eq. (3.3.30b), is the scattering
angle, Ī¯ is a unit 4 × 4 matrix, and the two parameters a and b are normalization factors
given, respectively, by the following two equations:
2 2
a= π , b= π . (3.3.31)
Fray () sin d Fdif () sin d
0 0
With the normalization factors given in Eqs. (3.3.31), it can be shown that the phase matrix
in Eq. (3.3.30b) is normalized such that the first phase matrix element P11 (i.e., the phase
function) satisfies the normalization condition defined in Eq. (2.4.34).
The conventional ray-tracing technique follows the postulation that the energy attenuated
by a scattering particle is equally partitioned into two parts: extinction associated with
diffraction and extinction due to Fresnel reflection and refraction. In this case, the extinction
efficiency, the ratio of extinction cross-section to the projected area of an ice particle, is
2 regardless of the particle’s size and shape. In addition, computation of the far field by
directly applying the ray-tracing technique leads to delta-transmission (Takano and Liou
1989a) in the forward direction, as is evident from the presence of a delta function. To
overcome these shortcomings, Yang and Liou (1995, 1996b, 1997) have developed two
improved geometric-optics approaches, described in the following Subsections 3.3.3 and
3.3.4.
3.3.3 Improved geometric-optics approach using the electromagnetic

equivalence theorem
In principle, the laws of geometric optics are applicable to light scattering by a particle whose
size is much larger than the incident wavelength so that rays can be localized. In addition
to the requirement of the localization principle, the conventional geometric ray-tracing
technique assumes that the energy attenuated by the scatterer may be decomposed into
equal extinction from diffraction and from Fresnel rays. Also, the Fraunhofer diffraction
formulation used in geometric ray-tracing does not account for the vector property of
electromagnetic fields. Finally, direct calculations of the far field by ray-tracing will produce
a discontinuous distribution of the scattered energy, such as the delta transmission noted by
Takano and Liou (1989a).
To circumvent a number of shortcomings in the conventional geometric-optics approach,

an improved method has been developed by Yang and Liou (1995, 1996a). The concept is
simple in that the energies determined from geometric ray-tracing at the particle surface are
collected and mapped to the far field. This differs from the conventional approach, which
collects energies produced by geometric reflections and refractions directly at the far field
through a prescribed solid angle (see Figure 3.14).
Tangential components of the electric and magnetic fields (see Section 2.2.2) on a surface
S that encloses the scatterer can be used to determine equivalent electric and magnetic
currents for the computation of a scattered far field on the basis of the electromagnetic
equivalence theorem (Schelkunoff 1943). In this theorem, the electromagnetic field detected
by an observer outside the surface would be the same as if the scatterer was removed and
replaced by equivalent electric and magnetic currents given by
J = ns × H, (3.3.32a)
M = E × ns , (3.3.32b)
where ns is the outward unit vector normal to the surface. For the far-field region, the
scattered electric field can be represented by the following integral equation:
3
eikr k 2 r r 4 r
Es (r) = M(r ) + × J(r ) exp −ikr · d 2 r , (3.3.33)
ikr 4π r S r r
where rr denotes the scattering direction, r is the reference position vector, r is the
position vector of the source point, and k is the wavenumber. The far-field solution can also
be determined by a volume integral involving the internal field.
By means of geometric ray-tracing, the electric field on the surface of a particle can
be evaluated after successive application of Fresnel reflection and refraction coefficients
parallel and perpendicular to a defined reference plane at the point of interaction, taking
into account the path length in 3-D geometry. If an ice crystal shape is of great complexity,
as with an aggregate, its surface can be defined as a cubic box so that the computation
of the electric field can be conducted on a regularly shaped surface. The electric field of
illuminated and shadowed sides can be defined as follows:

Ea (r) + Eb (r), illuminated side,
E(r) = (3.3.34a)
Eb (r), shadowed side,
where
∞

Ea (r) = Ei (r) + Er1 (r), Eb (r) = Etp (r). (3.3.34b)
p=2
In these equations, Ei is the incident electric field, Er1 is the electric field for external
reflection, and Etp are the electric fields produced by two refractions and internal reflections
(p ≥ 2). Because the transverse electromagnetic wave condition is implied in ray-tracing,
the magnetic field for each reflection and refraction for r outside the ice particle can be
obtained from the following equation:
p (r) = ep × Ep (r).
Hr,t r,t r,t
(3.3.35)
In practice, mapping the near-field solution to the far field can be accomplished in its
entirety for Ea in Eq. (3.3.34b). However, for Eb in Eq. (3.3.34b), the mapping must be
performed ray-by-ray so as to take into account full phase interferences in the determination
of the phase function.
We shall begin with Maxwell’s equations to derive integral equations for absorption and
extinction cross-sections. For a dielectric medium with an incident harmonic wave whose
time dependence is given by exp(iωt), Maxwell’s curl equations defined in Eqs. (2.2.7a)
and (2.2.7b) can be written as
∇ × H = ik(εr − iεi )E, (3.3.36a)
∇ × E = −ikH, (3.3.36b)
where k = ω/c. By defining S as the complex Poynting vector and using the preceding two
equations along with vector algebra, we obtain the following two equations:
c
S= E × H∗ , (3.3.37a)
4π
iω ωεi
∇ ·S= (εr E · E∗ − H · H∗ ) − E · E∗ , (3.3.37b)
4π 4π
where the asterisk denotes the complex conjugate. Taking the real part of Eq. (3.3.37b) and
integrating it over the region inside the particle leads to

3 2
Re ∇ · S(r )d r = Re nS · S(r )d r
V S
(3.3.38)
ω ∗ 3
=− εi (r )E(r ) · E (r )d r ,
4π V
where nS is the outward-pointing unit vector normal to the particle surface. Based on
the physical meaning of the Poynting vector and the principle of conservation of energy
(Jackson 1975), the surface integration term in Eq. (3.3.38) is the net rate at which elec-
tromagnetic energy intersects with the particle surface; that is, the energy absorbed by the
particle.
Furthermore, the incident electromagnetic flux is given by
c c
F0 = E0 · E∗0 = |E0 |2 . (3.3.39a)
4π 4π
Thus, the absorption cross-section of the particle can be expressed as

σa = −Re nS · S(r )d 2 r F0
S
(3.3.39b)
k ∗ 3
= εi (r )E(r ) · E (r )d r .
|E0 |2 V
In what follows, we shall derive an expression for the extinction cross-section similar
to Eq. (3.3.39b). In this connection, we consider the Poynting vector, which is defined in
Eq. (2.4.1), and note that it can be decomposed into incident S0 , scattered Ss , and extinction
Se components as follows:
S = S0 + Ss + Se , (3.3.40a)
where the complex extinction component of the Poynting vector is given by
c
Se = (E0 × H∗ + E∗ × H0 ). (3.3.40b)
4π
We note the following vector calculus identity:
∇ · (A × B) = B · (∇ × A) − A · (∇ × B) . (3.3.40c)
Hence, we obtain from Eqs. (3.3.40b) and (3.3.40c) the following relationship:
c
∇ · Se = ∇ · (E0 × H∗ + E∗ × H0 )
4π
c
= [H∗ · (∇ × E0 ) − E0 · (∇ × H∗ ) + H0 · (∇ × E∗ ) − E∗ · (∇ × H0 )].
4π
(3.3.40d)
Moreover, using Eqs. (3.3.36a) and (3.3.36b), we get
c
∇ · Se = [H∗ · (−ikH0 ) − E0 · (−ikε∗ E∗ ) + H0 · (ikH∗ ) − E∗ · (ikE0 )]
4π

∗ (3.3.40e)
iω ∗ iω
= (ε − 1)E∗ · E0 = − (ε − 1) E · E∗0 .
4π 4π
Further, by using Gauss’ divergence theorem denoted in Eq. (2.2.12), which transfers area
integral to volume integral, we have

Re ns · Se (r )d 2 r = Re ∇ · Se (r )d 3 r
S V

∗
iω ∗ 3
= −Re [ε(r ) − 1]E(r ) · E0 (r ) d r ,
V 4π
(3.3.40f)
and, noting that Re[(iA)∗ ] = −Im [A], we then obtain

2 ω ∗ 3
Re nS · Se (r )d r = Im [ε(r ) − 1]E(r ) · E0 (r )d r . (3.3.41a)
S 4π V
Consequently, the extinction cross-section is given by

ω ∗ 3
σe = − Im [ε(r ) − 1]E(r ) · E0 (r )d r F0
4π V
(3.3.41b)
k
= −Im [ε(r ) − 1]E(r ) · E∗0 (r )d 3 r .
|E0 |2 V
4 λ = 0.63 µm
L /a = 6

3

IGOM CGOM

2
FDTD
1

0
100 101 102 103

Size Parameter kL
Figure 3.16 Comparison of the extinction efficiency computed from conventional (CGOM) and
improved (IGOM) geometric-optics approaches and from the finite-difference time domain (FDTD)
method (see Section 4.1) as a function of size parameter. On the basis of the conventional approach,
the extinction efficiency is 2 based on equal energy partition between geometric and diffracted light
rays. The presentations use a wavelength of 0.63 µm and randomly oriented ice columns of a uniform
size identified in the figure (after Liou et al. 2000, with modifications).
Finally, when the ray-tracing technique is applied to obtain the surface field, one must
properly account for area elements from which externally reflected and transmitted localized
waves make a contribution to the surface field. If the cross-section of the incident localized
wave is σi , the area of the particle surface for external reflection is
σ1r = −σi (ni · ei )−1 . (3.3.42a)
For the transmitted rays, the area is given by
) *−1
σpt = −σi n1 · et1 (n1 · ei ) np · etp , p = 2, 3, 4, . . . , (3.3.42b)
where all unit vectors have been defined in Eqs. (3.3.1a) and (3.3.2b). The radius of a ray’s
cross-section is on the order of k −1 so that the phase change over this cross-section is not
significant and we may properly account for the phase interference of localized waves by
using phase information at the centers of the rays. Because phase variation over a ray’s
cross-section can be neglected, numerical results are not sensitive to the ray’s cross-sectional
shape. Hence, we may use a circular shape in the calculations.
In Figure 3.16, we first compare the extinction efficiency obtained from the conventional
geometric-optics method (CGOM) and that computed from the improved geometric-optics
method (IGOM). We then compare the results obtained from these two methods with those
determined from the finite-difference time domain (FDTD) method (see Section 4.1). On
the basis of the conventional approach, the extinction efficiency is 2, based on the equal
energy partition of geometric and diffracted light rays (see Subsection 3.3.1). The results
computed from IGOM and FDTD used randomly oriented solid columns of uniform sizes
with an aspect ratio of 6 and 0.63 µm wavelength in the computations. The size parameter
is defined by kL, with k = 2π/λ. The extinction ripples shown for IGOM are produced by
interference between geometric and diffracted rays, which converge to 2 for size parameters
approaching 1000. Because of computer time restrictions, FDTD computations were limited
to sizes smaller than 30, as denoted by crosses in the figure. The extinction efficiencies
obtained from IGOM and FDTD show excellent agreement in the size parameter range
20–30, but IGOM results deviate from FDTD for size parameters smaller than about 20.
Figure 3.17 compares the phase functions computed from IGOM and FDTD methods as
a function of the scattering angle for kL = 20 with aspect ratios of La = 6 (column) and 2
(plate). For the former, the ice crystal geometric cross-section is small, so the computation
of surface fields may not be sufficiently accurate. We see differences at scattering angles
larger than 70°. For the latter, because the ice crystal cross-section is increased by a factor
of 3, the accuracy of IGOM is greatly improved and the computed major features mimic
those produced by the FDTD method. Also shown in Figure 3.17 are the phase functions
computed by IGOM and FDTD for hollow columns and bullet rosettes. The depth of
the hollow pyramid is shown as d; the cross angle for the bullets is 90°; the tip length
of a bullet is t; and other parameters defining the geometry of the ice crystals are given
in the diagrams. The phase function patterns for solid- and hollow-column crystals are
similar, except in backscattering directions – a pronounced scattering maximum is shown
at a scattering angle of 155° in both cases. This feature does not appear in the case
of bullet rosettes. It appears that agreement of the phase function results computed from
two completely different approaches, IGOM and FDTD, is quite reasonable. The FDTD
method has inherent numerical limitations and becomes computationally expensive for size
parameters larger than 20 (see Section 4.1).
3.3.4 Ray-by-ray approach to extinction and absorption calculations

To further enhance the applicability of the geometric-optics method to small size parameters,
Yang and Liou (1997) developed a ray-by-ray approach, specifically for the calculation of
extinction and absorption cross-sections. The amplitude scattering matrix S(r), required
for single-scattering calculations, is the sum of the contributions from all localized rays as
follows:

S(r) = Sp (r), (3.3.43a)
γ p
where the vector r denotes the scattering direction. The first summation covers all the
incident rays impinging onto the sphere, denoted by γ , while the second summation is over
102 102
l=0.55μm l=0.55μm
kL=20 kL=20
Phase Function P11 101 L/a=6 101 L/a=2
Phase Function P11

100 100
10-1 10-1
FDTD FDTD
10-2 10-2
IGOM IGOM
10-3 10-3
0 60 120 180 0 60 120 180
Scattering Angle (deg.) Scattering Angle (deg.)
102 102
l=0.55μm l=0.55μm
kL=20 kL=5
101 d/L=0.25 101 d/L=0.5
L/a=6 a/L=0.25
Phase Function P11
Phase Function P11
100 100
10-1 10-1
FDTD
FDTD
10-2 10-2
IGOM
IGOM
10-3 10-3
0 60 120 180 0 60 120 180
Figure 3.17 Comparison of phase functions computed by IGOM and FDTD methods at a 0.55 µm
wavelength for solid and hollow columns, plates, and bullet rosettes as a function of the scattering
angle (after Yang and Liou 1996a, with modifications).
the internal localized rays, denoted by subscript p ( = 1, 2, 3 . . . ), based on the localization

principle. The amplitude scattering matrix for an individual ray can be written in the
form

k2 (1 − ε)
Sp (r) = Sp,q (r) = Kp Up {exp[iζp+1 (r)] − exp[iζp (r)]} ,
q
4π q m − r · ep q
(3.3.43b)
where the phase of a ray is defined by

⎛ ⎞

p−1
ζp (r) = k ⎝e0 · rQ1 + mj dj − r · rQp ⎠ , (3.3.43c)
j =1
and m denotes an average complex refractive index in a general inhomogeneous case and
can be evaluated by the Maxwell Garnett mixing rule (Chýlek et al. 1984). The term mj
represents the complex refractive index for the inhomogeneous layer in a sphere; ε is the
permittivity; k is the wavenumber; Q1 , Qp , and ep are defined in Figure 3.14; and dj is the
distance between the points Qj and Qj+1 , defined by |rQj +1 − rQj |. The summation over q
signifies the travel of a ray through homogeneous segments of an inhomogeneous particle.
The main objective is to effectively compute extinction and absorption cross-sections.
For this reason, we may simplify the preceding matrix formulation by replacing the factor
Kp Up by ±1, so that the coordinate transformation for polarization is neglected. The
diagonal elements of the scattering matrix Sp (r) can then be written in the form

k2 1−ε
Sjj,p (r) = ± {exp[iζp+1 (r)] − exp[iζp (r)]} , (3.3.44)
4π q m − r · ep q
where the ± corresponds to the sign of the cumulative product of Fresnel coefficients,
p−1
tj2 rj ; rj can take a + or − sign; and j = 1 or 2, denoting parallel or perpendicular
components, respectively.
The extinction cross-section is defined by the sum of the two diagonal elements in the
forward direction as follows [see also Eq. (2.4.45d)]:
2π
σe = Re[S1 (e0 ) + S2 (e0 )], (3.3.45a)
k2
where e0 denotes the incident direction and the absorption cross-section is given by
⎛ ⎞
∞
p−1
p−1 p−1 -
σa = exp ⎝−2k mi,j dj ⎠ [1 − exp(−2kmi,p dp )] t12 r1 + t22 r2 2,
γ p=1 j =1
(3.3.45b)
where mi,j (or p) represents the imaginary part of the refractive index for an inhomogeneous
particle. For a spherical particle, Eq. (3.3.44) is evaluated as if the path of twice-refracted
rays (p = 1) were not deflected at all, so that the phase of a ray in this equation is given by
ζ2 = 2ka cos τi (m − 1), which is the dominant exponential term for extinction, where τi is
the incident angle and a is the radius of a sphere. Note that ζl = 1, and ζp (p > 2) terms
vanish due to phase cancelation.
3.3.5 Contributions of surface waves to spheres: the edge effect

For a spherical ice particle, the extinction and absorption efficiencies and the radiation pres-
sure evaluated from the geometric-optics approach generally deviate from results computed
Incident wave front

surface wave
reflected wave
Incident localized wave

Q1
e0 refracted wave
e1
e3
Q2
reflected
wave
e2
diffracted wave

Q3
Figure 3.18 A conceptual ray-by-ray geometric-optics approach which includes surface-wave con-
tributions in addition to diffracted waves and localized geometric reflection and refraction (referred
to as GOS) for light scattering and absorption by a sphere (after Liou et al. 2010, with modifications).
from the exact Lorenz–Mie theory, due principally to the neglect of surface waves along
the edge of the spherical particle. These waves are produced by the interaction of incident
waves at grazing angles near the edges of a sphere and continuation of the wave motion
along its surface into the shadow region. If the sphere is relatively small, the waves may
move around and encompass the entire spherical surface (Figure 3.18). Also shown are
diffracted waves and waves undergoing geometric reflection and refraction. It is important
to state that surface and diffracted waves are not subject to the principle of localization for
light rays that undergo reflection and refraction, as noted previously.
Nussenzveig and Wiscombe (1980) and Nussenzveig (1992) have presented physical
equations for the calculation of surface waves based on complex angular momentum theory,
which makes use of the transformation of the Debye expansion of two scattering functions
in a complex domain. This procedure allows the mapping of localized incident rays into
a complex domain such that the Airy integral can be incorporated in the analysis. In the
remainder of this subsection, we try to capture key elements of the contribution of surface
waves to light scattering processes.
In order of size parameter x, we may express the surface-wave term for extinction in the
form
−1/2
Qext = c1 x −2/3 + 2Im[(m2 + 1)(m2 − 1) ]x −1 − c2 x −4/3
(3.3.46)
−3/2
− c3 Im[eiπ/3 (m2 − 1) (m2 + 1)(2m4 − 6m2 + 3)]x −5/3 ,
where the coefficients c1 = 1.992 39, c2 = 0.715 35, c3 = 0.664 13, and Im denotes the
imaginary part of the term. We see that the size parameter appears in the denominator of
Eq. (3.3.46), and for this reason Qext diverges when x approaches small values. Under
this condition, specific corrections must be made to ensure its physical continuity for
computational purposes.
The edge effect contribution to absorption is governed by the following two integrations:
2 ya yb
−1/3 −2/3 + − −
Qabs = 2 x ϕ(rj l )dy + [ϕ(rj l ) − ϕ(r̃j l )]dy , (3.3.47)
l=1 0 0
where l denotes two polarization components and the function ϕ is given by

ϕ(rj l ) = (1 − e−b )(1 − r2l )/(1 − r1l e−b ), (3.3.48)
where r2l and r1l are, respectively, the external and internal reflectivities for the state of
polarization l given by
rj l = |Rj l |2 , j, l = 1, 2; Rj,l = (−1)j (zj − uel )/(z + uel ), (3.3.49a)

z = cos θ, u = m cos θ , sin θ = m sin θ , el = 1,
(3.3.49b)
e2 = m−2 , z1 = z, z2 = z∗ ,
where z∗ denotes the complex conjugate of z and b = 4x Im(m cos θ + θ sin θ ). The term
rj±l in Eq. (3.3.47) can obtained from rjl such that z is replaced by
z± = −(2/x)1/3 eiπ/6 Ai (±ye2iπ/3 )/Ai(±ye2iπ/3 ), (3.3.50)
where Ai is the Airy function (Nussenzveig and Wiscombe 1980) and a prime on Ai signifies
a derivative, and θ is related to y by the following equation:
sin θ = 1 ± 2−1/3 x −2/3 y, (3.3.51)
where the + and − signs apply to the first and second integrals in Eq. (3.3.47), respectively.
The integration limits are given by
ya = 21/3 (m − 1)x 2/3 , yb = (x/2)2/3 . (3.3.52)
The term r̃j−l is obtained from rj−l such that z− is replaced by (2/x)1/3 y 1/2 .
The radiation pressure term required for calculation of the asymmetry factor for a
spherical particle can also be defined by the sum of two integrations, as follows:
2 ya yb
Qpr = −2−1/3 x −2/3 Re (ρl+ − τl+ + 1)dy + [(ρl− − ρ̂l− ) − (τl− − τ̂l− )]dy ,
l=1 0 0
(3.3.53)
where Re denotes the real part of the term, which is generally negative, and
∗
ρl = f1 (z)R2l R2l ,
τl = f1 (z)f2 e−b (1 + R1l
∗
)(1 + R1l ∗
)(1 + R2l
)(1 + R2l ∗
)(1 + R1l R1l f2 e−b )−1 ,
Rj 1 = (fj )−1 [m2 zj − u + (−1)j iM 2 ](m2 z + u + iM 2 )−1 ,
Rj 2 = (fj )−1 [(m2 + M 2 )zj − u + (−1)j iM 2 (1 − uzj )]
× [(m2 + M 2 )z + u + iM 2 (1 + uz)]−1 ,

f1 (z) = (1 + iz∗ )/(1 − iz), f2 = e−2iθ , (3.3.54)
with M 2 = m2 − 1. In all the parameters in Eq. (3.3.53) with ± upper indices, substitutions
denoted by Eqs. (3.3.50) and (3.3.51) are required. The terms ρ̂l− and τ̂l− are obtained from
√
ρl− and τl− such that z− is replaced by (2/x)1/3 ( y + i/4y).
3.3.6 The geometric-optics and surface-wave approach for spheres

A combination of the geometric-optics term and the surface-wave adjustment (referred to
as GOS) should, in principle, constitute a solution close to the “exact” solution derived
from the Lorenz–Mie (LM) theory such that
Qw (GOS) = Qw (GO) + Qw ≈ Qw (LM), (3.3.55)
where w denotes ext (extinction), abs (absorption), or pr (radiation pressure). In this equa-
tion, linearity between the geometric-optics (GO) and surface-wave () terms is implied.
The conventional geometric approach can be employed to calculate the asymmetry fac-
tor for homogeneous and layer spheres, which in combination with known extinction and
absorption coefficients can be used to evaluate a term referred to as radiation pressure
(van de Hulst 1957). Light rays can carry momentum as well as energy. The part of the
forward momentum that is removed from incident rays, which is not represented by the
forward momentum associated with scattered rays, is related to the hemispheric average of
the phase function weighted by cos θ , where θ is the zenith angle in Cartesian coordinates.
This term is referred to as the asymmetry factor. We can express the radiation pressure term
as follows:
Qpr (GOS) = Qext (GOS) − g(GOS)[Qext (GOS) − Qabs (GOS)], (3.3.56)
where g represents the asymmetry factor, Qext (= σe /πa 2 ) is the extinction efficiency,
Qabs (= σa /πa 2 ) is the absorption efficiency, and the index GOS denotes the terms evaluated
by the geometric-optics and surface-wave approach.
Based on Eq. (3.3.56), the asymmetry factor can be written in the form
Qext (GOS) − Qpr (GOS)
g(GOS) = . (3.3.57)
Qext (GOS) − Qabs (GOS)
The term Qext breaks down when x approaches zero, since the size parameter x appears
in the denominator in the right-hand side of Eq. (3.3.46). To investigate applicability of the
surface-wave term derived from complex angular momentum theory, we have conducted
an analysis of Qext in terms of the phase shift parameter, a function of both the refractive
index and the size parameter, defined as ρ = 2 |m − 1| x. The surface-wave adjustment term
Qext , which is normally on the order of 0.1, approaches an unrealistically large number by
a factor of more than 10 when ρ 2.86. The extinction efficiency results computed from the
ray-by-ray geometric-optics approach outlined above, without including the surface-wave
effect, match closely with those determined from the exact LM theory when ρ < 2.86, so
surface-wave adjustments are not required in this region.
Extinction Efficiency Q ext Single-Scattering Albedo v Asymmetry Factor g

1.0 1.0
4
0.9
LM 0.9
m = 1.3 − i 0.0
GOS
3
0.8 0.8
2 0.7
0.7
1 0.6 0.6
0 0.5 0.5
1.0 1.0
4
0.9 0.9
m = 1.3 − i 0.01
3
0.8 0.8
2
0.7 0.7
1 0.6 0.6
0 0.5 0.5
0.8 1.0
4
0.7 0.9
m = 1.3 − i 0.1
3
0.6 0.8
2
0.5 0.7
1 0.4 0.6
0 0.3 0.5
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
Size Parameter x Size Parameter x Size Parameter x
Figure 3.19 Comparison of extinction efficiency, single-scattering albedo, and asymmetry factor
as a function of size parameter between the results computed from Lorenz–Mie (LM) theory and
the geometric-optics and surface-wave (GOS) approach for a real refractive index of 1.3 and three
imaginary indices of 0.0, 0.01, and 0.1 (after Liou et al. 2010).
We have carried out a series of comparisons between the extinction efficiency Qext , the
single-scattering albedo , and the asymmetry factor g determined from GOS, and those
computed from LM equations for homogeneous spheres with size parameters ranging from
1 to 100. Figure 3.19 displays comparison results for mr = 1.3 and mi = 0.0, 0.01, and 0.1.
Both Qext and values computed from the GOS approach compare remarkably well with
the exact LM results. Maximum and minimum oscillations in the single-scattering patterns
are seen to be progressively damped when mi or absorption increases.
3.3.7 Application of the geometric-optics and surface-wave approach to ice crystals

3.3.7.1 Fundamentals
For application of the GOS approach to non-spherical ice crystals, we shall consider a gen-
eral aggregate particle comprised of a collection of core–shell spheres built by a stochastic
Surface wave
Reflected
photons
Incident photons
Refracted Diffracted wave

photons
Figure 3.20 A conceptual depiction of the geometric-optics and surface-wave approach to light
scattering by an ensemble of core–shell spheres (aggregates), representing general non-spherical
particles, including ice crystals. The geometric-optics components include the hit-and-miss Monte
Carlo photon tracing associated with external and internal reflections and refractions, and diffraction
based on Babinet’s optical principle for randomly oriented non-spherical particles. The surface waves
travel along the edge of the particles and propagate into shadow regions (after Liou et al. 2011 with
modifications).
process (Friedlander 2000), as illustrated in Figure 3.20. This diagram demonstrates the
concept of the traversing of surface waves, or the edge effects, which can touch the spheres
when they propagate in forward directions and can move into shadow regions of the aggre-
gate. Also shown are diffracted waves acting upon the projected cross-sectional area in
accordance with Babinet’s principle and passing around this area. In geometric reflec-
tion/refraction, we have employed the hit-and-miss Monte Carlo technique to treat the
interaction of light beams in terms of rays (see Subsection 3.1.3) in order to account for
the intricacy and complexity of an aggregate comprised of spheres with layered structures,
following the ray-by-ray approach described in Subsection 3.3.4.
3.3.7.2 Geometric optics

In conjunction with the conceptual depiction in Figure 3.20, consider a group of randomly
oriented ice crystals in space to replace the aggregate. Let A be the area of a square
sufficiently large to cover the geometric cross-section of an individual ice crystal and set
the square equal to L2 , where L denotes the crystal’s maximum dimension. Further, let Na
be the number of photons incident on the ice crystals, which is dependent on the particle’s
orientation, and Nt be the number of total photons used in the calculation. Thus, the effective
geometric shadow of an ice crystal on a plane perpendicular to the incident light beam,
whose propagation direction is along the z-axis, can be defined as follows:
As (α, β) = A [Na (α, β)/Nt ] (3.3.58)

where α and β are angles that denote the orientation of an aggregate in a two-dimensional
plane with reference to the incident light beam.
It follows that for an ensemble of randomly oriented ice crystals, the extinction and
absorption efficiencies averaged over all orientations can be written in the form

Qext,abs = σe,a (α, β) cos αdαdβ As (α, β) cos αdαdβ. (3.3.59)
α β α β
The original theoretical development of diffraction begins with Babinet’s principle,

which states that the diffraction pattern in the far field (Fraunhofer diffraction) from a
circular aperture is the same as that from an opaque disk or sphere of the same radius,
as discussed in Section 3.2. Following this principle and in order to apply it to randomly
oriented non-spherical particles, we may use the effective geometric cross-sectional area
defined in Eq. (3.3.58). It follows that the diffracted intensity Id for an ensemble of randomly
oriented ice particles can be expressed as
2

Id ∝ exp[−ikr(x, y, α, β) dxdy] cos αdαdβ, (3.3.60)

α β As
where r is the distance between a point in the shadow As and a point distant from the shadow.
Analytical solutions of Eq. (3.3.60) exist only for a number of simple non-spherical shapes,
such as column/plate, rectangular parallelepiped, and spheroid/ellipsoid (see Section 3.2).
However, “exact” numerical calculations can be carried out for irregular ice particles to
obtain the diffracted energy.
On the basis of the scattered intensities determined from geometric reflection/refraction
and diffraction for randomly oriented ice particles, the asymmetry factor g can be deter-
mined. We can then evaluate the term referred to as radiation pressure efficiency, the concept
of which can be applied to non-spherical particles as follows:
Qpr (GO) = Qext (GO) − g(GO)[Qext (GO) − Qabs (GO)] (3.3.61)
where GO stands for geometric optics and the extinction and absorption efficiencies Qext
and Qabs have been defined in Eq. (3.3.59).
3.3.7.3 Addition of surface waves

In the context of light scattering and absorption by a sphere, surface waves are generated by
interactions of incident waves at grazing angles near the sphere’s edge, and they propagate
along its surface into the shadow region, as shown in Figure 3.18. On the basis of theoretical
postulations and numerical calculations, Liou et al. (2011) demonstrated that a linear
combination (reflection/refraction and diffraction, denoted by GO) of the geometric-optics
term and coupled with the surface-wave adjustment (denoted by GOS) constitutes a solution
close to the exact Lorenz–Mie (LM) theory, so that
Qw (GOS) = Qw (GO) + f Qw ∼ Qw (LM), w = ext, abs, pr (3.3.62)

where Qw is the efficiency factor for extinction (ext), absorption (abs), or radiation pressure
(pr), and Qw is the surface-wave adjustment. Additionally, a parameter f has been added
into this equation for the case of non-spherical particles. In the previous analysis for spheres,
f was set to be 1. However, for large size parameters, greater than about 50, f is close to 0.
Once Qpr has been determined, the asymmetry factor from Eq. (3.3.57) applied to the case
of GOS can be expressed in the form
g(GOS) = [1 − Qpr (GOS)/Qext (GOS)]/ (GOS), (3.3.63)
where the single-scattering albedo is defined by Qsca /Qext and Qsca = Qext − Qabs .
Surface waves differ from diffraction, which originates from an incomplete wave front
due to blocking of the incident wave by a group of randomly oriented ice particles and
is governed by the geometric cross-sectional area facing the incident light beam. Surface
waves propagate along and touch ice crystal surfaces and must therefore be governed by
absorption in addition to scattering. The theoretical foundation for surface waves exists
only for spheres, as discussed in Subsection 3.3.5. For non-spherical particles, a physically
based adjustment must be developed in order to effectively account for the shape factor.
Based on physical arguments, it is postulated that irregular surfaces will suppress or dampen
the propagation of these waves and reduce the effectiveness of their interaction with ice
particles – slow down the waves, so to speak. By virtue of its smooth surface, a perfect
sphere is the most efficient geometry for surface-wave propagation. Therefore, the parameter
f introduced in Eq. (3.3.62) must be less than 1 for non-spherical ice particles; it must be
close to zero for an elongated particle since the probabilities for the forward and backward
propagation of surface waves would be limited in this case. Volume (three-dimensional)
must also play a key role in the effectiveness of surface-wave contributions.
In the following analysis, consider two types of volume. First, we may define a volume
for an ice crystal as follows:
-
Vv = n × 4πrv3 3, (3.3.64a)
where rv is the equivalent radius of a sphere with the same volume as the individual ice
particle, and n is the total number of ice crystals. We may also define a volume corresponding
to the geometric cross-sectional area As , defined in Eq. (3.3.58), such that
-
Va = n × 4πra3 3, (3.3.64b)
where πra2 = As . It is postulated that the correction parameter in Eq. (3.3.62) for non-
spherical particles may be defined by the ratio of two volumes as follows:
f = Vv /Va = (rv /ra )3 . (3.3.65)
It turns out that Vv (or rv ) is smaller than Va (or ra ), so that f is equal to or less than
1. For spheres, rv = ra , and f = 1; for elongated particles, rv ra , f 0. These limits
provide a theoretical basis and boundary conditions for the application of Eq. (3.3.62) to
ice particles.
3.3.7.4 Comparison with available FDTD and DDA results

The preceding theoretical development for non-spherical particles has been compared to
calculations of light scattering and absorption by column and plate crystals using the finite-
difference time domain (FDTD) method (Yang and Liou 1996a). Comparison has also been
made to calculations using the discrete dipole approximation (DDA) method originally
developed by Purcell and Pennypacker (1973) and later by Draine and Flatau (1994),
in which single-scattering results for small size parameters for column and plate can be
computed from existing computer programs. The DDA method calculates light scattering
and absorption by a target with an arbitrary geometry by defining the target as an array
of discrete dipoles located on a cubic lattice. Both the FDTD and DDA methods require a
substantial amount of computational effort for moderate size parameters, especially when
the 3-D orientation of the non-spherical particles is taken into account. For this reason,
it is computationally challenging to use FDTD and DDA for many practical applications
involving non-spherical particles with size parameters larger than about 20 (see Chapter 4
for further discussion).
For comparison purposes, wavelengths of 3.7 and 23 µm have been selected, which have
ice indices of refraction of 1.4005 − i 0.007 201 and 1.4424 − i 0.027 01, respectively.
Figure 3.21 compares the extinction efficiency, single-scattering albedo, and asymmetry
factor for randomly oriented columns (a) and plates (b) as a function of size parameter,
defined as xmax = krmax (= Lmax /2), where Lmax is the maximum dimension and k is the
wavenumber. Because of computational limitations, FDTD and DDA results are shown
for size parameters less than about 20. For columns, GOS results compare closely with
values computed from FDTD and DDA for both wavelengths for the three single-scattering
parameters. GOS results show slight systematic overestimates of extinction efficiency for
size parameters smaller than 2. For plates, GOS results compare closely with values
calculated from FDTD and DDA for single-scattering albedo as well as asymmetry factor.
Some deviations are shown in extinction efficiency curves; in particular, the GOS approach,
which is a linear combination of the geometric-optics and surface-wave approaches, does
not produce the fluctuations of the FDTD results at 23 µm.
3.3.8 Geometric-optics and surface-wave approach coupled with the

Rayleigh–Gans–Debye approximation
In the preceding presentations, we demonstrated that application of the improved geometric-
optics method (IGOM) can be extended to light scattering by ice particles with size parame-
ters of 20. Further, by incorporation of surface-wave contributions to account for the edge
effect, we have further extended the geometric-optics and surface-wave (GOS) approach
to size parameters on the order of 1. Results computed from FDTD and DDA numerical
methods have been used to calibrate those from GOS for light scattering by column and
plate crystals with size parameters from 1 to 20. Ice crystal sizes in the atmosphere
cover a range from a few µm to thousands of µm (see Figure 1.23) with size parameters
generally larger than 1 for visible and infrared wavelengths. In what follows, we present
(a) Extinction Efficiency Qext Single-Scattering Albedo ϖ Asymmetry Factor g
1.0 1.0
3
l = 3.7μm 0.8 0.8
2 0.6 0.6
Column
0.4 0.4
1
GOS
0.2 0.2
FDTD
DDA
0 0.0 0.0
1.0 1.0
3
0.8 0.8
l = 23 μm
2 0.6 0.6
0.4 0.4
1
0.2 0.2
0 0.0 0.0
1 10 100 1 10 100 1 10 100
Size Parameter xmax Size Parameter xmax Size Parameter xmax
(b) Extinction Efficiency Qext Single-Scattering Albedo ϖ Asymmetry Factor g

1.0 1.0
3
0.8 0.8
l = 3.7μm
2 0.6 0.6
Plate
0.4 0.4
GOS
1
FDTD
0.2 0.2
DDA
0 0.0 0.0
1.0 1.0
3
0.8 0.8
l = 23 μm
2 0.6 0.6
0.4 0.4
1
0.2 0.2
0 0.0 0.0
1 10 100 1 10 100 1 10 100
Size Parameter xmax Size Parameter xmax Size Parameter xmax
Figure 3.21 Comparison of extinction efficiency, single-scattering albedo, and asymmetry factor
for randomly oriented columns (a) and plates (b) as a function of size parameter xmax (= 2π rmax λ),
where rmax denotes one-half of the maximum dimension and λ is the wavelength, as computed from
the geometric-optics and surface-wave (GOS) approach for size parameters up to 100, and from the
finite-difference time domain (FDTD) and discrete dipole approximation (DDA) computer codes for
size parameters smaller than about 20 (after Liou et al. 2011).
an approximation that is applicable to size parameters smaller than 1, as in the case of
aerosols, some of which can serve as effective condensation and ice nuclei for the formation
and growth of ice crystals in the Earth’s atmosphere.
We shall first consider Rayleigh–Gans scattering for a spherical particle. This is an
extension of Rayleigh scattering to particle sizes larger than molecules by taking into
account the interference effect from all elements at different positions in a sphere for the
case when the real refractive index mr is close to 1.
The original Rayleigh–Gans scattering intensity equation was derived by Gans (1925)
in the form

9k 4 V 2 4
I RG = (m r − 1) 2
I0 G2 (u)(1 + cos2 ), (3.3.66)
32π2 r 2 9
where V is the volume of a sphere, k = 2πλ, λ is the wavelength, r is the distance from
the scattering particle to an observation point, I0 is the incident intensity, is the scattering
angle, and the factor G(u), which accounts for phase interferences, is given by
G (u) = 3 (sin u − u cos u) /u3 , (3.3.67)
where the parameter u = 2 ka sin (2). In Eq. (3.3.66), the polarizability α is approximated
by the term 2(mr −1)3.
In order to apply this to particles with a real index of refraction much larger than 1,
we may replace the factor (mr − 1)2 in the scattering intensity expression by the exact
polarizability formula given by
α 2 = (9/4)|(m2 − 1)/(m2 + 2)|2 , (3.3.68)
where m (= mr − imi ) is the complex refractive index. In this manner, the scattered intensity
can now be expressed in the form
2
9k 4 V 2 m2 − 1
I =
RG
I0 G2 (u)(1 + cos2 ). (3.3.69)
32π2 r 2 m2 + 2
Integrating Eq. (3.3.69) over all directions, we obtain the scattering cross-section as follows:
2 π
1 9k 4 V 2 m2 − 1
Csca =
RG
I r d =
RG 2
G2 (u)(1 + cos2 ) sin d. (3.3.70a)
I0 16π m2 + 2 0
Additionally, the asymmetry factor is defined by
π π
g =
RG RG
I cos sin d I RG sin d . (3.3.70b)
0 0
Because the interference effect does not affect absorption, the absorption cross-section
for Rayleigh–Gans scattering is similar to that for Rayleigh scattering, which is given by
2
m −1
RG
Cabs = −3kV Im , (3.3.71)
m2 + 2
where V is the volume and Im denotes the imaginary part of the bracket value.
For general application, we shall consider an aggregated particle (such as black carbon
or soot) consisting of Ns homogeneous spheres and define a form factor F as the ratio of
the scattered intensity for Ns spheres to that for a single sphere. We shall follow the form
factor developed by Debye (1915) in the form

Ns
2
Ns
Ns
Ns
Ns
Ns
Ns
F= |eiq·rj | = ei q·(rj −rk ) = cos(q · rj k ) = cos(|q||rj k | cos η),
j =1 j =1 k=1 j =1 k=1 j =1 k=1
(3.3.72)
where q is the vector corresponding to the scattered wave with a real value given by
|q| = 2k sin(/2), |rj k | = |rj − rk | = rj k , where rj and rk are position vectors of primary
spheres, and η is the angle between the q and rjk vectors. Each primary sphere is governed
by Rayleigh–Gans scattering, and to a good approximation it can be considered to be inde-
pendent of the other primary spheres. Further, we shall consider an ensemble of aggregates
randomly oriented in space so that
1 1
I RGD = I RG F dcosη ∼ = I RG F dcosη = I RG F̄ , (3.3.73)
0 0
where the mean form factor is given by

1 Ns
Ns 1
Ns
Ns
sin[2krj k sin(/2)]
F̄ = F dcosη = cos(|q||rj k | cos η) dcosη = .
0 j =1 k=1 0 j =1 k=1
2krj k sin(/2)
(3.3.74)
Accordingly, the scattering cross-section is given by

2 π
9k 4 V 2 m2 − 1
RGD
Csca = G2 (u)(1 + cos2 )F̄ sin d. (3.3.75)
16π m2 + 2 0
The absorption cross-section is simply Ns times the Rayleigh absorption cross-section,
given by
2
m −1
Cabs = Ns Cabs = −3Ns kV Im
RGD RG
. (3.3.76)
m2 + 2
The preceding equations, (3.3.73)–(3.3.76) constitute the Rayleigh–Gans–Debye (RGD)
approximation, applicable to size parameters x (= ka) smaller than 1.
In Figure 3.22, we show a comparison of the scattering phase matrix for randomly
oriented soot aggregates of Ns = 105, corresponding to a size parameter of 1.39, between
the RGD approximation and the superposition T-matrix method (see Section 4.2 for the T-
matrix approach). The phase function P11 shows a peak at a scattering angle of 0°, resulting
from diffraction, and becomes flat at backscattering directions. The degree of linear polar-
ization −P12 P11 has a maximum value of 1 at a scattering angle of 90°, which is the same
10 2 1.0
P22 /P11
0.8
RGD 0.6
T−matrix 0.4
−P12/P11
0.2
101
Phase Function P11
0.0
0 30 60 90 120 150 180
1.0
0.5
10 0
0.0
P43 /P11
P33 /P11
−0.5
P44/P11
10−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 3.22 Comparison of scattering phase matrix elements for a soot aggregate with Ns = 105
(xv = 1.39) between the Rayleigh–Gans–Debye (RGD) approximation and the superposition T-matrix
method (after Takano et al. 2013 with modifications).
as the Rayleigh scattering result, and P21 P11 = P12 P11 under the condition of random
orientation. Values of the matrix element P22 P11 are close to 1, whereas P43 P11 0.
The element P44 P11 is essentially equal to the element P33 P11 at all scattering
angles. Results computed for all matrix elements from the two approaches reveal close
comparison.
Based on the RGD approach, the scattering phase matrix for size parameters less than
about 1 may be expressed in the form [also see Eq. (2.4.58b)]
⎡ ⎤
P11 P12 0 0
⎢ P12 P11 0 0 ⎥
P(ka < 1) ≈ ⎢
⎣ 0
⎥. (3.3.77)
0 P33 0 ⎦
0 0 0 P33
Thus, the scattering phase matrix for randomly oriented small ice particles with size param-
eters less than 1 contains only three independent elements, similar to the Rayleigh scattering
case (Rayleigh 1871, Chandrasekhar 1950).
3.4 A unified theory of light scattering by ice crystals based on generalized

geometric optics
3.4.1 The essence of a unified theory of light scattering by ice crystals
In recognition of the inherent limitations of various approaches, including those pre-
sented above and in Chapter 4 for light-scattering calculations, it is unlikely that one
specific method can be satisfactorily employed to resolve intricate and complex scattering–
absorption problems involving ice crystals and snowflakes covering all possible size param-
eters from 0.1 to 1000, the numerous shapes displayed in Section 1.5, as well as external
and internal mixings of aerosol particles serving as ice nuclei or condensation nuclei. For
this reason, we have presented the concept of “a” unified theory of light scattering by ice
crystals (Liou et al. 2000), which combines the IGOM (improved geometric-optics method)
covering size parameters larger than 20 and the FDTD covering size parameters smaller
than 20. In light of the development of generalized geometric-optics approaches discussed
in preceding sections, we present in the remainder of this subsection “a” unified theory of
light scattering by ice crystals covering essentially all size parameters and shapes. We must
hasten to add that such a theory could “in principle” be developed by a combination of
numerical approaches to light scattering solutions as presented in Chapter 4.
It is pertinent herein to define the meaning of “exact” solutions within the context of
light scattering and radiative transfer. In Subsection 2.2.3, we demonstrated that “exact”
solutions exist only for sphere, circular cylinder, and spheroid, for which known coordinate
systems can be imposed, leading to analytic solutions of the electric field in the far field
in terms of a number of mathematical functions. In this context, we have also shown that
“exact” solutions can be derived for diffraction by sphere, rectangular, hexagonal cylinder,
and spheroid on the basis of Babinet’s principle. Numerical solutions developed for the
FDTD, DDA, and T-matrix approaches presented in Chapter 4 for light scattering problems
require a large number of numerical matrix inversions, resulting in substantial computer
time to achieve convergence and significant efforts to circumvent numerical instability.
Therefore, these numerical approaches are not within the purview of our definition of
“exact.”
In essence, we built “a” unified theory of light scattering by ice crystals based on
physical rationale and postulations within the context of first principles, which is critical
for understanding of and insight into the scattering, absorption, and polarization features
that occur in theoretical and numerical results. To demonstrate the unified theory, we shall
first use the spherical case for which the analytic solutions for extinction efficiency, single-
scattering albedo, and asymmetry factor can be derived from the exact Lorenz–Mie (LM)
theory, followed by numerical computations.
In Figure 3.23a, we present a comparison of single-scattering parameters Qext , co-albedo
1− , and g for a homogeneous ice sphere computed from LM and from the GOS/RGD
approximation as a function of size parameter x from 0.1 to 1000 associated with the
0.55 µm wavelength. In the Qext curve, we show the applicability of the CGOM, IGOM,
and GOS/RGD approaches in reference to exact LM results. In the context of CGOM
(a) (b)
5
3 IGOM
Qext
CGOM
2
1 GOS/RGD
0
10 −4 1
Ice l = 0.55 μm 0.8

10 −5
(1 - v) / v
GOS/RGD
LM 0.6 mcore = 2 - 1i, mshell = 1.5
10 −6 acore/ashell = 0.2
0.4
GOS/RGD
10 −7 0.2 LM
1-v v
10 −8 0
1
0.8
0.6
g
0.4
0.2
0
10 −1 100 10 1 10 2 10 3 10 −1 10 0 101 102 103
Size Parameter x Size Parameter x shell
Figure 3.23 Illustration of a unified theory of light scattering by ice crystals vis-à-vis GOS/RGD. (a) Comparison of the unified theory and exact
LM results for a homogeneous ice sphere in terms of the extinction efficiency Qext , the single-scattering co-albedo (1 − ), and the asymmetry
factor g as functions of the size parameter from 0.1 to 1000, using a wavelength of 0.55 µm in the calculations. In the Qext curve, limitations of
CGOM and IGOM are also displayed. (b) Same as (a), but for a core–shell sphere. Also, the single-scattering albedo is used in the comparison
and the size parameter is in reference to the spherical shell radius. The real and imaginary indices of refraction for the core and the shell employed
in the calculations are depicted in the figure, as is the core–shell ratio.
(conventional geometric-optics method), the extinction cross-section is a factor of 2 larger

than the geometric cross-section πr2 , as a result of the postulation of equal partition of
diffraction and geometric reflection/refraction processes. The IGOM is applicable to size
parameters 20–30. The results from GOS (for x 2) coupled with RGD (for x 2)
closely match those from LM, and differences cannot be distinguished between the two
curves presented. For x 2, Qsca is proportional to x4 , whereas Qabs is proportional to x
(van de Hulst 1957). For this reason, the pattern of the single-scattering co-albedo (1− )
is proportional to x−3 in this size parameter range. The sinusoidal oscillation in (1 − ) is
out of phase in comparison to Qext . When x 4, (1 − ) computed from LM increases
monotonically with sharp peaks due to resonances produced from non-linear interference
between diffraction, surface waves, and geometric reflection/refraction absent from GOS
based on a linear combination of these three components [see Eq. (3.3.55)]. The asymmetry
factor has sinusoidal oscillations that are in phase with Qext and we see close agreement of
g between the results from the unified theory based on the GOS/RGD approach and from
exact LM calculations.
Figure 3.23b compares the scattering and absorption results for a core–shell sphere
determined from GOS/RGD and the LM computational program. In this figure, the curves
for Qext , , and g for a sphere with core–shell structure are similar to those for a homo-
geneous sphere, depicted in Figure 3.23a. Minor differences stem from deviation in the
real refractive index (1.311 versus 1.5). However, when x 2, becomes almost constant
due to a large value of mi (core, assuming a value of 1) used in the calculations. This
constant behavior is quite different from the monotonic decrease for an equivalent sphere if
an average refractive index is derived from the Maxwell-Garnett mixing rule (Chýlek et al.
1984) and used to compute the single-scattering properties, suggesting that we must be
cautious in using an average refractive index as a surrogate for an inhomogeneous sphere
with a core–shell structure.
We wish to link the core–shell structure to black carbon (BC, or soot) in the atmosphere.
BC has been identified as the second most important anthropogenic global warming agent
in the atmosphere because it strongly absorbs solar radiation and concurrently acts as
cloud condensation nuclei (CCN) or ice nuclei for cloud formation. The two effects are
significantly influenced by its aging processes in the atmosphere, which transform BC from
external to internal mixing. Freshly emitted BC particles are mostly hydrophobic and shortly
after emission they form irregular aggregates. Hydrophobic BC can become hydrophilic
through coating by soluble materials during aging, including condensation of sulfate, nitrate,
and some organics, coagulation with pre-existing aerosols, and heterogeneous reactions
with gaseous oxidants. Aged BC particles would experience hygroscopic growth and could
be activated as CCN for the formation of ice particles in subfreezing temperatures. BC
particles in the atmosphere are mostly in the form of core (BC) and shell (mostly water).
Figure 3.24 illustrates a comparison of the single-scattering parameters Qext , , and g
for randomly oriented columns (a) and plates (b) as a function of size parameter among the
three approaches: GOS/RGD, FDTD, and DDA. The size parameter is defined as xmax =
kLmax 2, where Lmax denotes the maximum dimension and k is the wavenumber. Because
(a) (b)
5
Qext 2
0
1
0.8
0.6
Ice l = 3.7 μm Ice l = 3.7 μm
v
0.4
GOS/RGD
FDTD
0.2
DDA
0
1
0.8
0.6
g
0.4
0.2
0
10 −1 10 0 101 10 2 103 10 −1 10 0 101 10 2 103
Size Parameter xmax Size Parameter xmax
Figure 3.24 Illustration of a unified theory of light scattering by ice crystals vis-à-vis GOS/RGD. Comparison of the unified theory and FDTD and
DDA results for randomly oriented ice crystals with variable length-to-width ratios for (a) solid columns and (b) plates, in terms of the extinction
efficiency Qext , the single-scattering albedo , and the asymmetry factor g as functions of the size parameter (with reference to the maximum
dimension) from 0.1 to 1000, using a wavelength of 3.7 µm in the calculations (see also Figure 3.21).
10 4 1.0
l = 0.63 μm 0.8
De = 24 μm
0.6
50% +30% +20%
10 3 0.4 P22 /P11
smooth surfaces
0.2 −P12 /P11
0.0
Phase Function P11
10 2 −0.2
0 30 60 90 120 150 180
1.0
10 1
P44 /P11
0.5
10 0 P43 /P11
0.0
P33 /P11
10 −1 −0.5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 3.25 Scattering phase matrix elements computed from GOS for an effective size of 24 µm
composed of 50% bullet rosettes, 30% hollow columns, and 20% plates, all with smooth surfaces,
using a wavelength of 0.63 µm in the calculations.
of computational limitations, FDTD and DDA results are shown for size parameters less
than about 20. The curves for Qext , , and g are similar for columns and plates, although
minor differences can be seen. Comparison of the GOS/RGD results with those computed
from FDTD and DDA in the size parameter range 0.3–20 shows close agreement, providing
a validation of the GOS/RGD approach. Due to ice absorption at the 3.7 µm wavelength,
Qext , , and g values approach their asymptotic limits of 2, 0.5, and 1, respectively,
when the size parameter approaches 103 . In the following paragraphs, we present scattering
phase matrix elements for typical ice crystal sizes and shapes that occur in the Earth’s
atmosphere as computed from GOS.
Figure 3.25 illustrates phase matrix elements for light scattering by a group of ice crystals
having an effective size De of 24 µm, corresponding to cold cirrus comprised of 50% bullet
rosettes, 30% hollow columns, and 20% plates. In the calculations, we have assumed that
all ice particles have smooth surfaces and have used a visible wavelength of 0.63 µm. In
this case, the ice crystal size parameter is on the order of 120. In the phase function P11
we see inner and outer halo features at the scattering angles = 22° and 46°. Also, we
see another peak at a scattering angle of 10°, which is produced by rays transmitted (or
3.4 A unified theory of scattering by ice crystals 183
10 4 1.0
l = 0.63 μm 0.8
De = 24 μm
0.6
5 0% +30% +20%
10 3 0.4
rough sufaces P22 /P11
0.2 −P12 /P11
0.0
Phase Function P11
10 2 −0.2
0 30 60 90 120 150 180
1.0
10 1
P44/P11
0.5
10 0 P43 /P11
0.0
P33 /P11
10 −1 −0.5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 3.26 Same as Figure 3.25, except that all ice particles are assumed to have rough surfaces.
See Subsection 3.1.3 for the treatment of light scattering by roughed ice crystals.
refracted) through a pyramid face and a prism face involving bullet rosettes. A broad peak
at = 154° results from rays undergoing one or two internal reflections. With reference
to the degree of linear polarization LP = −P12 P11 , two local minima are seen at around
22° and 46°, corresponding to the halo features in P11 . At 50°, LP is positive, except
in backward directions. P22 P11 is close to 1 at forward directions and deviates from 1
at 30°. Deviation from 1 in this phase matrix element is an indication of scattering
by non-spherical ice particles. The element P44 P11 is larger than P33 P11 , due also to
non-sphericity, as is in the case for spheroids illustrated in Figure 2.20. The non-diagonal
element P43 P11 is closer to 0 than the diagonal elements.
In Figure 3.26, we show phase matrix elements when intense roughness (see Subsection
3.1.3 for light scattering by roughed ice crystals) is added to all ice particles. The surface
roughness results in substantial reduction of the 22° halo peak in the phase function. Also,
the halo features located at = 10° and 46° as well as the 170° peak largely disappear.
In the backscattering direction from 110° to 180°, an angular region important in remote
sensing applications, the phase function is essentially flat and featureless. The degree of
linear polarization −P12 P11 is closer to zero than in the smooth surface case. We also
note that matrix elements P22 P11 , P33 P11 , P44 P11 , and P43 P11 are smoothed out at
backward directions for 130°. In order to interpret the phase function results derived
from laboratory-generated ice crystal clouds, it is noted that a certain portion of rough ice
crystals must be included in theoretical calculations (see the following Subsection 3.4.2).
Also, the phase function curves determined by a number of researchers from cirrus clouds
in the atmosphere have illustrated that in general the 22° halo feature is relatively small and
that most of the fascinating halos and arcs have not been observed from aircraft platforms.
Finally, in Figure 3.27, we present the extinction efficiency Qext , single-scattering albedo
, and asymmetry factor g as a function of wavelength from 0.2 to 100 µm for randomly
oriented Koch snowflakes, having a spherical volume-equivalent radius of 100 µm. For
comparison purposes, also shown are these parameters for hexagonal plates as a function
of wavelength from 0.2 to 5 µm (solar or shortwave regions) taken from Liou et al. (2014).
As shown, Qext is close to 2 for solar wavelengths. But for thermal infrared or longwave,
Qext is less than 2, partly because of substantial absorption. The single-scattering albedo for
Koch snowflakes is similar to that for hexagonal plates at wavelengths smaller than 5 µm
due to the same volume used in light absorption and scattering. For wavelengths between
5 and 100 µm, due to substantial absorption, the single-scattering albedo is similar to those
for spheres and circular cylinders, as shown in Figures 2.17 and 2.18. The asymmetry factor
for Koch snowflakes is smaller than that for hexagonal plates at visible and near-infrared
wavelengths, consistent with the value at λ = 0.55 µm presented in Liou et al. (2014).
Variation of the asymmetry factor for λ> 5 µm is similar to the values obtained for sphere
and circular cylinder.
In summary, we have demonstrated that the unified theory of light scattering by ice
crystals on the basis of the GOS/RGD approach can be employed for accurate calculations
of the extinction efficiency, single-scattering albedo, and asymmetry factor for hexagonal
and irregular ice particles as well as internal mixing of absorbing aerosols (e.g., black
carbon and dust) using spherical particles as a proxy. These single-scattering properties
are most relevant to application in radiation parameterizations involving ice crystals and
snowflakes in the Earth’s atmosphere and for incorporation into computer models used in
climate studies, as illustrated in Section 6.2.
Furthermore, we have shown that GOS/RGD can be used for the calculation of the
scattering phase matrix. The elements, which comprise the phase function and linear
polarization information for ice particles, have been used to develop numerical algorithms
for satellite remote sensing of cirrus clouds in terms of their optical depth and ice crystal
size and shape by means of bidirectional reflected sunlight as a function of satellite orbit
and sun position. In Sections 5.2 and 5.3, we will demonstrate the importance of ice crystal
phase functions for reliable retrieval of ice cloud parameters. In addition, backscattering
and polarization data for ice crystals with respect to their orientation state and in reference
to spherical water droplets are important to lidar detection of cirrus cloud microphysics
(Section 5.5).
In Subsection 3.4.2, we discuss the development of a separate but associated experi-
mental program which provides an independent phase function and polarization datasets
determined from laboratory scattering and ice microphysics experiments, for the specific
purposes of comparing with and cross-checking theoretical results.
Extinction Efficiency
3
0
1.0
rsnow = 100 μm Single-Scattering Albedo
0.8
0.6
0.4
0.2
0.0
1.2
Asymmetry Factor
1.0
0.8
0.6
0.4
10 −1 10 0 101 10 2
Wavelength (μm)
Figure 3.27 Extinction efficiency, single-scattering albedo, and asymmetry factor as a function of
wavelength from 0.2 to 100 µm for Koch snowflakes with a spherical volume-equivalent radius of
100 µm. Also shown for comparison purposes are results for hexagonal plates covering the region
from 0.2 to 5 µm.
3.4.2 Laboratory scattering and ice microphysics experiments with ice crystals
Measurements of the angular scattering and polarization patterns of ice crystals have
been conducted in laboratory cloud chambers, beginning with experiments performed
by Huffman and Thursby (1969) and Huffman (1970). Subsequently, Liou and Lahore
(1974), Liou et al. (1976), Dugin and Mirumyants (1976), Sassen and Liou (1979a, 1979b),
Volkovitsky et al. (1980), Barkey and Liou (2001, 2006, 2008), and Barkey et al. (1999,
2002) have reported laboratory scattering and ice microphysics experiments and presented
pertinent scattering results for ice particles. In addition to laboratory experiments, a number
of scattering measurements have been conducted in cirrus clouds (C.-Labonnote et al. 2000,
Oshchepkov et al. 2000, Auriol et al. 2001, Jourdan et al. 2003, Shcherbakov et al. 2006).
In this subsection, we shall confine our discussion to laboratory scattering and ice cloud
experiments in support of the unified theory of light scattering by ice crystals as presented
above.
Desirable ice crystal sizes and shapes are difficult to generate and sustain for a time
period sufficiently long to perform light-scattering experiments. For this reason, Barkey
et al. (1999) carried out an experimental light-scattering program using hexagonal ice-like
crystals measured in the analog manner so that optical experiments could be performed over
a relatively long period of time for complex-shaped particles. The experiment consisted of a
polarized laser beam at λ = 0.63 µm and an array of 36 highly sensitive photodiode detectors
arranged between the scattering angles of 2.8° and 177.2°, mounted in a linear array on
a half-dome which could be rotated to vary the azimuthal angle. After careful calibration
and signal acquisition, this system was first used to measure the phase functions of a glass
sphere and a glass fiber configured to scatter light like an infinite cylinder. The experimental
results closely match those computed from the Lorenz–Mie and Rayleigh–Wait theories
(see Subsection 2.2.3). Subsequently, sodium fluoride (NaF) crystal, which has a refractive
index of 1.33, a value close to that of ice in the visible wavelength, was adopted for the
light-scattering experiments. The NaF crystal could be machined to a desirable shape with
a size on the order of a millimeter. It was mounted on top of a small pedestal, and its
orientation was controlled by a rotator. Angular integrations in the experiment could follow
the theoretical and computational procedures that were carried out for randomly oriented ice
particles. The phase functions determined from laboratory measurements and theoretical
results derived from IGOM for an aggregate and other crystal shapes assembled from NaF
columns showed close agreement. Also illustrated was excellent agreement between the
measured phase functions and theoretical results for a rough-surface plate having all eight
sides sanded with small scratches evenly distributed across the crystal surfaces. In the latter
case, the 22° and 46° features vanished and the phase function was essentially without
noticeable features.
The light scattering experiment, shown in Figures 3.28a and 3.28b, was conducted in
the cloud chamber at the Desert Research Institute at the University of Nevada, Reno.
It consisted of an insulated and refrigerated column approximately 4 m tall and 0.3 m
in diameter positioned above a larger cold chamber. Two dedicated chillers were used
to control the growth column temperature above and below the nucleation temperature,
while several thermocouples were placed along the length of the growth column. Water
droplets from an ultrasonic humidifier fell and cooled to the desired temperature in the
upper column; some of these were nucleated by a wire spring, and continuously cooled by
liquid nitrogen and presented to the ice cloud by a system of pulleys and a slowly rotating
(b)
From ultrasonic
water droplet
generator Water droplets
(a) Data-acquisition
Photodiode card and PC Continuous
detector/amplifiers computer Refrigerated nucleator
and insulated
Fiber light growth
guides chamber Dewar with
liquid nitrogen
Mirror Nucleated ice crystals

Refrigerated box
(not to scale)
Nephelometer
Forward
Diode Cloud scope
beam dump
laser Sample
(670 nm) guide tubes
Figure 3.28 (a) The physical structure of the polar nephelometer and its interaction with the laser
beam and the falling ice crystals confined to the center of the fiber-optic coupled detector array along
with the sample tube with a 3 mm inside diameter (after Barkey and Liou 2001). (b) The structure
of a cloud chamber for the generation of ice crystals. Water droplets are injected into the top of the
growth column and nucleated near the center. The cloud scope and polar nephelometer are placed in
the cold chamber at the bottom of the growth column (after Barkey et al. 2002).
electric motor. The water-vapor pressure could be controlled to certain degree by the rate at
which water droplets were injected into the crystal chamber, but ice crystal habit was mainly
controlled by the temperature at which the crystals were allowed to grow after nucleation.
The nucleated ice crystals continued to grow at the expense of water droplets as they fell
through the growth column and into the lower cold chamber where the polar nephelometer
was placed. Because the system was sealed, colder hence denser air fell to the bottom of
the lower cold box, which produced a temperature stratification that retarded the ice crystal
free-fall. Also, because of the slow fall speeds of the generated ice crystals, a suction
pump was connected to the sample outlet of the polar nephelometer to pull ice crystals
into the scattering volume. Simultaneously, the ice crystal evolution was monitored by a
cloud scope (a video microscope that observed the ice particles’ size and shape) positioned
approximately 20 cm from the nephelometer sample inlet.
The polar nephelometer system’s accuracy was first verified by measuring the scatter-
ing properties of water droplets generated by an ultrasonic humidifier at room temperature.
Water droplets falling through the upper sample tube were seen as a collimated and continu-
ous white stream of particles. Because there were no direct measurements of the droplet-size
distribution, several mean effective sizes and variances are generated by Lorenz–Mie calcu-
lations and compared with experimental results, which were adjusted to fit the normalized
102
101 Theory
Phase Function Experiment
100
10-1
10-2
0 60 120 180
Figure 3.29 Experimentally measured two-dimensional scattering intensity (normalized phase func-
tion) and the theoretically derived phase function (from Lorenz–Mie theory) for water droplets with
unpolarized incident light. The measurement error bar is shown at the scattering angle of 90° (after
Barkey and Liou 2001).
theoretical results by use of a multiplicative constant. A log-normal size distribution with a

mean effective radius of 3.75 µm and a variance of 0.1 produced the best match. The error
bar of 10% depicted in Figure 3.29 represents the system error. Overall, measured values
matched closely with theoretical results and reproduced the rainbow feature at a scattering
angle of 140°.
In the light scattering experiment involving ice crystals, the volume was defined by a
small cylinder 3 mm in diameter and 4 mm in length. Measured ice crystal concentration
was normally of the order of one particle per 2 mm3 , which corresponds to approximately
14 particles (20 µm sizes) in the scattering volume. The effect of multiple scattering on
nephelometer measurements can thus be neglected. The intensity of the light scattered into
the two-dimensional plane where the fiber light guides were positioned depends on the
number of crystals in the scattering volume and their size, shape, and orientation. These
experimental parameters vary considerably during experiments, which causes the intensity
of the measured results to vary widely from one measurement to the next. For this reason,
the scattering patterns over several hundred measurements were averaged, a process which
took approximately 2–4 minutes to complete. Ice crystals pulled into the scattering volume
were likely to be randomly oriented.
At least 190 measurements are needed to reduce the standard error of the mean to less
than 5%. In this manner, the averaged scattering pattern corresponds to an ice crystal with the
scattering properties of all the particles that existed during the time period of the experiment.
To obtain the phase function, we need the scattering cross-section in normalization processes
[see Eq. (2.4.34)]. For randomly√oriented simple hexagonal particles, the scattering cross-
section σ s is approximated by 3( 3a 2 + 2aL)/2 [Eq. (2.4.49)], where the ice crystal radius
a and length L were measured directly from the images, and absorption of ice at visible
wavelengths has been neglected. The phase functions presented in the following paragraphs
were determined from the sum P11 + P12 (see Subsection 2.4.3), which corresponds to
the two polarization states of the incident light for randomly oriented ice crystals in the
scattering volume.
Shown at the center of Figure 3.30 is a cloud-scope video image when the crystal
habit of the chamber cloud consisted predominantly of plates grown at a temperature of
approximately −6°C. The vertical white stripe in this image is due to glare from the cloud-
scope light source. The size distribution of the counted crystals is shown in the histogram.
The average aspect ratio L2a 0.8, where a is the radius of the hexagonal face and L is
the length of the column or the thickness of a plate. Fitted to the theoretical expectation
shown in the graph at the top, and based on the observed ice crystal microphysics, were
experimentally measured scattering properties from an average of 189 measurements taken
over a period of approximately 3 minutes. The error bars correspond to the 5% standard
error of the mean.
Figure 3.31 is similar to Figure 3.30, except that it involved a cloud event when mostly
hollow columns, grown between −5 and −7°C, were seen by the cloud scope. The 26
particles counted from this image consist of mostly hollow columns with an average aspect
ratio of approximately 2.1 and an average maximum dimension of approximately 36 µm.
However, regular columns, trigonal shapes, and plates can also be seen. The experimental
result was the average of 511 measurements taken over a period of approximately 4.5 min-
utes. The standard error of the mean for this result produces an approximate 3% error.
Normalized phase functions for simple hexagons and hollow columns calculated by IGOM
with various aspect ratios corresponding to the measured crystal sizes are combined by use
of the extinction cross-section determined from the measured crystal size and the counted
population to weight the scattering contributions of each particle. Approximately 80% of
the particles were assumed to have rough surfaces to account for the surface features that
can be seen on some of the larger plates in the cloud-scope image. Substantial surface
roughness was incorporated in the phase functions by random perturbations of the crystal
surface.
In Figure 3.31, approximately 10% of the phase functions used to produce theoretical
values were based on hollow column habit. The other crystal types were simple columns
and plates, of which 80% had rough surfaces. Overall, experimental results closely follow
theoretical values. Both the 22° and 46° halo intensity imprints are clearly displayed,
although less intense at 46° due to the presence of hollow columns; this is expected as the
46° halo feature is produced by two refractions through a 90° prism. For both results, a
large percentage of the theoretical results required crystals to have rough surfaces in order
for a reasonable fit to experimental results. This is most likely due to surface features on
the crystals that cannot be seen on the cloud-scope images because of the low resolution
of the video camera, or smaller particles not seen by the cloud scope. Also, we see that
102
Normalized Phase Function

Theory
101
Experiment
100
10-1
10-2
0 60 120 180
(deg.)
Figure 3.30 Two-dimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly plates are seen, as shown in the video cloud-scope image (middle
panel), along with an ice crystal count summarized in the histogram (bottom panel). The white area
in the middle image is due to the cloud-scope light source (after Barkey et al. 2002).
102
Normalized Phase Function

Theory
101
Experiment
100
10-1
10-2
0 60 120 180
Figure 3.31 Two-dimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly columns are observed, as shown in the video cloud-scope image
(middle panel), along with ice crystal count summarized in the histogram (bottom panel). The white
area seen in the middle of the image is due to the cloud-scope light source (after Barkey et al. 2002).
more measured light was scattered into angles in the ranges 10°–20° and 160°–175° than
in the theoretical results, which revealed that it is extremely difficult to completely match
the two results in view of numerous unknown factors involving ice crystal sizes and shapes
generated in cloud chambers and inherent limitations in theoretical calculations. Between
120° and 160°, the experimental results are not smooth in both cases due to systematic errors
in the calibration that are more noticeable at lower intensities. There is a slight difference
between the angular positions of both the 22° and the 46° intensity peaks because each fiber
optic sensed light scattered into an angle of approximately 2° and the angular positions of
the fiber optics were not at the points of maximum scattered intensity.
In summary, and in view of the preceding presentations, it is evident that the unified
theory of light scattering by ice crystals developed herein has been successfully used
to interpret phase functions produced from laboratory scattering and ice microphysics
experiments, thus providing new dimensions in the evolution of theoretical developments.
Furthermore, the well-known features of diffraction, 22° and 46° intensity maxima, and
large backscattering values that occur in the phase function measured in laboratory ice
cloud chambers match well to theoretical calculations.
3.4.3 Application to snow grains contaminated by black carbon (BC) and dust
In this subsection we apply the unified theory of light scattering by ice crystals, as presented
above, to absorption and scattering by snow grains with sizes on the order of 100–1000 µm
contaminated by BC (0.1–1 µm) and dust (1–10 µm). Liou et al. (2011) demonstrated
that a small BC particle on the order of 1 µm internally mixed with snow grains could
effectively reduce visible snow albedo by as much as 5–10%. Subsequently, Liou et al.
(2013) applied this approach to contaminated ice particles in contrail cirrus and investigated
the effects of internal and external mixing of BC in ice particles on the radiative forcings
and heating rates of contrail cirrus. Also presented was a parameterization of the spectral
extinction coefficient, single-scattering co-albedo, and asymmetry factor for contaminated
ice particles, for efficient application to radiative transfer calculations on the basis of the
comprehensive spectral database of the single-scattering properties of pure ice particles of
various shapes and sizes developed by Yang et al. (2013). In conjunction with this study, we
noted observations by Petzold et al. (1998, 1999) relative to plural numbers of BC particles
inside and outside of contrail ice crystals.
Multiple internal mixing of BC/dust in snow grains can be produced by nucleation,
accretion, riming, aggregation, and sintering during aerosol–cloud–precipitation processes
known as wet deposition. Internal mixing can also occur via the direct or dry deposition of
aerosols onto high-elevation snow, followed by successive snow events. Mountain glaciers
and polar ice sheets are excellent archives and imprints of atmospheric BC/dust particles
associated with wet and dry depositions. Ming et al. (2008) measured BC concentrations
in a 40 m ice core over the East Rongbuk Glacier in the northeast saddle of Mt. Everest,
which provided an historical record of BC deposition in the high Himalayas over the past
50 years. Additionally, Ming et al. (2009) collected BC concentrations from snow samples
taken at a number of selected glaciers in west China during 2004–2006 and found that BC
concentrations are higher at lower-elevation sites (Tienshan Mountains) than comparable
samples obtained on the Tibetan Plateau, likely because of the topography effect. Xu
et al. (2009) measured aerosol concentrations including BC in ice cores spanning five
Tibetan glaciers, showing an increasing deposition of anthropogenic aerosols during the
period 1998–2005. Hadley et al. (2010) reported one of the first direct measurements
documenting the efficient deposition of BC on Sierra Nevada snow packs while revealing
high BC concentrations in falling snow. Sterle et al. (2013) showed that BC concentrations
were enhanced seven-fold in surface snow compared to snowpack bulk values in the eastern
Sierra Nevada from February to May 2009. Ice cores extracted from high elevations in the
European Alps also displayed increasing BC concentrations from the era of the Industrial
Revolution to the twentieth century (Jenk et al. 2006, Thevenon et al. 2009). Using a single
particle soot photometer, Schwarz et al. (2013) showed that the modal volume-equivalent
radius for BC particles in snow is about 0.1 µm.
With reference to dust, Kumai (1977) investigated the long-term variation of atmospheric
aerosols in snow and deep ice core samples taken from Camp Century, Greenland, and
showed that a majority of these aerosols were silicate mineral dust with a mean size on
the order of 1 µm. Hammer et al. (1985) found dust in snow grains with a size on the
order of 0.1–2 µm in Greenland ice cores. Zdanowicz et al. (1998) measured and compared
the concentration, deposition rate, and size distribution of dust aerosols in snow pits on
the Penny Ice Cap, Baffin Island, Arctic Canada with earlier studies. Also, Zdanowicz
et al. (2006) measured the dustfall over snow packs in the St. Elias Mountains, Canada,
finding high dust concentrations in snow in April 2001. Kang et al. (2010) used ice cores to
reconstruct dust load history over the central Tibetan Plateau from 1940 to 2005 and found
that dust loading in this region has increased since the 1990s.
From the instant that snow hits the ground, it begins an endless process of metamorphism,
a term used in reference to physical changes in snow grains within a snowpack resulting from
differences in temperature and pressure. Using scanning electron microscopy in February
2002, Dominé et al. (2003) studied the natural metamorphism of snow samples collected
near Chamrousse, a ski resort town in the French Alps. The fresh snow was collected within
minutes of its fall and immersed immediately in liquid nitrogen to preserve its shape and
size. The sample pictures illustrate the wide variety of shapes and habits, including plates,
columns, column rosettes, stellar and dendritic crystals, and various irregular crystals, all of
which exhibit sharp angles. After 14 days of isothermal metamorphism, extensive changes
occurred such that the original shapes were substantially modified with no sharp edges
between two flat faces, leading to rounded crystals with different shapes. These snow
grains are referred to as old snow.
In view of these and other observations (e.g., Aoki et al. 2000), the shapes of hexag-
onal plate/column/bullet rosette and spheroid/sphere were used to model fresh and old
snow grains, respectively, for light scattering and absorption calculations (Figure 3.32).
Liou et al. (2014) have developed a stochastic approach to model the positions of BC/dust
(a) 100 μm 100 μm (b)
100 μm
100 μm
100 μm
100 μm 100 μm 100 μm
Soot Dust
(c) (d)
100 μm
100 μm
100 μm
30 μm
30 μm 30 μm 30 μm
Figure 3.32 The images on the left side are observed grain shapes for (a) fresh snow and (c) old snow
(Dominé et al. 2003). On the right side are corresponding parameterized grain shapes for (b) fresh
snow (plate, column, and bullet rosette) and (d) old snow (prolate and oblate spheroids and sphere)
with multiple internal and external mixing, using a stochastic procedure. The red dots represent either
soot or dust.
1 2
Two-Layer Model (rsnow = 100 μ m, rsnow = 1000 μ m), μ 0 = 0.5
(a)
BC (rBC = 0.1 μm)
1.0 Pure Snow

0.01 ppm 1.0
0.1 ppm
0.8
1 ppm
0.9
0.6
VIS NIR
0.4
0.8
0.2 0.4 0.6 0.8 1.0
0.2
Snow Albedo
0.0
(b)
Dust (rdust = 1 μm)
1.0 Pure Snow

2 ppm 1.0
10 ppm
0.8
100 ppm
0.9
0.6
VIS NIR
0.4
0.8
0.2 0.4 0.6 0.8 1.0
0.2
0.0
0.2 0.8 1.4 2.0 2.6 3.2 3.8 4.4 5.0
Wavelength (μm)
Figure 3.33 Snow albedo for a two-snow layer model (new snow on top of old snow) as a function
of wavelength for three BC/dust concentrations. New snow internally mixed with (a) BC or (b) dust
has an optical depth of 200, while pure old snow is considered to be semi-infinite. The insets display
snow albedo in the wavelength region 0.2–1 µm where substantial BC/dust absorption occurs (after
Liou et al. 2014).
internally mixed with three snow-grain types: hexagonal plate/column and spheroid (con-
vex) and bullet rosette (concave). Subsequently, light absorption and scattering analysis
can be carried out using the unified theory of light scattering by ice crystals, the theme of
Subsection 3.4.1, which includes the geometric-optics and surface-wave (GOS) approach
coupled with Monte Carlo photon tracing, to determine BC/dust single-scattering properties.
Figure 3.32 shows examples of external and internal positions of BC (or dust) particles for
these particles; the action of internal mixing absorbs substantially more light than external
mixing.
Figure 3.33 depicts spectral albedo (0.2–5 µm) by snow grains internally mixed with
BC/dust for a two-layer spectral snow model involving contaminated fresh snow on top
of old snow to investigate and understand the climatic impact of multiple BC/dust inter-
nal mixing associated with snow grain metamorphism, particularly over mountain/snow
topography. Concentrations of BC (Figure 3.33a) and dust (Figure 3.33b) are 0.01–1 ppm
(parts per million by mass) and 2–100 ppm, respectively. The impact of BC/dust absorption
is confined to wavelengths shorter than about 1.4 µm, beyond which ice absorption pre-
dominates. Based on the single-scattering properties determined from stochastic and light
absorption parameterizations and using the adding/doubling method for spectral radiative
transfer, the effect of snow-grain shape on absorption is relatively small, but its effect on for-
ward scattering is substantial. Due to a greater probability of intercepting photons, multiple
inclusions of BC/dust exhibit a larger absorption than an equal-volume single inclusion, and
the snow-grain shape plays a critical role in snow albedo calculations through its forward
scattering strength.
4
Other useful approaches to light scattering
by ice particles
During the past four decades, numerous computational methods and numerical approaches
have been developed to solve light scattering by non-spherical and inhomogeneous particles
for application to atmospheric and geophysical sciences, applied physics, computer science,
electrical engineering, and bioengineering. It is especially evident that the applicability and
accuracy of these approaches in terms of particle size range and morphology, established
by comparison with the “exact” Lorenz–Mie theory, has been enhanced as a result of
the availability of supercomputers as well as clever technical innovation. Most of these
developments have been reviewed from various perspectives by Liou and Takano (1994),
Wriedt (1998), Kokhanovsky (1999), Mishchenko et al. (2000b), Liou (2002), Kahnert
(2003), and Yang and Liou (2006). In particular, the finite-difference time domain (FDTD)
method, the T-matrix method, and the discrete dipole approximation (DDA) have been
extensively used for a number of practical applications. The following discussions will
be confined to these three methodologies within the context of their applications to light
scattering by ice crystals and in association with the geometric-optics, diffraction, and
surface-wave approaches presented in Chapter 3.
4.1 Finite-difference time domain method

4.1.1 A brief review
It appears that Yee (1966) pioneered the development of the FDTD method for application
to electromagnetic wave propagation and scattering problems. He demonstrated in a concise
paper the basic FDTD concept for simulating the propagation of an electromagnetic wave in
the near-field zone. However, several important aspects related to the application of FDTD
to light scattering problems, including the boundary condition imposed on spatial-domain
truncation and mapping of the near field to the far field, were not properly addressed.
From an historical standpoint, the FDTD approach had received insufficient recognition
until substantial efforts were made by electrical engineers and computational physicists in
the late 1970s and early 1980s (e.g., Taflove and Brodwin 1975, Holland 1977, Kunz and
Lee 1978, Taflove 1980, Kunz and Simpson 1981, Umashankar and Taflove 1982). After
their pioneering contributions, FDTD has been applied to a wide range of electromagnetic
197
198 Other useful approaches to light scattering
Artificial Absorbing Layer
Connecting Surface
Scattering Particle
Total-field Region
Scattered-field Region
Figure 4.1 A conceptual diagram for the computation, by means of the finite-difference time domain
(FDTD) method, of the near field for electromagnetic scattering by a non-spherical particle illuminated
by plane waves. The connecting surface (or the Huygens surface) divides the computational domain
into total-field and scattered-field regions.
problems, including antenna scattering, numerical modeling of microstrip structures, elec-

tromagnetic absorption by human tissues such as red blood cells, and scattering of light
by atmospheric particles (e.g., Taflove and Umashankar 1990, Taflove 1995, Yang and
Liou 1996a and 1996b, Sun et al. 1999, Yang et al. 2000b, Lu et al. 2005). In particular,
with the development of the perfectly matched layer (PML) absorbing boundary condition
(Berenger 1994) and further improvements, such as the uniaxial anisotropic PML (UPML)
(Gedney 1996) and convolution PML (CPML) boundary conditions (Roden and Gedney
2000), the FDTD approach has become a powerful and robust numerical tool to facilitate
simulation of the interaction of an arbitrarily oriented structure or particle with incident
electromagnetic waves.
The most computationally efficient implementation of the FDTD technique can be real-
ized through the use of a Cartesian grid mesh, as schematically illustrated in Figure 4.1.
Specifically, the spatial region comprised of a scattering particle is truncated and subse-
quently discretized in terms of a number of rectangular cells. Appropriate values of permit-
tivity, permeability, and conductivity are assigned to the grid cells to specify electromag-
netic characteristics of the particle and the surrounding medium. A direct implementation
of Maxwell’s curl equations is invoked to solve the temporal variation of electromag-

netic waves in the truncated spatial domain within which the scatterer is imbedded. More
specifically, the finite-difference discretization is performed for Maxwell’s curl equations
in both time and space. The propagation of excited waves (i.e., scattered waves) in the time
domain can be simulated with the finite-difference analog of Maxwell’s equations through
a time-marching iterative procedure.
In numerical computation, the scattering of an incident electromagnetic wave by a
particle must be confined to a finite spatial domain. In order to apply the FDTD tech-
nique, an appropriate boundary condition must be imposed such that the simulated field
within the truncated spatial domain would be the same as that for the unbounded case.
As a matter of fact, the implementation of an optimal boundary condition to suppress
spurious reflections is critical to practical computations based on the FDTD method in
terms of their numerical accuracy and stability. Substantial technical efforts have been
expended to minimize artificial reflections from boundaries of the truncated FDTD compu-
tational domain (e.g., Moore et al. 1988, Blaschak and Kriegsmann 1988, Yang and Liou
1998b). However, it appears that the use of an artificial absorbing medium, known as the
PML medium (Berenger 1994), may be the most efficient approach to implementing an
absorbing boundary condition to effectively truncate the spatial domain. The absorbing
boundary condition must be imposed only for outgoing waves. For this reason, a connect-
ing surface (Taflove 1995), or the so-called Huygens surface (Merewether et al. 1980),
is defined in the computational domain, as shown in Figure 4.1. Inside the connecting
surface, the total field is simulated, while the scattered (outgoing) waves are simulated
outside the connecting surface, along with initiation of the incident wave at the connecting
surface. Also, a scattered-field FDTD scheme (i.e., the incident field is subtracted from
the total field), which uses the scattered field within the entire computational domain,
has been reported by Britt (1989) and Yang and Liou (1995). This technique, however,
appears to be less computationally efficient compared with the counterpart depicted in
Figure 4.1.
The solution of the finite-difference analog of Maxwell’s curl equations is in the time
domain. To obtain the frequency response of the scattering particle to illumination by the
incident wave, an appropriate transformation of the relevant electromagnetic field values
from the time domain to the frequency domain is required. The discrete Fourier transform
technique can be employed to obtain the frequency spectrum of time-dependent signals
if a pulse is used as an initial excitation. In practice, selection of an appropriate pulse is
required to avoid numerical aliasing and dispersion. Furthermore, a mapping of the near-
field result to the far field must be performed in order to derive the particle scattering and
polarization properties, namely, the scattering phase matrix, extinction cross-section, and
single-scattering albedo. A surface integration or a volume integration technique can be
employed to obtain the far-field solution. Several computational issues exist in performing
FDTD numerical calculations, including the staircasing effect in approximating the particle
shape as well as the determination of appropriate white space between the particle and the
artificial boundary to balance computational demand and numerical accuracy.
4.1.2 Numerical discretization of Maxwell’s curl equations

To present the basic principle of the FDTD technique, we shall consider the source-
dependent form of Maxwell’s curl equations in the time domain for a non-ferromagnetic
medium with μ = 1, a valid simplification for ice crystals, in the forms [see Eqs. (2.2.1a),
(2.2.1.b), and (2.2.4)]
ε ∂E(r, t) 4π
∇ × H(r, t) = + σ E(r, t), (4.1.1a)
c ∂t c
1 ∂H(r, t)
∇ × E(r, t) = − , (4.1.1b)
c ∂t
where ε and σ are, respectively, the permittivity and conductivity and both are real quanti-
ties. To construct a computationally efficient finite-difference scheme corresponding to the
preceding two equations, Yang and Liou (1996b) introduced a new variable, τ = 4πσ/ε,
based on which Eq. (4.1.1a) can be expressed in the form

ε ∂E(r, t)
∇ × H(r, t) = + τ E(r, t) . (4.1.2)
c ∂t
We can use Eqs. (4.1.1b) and (4.1.2) to construct the finite-difference analog of
Maxwell’s curl equations to circumvent the use of complex quantities even if the scat-
tering particle is absorptive, i.e., the corresponding complex refractive index has a non-zero
imaginary part. To discretize Eq. (4.1.2) in time, we multiply both sides by a factor exp(τ t)
and rearrange the resultant terms in the form
∂E(r, t) c
exp(τ t) + τ exp(τ t)E(r, t) = exp(τ t)∇ × H(r, t). (4.1.3)
∂t ε
Upon using the chain rule for derivatives, we can rewrite this in the form
∂[exp(τ t)E(r, t)] c
= exp(τ t) ∇ × H(r, t). (4.1.4)
∂t ε
Furthermore, consider the second-order difference scheme, referred to as the leap-frog
scheme and usually used in finite-difference analysis in computational mathematics, for
the derivative of a function f with its argument t, in the form

df (t) f [(n + 1)t] − f (nt)
≈ + O(t 2 ) = (f n+1 − f n )/t + O(t 2 ).
d t t=(n+1/2)t t
(4.1.5)
In the preceding equation, the superscripts n + 1 and n indicate that the associated quantities
are evaluated at the time steps t = (n + 1)t and t = nt. The term O(t 2 ) indicates that
the truncation error of this leap-frog scheme is of the second order. Applying this scheme
to Eq. (4.1.4) at time step t = (n + 1/2)t, we obtain

∂[exp(τ t)E(r, t)] c
= exp(τ t) ∇ × H(r, t) . (4.1.6a)
∂t t=(n+1/2)t ε t=(n+1/2)t
We may rewrite this equation in a more explicit form as follows:

exp[τ (n + 1)t]En+1 (r) − exp(τ nt)En (r) c
= exp[τ (n + 1/2)t] ∇ × Hn+1/2 (r).
t ε
(4.1.6b)
Simplifying and rearranging Eq. (4.1.6b), we obtain an explicit scheme for the finite-
difference analog of Eq. (4.1.4) for simulating the temporal variation of the electric field,
in the form
ct
En+1 (r) = exp(−τ t)En (r) + exp(−τ t/2) ∇ × Hn+1/2 (r). (4.1.7)
ε
Similarly, the finite-difference analog of Maxwell’s curl equation for the magnetic field can
be expressed as
Hn+1/2 (r) = Hn−1/2 (r) − ct∇ × En (r). (4.1.8)
Note that Eqs. (4.1.7) and (4.1.8) can also be derived by directly integrating the dif-
ferential forms of Eqs. (4.1.4) and (4.1.1b). For example, if we integrate both sides of
Eq. (4.1.1b) from t = (n − 1/2)t to t = (n + 1/2)t, we obtain
(n+1/2)t
1 (n+1/2)t ∂H(r, t)
∇ × E(r, t)dt = − dt, (4.1.9)
(n−1/2)t c (n−1/2)t ∂t
which can be approximated in the form
1
∇ × En (r)t ≈ − [Hn+1/2 (r) − Hn−1/2 (r)]. (4.1.10)
c
Note that rearranging the terms in Eq. (4.1.10) leads to Eq. (4.1.8).
For numerical FDTD simulations, the vector forms of Eqs. (4.1.7) and (4.1.8) must be
decomposed into their scalar components. For example, the governing equations for the
z-components of the electric and magnetic fields can be obtained from these two equations
as follows:
n+1/2 n+1/2

ct ∂Hy (r) ∂Hx (r)
Ez (r) = exp(−τ t)Ez (r) + exp(−τ t/2)
n+1 n
− ,
ε ∂x ∂y
(4.1.11)

∂Eyn (r) ∂Exn (r)
Hzn+1/2 (r) = Hzn−1/2 (r) − ct − . (4.1.12)
∂x ∂y
To discretize the spatial derivatives in the preceding equations, Yee (1966) suggested a
configuration for the electromagnetic field components on a rectangular cell, as schemat-
ically illustrated in Figure 4.2. The cell center is located at (I x, J y, Kz) in which
x, y, and z are cell dimensions along the x-, y-, and z-axes, respectively. It turns
out that the staggered configuration, shown in Figure 4.2, for electromagnetic field vec-
tor components inherently guarantees that electromagnetic boundary conditions at cell
(I -
/ 2) 1/
2,
K+1 J+
2, Ey 1/2
1/ ,K
J- Hz +1
2, /2
1/ )
(I- Ex Ex
Ez z
Ez Ey
Ez Hy
(I, J, K)
Ez y
Hx Ey
x
(I + Ex
1/ )
2, Ey Ex 1 /2
J- K-
1/ 2,
2,
K- 1/
1 J+
/2 2,
) 1/
(I +
Figure 4.2 Locations of various field components at staggered locations on a cubic cell, following a
finite-difference scheme suggested by Yee (1966). The E field components are tangential to the cell
edges, while the H field components are normal to the cell faces.
interfaces are such that the tangential components of E-field and the normal components
of H-field are continuous at interfaces.
In the FDTD formalism, the spatial location (I x, J y, Kz) is concisely represented
in terms of indices (I, J, K), and any variable, such as the z-component of the electric field
vector as a function of space and time, is denoted in the form
Ezn (I + 1/2, J + 1/2, K) = Ez [(I + 1/2)x, (J + 1/2)y, Kz, nt]. (4.1.13)
With the preceding notation and the field component configuration depicted in Figure 4.2,
application of the second-order central difference scheme to the spatial derivatives defined
in Eqs. (4.1.11) and (4.1.12) leads to
Ezn+1 (I + 1/2, J + 1/2, K)

= exp[−τ (I + 1/2, J + 1/2, K)t]Ezn (I + 1/2, J + 1/2, K)
ct
+ exp[−τ (I + 1/2, J + 1/2, K)t/2]
ε(I + 1/2, J + 1/2, K)

1 ) n+1/2 *
× Hy (I + 1, J + 1/2, K) − Hyn+1/2 (I, J + 1/2, K)
x

1 ) n+1/2 *
− H (I + 1/2, J + 1, K) − Hx n+1/2
(I + 1/2, J, K) , (4.1.14)
y x
Hzn+1/2 (I, J, K + 1/2)

1 ) n
= Hzn−1/2 (I, J, K + 1/2) − ct Ey (I + 1/2, J, K + 1/2)
x
* 1 ) n
− Eyn (I − 1/2, J, K + 1/2) − E (I, J + 1/2, K + 1/2)
y x

*
− Exn (I, J − 1/2, K + 1/2) . (4.1.15)
Equations (4.1.14) and (4.1.15) represent an explicit finite-difference scheme; that is, the
temporal variations of the electromagnetic field values at the grid points can be simulated via
a straightforward time-marching iterative procedure without invoking the matrix inversion
required by an implicit finite-difference scheme.
For a given complex refractive index m = mr − imi , where mr and mi are, respectively,
the real and imaginary parts of the refractive index [both are positive real numbers; see
Eq. (2.3.14)], we must evaluate the corresponding permittivity and conductivity. To do so,
we shall consider a time-harmonic electromagnetic wave in the form
E(r, t) = eikct E(r), (4.1.16a)
H(r, t) = eikct H(r). (4.1.16b)
In the preceding equations, k is the circular wavenumber associated with the incident
wave, given by k = 2π/λ, and λ is the incident wavelength. The assumption of the time-
dependence factor exp(ikct) leads to a negative value for the imaginary part of the refractive
index. Substituting Eqs. (4.1.16a) and (4.1.16b) into Eq. (4.1.2) and simplifying the resultant
expression, we obtain
τε
∇ × H(r) = ik ε − i E(r) = ik ε̃E(r), (4.1.17)
kc
where the effective complex permittivity ε̃ is given by
ε̃ = ε − iτ ε/(kc). (4.1.18)
It is clear that, for a given pair of permittivity ε and conductivity σ , the effective
complex permittivity ε̃ is dependent on the frequency, ω (= kc), of the electromagnetic
waves. Furthermore, for a non-ferromagnetic medium, we have the following relationship:

ε̃ = (mr − imi )2 = m2r − m2i − i2mr mi . (4.1.19)
Comparing Eqs. (4.1.18) and (4.1.19), we obtain
ε = m2r − m2i . (4.1.20a)
2mr mi kc
τ= . (4.1.20b)
m2r − m2i
The dielectric properties of the scattering particle and surrounding medium are given in
discrete form through τ (I + 1/2, J + 1/2, K) and ε(I + 1/2, J + 1/2, K) in Eq. (4.1.14).
Because of the discretization of the parameters τ and ε, a pronounced shortcoming of the

FDTD technique is the staircasing approximation of the scattering particle in a Cartesian
grid mesh. To alleviate the staircasing effect, various techniques have been suggested to
evaluate permittivity values involved in FDTD equations (e.g., Yang and Liou 2000, Yang
et al. 2001), including the use of effective permittivity values for grid cells based on the
Maxwell-Garnett (1904) theory.
In terms of truncation errors, Eqs. (4.1.14) and (4.1.15) represent a second-order
scheme in both time and space. Furthermore, to avoid numerical instability, the choices
of x, y, z, and t values cannot violate the Courant–Friedrichs–Lewy (CFL) condi-
tion (Courant et al. 1928). Specifically, it was shown by Taflove and Brodwin (1975) that
the following condition is mandatory in order to achieve numerical stability of the FDTD
numerical scheme:
1
ct ≤ . (4.1.21)
1/x 2 + 1/y 2 + 1/z2
It should be pointed out that the finite-difference expressions in Eqs. (4.1.14) and (4.1.15)
along with their counterparts for the x and y components of electromagnetic fields constitute
an explicit numerical form; that is, the propagation of an electromagnetic wave can be
simulated through a time-marching iterative procedure without invoking the solution of a
set of linear equations involved in an implicit numerical scheme. For practical simulations,
the spatial increment needs to be smaller than approximately 115 120 of the incident
wavelength such that spatial variations of the field components are negligible over the
lengths of cell dimensions.
4.1.3 Absorbing boundary conditions and electromagnetic fields

in the frequency domain
An appropriate boundary condition for the truncation of the spatial domain is essential to the
successful implementation of the FDTD technique. Among various techniques developed
for the imposition of boundary conditions in conjunction with application of the FDTD
technique to electromagnetic scattering problems, the innovative perfectly matched layer
(PML) boundary condition developed by Berenger (1994) has been an effective and robust
approach to the suppressing of artificial reflections from boundaries of the truncated FDTD
computational domain. Technically, the absorption of outgoing waves by the PML method
is based on absorption by an artificial absorbing medium located at boundary layers in
the computational domain, backed by a perfectly conducting surface that is numerically
realized by setting the field values at outermost boundary-layer grid points to zero. The
conventional absorbing boundary condition is to specifically define the wave impedance
of the medium so that it matches that of free space. Such a simple matching approach
produces substantial non-zero reflections when a scattered wave impinges obliquely on an
absorbing medium. To overcome the disadvantage of the conventional method, the PML
absorbing medium can be selected such that the wave decay due to absorption is imposed
on field components parallel to their boundary layers. To achieve this goal, each Cartesian
component of the electromagnetic field is split into two parts as follows:
(Ex , Ey , Ez ) = [(Ex2 + Ex3 ), (Ey1 + Ey3 ), (Ez1 + Ez2 )], (4.1.22a)
(Hx , Hy , Hz ) = [(Hx2 + Hx3 ), (Hy1 + Hy3 ), (Hz1 + Hz2 )], (4.1.22b)
where the subscripts 1, 2, and 3 denote the components of the electric (or magnetic) field,
which are respectively associated with the spatial derivatives of magnetic (or electric) field
components along the x, y, and z directions. The splitting of field components leads to 12
scalar equations which govern the propagation of electromagnetic waves within boundary
layers. For example, the PML boundary condition equations for the Ex and Hx components
at a boundary perpendicular to the z-axis are given by
exp[−τz (z)t] ∂ ∂(Hy1 + Hy3 )
{exp[τz (z)t]Ex3 } = − , (4.1.23a)
c ∂t ∂z
1 ∂Ex2 ∂(Hz1 + Hz2 )
= , (4.1.23b)
c ∂t ∂y
exp[−τz (z)t] ∂ ∂(Ey1 + Ey3 )
{exp[τz (z)t]Hx3 } = , (4.1.23c)
c ∂t ∂z
1 ∂Hx2 ∂(Ez1 + Ez2 )
=− . (4.1.23d)
c ∂t ∂y
In Eqs. (4.1.23a) and (4.1.23b), τz (z) is defined at boundary layers near z = 0 and
z = zmax , where zmax represents the computational domain dimension along the z-axis, as
follows:

τmax |(z − LPML )/LPML |p , for 0 ≤ z ≤ LPML ,
τz (z) = (4.1.24)
τmax |[LPML − (zmax − z)]/LPML | , for zmax − LPML ≤ z ≤ zmax ,
p
where LPML denotes the PML medium thickness, p is usually selected between 2 and 3
(Lazzi and Gandhi 1997) and τ max denotes the maximum absorption at z = 0 and z =
zmax , which can be determined by specifying the reflectance of boundary layers at a normal
incidence. From Eq. (4.1.24), the absorbing strength of the PML medium can be seen
to gradually increase from the interior PML medium boundary (e.g., z = LPML ) to the
exterior PML medium boundary (e.g., z = 0) in the case of the boundary near z = 0.
Various numerical experiments and sensitivity studies reported in the literature (e.g., Katz
et al. 1994) have proven that PML boundary conditions are superior to other absorbing or
transmitting boundary conditions such as the one derived from the one-way wave equation
(Mur 1981).
Values of the near field in the time domain can be computed using a time-marching
iterative procedure based on the finite-difference analog of Maxwell’s curl equations
and boundary condition equations. To obtain the single-scattering properties that are the
inherent optical characteristics of the particle in the frequency domain, time-dependent
field values must be transformed to their corresponding counterparts in the frequency

domain. In practice, we may select an incident pulse as the initial excitation in the com-
putation and employ the discrete Fourier transform to obtain frequency-domain counter-
parts of the simulated field. Let Ez be the z-component of the electric field at location
[(I + 1/2)x, (J + 1/2)y, Kz] and its value at the time step n be Ezn . The time varia-
tion of Ez can then be written as
Ez [(I + 1/2)x, (J + 1/2)y, Kz, t]

N
= Ezn [(I + 1/2)x, (J + 1/2)y, Kz]δ(t − nt), (4.1.25)
n=0
where δ is the Dirac delta function and the maximum time step N is chosen such that
the field in the time domain is reduced to a small value. Accordingly the corresponding
spectrum in the frequency domain is given by
Ez [(I + 1/2)x, (J + 1/2)y, Kz, k]

∞ N

= Ez [(I + 1/2)x, (J + 1/2)y, Kz]δ(t − nt) exp(ikct)dt
n
−∞ n=0

N
= Ezn [(I + 1/2)x, (J + 1/2)y, Kz] exp(ikcnt). (4.1.26)
n=0
To avoid aliasing and numerical dispersion and to obtain a correct frequency spectrum,
the maximum wavenumber or the minimum wavelength, for the region in which the fre-
quency response of the scattering is evaluated, is bounded. In any finite-difference equation,
it is required that the wavelength of a simulated wave be larger than the grid size. In addi-
tion, field values in the frequency domain obtained by this procedure must be normalized
by the counterpart from Fourier transform of the incident wave at the center of the grid
mesh such that the frequency response of the scattering particle is scaled with reference to
an incident harmonic wave with unit amplitude.
To initialize the FDTD simulation using time-marching iterations, a Gaussian pulse can
be used as the incident wave, for example, the incident electric field polarized along the x
direction, given by
n
Ex,inc = exp[−(n/w − 5)2 ]. (4.1.27)
Using the trial and error method, Yang and Liou (1995) found w = 30 to be an optimal
value to avoid the numerical dispersion that can occur in the pulse propagation involved
in the simulation of light scattering by non-spherical ice crystals. Note that in Eq. (4.1.27)
the pulse peak is shifted by 5w such that the incident signal begins at a small value on the
order of ∼ 10−11 .
4.1.4 Near-to-far-field mapping and optical property calculations

The frequency-domain near-field values computed from the FDTD algorithm and the dis-
crete Fourier transform must be mapped to their far-field counterparts. Either a surface-
integral method (SIM) or a volume-integral method (VIM) can be employed to map elec-
tromagnetic signals from the near-field regime to the far-field regime. For large indices
of refraction, the SIM allows a much coarser grid mesh resolution than the VIM in terms
of similar numerical accuracy (Zhai et al. 2007). These two approaches are in principle
equivalent; however, with reference to the optical properties of ice crystals in solar and
thermal infrared spectral regions, we found that the VIM technique is most efficient within
the framework of FDTD, particularly in terms of the accuracy of simulating backscatter
and phase matrix elements associated with polarization. Following is a brief summary of
the basic physical principle and mathematical framework for both SIM and VIM. In the
far field (or radiation zone), kr → ∞, the scattered far field for the electric vector may be
written in terms of a volume integral or a surface integral equation in the forms
k 2 exp(ikr)
VIM: Es (r) =
4πr

) 2 *
× mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r ) {E(r ) − η[η · E(r )]}
V
× exp(−ikη · r )d 3 r , (4.1.28)

k 2 exp(ikr)
SIM: Es (r) = η × {ns (r ) × E(r ) − η × [ns (r ) × H(r )]}
−ik4πr s
× exp(−ikη · r )d 2 r , (4.1.29)
where η = r/r denotes the scattering direction. In Eq. (4.1.28) the integration is over the
particle interior volume, whereas in Eq. (4.1.29) the integration is performed over a surface
enclosing the scattering particle, and ns is a unit vector that is locally normal to the surface.
Note that the scattered field given by Eq. (4.1.28) or (4.1.29) is exact if the near field within
the particle interior or on the aforementioned surface is precisely known. To compute the
scattering phase matrix, the scattered field must be decomposed into components parallel
and perpendicular to the scattering plane in the form
Es (r) = X// Es,// (r) + X⊥ Es,⊥ (r), (4.1.30)
where X// and X⊥ are, respectively, the unit vectors parallel and perpendicular to the
scattering plane and satisfy the relationship X⊥ × X// = η. We can express the scat-
tered E-field in terms of a 2 × 2 amplitude scattering matrix containing four elements
Si (i = 1, 2, 3, 4) in the form

Es,// (r] exp(ikr) S2 S3 Eo,//
= , (4.1.31)
Es,⊥ (r) −ikr S4 S1 Eo,⊥
where Eo,// and Eo,⊥ are the incident E-field components defined with respect to the
scattering plane. In the FDTD method, the incident wave is defined with respect to the
incident coordinate system whose z-axis is aligned with the incident direction, and is
represented by Eo,x and Eo,y . To map the incident E-field vector with respect to the incident
coordinate system to that with respect to the scattering plane, the following transformation
is required:

Eo,// X⊥ · x −X⊥ · y Eo,y
= . (4.1.32)
Eo,⊥ X⊥ · y X⊥ · x Eo,x
In this equation, x and y are unit vectors along the x- and y-axes of the incident coordinate
system. We can then perform an inversion to obtain the equivalent form, as follows:

Eo,y X⊥ · x X⊥ · y Eo,//
= . (4.1.33)
Eo,x −X⊥ · y X⊥ · x Eo,⊥
To compute the scattering amplitude matrix, we must consider two polarization states
for the incident wave, specifically with polarization configurations of (Eo,x = 1, Eo,y = 0)
and (Eo,x = 0, Eo,y = 1). Using the SIM or VIM scheme in conjunction with the two
polarization cases and performing a number of vector algebraic calculations, we can obtain
the amplitude scattering matrix. Specifically, we may define the following equations in
conjunction with either SIM or VIM in the following forms:
SIM scheme:

F//,x k2 X⊥ · [ns × E(r )] + X// · [ns × H(r )]
=
F⊥,x 4π s X⊥ · [ns × H(r )] − X// · [ns × E(r )]

× exp(−ikη · r )d 2 r , (4.1.34a)
Eo,x =1,Eo,y =0

F//,y k2 X⊥ · [ns × E(r )] + X// · [ns × H(r )]
=
F⊥,y 4π s X⊥ · [ns × H(r )] − X// · [ns × E(r )]

× exp(−ikη · r )d 2 r , (4.1.34b)
Eo,x =0,Eo,y =1
VIM scheme:

F//,x ik 3 ) 2 * X// · E(r )
= mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r )
F⊥,x 4π V X⊥ · E(r )

× exp(−ikη · r )d 3 r , (4.1.35a)
Eo,x =1,Eo,y =0

3

F//,y ik ) 2 * X// · E(r )
= mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r )
F⊥,y 4π V X⊥ · E(r )

× exp(−ikη · r )d 3 r . (4.1.35b)
Eo,x =0,Eo,y =1
From Eqs. (4.1.28), (4.1.29), (4.1.33), (4.1.34a), (4.1.34b), (4.1.35a), and (4.1.35b) and
after analysis, we obtain

Es,// exp(ikr) F//,y F//,x Eo,y
=
Es,⊥ −ikr F⊥,y F⊥,x Eo,x

exp(ikr) F//,y F//,x X⊥ · x X⊥ · y Eo,//
= . (4.1.36)
−ikr F⊥,y F⊥,x −X⊥ · y X⊥ · x Eo,⊥
Comparing Eqs. (4.1.31) and (4.1.36), we obtain the amplitude scattering matrix in the
form

S2 S3 F//,y F//,x X⊥ · x X⊥ · y
= . (4.1.37)
S4 S1 F⊥,y F⊥,x −X⊥ · y X⊥ · x
With the scattering amplitude matrix given in Eq. (4.1.37), it is straightforward to

evaluate the scattering phase matrix using the formalism outlined in Subsection 2.4.3,
where we showed that, for ice particles oriented randomly with an equal number of mirror-
imaging positions in space, the scattering phase matrix has a block-diagonal structure with
eight non-zero elements of which only six are independent. Moreover, the absorption and
extinction cross-sections σabs and σext , respectively, can be computed from the near field
within the interior of ice particles via the following equations:

k
σabs = 2mr (r )mi (r )E(r ) · E∗ (r )d 3 r , (4.1.38)
Einc · E∗inc V

k ) 2 2
* ∗ 3
σext = Im mr (r ) − mi (r ) − 1 E(r ) · Einc (r )d r , (4.1.39)
Einc · E∗inc V
where an asterisk denotes complex conjugation and Im{ } in Eq. (4.1.39) signifies the
imaginary part of the related variables. Equations (4.1.38) and (4.1.39) can be derived from
the electromagnetic energy conservation principle involving the Poynting vector associated
with light scattering processes, as illustrated in Yang and Liou (1996b).
4.1.5 Validation of the method and some pertinent results

The accuracy of the FDTD method can be checked using a number of canonical cases,
for instance, infinite circular cylinders and spheres (see Subsection 2.4.6) for which exact
solutions are available. In general, when the size of grid cells is on the order of 120
of the incident wavelength, solutions are in excellent agreement with their corresponding
analytical counterparts for cases involving ice indices of refraction at visible and infrared
wavelengths. Accuracy improvement could be achieved by decreasing the ratio of the grid
size to the incident wavelength; however, this improvement would be at the expense of
increasing computational effort. In what follows, we present several representative results
to illustrate the numerical performance of the FDTD method and its general applicability
to light scattering by ice particles.
103 1
Absolute Error -P43/ P11 Absolute Error P33/ P11 Absolute Error -P12/ P11
Lorenz-Mie 0.5
FDTD 0.0
102
Phase Function P11
-0.5
x = 10 -1
0.3
10 1 l/Δs = 30
0.15
m = 1.5015 - i 0.067
0.0
100
-0.15
1
-1 0.5
10
0.0
-0.5
102 -1
0.3 0.3
Relative Error (%) Absolute Error
0.15
0.15 0.0
0.0 -0.15
-0.3
-0.15 1
0.5
-0.3
0.0
30 -0.5
15 -1
0.4
0.0 0.2
0.0
-15
-0.2
-30 -0.4
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Figure 4.3 Non-zero phase matrix elements P11 in terms of absolute and relative (%) errors, and
−P12 , P33 , and −P43 normalized by P11 in terms of absolute errors, computed from the FDTD method
in comparison to the Lorenz–Mie solution for an ice sphere at a wavelength of 25 µm (after Yang
et al. 2004).
Figure 4.3 shows a comparison of the non-zero phase matrix elements computed with
the FDTD method and the Lorenz–Mie results for an ice sphere at a far-infrared wavelength
λ = 25 µm with a refractive index of m = 1.5015 − i 0.067. In the past two decades, far-
infrared spectral information has received substantial attention in the atmospheric research
community (e.g., Mlynczak et al. 2002). In particular, its signatures have been demon-
strated to effectively facilitate the retrieval of microphysical and optical properties of thin
cirrus clouds (Yang et al. 2003b). For the FDTD simulation shown in Figure 4.3, 5872
time-marching iterations were performed to map the near field from the time domain to
the frequency domain through the discrete Fourier transform. The left-hand column of
Figure 4.3 shows FDTD and Lorenz–Mie phase functions along with both absolute and rel-
ative errors of the FDTD results. For the FDTD phase function displayed here, the largest
absolute errors are found in forward directions with scattering angles smaller than 20°,
whereas the largest relative errors are noted at scattering angles near 160°. The right-hand
column of Figure 4.3 depicts a comparison between FDTD and Lorenz–Mie solutions for
102 1
Absolute Error -P43/ P11 Absolute Error P33/ P11 Absolute Error -P12/ P11
Lorenz-Mie 0.5
FDTD 0.0
Phase Function P11
-0.5
-1
101 x=5 0.1
l/Δs =165 0.05
m = 8.2252 - i 1.6808 0.0
-0.05
100 -0.1
1
0.5
0.0
-0.5
-1
10 -1
0.25 0.1
Relative Error (%) Absolute Error
0.05
0.0 0.0
-0.05
-0.25 -0.1
1
0.5
-0.5
0.0
10 -0.5
5 -1
0.1
0.0 0.05
0.0
-5
-0.05
-10 -0.1
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Figure 4.4 Comparison between FDTD and Lorenz–Mie solutions for a sphere with a large refractive
index. The presentation in this figure has the same format as in Figure 4.3 (after Yang et al. 2004).
the other phase matrix elements related to the polarization state of scattered light. Because
these phase matrix element values can be zero, errors in the FDTD results are only quan-
tified in terms of absolute errors in the right-hand column of Figure 4.3. Overall, FDTD
results for phase matrix elements are quite accurate with respect to exact Lorenz–Mie
values.
To demonstrate the applicability of the FDTD method for an optically dense medium
with a large refractive index, Figure 4.4 shows a comparison between FDTD and Lorenz–
Mie solutions for the scattering phase matrix of a sphere with a size parameter of 5 and
a refractive index of 8.2252 − i1.6808, which represents the value for liquid water at a
temperature of 300 K at a microwave wavelength of 3.2 cm. Because an optically dense
medium is involved in this case, a very fine grid-mesh resolution (s = λ165) was
used and, in this case, 7500 time-marching steps were required to obtain a convergent
solution. In terms of the relative error, deviations of FDTD phase function results from
exact Lorenz–Mie values are less than 5% across the entire scattering angle domain. For
other phase matrix elements, FDTD solutions closely match Lorenz–Mie counterparts.
The comparisons shown in Figures 4.3 and 4.4 clearly demonstrate the applicability and
1
102 ADDA 0.8 P22/P11
FDTD 0.6
0.4
0.2
101 kL = 30, l = 0.66 μm
1
m = 1.3078 - i1.66x10-8 P33/P11
0.5
P11
0
100
-0.5
-1
0.4
10-1 0.2 P34/P11
0
-0.2
10-2 -0.4
1
P44/P11
P12/P11
0.5 0.5
0 0
-0.5 -0.5
1
102 ADDA 0.99
FDTD 0.98
0.97
P22/P11
0.96
101
kL = 30, l = 12 μm 1
m = 1.28 - i0.41332 0.5
P33/P11
P11
0
100
-0.5
-1
0.5 P34/P11
10-1
0
-0.5
10-2 1
1
P12/P11
0.5
0 0 P44/P11
-0.5
-1 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.5 Comparison between phase matrix elements calculated from FDTD and DDA at 0.66
and 12 µm wavelengths. The Amsterdam DDA (ADDA) code developed by Yurkin et al. (2007) was
used for the present DDA simulation (after Yang et al. 2013).
accuracy of FDTD for the scattering of light by a dielectric sphere with a moderate or large
refractive index.
Applicability of the FDTD method to light scattering by hexagonal particles is illustrated
in Figure 4.5, in which a comparison between FDTD and DDA solutions was made for the
scattering phase matrix of randomly oriented hexagonal ice crystals at 0.66 µm (visible) and
Extinction Efficiency Qext 4
l = 0.55 μm
3 L/a = 6
FDTD
1
Ray-by-ray
0
6 10 100 1000
4 1
Single-Scattering Albedo
Extinction Efficiency Qext
l = 3.7 μm
l = 3.7 μm
3
2 0.75
0 0.5
6 10 100 1000 6 10 100 1000
Size Parameter kL Size Parameter kL
Figure 4.6 Comparison of extinction efficiency and single-scattering albedo computed by FDTD
and by the ray-by-ray geometric-optics approach for randomly oriented column crystals with an
aspect ratio of 6 using 0.55 µm (essentially without absorption) and 3.7 µm wavelengths (after Yang
and Liou 1997).
12 µm (thermal IR) wavelengths. Herein, we note that FDTD and DDA are substantially
different in terms of their fundamental approaches: The former is a time-domain method,
whereas the latter is a frequency-domain method. Figure 4.5 illustrates that FDTD results
closely match their DDA counterparts, a confirmation of the applicability and accuracy of
the FDTD approach to light scattering by ice crystals from the perspective of another well
validated and numerically rigorous method.
Figure 4.6 illustrates the extinction efficiency and single-scattering albedo computed
from FDTD and the ray-by-ray geometric-optics algorithm (Yang and Liou 1997) dis-
cussed in Chapter 3. At a size parameter of approximately 20, results computed with the
two methods converge. The ray-by-ray algorithm is much more computationally efficient
than FDTD, particularly for computing extinction and absorption cross-sections. Thus,
results shown in Figure 4.6 provide an approximate accuracy guideline in terms of deter-
mining extinction and absorption characteristics; that is, FDTD can be effectively applied
to size parameters smaller than 20, whereas the ray-by-ray approach can be employed
2
10
1 branch
3 branches
4 branches
1 5 branches
10
Phase Function
6 branches
8 branches
9 branches
0
10 10 branches
12 branches
-1
10
x = 10, l = 11 μm
-2
10
0 60 120 180
Figure 4.7 Phase functions for randomly oriented single-bullet and bullet-rosette ice crystals as a
function of the scattering angle at a wavelength of 11 µm with an index of refraction m = 1.0925 −
i 0.248. The size parameter is defined with respect to the length of a bullet element (after Yang et al.
2004).
for larger size parameters. The convergence of ray-by-ray and FDTD approaches at a size
parameter of 20 and an aspect ratio of La = 6 for ice columns illustrates that the edge
effect (see Subsections 3.3.5–3.3.7), which was not included in the former, appears to be
insignificant.
The FDTD method has been employed to study the optical properties of complex ice
crystals. For example, Figure 4.7 shows phase functions of randomly oriented single-bullet
and bullet-rosette ice crystals at an infrared wavelength of 11 µm. The angle between the
pyramidal faces of a bullet element and the element axis is 26.5°. The phase function
is quite sensitive to the number of ice bullet elements. In particular, the phase function
of a multi-branched rosette with more than eight branches displays maximal features in
the side-scattering direction at 60°. Although the radiative properties of ice crystals are
dominated by absorption at λ = 11 µm, the scattering effect is not negligible (Chou et al.
1999). Thus, using an appropriate scattering phase function for multi-branched rosette ice
crystals, representative of the morphology in some cirrus clouds, may enhance the accuracy
in downstream remote sensing applications (Heymsfield et al. 2002).
Figure 4.8 depicts phase functions for six ice crystal habits with a small size parameter kD
of 20, where D is the maximum dimension, to illustrate ice crystal shape effect. The imprint
of halo peaks in the phase function pattern largely disappears for small ice crystals. The
phase function patterns for small plates and columns are rather smooth, particularly in the
scattering region from 90° to 180°. Furthermore, the phase function of droxtal ice crystals
displays substantial variation in the scattering angle domain due to phase interference
effects that have not been smoothed out in random orientation average.
102 102 102
101 101 101
100 kD = 20 100 kD = 20 100 kD = 20
10-1 10-1
10-1
Phase Function
10-2 10-2
10-2
0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180
102 102 102
101 101 101
ka = 10 kL = 20 kL = 20
100 L/a = 0.5 100 100
L/a = 8 L/a = 8
10-1 10-1 10-1
10-2 10-2 10-2

0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)
Figure 4.8 Comparison of phase functions computed from the FDTD method for ice crystal shapes
commonly observed in ice clouds. The parameter D is the maximum dimension for droxtal, bullet
rosette, or aggregate. For plate and column, a denotes the half-width and L is the length (column) or
depth (plate), k = 2π/λ, and λ is the wavelength (after Yang and Liou 2006).
4.1.6 Pseudo-spectral time domain method

Stemming from the traditional FDTD, Liu (1997) employed the pseudo-spectral method,
instead of the finite-difference method, to approximate the spatial derivatives in Maxwell’s
equations; this is frequently referred to as the pseudo-spectral time domain (PSTD) method.
A comprehensive review of the PSTD method can be found in Panetta et al. (2013). In this
subsection, we shall capture the essence of the PSTD approximation in terms of spatial
derivatives of the form
⎡ ⎤
1 2πI l i ⎣
N/2−1 N/2−1
∂F n (I, J, K)
F n (m, J, K)e− N i ⎦,
2πlm
= e N 2πkl i (4.1.40)
∂x N l=−N/2 m=−N/2
where N is the total number of grid points in the x-axis direction and kl is the Fourier spectral
variable given by kl = l(Nx). Unlike FDTD, which uses the “Yee cell” (Figure 4.2),
PSTD employs a centered grid scheme in which all the field components are specified at the
centers of grid cells. PSTD provides a higher degree of accuracy and smaller numerical
dispersion errors, leading to the enhancement of computational applicability and efficiency
due to the use of a relatively coarse grid mesh and fast Fourier transform. Furthermore, the
CFL condition for PSTD, which differs from the FDTD counterpart, is given by (Liu 1997)
2
ct ≤ . (4.1.41)
π 1/x + 1/y 2 + 1/z2
2
Chen et al. (2008) applied PSTD to compute the single-scattering properties of ice
crystals with small to medium size parameters by alleviating the Gibbs phenomenon and
parallelizing PSTD implementations. As an example, Figure 4.9 depicts phase functions
computed by the PSTD method compared with those from the Lorenz–Mie theory for size
parameters 30 and 50 in an optically soft case (m = 1.05 − i 0) and in a case involving a
moderate refractive index (m = 1.6 − i 0). For the optically soft case, a grid spatial resolution
of 10 was used, whereas the grid resolution was increased to 16 for the moderate case.
It is clear from Figure 4.9 that an excellent agreement between PSTD and Lorenz–Mie
results has been achieved, although slight differences are observed at scattering angles
of 120° and 160° at which phase function minima occur. Also, PSTD phase functions
deviate slightly from their Lorenz–Mie counterparts at scattering angles near 180°.
Liu et al. (2012a) further improved the PSTD method using an empirical approach to
mitigate the effect associated with the Gibbs phenomenon caused by medium discontinuity.
An illustrative explanation of the Gibbs phenomenon can be found in Panetta et al. (2013).
Furthermore, Liu et al. (2012b) applied PSTD to moderately large size parameters. For
example, Figure 4.10 shows the PSTD solution for the optical properties of a sphere with
a size parameter of 200 in comparison with Lorenz–Mie results. The refractive index of
the sphere used was 1.312 − i 1.489 × 10−9 . The relative error in the normalized phase
function and the absolute errors in the ratios P12 P11 , P33 P11 , and P34 P11 are illustrated
in Figure 4.10. Overall, PSTD and Lorenz–Mie results are in close agreement.
Figure 4.11 depicts a comparison between PSTD and T-matrix solutions for the phase
function associated with an oriented spheroid with a refractive index of 1.381 − i 3.9 × 10−9
for finite circular cylinder and spheroid cases. The cylinder size is specified by the radius of
its cross-section r and by its length L, while the spheroid size is determined in terms of its
polar radius b and equatorial radius a. For T-matrix results, the T-matrix code developed by
Mishchenko and Travis (1998) was used, while for PSTD simulations, a grid resolution of
12 was employed. Additionally, the phase functions in Figure 4.11 are defined with respect
to the plane containing the incident direction and the particle’s symmetry axis. PSTD results
for finite cylinders and spheroids closely match T-matrix results.
Figure 4.12 illustrates the phase functions of ice crystals defined by two shapes, hollow
hexagonal column and aggregate, computed by the PSTD method. The orientation of the
ice crystals in the laboratory coordinate system is shown in the upper diagrams. The phase
function of ice crystals with a size parameter of 20 was calculated using the PSTD method
with a spatial resolution of 14 at a wavelength of 3.7 µm. In the simulation, the incident
106
(a) 2πr/l = 30 (b) 2πr/l = 30
m = 1.05 m = 1.6
105
Δs = l /10 Δs = l /16
104
Phase Function
103 Lorenz-Mie
PSTD
102
101
100
10-1
10-2
107
(c) 2πr/l = 50 (d) 2πr/l = 50
106 m = 1.05 m = 1.6
Δs = l /10 Δs = l /16
105
Phase Function
104
103
102
101
100
10-1
10-2
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.9 Comparison of phase functions computed by the PSTD method and by the Lorenz–Mie
theory for spheres with size parameters of 30 and 50 (after Chen et al. 2008).
direction was defined by θi = 45◦ and φi = 0◦ , where θi and φi are the incident zenith
and azimuthal angles specified with respect to the laboratory coordinate system. Phase
functions were computed for θ s = 0°–180° and φ s = 0°. PSTD solutions are compared
with the corresponding results computed by the FDTD method. The spatial resolutions
for PSTD and FDTD simulations were 12 and 32, respectively. The phase functions com-
puted from the PSTD and FDTD methods for each geometry are compared in the graphs,
Figure 4.12a for a hollow column ice crystal and Figure 4.12b for an aggregate. The aspect
2
4
10
Phase Function, P11
Lorenz−Mie
Relative Error
2 1
10 PSTD
10 0
0
−2
10
−1
1.0 1.0
P12 /P11
0.0 0.0
−1.0 −1.0
1.0 1.0
Absolute Error
P33 /P11
0.0 0.0
−1.0 −1.0
1.0 1.0
P34 /P11
0.0 0.0
−1.0 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.10 Non-zero phase matrix elements computed using PSTD for a sphere with a size param-
eter of 200 and a refractive index of 1.312 − i 1.489 × 10−9 . The relative errors in the phase function
(P11 ) and absolute errors in the terms P12 P11 , P33 P11 , and P34 P11 are displayed in the right panels
(after Liu et al. 2012b).
ratio of the hollow column is 2aL = 0.7, where a and L are the semi-width and length of
the column, respectively. The depth of cavities at each end of the column is given by d =
0.25L. The aggregate geometry definition for a given maximum dimension D of an ice crys-
tal follows that of Yang and Liou (1998a). From Figure 4.12a, the PSTD results agree well
with their FDTD counterparts in all scattering directions, except for slight differences at
several scattering angles. However, the FDTD method requires a much finer grid resolution
than the PSTD method in order to achieve similar accuracies.
104 106
(a) 2πb/l = 20 (b) 2πb/l = 50
b/a = 2 b/a = 2
103 105
102 104
Phase Function
T-matrix
PSTD
101 103
100 102
10-1 101
10-2 100
105 107
(c) 2πr/l = 20 (d) 2πr/l = 50
L/r = 1 106 L/r = 1
104
105 o
o 30
Phase Function
30
103
104
103
102
102
101
101
100 100
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.11 Comparison of PSTD and T-matrix results for the phase functions of spheroids and
circular cylinders. The index of refraction used is m = 1.381 − i 3.9 × 10−9 and the size parameters
are 20 and 50 with an aspect ratio of 2 for the spheroid and 1 for the circular cylinder (after Chen
et al. 2008).
Figure 4.13 shows the non-zero phase matrix elements for randomly oriented hexagonal
columns with a size parameter of 100. With moderate size parameters, the PSTD results
can only be compared with those given by geometric-optics methods. Here, the solutions
based on the improved geometric-optics method (IGOM) (Yang and Liou 1996a) are used
Z Z
(a) (b)
Y
Y
X
X
103 103
Lπ/l = 20 Dπ/l = 20
102 l = 3.7 μm 102 l = 3.7 μm
m = 1.4005 - i 7.2x10-3 m = 1.4005 - i 7.2x10-3
Phase Function
101 101
FDTD
PSTD
100 100
10-1 10-1
10-2 10-2
10-3 10-3
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.12 Comparison of FDTD and PSTD results for the phase function as a function of scattering
angle for hollow column and aggregate ice crystals. The input parameters are shown in the figure
(after Chen et al. 2008).
for the comparison, and the IGOM results closely agree with their PSTD counterparts.
The two methods give almost the same overall variation patterns, particularly, for P11 ;
however, the PSTD solutions for P12 P11 , P22 P11 , P33 P11 , P34 P11 , and P44 P11 show
pronounced variations versus scattering angle, and the IGOM results are relatively smooth.
The differences occur because the PSTD simulation rigorously takes into account the phase
interference of the electromagnetic waves. Moreover, Figure 4.13 clearly identifies strong
scattering peaks at 22° and 46° in the phase function computed from both PSTD and IGOM
methods for a size parameter of 100.
The single-scattering properties (extinction efficiency, single-scattering albedo, and
asymmetry factor) of randomly oriented hexagonal columns and plates computed by FDTD
and PSTD are illustrated in Figure 4.14. The indices of refraction of ice at wavelengths
of 3.7 µm and 23 µm are 1.4005 − i 0.007 201 and 1.4424 − i 0.027 01, respectively. The
results obtained by FDTD and PSTD are shown to agree with each other quite well.
Due to its flexibility in handling particle geometry and inhomogeneity, the FDTD method
has been widely used to compute the single-scattering properties of non-spherical ice
4
10 0.4
PSTD
Phase Function, P11
IGOM 0.2
2
10 x = 100
12 11
P /P
0
0
10
−0.2
−2
10 −0.4
1 1
0.5
0.5
33 11
22 11
P /P
P /P
0
0
−0.5
−0.5 −1
0.2 1
0 0.5
34 11
44 11
P /P
P /P
−0.2 0
−0.4 −0.5
−0.6 −1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.13 Non-zero phase matrix elements of randomly oriented hexagonal columns as a function
of scattering angle using a size parameter of 100, simulated by PSTD and IGOM. The hexagonal
column has a diameter-to-length ratio of 1.0 and a refractive index of 1.312 − i 1.489 × 10−9 (after
Liu et al. 2012b).
crystals and aerosols (Yang and Liou 1996b, 2000; Fu et al. 1998; Sun et al. 1999; Yang
et al. 1997, 2000b, 2005; Baran 2003; Baum et al. 2007; Ishimoto et al. 2012). Application
of the more powerful PSTD method to complex ice crystals and aerosols appears to be
quite promising, as shown in the work of Chen et al. (2008) and Liu et al. (2012b). With
continued numerical and technical advances in FDTD and PSTD as well as an increase in
supercomputer power, it is anticipated that these two methods will play an important role
in solutions for electromagnetic scattering and absorption by atmospheric particles which
exhibit complex geometry and internal inhomogeneity.
(a ) (b)
Single-Scattering Albedo Extinction Efficiency l = 3.7μm l = 23μm l = 3.7μm l = 23μm
3
2.5
2
1.5
1 FDTD FDTD
PSTD PSTD
0.5
0
1
0.8
0.6
0.4
0.2
0
1
Asymmetry Factor
0.8
0.6
0.4
0.2
0
100 101 102 100 101 102 100 101 102 100 101 102
Size Parameter xmax Size Parameter xmax
Figure 4.14 The extinction efficiency, single-scattering albedo, and asymmetry factor for (a) ran-
domly oriented hexagonal columns and (b) plates as a function of size parameter xmax at a wavelength
of 3.7 µm (m = 1.4005 − i 0.007 201) and 23 µm (m = 1.4424 − i 0.027 01), where xmax is defined
as 2πrmax λ, rmax denotes one-half of the maximum dimension, and λ is the wavelength (courtesy of
Chao Liu).
4.2 T-matrix numerical method

4.2.1 Fundamentals of the T-matrix approach
The incident and scattered electric fields can be expanded in terms of suitable vector
wave functions. The expansion coefficients associated with the scattered wave can be
related to their counterparts for the incident wave through a transmitting matrix, referred
to as the T-matrix, which contains all the inherent single-scattering characteristics of the
corresponding scattering particle. Waterman (1971) developed a technique known as the
extended boundary condition method (EBCM) to solve the T-matrix. The EBCM has been
thoroughly explained by Barber and Yeh (1975). Mishchenko and Travis (1998) have
implemented EBCM to efficiently compute the single-scattering properties of a number
of axially symmetric non-spherical particles, including spheroids, circular cylinders, two-

sphere clusters, and Chebyshev particles.
The EBCM technique has been found to be numerically unstable in calculations for
particles with large size parameters, and the maximum convergent size parameter has
been shown to be sensitively dependent on the refractive index and particle geometry.
In the early development of EBCM, Iskander et al. (1983) further improved Waterman’s
EBCM in terms of the numerical stability for spheroidal particles. Additionally, Mugnai
and Wiscombe (1986, 1989) and Wiscombe and Mugnai (1988) applied EBCM to non-
spherical shapes with smooth surfaces that could be defined by Chebyshev polynomials.
More efficient approaches that overcome the numerical instability problem in computing the
T-matrix for highly elongated particles have subsequently been developed. With computer
coding improvements, in particular, of the analytical orientation-average algorithm and
the computation of relevant special functions, the T-matrix method can be applied to
axially symmetric homogeneous non-spherical particles with size parameters as large as 180
(Mishchenko et al. 2000a). However, EBCM has not been frequently used for application
to non-symmetric and inhomogeneous particles due to inadequate numerical performance.
The T-matrix method commonly begins with the expansion of the incident and scattered
fields with respect to vector spherical wave functions in the forms
∞
n
Ei (R) = [amn RgMmn (kR) + bmn RgNmn (kR)], (4.2.1a)
n=1 m=−n
∞
n
Es (R) = [pmn Mmn (kR) + qmn Nmn (kR)], (4.2.1b)
n=1 m=−n
where RgMmn , RgNmn , Mmn , and Nmn are vector spherical wave functions that satisfy the
vector Helmholtz equation; Mmn and Nmn involve spherical Hankel functions and form a
basis for expanding the scattered field satisfying the Somerfield radiation condition; and
RgMmn and RgNmn are regular spherical wave functions obtained from spherical Bessel
functions. Because of the linearity of Maxwell’s equations and boundary conditions, the
relationship between the scattered coefficients (pmn and qmn ) and incident coefficients (amn
and bmn ) must be linear and can be expressed by a T-matrix as

∞
n
) 11 *
pmn = Tmnm n am n + T12
mnm n bm n , (4.2.2a)
n =1 m =−n

∞
n
) 21 *
qmn = Tmnm n am n + T22
mnm n bm n . (4.2.2b)
n =1 m =−n
In compact matrix form, we can write

11

p a T T12 a
=T = . (4.2.3)
q b T21 T22 b
The preceding equation can also be written in a more detailed matrix form as
⎡ ⎤ ⎡ ⎤⎡ ⎤
p1 T11 11 T11 12 ... ... T11 1,lmax T12 1,lmax a1
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ q1 ⎥ ⎢ T11 21 T11 22 ... ... T21 1,lmax T22 1,lmax ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ b1 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ ... ⎥ ⎢ ... ... ... ... ... ... ⎥⎢ ... ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ . (4.2.4)
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ ... ⎥ ⎢ ... ... ... ... ... ... ⎥⎢ ... ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ p ⎥ ⎢ T11 T12 T11 T12 ⎥⎢ ⎥
⎣ lmax ⎦ ⎣ lmax ,1 lmax ,1 ... ... lmax ,lmax lmax ,lmax ⎦ ⎣ almax ⎦
qlmax T21
lmax ,1 T22
lmax ,1 ... ... T21
lmax ,lmax T22
lmax ,lmax blmax
In Eq. (4.2.4), double indices (mn) are combined into a single index l defined as
n(n + 1) + m. Equation (4.2.3) or (4.2.4) forms the foundation of the T-matrix approach.
If the T-matrix for a given scattering particle is known, the scattering amplitude matrix can
be calculated in a straightforward manner. To illustrate the computational procedure, we
shall consider the amplitude scattering matrix of the Lorenz–Mie type, given by

S2 0
S= , (4.2.5)
0 S1
where the two diagonal elements are defined as follows:
∞
2n + 1
S2 = [an τn (cos θ ) + bn πn (cos θ )], (4.2.6)
n=1
n(n + 1)
∞
2n + 1
S1 = [an πn (cos θ ) + bn τn (cos θ )]. (4.2.7)
n=1
n(n + 1)
In Eqs. (4.2.6) and (4.2.7), θ is the scattering angle, and the two coefficients are related to
the T-matrix elements by
∞ 5
2n + 1 n −n ) 21 *
an = − i T1n1n + T1n1n
22
, (4.2.8)
n =1
2n + 1
∞ 5
2n + 1 n −n ) 11 *
bn = − i T1n1n + T1n1n
12
. (4.2.9)
n =1
2n + 1
When the coefficients an and bn in Eqs. (4.2.8) and (4.2.9) are known, the amplitude
scattering matrix and, subsequently, the phase matrix can be computed.
The T-matrix contains the complete inherent optical properties of the scattering parti-
cle and is independent of the orientation of the particle with respect to the direction of
the incident light. This unique feature of the T-matrix discovered by Mishchenko (1990)
provides a computationally efficient and exact approach to calculating single-scattering
properties averaged over random orientations. For example, the extinction and scattering
cross-sections can be obtained for T-matrix elements through explicit summations given by
Mishchenko et al. (2002) as follows:
2π
lmax
lmax
11 2 12 2 21 2 22 2
Csca = T + T + T + T , (4.2.10)
k l=1 l =1 ll
2 ll ll ll
2π
lmax
11
Cext = − 2
Re Tll + Tll22 . (4.2.11)
k l=1
The EBCM approach to determining the T-matrix is to first find the field scattered
by an object bounded by a closed surface in terms of an integral equation that involves
the unknown surface field on the exterior of the closed surface, which is related to the
internal field through boundary conditions. The internal field within the particle defined by
a refractive index m relative to the surrounding medium is defined by
∞
n
E (R) =
int
[cmn RgMmn (mkR) + dmn RgNmn (mkR)] , (4.2.12)
n=1 m=−n
where cmn and dmn are unknown coefficients (the symbol for the refractive index should not
be confused with the subscript m). By using boundary conditions at the scatterer surface,
which requires the continuity of tangential components of electric and magnetic fields, we
obtain the following matrix equation that relates scattered and internal fields:

p RgQ11 RgQ12 c
=− . (4.2.13)
q RgQ21 RgQ22 d
The elements in the RgQ matrix are surface integrals of the products of vector spherical
wave functions that depend only on the particle’s size, shape, and refractive index. Moreover,
from the boundary condition and the cancelation of incident and scattered fields within the
particle with reference to equivalent surface currents, we can show that

11

a Q Q12 c
= , (4.2.14)
b Q21 Q22 d
where the Q elements are also given by integrals over the surface of the particle and depend
only on its geometric characteristics. The unknown expansion coefficients of the internal
field, c and d, can be expressed in terms of known coefficients, a and b, by inverting the
matrix equation (4.2.14). From Eqs. (4.2.3), (4.2.13), and (4.2.14), we obtain the T-matrix
in the form
T = −Rg Q Q−1 , (4.2.15)
where the T-matrix is independent of the incident field represented by coefficients a and b.
We present Figure 4.15 to show the phase function P11 and the degree of linear polar-
ization −P12 P11 computed for a sphere and for three spheroids with aspect ratios 1.5, 2.0,
and 2.5. The same size parameter was used for the sphere and spheroids, defined in terms
of a volume-equivalent sphere of 35. The simulation was conducted by using the EBCM
T-matrix computational program developed by Mishchenko and Travis (1998). Note that
104 1
sphere
3
10 a/b = 1.5
a/b = 2.0 0.5

102 a/b = 2.5
xv = 35
-P12/P11
P11
1 0
10
100
-0.5
-1
10
10-2 -1
0 45 90 135 180 0 45 90 135 180
Figure 4.15 T-matrix computations of the phase function P11 and the degree of linear polariza-
tion −P12 P11 as a function of the scattering angle for a sphere and three spheroids defined by a
volume equivalent spherical size parameter, xv . The T-matrix computational program developed by
Mishchenko and Travis (1998) was used in these simulations. The refractive index employed in the
calculation for these four particles was 1.53 − i 0.008.
−P12 P11 represents the degree of linear polarization in the case involving an unpolar-
ized incident light. Figure 4.15 clearly illustrates deviations between the scattering and
polarization properties of spheroids and a sphere.
4.2.2 Invariant imbedding T-matrix method

In this subsection, we present an alternative to the conventional T-matrix to light scattering
computations which follows the invariant imbedding principle. This principle has been
used to facilitate radiative transfer computations for external emergent quantities associ-
ated with reflection and transmission of a system (e.g., Chandrasekhar 1950). It appears
that application of the invariant imbedding principle to the single-scattering of light by a
particle has been accomplished by Johnson (1988), who used this principle to compute
T-matrix elements associated with the single-scattering properties of dielectric particles in
a number of axially rotational symmetry cases for small size parameters. Johnson’s work,
however, has not been noted and cited by a number of excellent books which discuss the
T-matrix approach (Barber and Hill 1990, Mishchenko et al. 2002, Doicu et al. 2006), nor
by the comprehensive T-matrix reference database (Mishchenko et al. 2010a, Zakharova
(a) (b) (c)
(d) (e) (f)
Figure 4.16 A conceptual representation of a 2-D plate as an example of the invariant imbedding
T-matrix (II-TM) method. (a) The 2-D plate shape. (b) This is circumscribed by an imaginary
inhomogeneous sphere; the portion of the sphere exterior to the 2-D plate is empty. (c) Discretization
of this inhomogeneous sphere into multi-layers. (d) Computation of the T-matrix using an internal
sphere whose T-matrix can be obtained from the Lorenz–Mie theory. (e) An intermediate step in
computing the T-matrix via iterative procedures based on the II-TM approach. (f) The final particle
geometry, a 2-D plate, after completing II-TM computations.
et al. 2012). Also, according to the ISI Web of Knowledge, Johnson’s work in 1988 was
cited only five times by other researchers as of February 8, 2013. We further note that John-
son’s exploration of the invariant imbedding principle for incorporation into the T-matrix
approach can be traced back to his study of quantum scattering (Johnson and Secrest
1968).
Figure 4.16 illustrates a conceptual procedure for application of the invariant imbedding
principle to T-matrix calculations (hereafter referred to as the II-TM approach) for scattering
by an arbitrarily shaped particle: (a) A 2-D plate is used as an example. (b) Scattering by
this plate can be thought of as scattering by an inhomogeneous circumscribed sphere
that is partially empty, such that the refractive index of the empty portion is 1. (c) This
inhomogeneous sphere is then discretized into a multi-layered sphere, so the problem
of light scattering by a non-spherical particle is transformed into that of light scattering
by a multi-layered inhomogeneous sphere. (d) The T-matrix is computed for an internal
sphere using Lorenz–Mie theory; for computational efficiency, we may select a sphere that
is inscribed within the 2-D plate. (e) An intermediate step is to compute the T-matrix by
means of an iterative procedure based on the II-TM approach. (f) The final particle geometry
when iterative II-TM computations are completed would be the original 2-D plate.
The principle of II-TM is to obtain the T-matrix of an inhomogeneous sphere with p
layers based on the T-matrix of an inhomogeneous sphere with p − 1 layers. As an initial
point for iterative computations, the T-matrix at the origin is zero. The II-TM recursive
4
10 0.5
11
II-TM
-P /P
3
EBCM 0
12
10
c
2 -0.5
10 a
1
0.5
11
P /P
11
1
10 0
P
22
m = 1.311 -0.5
10
0
kc = 2ka = 95 -1
1
0.5
11
-1
P /P
10 0
33
-0.5
-2
10 -1
0.5 1
0.5
11
11
P /P
P /P
0 0
44
43
-0.5
-0.5 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.17 Six independent scattering phase matrix elements simulated by the EBCM method
(Mishchenko and Travis 1998) and the new II-TM approach for randomly oriented spheroids as a
function of the scattering angle (after Bi et al. 2013).
equation is given as follows:

T(rp+1 ) = Q11 (rp+1 ) + [I + Q12 (rp+1 )][I − T(rp )Q22 (rp+1 )]−1 T(rp )[I + Q21 (rp+1 )],
(4.2.16)
where T(rp ) is the T-matrix composed of p layers, I is the identity matrix, and Qij are the
matrices involving integration over inhomogeneous spherical layers. To speed the II-TM
computation, the method of separation of variables (Bohren and Huffman 1983) can be
employed to compute the T-matrix of a sphere inscribed within the non-spherical particle
to reduce the number of iterations of Eq. (4.2.16); this equation can be derived from an
electromagnetic volume integral equation as follows (Johnson 1988, Bi et al. 2013):

) 2 *↔
E(r) = E (r) + k
inc 2
m (r) − 1 G(r − r ) · E(r )d 3 r , (4.2.17)
V
inc
where E is the total field, E is the incident electric field, m is the refractive index, V is
↔
the particle volume, and G(r − r ) is the dyadic Green function.
Since Johnson’s work in 1988, numerical techniques and computing power have sig-
nificantly advanced. We have revisited the II-TM method for single-scattering computa-
tions. Based on a contemporary numerical implementation (Bi and Yang 2014), the II-TM
approach can apply to size parameters substantially larger than its EBCM-TM counter-
part. To validate the II-TM implementation, Figure 4.17 compares the scattering phase
105 1 1
II-TM
c IGOM -P12 /P11
a 0.5 P22 /P11
0.5
P11
m = 1.311 0
ka = 150, kc = 300
0
100 -0.5
-1 -0.5
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Figure 4.18 Comparison of phase matrix elements simulated using the II-TM method and using an
improved geometric-optics method (IGOM) (Yang and Liou 1996b, Yang et al. 2005) for randomly
oriented spheroids with a size parameter of 300 (after Bi et al. 2013).
matrices computed from II-TM and EBCM-TM computational codes for randomly ori-
ented spheroids with a size parameter of 95, specified in terms of the semi-major axis, and
an aspect ratio of 2. It is evident that the results for six scattering phase matrix independent
elements computed from these two numerical codes are indistinguishable.
The II-TM program was further used to compute randomly oriented spheroids with an
unprecedentedly large size parameter of 300. In this case, the EBCM-TM code developed
by Mishchenko and Travis (1998) was unable to achieve a convergent solution. Figure 4.18
shows a comparison of the results of three scattering phase matrix elements computed from
the II-TM technique with the results of an improved geometric-optics method (IGOM)
developed by Yang and Liou (1996a), which is primarily applicable to large size parameters
(see Subsection 3.3.3). The comparison depicted in Figure 4.18 illustrates the potential
capability of the II-TM technique for application to large non-spherical particles of various
shapes.
Bi and Yang (2014) further improved the II-TM method for computing the single-
scattering properties of a sample of randomly oriented particles such that the principle
of reciprocity can be applied. In this case, the scattering phase matrix comprises ten
independent parameters [see Eq. (2.4.55)]. To simplify the following presentation, we will
use the notation of van de Hulst (1957) for the presentation of scattering phase matrix
elements in the form
⎡ ⎤
a1 b1 b3 b5
⎢ b1 a2 b4 b6 ⎥
P =⎢
⎣ −b3
⎥. (4.2.18)
−b4 a3 b2 ⎦
b5 b6 −b2 a4
In terms of the circular-polarization representation, the scattering phase matrix for a

macroscopically isotropic and mirror-symmetric medium may be expressed in the form
(Mishchenko et al. 2002)

⎡ ⎤
(a2 + a3 ) − 2ib4 (b1 + b6 ) + i(b2 − b3 ) (b1 − b6 ) − i(b2 + b3 ) a2 − a3
⎢(b + b ) + i(b − b ) a1 + a4 + 2b5 a1 − a4 (b1 + b6 ) − i(b2 − b3 )⎥
⎢ 1 6 2 3 ⎥
P = 1/2 ⎢
c
⎥.
⎣(b1 − b6 ) − i(b2 + b3 ) a 1 − a4 a1 + a4 − 2b5 (b1 − b6 ) + i(b2 + b3 )⎦
a2 − a3 (b1 + b6 ) − i(b2 − b3 ) (b1 − b6 ) + i(b2 + b3 ) (a2 + a3 ) + 2ib4
(4.2.19)
In Eqs. (4.2.18) and (4.2.19), the notations a1−4 and b1−6 should be distinguished from the
expansion coefficients defined in Eqs. (4.2.6) and (4.2.7).
Bi and Yang (2014) further showed that scattering phase matrix elements can be
expressed in the following forms:
∞

a1 (θ ) = α1s d00
s
(cos θ ), (4.2.20a)
s=0
∞
s s
a2 (θ ) = α2 + α3s d22 (cos θ ), (4.2.20b)
s=0
∞
s s
a3 (θ ) = α2 − α3s d2−2 (cos θ ), (4.2.20c)
s=0
∞

a4 (θ ) = α4s d00
s
(cos θ ), (4.2.20d)
s=0
∞

b1 (θ ) = β1s d02
s
(cos θ ), (4.2.20e)
s=0
∞

b2 (θ ) = β2s d02
s
(cos θ ), (4.2.20f)
s=0
∞

b3 (θ ) = β3s d02
s
(cos θ ), (4.2.20g)
s=0
∞

b4 (θ ) = β4s d22
s
(cos θ ), (4.2.20h)
s=0
∞

b5 (θ ) = β5s d00
s
(cos θ ), (4.2.20i)
s=0
∞

b6 (θ ) = β6s d20
s
(cos θ ), (4.2.20j)
s=0
s
where di,j (i, j = 0, 2, −2) denote the amplitude functions as functions of cos θ , and
θ (= ) is the scattering angle. The coefficients α i (i = 1–4) and β j (j = 1–6) are given by
α1 = (g0,0 + g−0,−0 + 2g0,−0 )/2, (4.2.21a)
α2 = Re{g2,2 } + g2,−2 , (4.2.21b)
α3 = Re{g2,2 } − g2,−2 , (4.2.21c)
α4 = (g0,0 + g−0,−0 − 2g0,−0 )/2, (4.2.21d)
β1 = Re{g0,2 } + Re{g−0,2 }, (4.2.21e)
β2 = Im{g0,2 } − Im{g−0,2 }, (4.2.21f)
β3 = −Im{g0,2 } − Im{g−0,2 }, (4.2.21g)
β4 = −Im{g2,2 }, (4.2.21h)
β5 = (g0,0 − g−0,−0 )/2, (4.2.21i)
β6 = Re{g0,2 } − Re{g−0,2 }. (4.2.21j)
The terms gi,j (i, j = 0, −0, 2, −2) are expressed in the forms
∞

n+s
min(n,i)
s
g00 = i1
hsni Cn1s0 im
Cnms0 00
Dmni , (4.2.22a)
n=1 i=max(1,|n−s|) −min(n,i)
∞

n+s
min(n,i)
s
g0−0 = hsni (−1)n+i+s Cn1s0
i1 im
Cnms0 0−0
Dmni , (4.2.22b)
∞

n+s
min(n,i)
−0−0
s
g−0−0 = i1
hsni Cn1s0 im
Cnms0 Dmni , (4.2.22c)
∞

n+s
min(n,i)
s
g22 = i1
hsni Cn−1s2 i2−m
Cn−ms2 22
Dmni , (4.2.22d)
∞

n+s
min(n,i)
s
g2−2 = hsni (−1)n+i+s Cn−1s2
i1 i2−m
Cn−ms2 2−2
Dmni , (4.2.22e)
∞

n+s
min(n,i)
−2−2
s
g−2−2 = i1
hsni Cn−1s2 i2−m
Cn−ms2 Dmni , (4.2.22f)
∞

n+s
min(n,i)
s
g02 =− i1
hsni Cn1s0 i2−m
Cn−ms2 02
Dmni , (4.2.22g)
∞

n+s
min(n,i)
−02
s
g−02 =− hsni (−1) n+i+s i1
Cn1s0 i2−m
Cn−ms2 Dmni , (4.2.22h)
√
where hsni = (2n + 1)/(2i + 1)(2s + 1)π/(k Csca ), k = 2πλ, and λ is the wavelength,
2
Csca is defined in Eq. (4.2.10), Cnms1−m

i1
denote Clebsch–Gordan coefficients (Mishchenko
et al. 2002). The D-terms are defined as

∞

n1
∗
00
Dmni = (2n1 + 1) B 3 kmnn1 (B 3 kmin1 ) , (4.2.23a)
n1 =|m−1| k=−n1
∞

n1
∗
0−0
Dmni = (2n1 + 1) B 2 kmnn1 (B 2 kmin1 ) , (4.2.23b)
n1 =|m−1| k=−n1
∞

n1
−0−0 ∗
Dmni = (2n1 + 1) B 1 kmnn1 (B 1 kmin1 ) , (4.2.23c)
n1 =|m−1| k=−n1
∞

n1
∗
22
Dmni = (2n1 + 1) B 1 kmnn1 (B 3 −k2−min1 ) , (4.2.23d)
n1 =|m−1| k=−n1
∞

n1
∗
2−2
Dmni = (2n1 + 1) B 4 kmnn1 (B 2 −k2−min1 ) , (4.2.23e)
n1 =|m−1| k=−n1
∞

n1
−2−2 ∗
Dmni = (2n1 + 1) B 3 kmnn1 (B 1 −k2−min1 ) , (4.2.23f)
n1 =|m−1| k=−n1
∞

n1
∗
02
Dmni = (2n1 + 1) B 2 kmnn1 (B 3 −k2−min1 ) , (4.2.23g)
n1 =|m−1| k=−n1
∞

n1
−02 ∗
Dmni = (2n1 + 1) B 1 kmnn1 (B 4 −k2−min1 ) . (4.2.23h)
n1 =|m−1| k=−n1
j
The terms Bkmnn1 (j = 1, 2, 3, 4) can be expressed in the forms
j
1
n+n

n ,1 j
Bkmnn1 = Cn,m,n1 ,1−m
Aknn n1 , j = 1, 2, 3, 4, (4.2.24)
n =max(1,|n−n1 |)
where
j

i n −n n
n ,m1 +k j
Aknn n1 = √ Cn,m T
1 ,n1 ,k m1 ,n,m1 +k,n
, (4.2.25a)

2n + 1 m1 =−n
Tm11 ,n,m1 +k,n = Tm111 ,n,m1 +k,n + Tm121 ,n,m1 +k,n + Tm211 ,n,m1 +k,n + Tm221 ,n,m1 +k,n , (4.2.25b)
Tm21 ,n,m1 +k,n = Tm111 ,n,m1 +k,n + Tm121 ,n,m1 +k,n − Tm211 ,n,m1 +k,n − Tm221 ,n,m1 +k,n , (4.2.25c)
Tm31 ,n,m1 +k,n = T−m

11
1 ,n,−m1 −k,n
− T−m ,n,−m −k,n − T−m ,n,−m −k,n + T−m ,n,m −k,n ,
12
1 1
21
1 1
22
1 1
(4.2.25d)
Tm41 ,n,m1 +k,n = 11
T−m 1 ,n,−m1 −k,n
− 12
T−m 1 ,n,−m1 −k,n
+ 21
T−m 1 ,n,−m1 −k,n
− 22
T−m 1 ,n,m1 −k,n
.
(4.2.25e)
(a) (b)
3 4
10 10
II-TM II-TM
DDA IGOM
3 D
10
l = 0.66 μm
2
10
-8 L
m = 1.3078 - i 1.66 x 10
Phase Function
2πL /l = 2πD/l = 30 2
10 2πL/l = 2πD/l = 150
1
10
1
10
0
10
0
10
-1 -1
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.19 Comparison of phase functions computed from the II-TM and DDA methods for (a)
a short column with a size parameter of 30, and (b) the II-TM and IGOM methods for a large size
parameter of 150. Both calculations use a wavelength of 0.66 µm and a refractive index of m =
1.3078 − i 1.66 × 10−8 (after Bi and Yang 2014).
s s s s
In Eq. (4.2.22a)–(4.2.22h), the terms g00 , g0,−0 , g−0,−0 , and g2,−2 are generally real numbers,
s s s s
while the terms g2,2 , g−2,−2 , g0,2 , and g−0,2 are generally complex numbers.
In Figure 4.18, we illustrated the applicability of the II-TM method to the computation
of scattering phase matrix elements for a large size parameter of 300. To check the II-TM
method’s numerical implementation, we have compared the phase function computed from
II-TM and from DDA (Purcell and Pennypacker 1973, Draine and Flatau 1994, Yurkin
et al. 2007; also see the paragraphs below for further discussion) for a short column with
a size parameter of 30 using a wavelength of 0.66 µm, as shown in Figure 4.19a. The
phase functions computed from II-TM and a volume-integral based IGOM (Yang and Liou
1997, Bi et al. 2011b) for a short column with a size parameter of 150 are compared in
Figure 4.19b. It is evident that the II-TM result agrees well with the DDA counterpart for
a size parameter of 30, and with IGOM for a large size parameter of 150. In the latter case,
geometric-optics features of 22° and 46° halos and a maximum near the scattering angle of
150° are clearly produced by both approaches.
For randomly oriented particles without a plane of symmetry, 16 scattering phase matrix
elements contain ten independent parameters, as indicated previously. Shown in Figure 4.20
are ten scattering phase matrix elements of an aggregate of eight hexagonal columns with
different sizes and aspect ratios. Although each hexagonal column has a plane symmetry,
II-TM ADDA
3
10 1 0.01
0.8 P22/P11 0.005
P
13
/P
11
2
10 0.6
Phase Function
l = 0.66 μm 0
0.4
10
1
D = 7 μm 0.2 -0.005
1 0.02
10
0 P /P P /P
0.5 33 11 14 11
0.01
-1 0
10
0
-0.5
-2 -0.8
10 -0.01
0.15 1
P /P P /P
44 11 0.01 23 11
0.1 0.5
0
0.05
0
-0.01
11
0.0
/P
-0.4
0.2
12
-0.05 P /P P /P
P
0.1 43 11 0.01 24 11
-0.1
0 0
-0.15 -0.1 -0.01
-0.2 -0.2 -0.02
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Figure 4.20 Comparison of ten independent elements of the scattering phase matrix for an aggregate
of eight columns, computed from the II-TM and ADDA methods (after Bi and Yang 2014).
aggregating processes lead to the destruction of symmetry, resulting in non-zero values

for P13 , P14 , P23 , and P24 (the third column in Figure 4.20). We also note that P13 P11 ,
P14 P11 , P23 P11 , and P24 P11 are relatively small. Based on a comparison between II-TM
and ADDA results in Figure 4.20, we find that the agreement between the two is much
closer for P11 , P22 , P33 , and P44 due to their relatively larger values than for the other six
non-zero elements.
The II-TM method can effectively apply to arbitrarily shaped non-spherical particles,
so particle shapes can be incorporated into the computational program without introducing
cumbersome procedures. For example, Figure 4.21 compares phase functions computed
from II-TM and ADDA for six ice crystal habits: droxtals, bullet rosettes, aggregates of
eight columns, hollow columns, hollow bullet rosettes, and aggregates of ten plates. The
wavelength of the incident light used in the calculations is 0.66 µm and the maximum
dimension of the ice crystals is 3 µm, resulting in a size parameter of 28.56.
Using the II-TM method, accuracies of the single-scattering properties of hexagonal
ice crystals with different sizes and aspect ratios can be investigated. The upper panels of
2 3 2
10 10 10
2
1 10 1
10 10
Phase Function
1
10
0 0
10 10
0
10
-1 -1
10 -1 10
ADDA 10
II-TM
-2 -2 -2
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
2 2 2
10 10 10
1 1
10 10
1
Phase Function
10
0 0
10 10
0
10
-1 -1
10 10
-2 -2 -1
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Figure 4.21 Comparison of the phase functions of six ice habits (droxtal, solid bullet rosette, aggre-
gate of eight columns, hollow column, hollow bullet rosette, and aggregate of ten plates) computed by
the ADDA and II-TM methods. The wavelength used was 0.66 µm and the particle size was defined
in terms of the maximum dimension of 3 µm (after Bi and Yang 2014).
Figure 4.22 shows the phase functions for hexagonal ice crystals with different sizes and
aspect ratios. For clarity, the phase functions of ice crystal of sizes 5 and 10 µm are shifted
by a factor of 10. To show the enhancement of ice halos when the ice crystal increases in
size, we define an indicator such that ξ = P11 (22◦ )/Pref (22◦ ), where the reference value
Pref (22◦ ) is calculated from interpolation of the phase function at scattering angles of 15°
and 40°. When the crystal size is small, the indicator fluctuates near unity, produced by
Fraunhofer diffraction. When the crystal size becomes larger than some threshold value,
the indicator is always larger than unity and increases with crystal size, indicating the
enhancement of ice halos associated with the refraction of light. As shown, halos can be
more easily observed for short hexagonal ice particles (DL = 1.0).
The II-TM approach can be employed to accurately model lidar polarimetric measure-
ments. As an example, Figure 4.23 shows the depolarization ratio of a linearly polarized
6 6 6
10 10 10
D = 5 μm
D = 10 μm
D = 20 μm D
4 4 4
10 10 L 10
D/L = 0.5 D/L = 1.0 D/L = 2.0

P11
2 2 2
10 10 10
0 0 0
10 10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180

4 4 4
3.5 3.5 3.5

P11(22o)/<P11,ref>
3 3 3
2.5 2.5 2.5
2 2 2
1.5 1.5 1.5
1 1 1
0.5 0.5 0.5
0 0 0
0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20
Size (μm) Size (μm) Size (μm)
Figure 4.22 Phase functions of hexagonal ice crystals with various sizes and aspect ratios (upper
panels) as a function of the scattering angle. The lower panels depict the enhancement of the 22° halo
using an indicator as a function of size, as discussed in the text.
light beam backscattered from randomly oriented hexagonal ice crystals with particle sizes
ranging from 0.1 to 20 µm. Three particle aspect ratios were selected for the simulations in
order to show the shape dependence of the depolarization ratios. The particles are assumed
to be randomly oriented in space. The incident wavelength was 0.532 µm, at which the
refractive index of ice is 1.3116 − i 1.48 × 10−9 (Warren and Brandt 2008). The depolar-
ization ratio was found to increase to a maximum, then decrease and become more stable
as the crystal size increased.
0.8
D/L = 0.5
D
0.6 1.0 L
Depolarization Ratio
2.0
0.4
0.2
0
0 5 10 15 20
Ice Crystal Size (μm)
Figure 4.23 Linear depolarization ratio of a hexagonal ice crystal with three different aspect ratios
at 0.532 µm, as a function of ice crystal size.
Figure 4.24 shows the asymmetry factor, the extinction efficiency, and the single-
scattering albedo computed from II-TM for randomly oriented hexagonal ice crystals with
size parameters (k1 L) ranging from 1 to 150. In the simulation, the aspect ratio of the hexag-
onal particle was assumed to be unity (i.e., L = 2a). As the size parameter increases, differ-
ences between the asymmetry factor computed from II-TM and from IGOM are smaller,
as IGOM becomes increasingly accurate. Because ice crystals are almost non-absorptive
at 0.66 µm, the interference between diffraction and forward scattering causes oscillation
of the extinction efficiency with respect to ice crystal size parameter. The behavior of the
extinction efficiency in terms of its asymptotic value and oscillation pattern computed from
IGOM was similar to T-matrix results. For absorptive particles, when the ice crystal size
parameter is larger than a certain value, internal reflection of geometric-optics waves can be
properly neglected. In this case, the contribution to scattering was essentially from external
reflection and diffraction. The asymptotic extinction efficiency can be explicitly proven to
be 2 in the framework of geometric optics (equal partition of geometric reflection/refraction
and diffraction; see Subsection 3.3.2).
To illustrate applicability of the II-TM method to extremely complicated particle geome-
tries, Figure 4.25 shows solutions for the phase function and the degree of linear polarization
of two types of oceanic particles, coccolith and coccolithophore, which are important in
the discussion of the global carbon cycle (Bi and Yang 2015). For comparison, the DDA
solutions are also shown. The size parameter defined in terms of the radius of a volume-
equivalent sphere is 7.23 for a coccolith and 34.25 for the coccolithophore. The refractive
index of coccolith relative to water is 1.20 and the relative refractive index of the spherical
1 4
II-TM 1
0.8 IGOM
3 0.8
0.6 a 0.6
2
L
0.4 l = 0.66 μm 0.4
L = 2a 1
0.2 0.2
0 0
0
1 10 100 500 1 10 100 500
Single-scattering Albedo
1 4 0.8
Extinction Efficiency
Asymmetry Factor
0.8
3 0.6
0.6 l = 8.50 μm
2 0.4
0.4
1 0.2
0.2
0
0 0
1 10 100 500 1 10 100 500
1 2 1
0.8 0.8
0.6 l = 11.00 μm 0.6

1
0.4 0.4
0.2 0.2
0
0 01
1 10 100 500 10 100 500
Size Parameter k1L Size Parameter k1L
Figure 4.24 The asymmetry factor, the extinction efficiency, and the single-scattering albedo for a
short column as a function of size parameter, computed using the II-TM and IGOM methods for
wavelengths of 0.66, 8.50, and 11 µm (after Bi and Yang 2014).
coccolithophore cell is 1.03 − i 0.005. As shown in Figure 4.25, the results from two fun-
damentally different numerical methods agree closely with differences produced by shape
definition and orientation-average process of the DDA method. In the DDA orientation-
average procedure, 64 α, 65 β, and 32 γ angle values were used, where α, β, and γ are the
three Euler angles defined according to the x-y-z convention.
4.3 Discrete dipole approximation

4.3.1 Fundamentals of the DDA approach
The discrete dipole approximation (DDA) is a numerical technique for the calculation of
scattering and absorption properties of arbitrarily shaped particles. A particle is modeled
by an array of N-point dipoles at positions ri with polarizabilities α i . The method has been
widely employed for studying the optical properties of various particles, particularly those
103 10
4
xv = 7.23 DDA xv = 34.25

II-TM
102
2
10
Phase Function
1
10
0
10
0
10
-2
-1
10
10
-2 -4
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
1 1
P12/P11
0 0
-1 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 4.25 Comparison of phase function and P12 P11 for a sample of randomly oriented coccolith
(left column) and coccolithophore (right column) as a function of the scattering angle, computed
using the DDA and II-TM methods (courtesy of Lei Bi).
lacking rotational symmetry, having inhomogeneous chemical composition, or composed

of anisotropic material where an analytical or semi-analytical solution does not exist.
The DDA was originally introduced by Purcell and Pennypacker (1973) to study the
optical properties of interstellar dust grains. The basic idea of DDA appears to have been
developed by DeVoe (1964), who studied the optical properties of molecular aggregates.
Simply stated, the particle in this method is assumed to be a finite array of polarization
points (see the example of a hexagonal particle shown in Figure 4.26a), with each point
responding to the local electric field, the sum of the incident field and the field arising from
the other dipoles, as shown in Figure 4.26b. As a result of this arrangement, the amplitudes
of all dipole moments are self-consistent.
The electric dipole moment is proportional to the total electric field E(r) and the local
exciting field Eexc (r), defined as the superposition of the incident field and the fields
produced by other dipoles in the form
P(r) = χ (r)E(r) = αEexc (r), (4.3.1)

(a) (b)
k E
0,k
j E
0, j
i E
0,i
Figure 4.26 (a) A schematic geometry to illustrate the approximation of a hexagonal aggregate in
terms of discretized dipoles. (b) An example of a three-dipole system (red dots) define by i, j, k indices
and associated incident electric vectors.
where the susceptibility χ (r) is given by
m2 − 1
χ (r) = d 3 (4.3.2)
4π
and the polarizability α is related to the refractive index m through the Clausius–Mossotti
(or Lorentz–Lorenz) relation (Lorentz 1880, Lorenz 1880) as follows:
3 m2 − 1
α = d3 , (4.3.3)
4π m2 + 2
where d is the dipole length. The electric dipole field at the position of the jth dipole,
radiated by the ith electric dipole moment Pi , is given by (Jackson 1975):
% & % &
k 2 exp(ikR) R̂ R̂ 1 − ikR 3R̂ R̂
E(R = rj − ri ) = Aij · Pi = I− 2 − I− · Pi ,
R R (kR)2 (kR)2
(4.3.4)
where k is the wavenumber. It follows that the fundamental equation for the induced dipoles
is given by
% &

Pi = αi E0,i + Aij · Pj , (4.3.5)
i
=j
where E0,i is the incident field and Aij is defined in Eq. (4.3.4).
Once the amplitudes of dipole moments in Eq. (4.3.5) are solved, it can be shown
that the extinction and absorption cross-sections are given by the following equations
[see Eqs. (4.1.38) and (4.1.39)]:

4πk ∗
N
k ∗ 3
σext = Im (ε − 1) E0 (r ) · E(r ) d r = Im E0,j · Pj , (4.3.6a)
|E0 |2 V |E0 | j =1
2

4πk ∗
N
k
σabs = Im(ε)E∗ (r ) · E(r )d 3 r = Im Ej · Pj , (4.3.6b)
|E0 |2 V |E0 | j =1
2
where ε is the permittivity and Eq. (4.3.1) is used in these derivations. By neglecting higher-
order terms in Eq. (4.3.1) or Eq. (4.3.4), the scattered far field can be determined and is
given by
exp(ikr)
Esca (r) |r→∞ = (−ik 3 )(Ī − r̂ r̂) · Pj exp(ikrj · r̂). (4.3.7)
−ikr j
To compute the scattering amplitude matrix and subsequently the phase matrix, the
scattered field must be decomposed into components parallel and perpendicular to the
scattering plane:
Esca (r) = α̂Eα + β̂Eβ . (4.3.8)
In matrix form, we have

Eα exp(ikr) E0,α
= S , (4.3.9)
Eβ −ikr E0,β
where

fx,α fy,α β̂ · êx β̂ · êy
S= , (4.3.10)
fx,β fy,β−β̂ · êy β̂ · êx

Pj · α̂
fy,α
= −ik 3
exp(ikrj · r̂)|E0,x =0,E0,y =1 , (4.3.11)
fy,β Pj · β̂
j =1

Pj · α
fx,α
= −ik 3
exp(ikrj · r̂)|E0,x =1,E0,y =0 . (4.3.12)
fx,β Pj · β
j =1
From Eqs. (4.3.11) and (4.3.12), we note that Eq. (4.3.1) must be solved twice, correspond-
ing to two cases: (a) E0,y = 1, E0,x = 0 and (b) E0,y = 0, E0,x = 1.
Equation (4.3.2), which was established by Purcell and Pennypacker (1973), is to a
certain degree based on physical rationale. In later developments, DDA was rigorously
refined on the basis of mathematical analysis starting from an electric volume integral
equation in the form

E(r) = E0 (r) + k 2 (ε − 1)Ḡ(r − r ) · E(r)dv, (4.3.13)
V
where E0 is the incident field and Ḡ is the dyadic Green function satisfying the following
differential equation:
∇ × ∇ × G(r − r ) − k 2 G(r − r ) = Īδ(r − r ), (4.3.14)
where δ(r − r ) is the Dirac delta function. Direct computation of the integral in Eq. (4.3.13)
is problematic because of the inherent singularity problem associated with the dyadic Green
function, given by
1 exp(ikR) Ā(R = r − r )
Ḡ(R = r − r ) = [k 2
Ī + ∇∇] = , (4.3.15)
4πk 2 R 4πk 2
which contains a 1/R 3 dependence term in the form

1 Ī 3
Ḡs (R = r − r ) = − + R̂ R̂ . (4.3.16)
4πk 2 R3 R3
To treat the singularity problem for appropriate numerical computations, the volumetric
equation for R → 0 can be transformed to a surface integration equation. Based on the
assumption that the electric field is constant within a small volume, Eq. (4.3.13) can then
be rewritten for numerical calculations as follows:

E(r) ≈ E0 (r) + k 2 (ε − 1)Ḡ(r − r ) · E(r )dv + (ε − 1)[M̄(r) + L̄(r)] · E(r),
V −Vε
(4.3.17)
where Vε is an elementary volume and M̄(r) and L̄(r) are integrals defined by
⎧ ⎫
⎨ ⎬
M̄(r) = k 2 [Ḡ(r − r ) − Ḡs (r − r )]d 3 r , (4.3.18)
⎩ ⎭
Vε
⎧ ⎫
⎨ ⎬
L̄(r) = k 2 Ḡs (r − r )d 3 r . (4.3.19)
⎩ ⎭
Vε
From the divergence theorem (Arfken 1985), Eq. (4.3.19) is equivalent to

¯
1 I 3 1 R̂ 1 n̂R̂
L̄ = − + R̂ R̂ dvε = − ∇ dvε = − ds. (4.3.20)
4π R3 R3 4π R2 4π s R 2
Vε Vε
If the differential volume is a cube or sphere, it can be proven that L̄ = −I/3 (Yaghjian
1980). Based on Eq. (4.3.17), the source of the field (or the exciting field) at the position
r, i.e., the contribution from the incident field and the scattered field from the remaining
volume, is given by
Eexc (r) = [Ī − (ε − 1)(M̄ + L̄)] · E(r). (4.3.21)
With Eq. (4.3.1), the polarizability can now be defined in the form
V (ε − 1)
α= . (4.3.22)
4π[Ī − (ε − 1)(M̄ + L̄)]
Strong and weak forms of DDA have been proposed by Lakhtakia (1992) to distinguish
the treatment of the singularity problem associated with the dyadic Green function depicted
in Eq. (4.3.15). The weak-form DDA corresponds to the Clausius–Mossotti relation (i.e.,
M̄ = 0̄), while the strong-form DDA indicates non-zero M̄ stemming from the finite ele-
mentary volume used in the calculation. In the case of the strong-form DDA, several
polarizability relations (see Yurkin et al. 2007) are formulated instead of the well-known
Clausius–Mossotti polarizability based on different approximations to calculate M̄. A few
polarizability relations based on different mathematical analysis have also been discussed
by Yurkin et al. (2007). A number of DDA implementations have recently been presented
for the solution of light scattering by dielectric particles, such as the DDSCAT code devel-
oped by Draine (1988) and Draine and Flatau (1994) and the ADDA method developed by
Yurkin and Hoekstra (2011).
An apparent advantage of DDA is that it is flexible in handling the complex geometry of
a particle because it is not necessary to impose electromagnetic boundary conditions at the
particle surface. The major limitation is associated with size parameter, which determines
the number of dipoles and the associated computer memory requirement. With state-of-
the-art computers, DDA can efficiently obtain scattering and absorption properties when
particle size parameters are smaller than 20. For large size parameters, computational
resources such as memory and the number of processors increase dramatically with particle
size parameter. However, with the development of computer clusters, the light scattering
properties of large size parameters can be obtained by the use of parallel computation. For
example, the single-scattering properties of a sphere with a size parameter of 320 and a
refractive index of 1.05 have been obtained using DDA with 512 processors and 698 Gb of
memory (Yurkin and Hoekstra 2011).
Due to its inherent errors in defining shape and the nature of approximating polarizability
relations, DDA can be viewed as an approximate or “semi-exact” method. The accuracy of
DDA has been examined by comparison of its results with those computed from analytical or
semi-analytical methods such as the Lorenz–Mie theory for spheres and the EBCM numer-
ical method for spheroids and circular cylinders. Figure 4.27 compares the phase function
of a sphere computed from the Lorenz–Mie theory and from DDA in which the ADDA
code was employed in the calculation. Comparisons of the phase functions of spheroids and
cylinders computed by T-matrix and DDA methods are also given in this figure. The size
parameter, defined in terms of the semi-maximum dimension, was 20, while the refractive
index used was 1.33. Relative errors are displayed in the panels below the corresponding
phase functions to illustrate the accuracy of DDA results.
4 2 3
10 10 10
Lorenz-Mie T-matrix T-matrix
ADDA ADDA ADDA
2 1 2
Phase Function
10 20 10 20 10
20
10 10
0 0 1
10 10 10
-2 -1 0
10 10 10
-4 -2 -1
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
0.5 0.15 0.15
Relative Error
0.1 0.1
0.25
0.05 0.05
0
0.0 0.0
-0.25 -0.05 -0.05

0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Figure 4.27 Comparison of the phase function of a sphere computed from the Lorenz–Mie theory
and from ADDA, and of the phase functions of a spheroid and a circular cylinder computed from
T-matrix and from ADDA. In ADDA simulations, the number of dipoles per wavelength used was
13.8203.
4.3.2 Comparison among DDA, FDTD, and PSTD

We may carry out a comparison between the performances of DDA and FDTD because
these two numerical methods have similar applicability with reference to DDA in terms of
shape-independence and comparable computer memory requirements. Such a comparison
study can provide the best approach in selecting the optimized numerical method for
practical calculations. Yurkin et al. (2007) reported a comparison between DDA and FDTD
for the case of spheres with indices of refraction ranging from 1.02 to 2 and size parameters
up to 80. The comparison exercise revealed that DDA appears to be faster than FDTD when
the refractive index was smaller than 1.4.
The PSTD technique, discussed in Subsection 4.1.6, is more advantageous than FDTD
because the coarse spatial resolution can be used to discretize particle domain. Liu et al.
(2012a) reported a comparison between the numerical performance of the DDA and PSTD
techniques and found DDA to be faster than PSTD when the indices of refraction were
smaller than a certain value for a fixed size parameter (Figure 4.28). As shown, PSTD is
more efficient for particles with large indices of refraction, whereas DDA outperforms for
cases involving small indices of refraction. Furthermore, we notice that in the literature
DDA has seldom been employed in calculations when the refractive index is larger than
2 because of slow convergence or failure to converge. Some progress in extending the
application of DDA to indices of refraction larger than 2 has been made by Yurkin et al.
(2010). However, the FDTD/PSTD method has no limitation concerning the use of the
2.0
1.8
PSTD
Refractive Index m
1.6
1.4
DDA
1.2
0 20 40 60 80 100
Size Parameter x
Figure 4.28 The relative performance of PSTD and DDA on the (x, m) domain. The green area
indicates that DDA is computationally more efficient than PSTD, whereas the blue region shows that
PSTD outperforms DDA (after Liu et al. 2012a).
refractive index in numerical calculations, although a fine grid resolution is required for
large indices of refraction.
Examples of DDA computations for the six phase matrix elements of hexagonal particles
are presented in Figure 4.29. The refractive index used was 1.33, along with a size parameter
of 50 (defined in terms of height), while the aspect ratio was kept at unity. We have compared
DDA computations with their counterparts simulated by the PSTD method and the efficient
geometric-optics hybrid (EGOH) method developed by Bi et al. (2011b), which is an
improved form of the ray-by-ray integration (RBRI) technique (Yang and Liou 1997).
In this method, beams with polygon-shaped cross-sections are traced within a scattering
particle, while the near field associated with the beams is analytically mapped to their
far-field counterparts based on a rigorous volume-integral electromagnetic relation. It is
computationally more efficient than RBRI and can represent a “benchmark” within the
framework of geometric optics. The two numerical methods (DDA and PSTD) produce
very close results. The approximate results determined from EGOH demonstrate similar
results for phase matrix elements as compared with those obtained from DDA and PSTD
numerical methods.
4
10 0.2
ADDA
PSTD 0.1
2 EGOH
10
P12 /P11
0
P11
0 -0.1
10
-0.2
-2
10 -0.3
0 45 90 135 180 0 45 90 135 180
1 1
0.8
0.5
P33 /P11
P22 /P11
0.6
0
0.4
-0.5
0.2
0 -1
0 45 90 135 180 0 45 90 135 180
1
0.4
0.5
P44 /P11
P43 /P11
0.2
0 0
-0.5 -0.2
0 45 90 135 180 0 45 90 135 180
Figure 4.29 Comparison of phase matrix elements for randomly oriented hexagonal ice columns as
a function of the scattering angle, computed using ADDA, PSTD, and EGOH methods.
In summary, various combinations of DDA, FDTD, PSTD, and EGOH methods can be
employed in principle to study the single-scattering properties, particularly the scattering
phase matrix, of non-spherical particles covering the size parameter range from Rayleigh
to geometric-optics domains.
5
Application of light scattering by ice crystals
to remote sensing
Remote sensing is a big interdisciplinary field involving science, engineering, and tech-
nology. However, we shall confine our presentations to the book’s theme, namely light
scattering by ice crystals. The subject of the remote sensing of ice clouds in the Earth’s
atmosphere and associated supporting material will be the focus of this chapter.
Remote sensing is differentiated from in situ measurements in that specific observations
are made outside of the medium. It involves the interpretation and inversion of radiation
signals measured some distance away, which are characterized by specific wavelengths
sensitive to some physical properties of the medium. In the case of ice clouds, the interpre-
tation and inversion processes require the use of fundamental light scattering and radiative
transfer theories. The general principle of the inversion theory is that the pieces of informa-
tion content that can be recovered from a target, such as clouds, cannot be larger than the
number of channels (wavelengths) employed for remote sensing. Simply put, initial efforts
must be greater than expectations.
Remote sensing can use the radiation sources readily available in nature – radiation
emitted from the sun and from the Earth–atmosphere system; this is referred to as passive
remote sensing. Many sensors in current satellites carry spectral radiometers operating
at visible and near-IR wavelengths to detect clouds and aerosols and make inferences
about their physical and radiative properties based on the sunlight reflected from the
cloud and aerosol particles. Because air (N2 and O2 ), trace gases, aerosols, and clouds
coexist in the atmosphere, along with reflecting surfaces, direct inversions from the basic
radiative transfer principle are not feasible and many numerical procedures, referred to as
algorithms, are required for the determination of these properties from suitable radiometers
or spectrometers. Independent validations become essential to the success of space remote
sensing.
Active remote sensing uses radiation generated by artificial means such as lasers and
microwaves, leading to the development of lidar and radar systems. This type of remote
sensing is primarily concerned with backscattering involving a system of collimated trans-
mitter and detector. In recent years, significant advances in the development of high-energy
lasers and mm-wave sources have led to the successful deployment of optical lidar and
94 GHz systems in space. These systems are collecting important information on clouds in
general and high clouds in particular based on the backscattering principle.
247
248 Application of light scattering to remote sensing
In this chapter, we introduce the remote sensing of ice clouds in the Earth’s atmosphere,
including discussions on atmospheric composition and structure, atmospheric absorption
spectra, the sun–satellite geometric configuration and radiative transfer, the bidirectional
reflectance spectra of ice clouds, and the A-Train satellite constellation. From this base, we
then present the subject of retrieving the optical depth and ice crystal size of cirrus clouds
using reflected visible and near-IR radiation, and illustrate the importance of the phase
function (scattering pattern) of ice crystals. A discussion follows on detecting thin cirrus
and vertical sizing of cirrus cloud layers. We next tackle the subjects of remote sensing of
ice clouds using reflected polarization and the principle of backscattering depolarization
based on geometric optics to differentiate non-spherical ice crystals and spherical water
droplets. Lastly, we present reflected line spectra at the 1.38 µm band and the oxygen
A-band for the potential determination of the composition and optical properties of high
clouds.
5.1 Introduction to remote sensing of ice clouds in the Earth’s atmosphere

5.1.1 Atmospheric composition and structure
Before proceeding to a discussion about the remote sensing of ice clouds imbedded in
the Earth’s atmosphere, we shall discuss atmospheric composition and thermal structure.
The present atmosphere of the Earth is composed of two groups of gases, one with nearly
permanent concentrations and the other with variable concentrations. Permanent gases such
as nitrogen molecules (N2 , 78.084%), oxygen molecules (O2 , 20.948%), and argon (Ar,
0.934%) account for more than 99.96% of the atmosphere by volume, and have virtually
constant volume ratios up to an altitude of 60 km.
Another permanent constituent, carbon dioxide (CO2 ), with a concentration of
400 ppm in 2014, has been increasing by about 0.4% per year as a result of fossil
fuel combustion, absorption and release of CO2 by the oceans, and photosynthesis. Obser-
vations of the atmospheric methane (CH4 ) concentration, with a present value of 1.7 ppm,
indicate that this gas is increasing by 1–2% per year and may have been increasing for
a long period of time due to greater biogenic emissions associated with rising human pop-
ulations and rice paddies. Changes in carbon monoxide (CO) concentration appear to be
closely associated with deforestation, biomass burning, and modification of CH4 sources.
There is also evidence of an increase in nitrous oxide (N2 O) concentrations by 0.2% per
year, attributable to fossil fuel combustion and fertilizer denitrification. CO2 , CH4 , and N2 O
have been identified as key greenhouse gases and man-made perturbations dating from the
era of the Industrial Revolution. These gases play an essential role in global warming and
climate change.
Water vapor (H2 O), a highly variable gas, is the major radiative and dynamic element in
the Earth’s atmosphere. Its concentration varies significantly with both space and time and is
governed by the local hydrological cycle via evaporation, condensation, and precipitation,
as well as by large-scale transport processes. H2 O concentrations decrease rapidly with
5.1 Remote sensing of ice clouds in the atmosphere 249
pressure and, to a lesser degree, with latitude. More than 50% of H2 O is concentrated below
850 hPa, while more than 90% is confined to the layers below 500 hPa. Variability
of the vertical H2 O concentration shows a bimodal distribution with a maximum in the
subtropics of both hemispheres below 700 hPa, and is very small in the equatorial region
and poleward of 60°. The H2 O concentration is relatively small, 3–4 ppm, in the lower
stratosphere, and is controlled by the tropical tropopause temperature and the formation
and dissipation of cirrus anvils due to outflows from cumulonimbus clouds in the tropics.
Ozone (O3 ) concentrations also vary significantly with space and time, and are detected
principally at altitudes from 15 to 30 km in an area known as the ozone layer. The
maximum ozone concentration occurs at 20–25 km, depending on latitude and season.
O3 is created by three-body collisions involving O2 , O, and a third body (N2 or O2 ) and
governed by numerous photochemical processes associated with solar ultraviolet radiation.
The absorption of deadly solar ultraviolet radiation by the ozone layer is essential to the
survival of life on Earth. Many photochemical reactions associated with O3 involve H2 O,
CH4 , and CO. The total O3 concentration is a function of latitude and season, with maximum
concentrations occurring during the polar night.
Nitrogen oxides (NOx ; NO or NO2 ) are important in the determination of both tropo-
spheric and stratospheric O3 concentrations. Atmospheric NOx are emitted by transportation
and combustion processes at the surface and by high-flying aircraft in the upper troposphere
and lower stratosphere. In the stratosphere, the major source of NOx is the dissociation
of N2 O via an excited oxygen atom. In the lower atmosphere, the major source of NOx
is the anthropogenic combustion of fossil fuels and biomass burning. Chlorofluorocarbons
(CFCs) have also been recognized as presenting a potential threat to the ozone layer. Large
amounts of these chemicals are produced by industry and are used in solvents, refrigerants,
and spray-can propellants. Principal chlorofluorocarbons include CFCl3 (CFC-11), CF2 Cl2
(CFC-12), and CF3 Cl (CFC-13). Sulfur dioxide (SO2 ) in the stratosphere is largely pro-
duced by volcanic eruptions; this along with other sulfur-based gases are primary precursors
of stratospheric aerosols. Emissions of SO2 from the surface are important in the formation
of tropospheric aerosols.
The atmosphere also contains various kinds of aerosols, clouds, and precipitation, which
are highly variable in space and time. In Sections 1.1 and 1.2, we introduced the vertical
and horizontal structures of clouds in general and cirrus clouds in particular. Size and shape
distributions of ice crystals were also presented. Fortunate clouds can produce precipitation
in the form of raindrops, various types of snowflakes, and hailstones with sizes on the
order of mm or cm. We further pointed out that the key to trigger the formation of clouds
in moist atmospheres is the existence of condensation and ice nuclei, which are a part
of ubiquitous aerosol particles ranging in size from 10−3 to 20 µm. These aerosols
are known to be produced by natural processes as well as by human activity. Natural
aerosols include volcanic dust, smoke from forest fires, particles from sea spray, wind-
blown dust, and small particles produced by the chemical reactions of natural gases. Primary
man-made aerosols include particles directly emitted during combustion processes and
particles formed from gases emitted during combustion. Aerosol concentration varies with
locality; the largest concentrations generally occur in urban and desert areas. Under normal
conditions, the background aerosol concentration gives rise to a visibility of 20–50 km.
Aerosol concentrations generally decrease rapidly with height in the troposphere. Thin
layers of aerosols have been observed to persist for a long period at some altitudes of the
stratosphere.
Remote sensing of the atmosphere and surfaces using infrared and microwave tech-
nologies requires an understanding of temperature profiles. The commonly used standard
vertical temperature profile represents a typical condition in midlatitude regions. The tro-
posphere is characterized by a decrease from a mean surface temperature of 288 K to
220 K with respect to height with a typical lapse rate of 6.5 K km−1 . The temperature
structure in this layer is a consequence of radiative balance and convective energy transport
from the surface to the atmosphere. Virtually all water vapor, clouds, and precipitation
are confined to this layer. Moreover, the lowest 1 km or so of the atmosphere, referred
to as the planetary boundary layer (PBL), differs from the remaining troposphere in that
interactions with the surface are strong and significant. The stratosphere is characterized
by an isothermal layer from the tropopause to about 20 km, above which the temperature
increases to the stratopause, with a value of 270 K. The state of the stratosphere is primar-
ily determined by the absorption of solar fluxes by O3 and through the emission of infrared
fluxes by CO2 . As in the troposphere, temperatures in the mesosphere decrease with height
from about 50 to 85 km. Above this height and extending upward to an altitude of several
hundred kilometers lies the thermosphere, where temperatures range from 500 K to as high
as 2000 K, depending on the level of solar activity.
5.1.2 Atmospheric absorption spectrum

The upper panel of Figure 5.1 depicts solar (or shortwave) and thermal infrared (IR or
longwave) spectral irradiances at the top of the Earth’s atmosphere (TOA) as a function
of wavelength and wavenumber. The solar spectrum (red curve) is computed from Planck
emission associated with the sun’s photospheric temperature of 5754 K and attenuated to
TOA corresponding to a solar constant of 1366 W m−2 , while the red area represents the
observed solar fluxes available at the surface. The IR spectrum (blue curve) is computed
from the global equilibrium temperature of the Earth–atmosphere system of 254 K, the
peak of which is normalized with reference to the solar spectral peak, while the blue
area denotes the IR fluxes absorbed in the atmosphere and available at TOA. The solar and
thermal IR spectra cover wavelength ranges from 0.1 to 5 and 5 to 100 µm, respectively,
with a slight overlap between 3 and 5 µm. For this reason, a correction must be made
if a thermal IR technique is used for space remote sensing of the atmosphere and the
surface.
The middle panel of Figure 5.1 shows absorption by H2 O produced by vibration–rotation
transitions located at 0.72, 0.82, 0.94, 1.1, 1.38, 1.87, 2.7, and 3.2 µm in the solar spectrum
and at 6.3 µm in the thermal IR spectrum. H2 O rotational lines cover essentially the entire
solar and thermal IR spectra, but these absorption lines are strongest at wavelengths longer
10 -1 10 0 10 1 10 2
SpectralIrradiance
Radiance
10−1 10 0
Solar Radiation (shortwave) 1
Thermal Infrared10Radiation (longwave) 10 2
Sun, 5754K (1366 W/m2) at TOA Earth, 254K
Spectral
UV Visible Infrared
100
Major Components
Percent
75
50
25
0
H2 O
Major Components
CO2
O2, O3
Percent
CH4
Absorption
and Scattering N 2O
Rayleigh Scattering
−1
10-1
10 10 0 101 1022
10
Wavelength (μm)
10 5 10 4 10 3 10 2
Wavenumber (cm −1)
Figure 5.1 Top panel: Spectral irradiances for the solar (red) and thermal infrared (blue, scaled with
reference to solar) regions as a function of wavelength and wavenumber. The red area represents the
observed solar fluxes available at the surface, while the blue area denotes the IR fluxes absorbed in the
atmosphere and available at TOA. Middle panel: Absorption by H2 O, CO2 , O2 , O3 , CH4 , and N2 O,
and Rayleigh scattering. Bottom panel: Total atmospheric absorption in terms of the percentage of
the entire atmosphere corresponding to the U.S. 1976 Standard Atmosphere; 100% and 0% indicate,
respectively, that the atmosphere is opaque to radiation and is clear (total transmission) to radiation.
than 12 µm, known as the rotational band of water vapor. Present satellites do not have the
capability to measure this band, which produces substantial cooling rates in the atmosphere.
Absorption due to CO2 is largely produced by vibrational transitions with spectral bands
located at 1.6, 2.0, 2.7, and 4.3 µm in the solar spectrum. Weak 1.6, 2.0, and 2.7 µm com-
bination bands have been used to retrieve CO2 concentrations by the OCO2 satellite. The
strong CO2 15 µm band in the thermal IR spectrum is produced by two modes of perpen-
dicular vibration associated with the ν 2 fundamental, combination bands, and hot bands; it
is most pronounced in trapping the emission from the surface and lower atmospheres. This
band is the source of about half of the radiative forcing due to anthropogenic greenhouse
warming since the era of the Industrial Revolution. Additionally, this band has been used
to retrieve temperature profiles on a routine basis.
Absorption of O2 and O3 in the solar spectral region is produced by electronic–vibrational
transitions chiefly occurring in the ultraviolet and visible regions. For O2 , absorption occurs
Figure 5.2 Atmospheric transmission as a function of wavelength, wavenumber, and frequency

covering the ranges 10−1 –105 µm, 105 –10−1 cm−1 , and 105 –101 GHz, respectively. In the solar and
thermal infrared regions, these spectra depict the upside-down pattern presented in Figure 5.1.
primarily at wavelengths shorter than 0.2 µm. The O2 0.762 µm A-band has the imprint of
the upper troposphere and high-level clouds. In the solar spectrum, O3 displays absorption
at wavelengths shorter than 0.3 µm. O3 also displays absorption at the 9.6 µm vibration–
rotation band. A number of weaker absorption bands of CH4 and N2 O appear in the solar
spectrum. Both molecules display strong vibration–rotation lines in the vicinity of 7.7 µm.
In addition to CO2 , O3 , CH4 , and N2 O are also known as greenhouse gases. Lastly, at
wavelengths less than 0.7 µm, attenuation due to molecular (Rayleigh) scattering becomes
significant.
The bottom panel of Figure 5.1 illustrates the total atmospheric absorption produced by
the various molecules discussed above in terms of the percentage of the entire atmosphere
corresponding to the U.S. 1976 Standard Atmosphere; 100% and 0% indicate, respectively,
that the gas is opaque to radiation and is clear (total transmission) to radiation. This
information is essential to the development of space remote sensing of the atmosphere,
oceans, and land surfaces, employing UV, visible, and IR techniques. Note that Rayleigh
scattering occurs primarily in the UV and visible regions.
Figure 5.2 illustrates atmospheric transmission, the inverse of atmospheric absorption
presented in the bottom panel of Figure 5.1. When scattering can be neglected, (percentage
transmission) = 100% – (percentage absorption). However, this figure depicts additional
information on microwave transmission as a function of frequency in GHz in addition to
wavelength (λ) and wavenumber (ν), where ν (in cm−1 ) = 1λ (in µm) and 30 GHz is
equivalent to 1 cm wavelength.
In the microwave region, only H2 O and O2 exhibit significant absorption lines. Below
40 GHz, only the weakly absorbing pressure-broadened 22.235 GHz H2 O line is dominant.
Out
In
π-Θ
q q0 H2O, O3, CO2, etc.
Aerosols
Surface (land, ocean)
f f0
Δf
Figure 5.3 Geometry governing bidirectional reflectance (reflection function) involving the sun, a
satellite, and an ice cloud. “In” and “Out” denote the incoming (θ 0 , φ 0 ) and outgoing (θ, φ) directions,
respectively. The 3-D angle between the sun and the satellite is defined by π − , where is the
scattering angle. In addition to clouds, the atmosphere contains absorbing molecules and aerosols,
while the surface has important effects on satellite remote sensing of clouds and aerosols.
This resonance absorption line is produced by transitions between nuclear spin rotational
states. At 31.4 GHz, air is relatively transparent, the window between resonance water
vapor lines. O2 has a magnetic dipole moment arising from the combined spins of two
unpaired electrons in their electronic ground state. Changes in electronic spin orientation
relative to molecular rotation orientation produce a strong band of magnetic dipole tran-
sitions near 60 GHz, which has been used for temperature retrieval, particularly in cloudy
conditions, and a single transition at 118.75 GHz. For frequencies larger than 120 GHz,
H2 O absorption again becomes substantial due to the strongly absorbing line located at
183 GHz. Transitions at 35 GHz (<1 cm wavelength) and 94 GHz (close to 0.3 cm wave-
length) have been used for cloud detection. The deployment of an active 94 GHz instrument
on a satellite platform has shown to be extremely successful for profiling clouds in general
and ice clouds in particular.
5.1.3 Relative position of the sun–satellite system and radiative transfer

Consider the sun–satellite system depicted in Figure 5.3, where the solar beam with its
zenith and azimuthal angles are along a meridional plane, referred to as the incoming
direction. The position of the satellite is along another meridional plane, not necessarily
the same as the sun’s position, and is referred to as the outgoing direction. From these four
angles we can define the scattering angle depicted in the figure. In addition to clouds, the
atmosphere contains various absorbing gases and aerosols as discussed in Subsection 5.1.2.
Also, surface albedo must be accounted for in satellite remote sensing of the atmosphere
using solar reflection techniques. In the following paragraphs, we briefly introduce the
radiative transfer required to obtain a solution of the reflected intensity at the satellite
position (TOA) for remote sensing applications.
It suffices to use scalar radiative transfer for illustration purposes, although application to
vector radiative transfer can be followed via Eq. (2.5.16) for polarization analysis. Referring
to Subsection 2.5.1, the basic scalar radiative transfer equation defining the diffuse scattered
intensity I, neglecting emission contributions, can be written as follows:

dI (τ ; μ, φ)
μ = I (τ ; μ, φ) − I (τ, μ , φ ) P (μ, φ; μ , φ )dμ dφ
dτ 4π
4π (5.1.1)
−τ /μ0
− F P (μ, φ; −μ0 , φ0 )e ,
4π
where τ is the optical depth, ϖ is the single-scattering albedo, P is the phase function, F
is the incident solar flux, and other notations have been defined in Subsection 2.5.1.
Furthermore, the non-dimensional reflection function (or bidirectional reflectance dis-
tribution function, BRDF) can be defined as
R(μ, φ; μ0 , φ0 ) = πI (0; μ, φ)/μ0 F , (5.1.2)
where I (0; μ, φ) represents the reflected intensity at TOA as a function of the zenith and
azimuthal angles depicted in Figure 5.3, where the scattering angle is also defined with
reference to the sun–satellite geometric configuration. The solution for I (0; μ, φ) requires a
tedious radiative transfer program, which takes into account interactions of the atmosphere,
clouds, and the surface. Under the condition that the atmosphere and clouds are optically
thin, so that single scattering events predominate, we have the following solution:
τ
R(μ, φ; μ0 , φ0 ) P (μ, φ; −μ0 , φ0 ), (5.1.3)
4μμ0
where −μ0 denotes the cosine of the downward solar beam. This approximation is appli-
cable to optically thin cirrus and aerosols with optical depths < 0.1. It is clear that the
bidirectional reflectance is directly proportional to the phase function, single-scattering
albedo, and optical depth. Therefore, to determine cloud and aerosol optical depths, the
single-scattering albedo and the phase function, which is a function of incoming and out-
going directions, are critical to inversion and must be known a priori.
5.1.4 Bidirectional reflectance spectra of ice

The remote sensing of ice clouds from space using reflected sunlight requires a fundamental
understanding of the radiative properties of ice clouds. Determination of these properties
0 1.0
10
m 0 = 0.5, m = 1
De = 10 μm CO2
De = 42 μm
0.8
−1 De = 124 μm O3
10
Bidirectional Reflectance
H2 O
Absorptance
0.6
−2
10
0.4
−3
10
0.2 H2 O
−4 0.0
10
0 1 2 3 4 5 0 1 2 3 4 5
Wavelength (μm) Wavelength (μm)
Figure 5.4 Spectral bidirectional reflectance (on a logarithmic scale) and absorptance as functions
of wavelength for cirrus clouds with three mean effective ice crystal sizes of 10, 42, and 124 µm
corresponding to the positions of the sun (µ0 = cos θ 0 = 0.5) and the detector (µ = 1) in the principal
plane, i.e., independent of the azimuthal angle. Absorption bands due to ozone, water vapor, and
carbon dioxide are identified in the right-hand panel.
must begin with knowledge of the complex refractive index of ice as a function of wave-
length (see Figure 2.8 for ice). The imaginary index of refraction for ice varies rapidly in
the solar and near-IR bands, resulting in significant variation in the absorption properties
of ice in these bands.
In the near-IR region, absorption of solar radiation is primarily due to water vapor, which
is strong in the band centers and falls off sharply away from these centers. Absorption by ice,
on the other hand, is more constant across each band, being less intense than water vapor in
the band centers but more intense away from the centers. Also, the maximum absorption due
to ice does not coincide exactly with the water-vapor band centers. There are regions in the
near IR where water vapor absorption dominates, whereas in other regions ice absorption is
more significant. The reflection of sunlight from ice clouds depends on their optical depth
and ice crystal size, as well as the amount of water vapor and surface albedo.
Figure 5.4 depicts the spectral BRDF [defined in Eq. (5.1.2)] and absorptance for three
mean effective ice crystal sizes [see Eq. (1.6.1) for definition] as a function of wavelength.
The computation uses parameters corresponding to the U.S. 1976 Standard Atmosphere
(McClatchey et al. 1972, Liou 2002), a cosine of the solar zenith angle of 0.5, a cosine of
the emergent angle of 1, an ice water path of 2 g m−2 , and a cloud thickness of 2 km. To
180 160 140 120 100 80 180 160 140 120 100 80
0.6 0.6 180 160 140 120 100 80
0.6
Observations Observations Observations

irregular crystal irregular crystal irregular crystal
column/plate column/plate column/plate
0.5 ice sphere 0.5 ice sphere 0.5 ice sphere
t c = 2.7± 0.5, Cs t c = 1.5 ± 0.37, Cs t c = 3.0 ± 0.7, Cs

0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
q 0 = 59.4 o q 0 = 57.6 o q 0 = 48.4 o
0.1 0.1 0.1

−60 −40 −20 0 20 40 60 −60 −40 −20 0 20 40 60 −60 −40 −20 0 20 40 60
Nadir View Angle (deg.)

Figure 5.5 Comparison of measured and computed BRDFs for cirrus clouds as a function of viewing (nadir) angle from −60° to +60°,
along with the solar zenith angle and the scattering angle. Spinhirne et al. (1996) conducted observations over Oklahoma on November
24 and 25, 1991 using a scanning radiometer aboard ER-2. Theoretical interpretations used a mean effective ice crystal size of 42 µm and
three ice crystal models: spheres, defined hexagonal plates and columns, and irregular ice aggregates with rough surfaces (after Liou et al.
2000).
include line absorption in multiple scattering atmospheres, we have used a spectral interval
of 50 cm−1 containing 30 cumulative probability functions covering the spectrum from
0.2 to 5 µm, based on the correlated k-distribution method for radiative transfer (Fu and
Liou 1992a, Liou 2002; see Subsection 6.1.3 for additional references). H2 O absorption
bands located at 3.2, 2.7, 1.87, 1.38, 1.1, 0.94, 0.82, and 0.72 µm are clearly identified, as
well as the 4.3 µm CO2 band and O3 UV bands discussed in Subsection 5.1.2. Bidirectional
reflectances are shown to have maxima in between H2 O absorption bands. Small ice crystals
reflect more solar radiation. Sufficient ice crystal size information is clearly identified in
the near-IR bidirectional reflection spectrum.
Hansen and Pollack (1970), who have interpreted spectral near-IR reflectances of clouds
measured from aircraft, have suggested the possibility of using BRDF variation in the
near-IR region to infer the physical properties of clouds. Twomey and Seton (1980) have
also suggested the inference of optical depth and mean radius from spectral reflectance
measurements. The preceding suggestions are based on the principle of radiative transfer,
i.e., that the reflection of clouds at a visible wavelength with little absorption is primarily a
function of their optical depth. Whereas at a near-IR wavelength, at which the absorption
of water vapor in the clouds is small, their reflection is primarily dependent on particle size.
See Section 5.2 for further discussion of this topic.
In Figure 5.5, we compare the measured and computed BRDFs for cirrus clouds. The
observations were conducted over Oklahoma on November 24 and 25, 1991 using a scanning
radiometer aboard ER-2 by Spinhirne et al. (1996), who also derived the best-fit cloud
optical depths (displayed in the figures) and surface albedos from concurrent lidar and
spectral radiometric measurements. For theoretical interpretation, we have used cirrostratus
with a mean effective size of 42 µm and three ice crystal models: spheres, defined hexagonal
plates and columns, and irregular ice aggregates with rough surfaces. BRDF is presented
as a function of the viewing (nadir) angle from −60° to +60°, along with the solar zenith
angle and the scattering angle. The measured data appear as blue dots, in terms of mean and
standard deviation. First, for spheres, the theoretical patterns computed from the Lorenz–
Mie phase function do not match the observed data in terms of the overall BRDF magnitudes
and detailed scattering patterns. The spherical case has a small peak at 140° associated
with the first rainbow feature reduced by multiple scattering. Second, for the defined
plate/column case, we find that the side scattering at 130° and the maximum feature
at 150–160° deviate from observed values. It is possible that the measured data, which
were obtained from viewing angles 20° apart, may have missed the scattering maximum.
Third, the best match is for the case involving irregular ice particles. In view of these
interpretations, it appears that ice crystals in these developed cirrus clouds must contain a
combination of hexagonal and irregular ice particles.
5.1.5 The A-Train satellite constellation

For the remote sensing of clouds and aerosols, the most important satellite system is the
NASA A-Train satellite constellation, whose positions are displayed in Figure 5.6. The
Afternoon Constellation adopted the name “A-Train” in reference to Billy Strayhorn’s
A-Train Constellation
295.5 sec.
GCOM-W1
73 sec.
Aqua 101 sec.
17.5 sec. CALIPSO
OCO 2
CloudSat
CERES
CALIOP AIRS
AMSR-E
AMSU-A
AMSR2
IIR
2 min. CPR MODIS
WFC
4 min. PARASOL
Aura
MLS
TES POLDER
HIRDLS
CMI
Figure 5.6 Relative NASA satellite positions. The Afternoon Constellation adopted the name “A-
Train” – in reference to Billy Strayhorn’s jazz standard, “Take the A Train” – because Aqua (launched
May 4, 2002) and Aura (launched July 15, 2004) were the first to join the fleet and both of their names
start with the letter “A,” and also because the local-time equatorial crossing is in the afternoon (after
NASA 2012).
jazz standard, “Take the A Train.” The Aqua satellite (launched May 4, 2002) and the
Aura satellite (launched July 15, 2004) were the first to join the fleet, both of their names
start with the letter “A,” and both orbits correspond to a local-time equatorial crossing in
the afternoon (NASA 2012). These satellites cross the Equator within a few minutes of
one another at around 1:30 pm local time. Together, their overlapping science instruments
give a comprehensive picture of Earth weather and climate. In the following paragraphs,
we provide a brief review of these satellites in accordance with their positions.
The Orbiting Carbon Observatory (OCO) was a NASA Earth System Science Pathfinder
Project (ESSP) mission designed to make precise, time-dependent global measurements of
atmospheric CO2 from an Earth-orbiting satellite. Due to a launch vehicle payload firing
anomaly on February 24, 2009, OCO unfortunately failed to reach orbit. The replacement
OCO2 was successfully launched on July 2, 2014. OCO2 is equipped with spectrometers
that measure the reflected sunlight intensities at three wavelength bands – the weak 1.6 µm
and relatively stronger 2.2 µm CO2 bands and the 0.76 µm O2 A-band.
The Global Change Observation Mission (GCOM-W1) is the first in the GCOM-W
series, designed to observe the water cycle. This satellite carries the Advanced Microwave
Scanning Radiometer 2 (AMSR2) instrument, the successor to the AMSR-E carried by

Aqua. This microwave radiometer will observe precipitation, water vapor, and wind velocity
above the ocean, as well as sea-water temperature and water level on land areas, and, lastly,
snow depth.
The Aqua satellite carries the following instruments: (1) Advanced Microwave Scanning
Radiometer-EOS (AMSR-E), which measures cloud properties, sea surface temperature,
near-surface wind speed, radiative energy flux, surface water, ice, and snow; (2) Moderate
Resolution Imaging Spectroradiometer (MODIS), which also measures cloud and aerosol
properties, radiative energy flux, land cover and land-use change, fires, and volcanoes;
(3) Advanced Microwave Sounding Unit (AMSU-A), which measures atmospheric tem-
perature and humidity; (4) Atmospheric Infrared Sounder (AIRS), which measures atmo-
spheric temperature and humidity and land and sea surface temperatures; and (5) Clouds and
the Earth’s Radiant Energy System (CERES), which measures broadband radiative energy
flux.
The CloudSat satellite carries Cloud Profiling Radar (CPR), a 94 GHz nadir-looking
radar, which measures the power backscattered by clouds and light precipitation as a
function of distance. It is particularly useful for the detection of layered cloud systems,
since the 94 GHz radar power penetrates through high-level clouds.
The Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO)
carries (1) Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), which provides
high-resolution vertical profiles of aerosols and clouds; (2) a Wide Field Camera (WFC),
which was selected to match band 1 of the MODIS instrument on the Aqua satellite; and
(3) an Imaging Infrared Radiometer (IIR), which is used to detect cirrus cloud emissivity
and particle size. The CALIOP laser beam is aligned with the center of the IIR image to
optimize joint CALIOP/IIR observations.
The Glory satellite, which was designed for remote sensing of aerosols and clouds,
particularly for detection of absorbing aerosols and their single-scattering albedo for climate
forcing studies, failed during launch on March 4, 2011. It would have flown between
CALIPSO and Aura.
The Polarization & Anisotropy of Reflectances for Atmospheric Sciences coupled with
Observations from a Lidar (PARASOL) is a French-built Earth observing research satellite.
It carries an instrument called POLarization and Directionality of the Earth’s Reflectances
(POLDER), which studies the radiative and microphysical properties of clouds and aerosols.
It was launched on December 18, 2004 and moved to another (lower) orbit on December
2, 2009.
Aura (Latin for breeze) was launched on July 15, 2004. It carries the following active
instruments: (1) Microwave Limb Sounder (MLS), which measures emissions from ozone,
chlorine, and other trace gases, and clarifies the role of water vapor in global warming;
(2) Ozone Monitoring Instrument (OMI), which uses ultraviolet and visible radiation to
produce daily high-resolution maps; and (3) Tropospheric Emission Spectrometer (TES),
which measures tropospheric ozone, carbon monoxide, methane, and nitrogen oxides in
infrared wavelengths.
Most relevant to the detection and quantification of clouds and aerosols in general
and cirrus clouds in particular are MODIS (global coverage with its scan capability) and
CloudSat and CALIPSO (nadir-looking and push-bloom track). In what follows, we present
the fundamentals of retrieval of the optical depth and ice crystal size of cirrus clouds using
two channels in MODIS.
5.2 Determination of cirrus cloud optical depth and ice crystal size
5.2.1 Definition of optical depth and mean effective size
First, we define the particle size that can be inferred from reflected sunlight using water
clouds containing spherical droplets. Based on the remote sensing principle discussed in
the introduction to this chapter, determination of the detailed droplet size spectrum from
space, employing currently available instruments, has yet to be developed and verified.
Nevertheless, we may define a mean effective radius ae and the corresponding effective
variance v e to represent the actual size distribution, taking into account the scattering
property of spherical droplets, such that they scatter an amount of light in proportion to
their cross-sectional area as follows:
,
ae = a · πa 2 n(a)da πa 2 n(a)da, (5.2.1a)
,

ve = (a − ae )2 · πa 2 n(a)da ae2 πa 2 n(a)da , (5.2.1b)
where the integration limits throughout this subsection are from a (min) to a (max), πa2 is
the geometric cross-section of a water droplet, and n(a) denotes the droplet size distribution
in units of cm−3 µm−1 . The mean effective radius accounts for droplet cross-section as a
weighting factor. Once the droplet size distribution is known, e.g., a gamma distribution,
we may also define an effective variance. However, based on numerous radiative transfer
calculations, Hansen and Travis (1974) demonstrated that the reflection and transmission
of sunlight are primarily dependent on the mean effective radius if the effective variance
is given, regardless of the detailed distribution of droplet sizes. For practical retrieval,
the effective variance has usually been predetermined using available observations (e.g.,
v e = 0.1). In the following, we relate the mean effective radius to the amount of vertically
integrated liquid water in the cloud, referred to as the liquid water path (LWP, g cm−2 ), and
the optical depth.
The liquid water content (LWC, g cm−3 ) for spherical droplets is defined in the following
way:

4π
LWC = V ρl n(a)da = ρl a 3 n(a)da, (5.2.2)
3
where the density of water ρl ≈ 1 g cm−3 . Thus, LWP = LWC × z (cloud thickness). The
optical depth is defined as

τ = βe · z = Qext πa 2 n(a)da · z, (5.2.3)
5.2 Cirrus cloud optical depth and ice crystal size 261
where βe is the extinction coefficient in units of (length)−1 , Qext is the efficiency factor for
extinction (Figure 3.23), a function of the droplet radius, wavelength, and refractive index.
For wavelengths in the visible, Qext 2 for cloud droplets. By combining the preceding
three equations and noting that ρl ≈ 1 g cm−3 , we obtain
3
ae ≈ LWP/τ , (5.2.4)
2
which relates the optical depth, LWP, and the mean effective droplet size. This equation has
an important physical implication. Consider two clouds having the same LWP. The cloud
that contains a smaller mean effective droplet size will have a bigger optical depth, leading
to more reflected sunlight.
Ice crystals are much more complicated than spherical water droplets with respect to
their scattering and radiative properties, due to non-sphericity and a substantially larger
size spectrum. We may define the mean effective ice crystal size to represent ice crystal
size distribution in the form [see Eq. (1.7.2)]

3
De = V n(L)dL A n(L)dL, (5.2.5)
2
where L is the maximum dimension of the ice crystal and the integration is from L (min)
to L (max), V is the volume, A is the geometric projected area of the ice crystal on a
surface perpendicular to the incident light beam, n(L) is the ice crystal size distribution,
and a constant factor of 32 is added so that a spherical particle is the limiting case. In this
manner, the shape of irregular ice crystals is approximately accounted for. Further, the ice
water content (IWC) for a given ice crystal size distribution is defined by

IWC = V ρi n(L)dL, (5.2.6)
√
where ρi is the density of ice. The volume of a hexagonal ice crystal is given by 3 3LD 2 /8,
with D the ice crystal width. Moreover, for randomly oriented ice crystals
√ within the limits
of geometric optics (Qext 2), the extinction cross-section is 3D( 3D/4 + L)/2 [see
Eq. (2.4.49) and the discussion immediately below that equation]. With these relationships
and following the principle introduced for water droplets, we may also relate the optical
depth and the mean effective ice crystal size, as follows:
τ∼
= IWP(a + b/De ), (5.2.7)
where the ice water path IWP = IWC × z, and the coefficients a ∼ = −6.656 × 10−3 and
b∼= 3.686 for ice columns.
In what follows, we present the principle of using reflected sunlight at visible and
near-infrared wavelengths to retrieve cirrus cloud microphysical and optical properties,
represented by mean effective ice crystal size and optical depth, respectively. Implications
for climate and climate sensitivity analysis will be presented in Chapter 6.
5.2.2 Two-channel inversion technique

We first present some background about this technique for the determination of optical
depth and mean effective ice crystal size, followed by an introduction to the MODIS
spectrum for application to the retrieval of these two parameters. Solar reflection methods
are based on the principle that cloud reflection at a non-absorbing channel in the visible
spectrum is primarily a function of cloud optical depth, whereas cloud reflection at an
absorbing channel in the near IR is mainly a function of cloud particle size (e.g., see
King et al. 1997).
Application of the solar reflection method was demonstrated by Hansen and Pollack
(1970), who attempted to explain spectral variation in aircraft reflectivity measurements
(Blau et al. 1966) in terms of particle phase and size, based on a set of theoretical computa-
tions. From radiative transfer analysis, Twomey and Seton (1980) showed that mean droplet
radius and optical depth can be determined for optically thick clouds from simultaneous
reflectance measurements in the visible and near IR. Application of this multispectral cor-
relation technique using 0.75, 1.0, 1.2, and 2.25 µm aircraft reflectance measurements was
not entirely successful (Twomey and Cocks 1982). However, after use of an improved and
modernized radiometer as well as addition of a 1.66 µm wavelength, Twomey and Cocks
(1989) showed that optical depth and particle size can be inferred by means of a two-channel
technique. Nakajima and King (1990) further showed that a tri-spectral approach (0.75,
1.6, and 2.2 µm) can be used to simultaneously retrieve the optical depth and mean effec-
tive radius of water clouds and that the addition of a third channel was desirable to remove
uncertainties produced by the monochromatic variation of size and single-scattering albedo.
They also found that the retrieved mean effective radius corresponds to an undefined upper
portion of the cloud, implying the need to make some adjustments in order to estimate the
mean effective radius representative of the entire cloud. Later analysis of aircraft observa-
tions at 0.75, 1.65, and 2.16 µm (Nakajima et al. 1991) over stratocumulus clouds produced
an excellent correlation between the remotely sensed, center-adjusted mean effective radii
and in situ observations. Wielicki et al. (1990) estimated particle sizes for water and ice
clouds on the basis of Landsat observations at 0.83, 1.65, and 2.21 µm. Good agreement
was found between observed and retrieved mean effective radii for water clouds; however,
large discrepancies existed for ice clouds. The major problems involving ice clouds are
associated with uncertainty in the scattering and absorption properties of non-spherical
particles (e.g., Wielicki et al. 1995, Liou 1986).
In view of intricate ice crystal morphologies and sizes, coupled with the complexity
of atmospheric absorption properties and surface reflection, a direct solution for optical
depth and mean effective ice crystal size from two-channel measurements at the top of the
atmosphere is not possible. However, one may follow a numerical procedure to recover
information content based on the following optimization principle:
(1) First, obtain single-scattering properties and phase functions computed from a light-
scattering theory employing a priori (an initial guess) a set of particle size and shape
distributions in terms of their mean values.
1.0
t c (0.75 μm) = 16
0.8 re = 4 μm
re = 8 μm
re = 12 μm
Spherical Albedo
re = 16 μm
0.6
re = 20 μm
re = 20 μm + vapor
0.4
0.2
0.0
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Wavelength (μm)
Figure 5.7 Cloud spherical albedo, a non-dimensional parameter ranging from 0 to 1, as a function
of wavelength for a number of cloud droplet mean effective radii having a modified gamma size
distribution with an effective variance ve of 0.111 [see Eq. (5.2.1b)], cloud optical depth of 16 at
0.75 µm, and saturation water vapor of 0.45 g cm−2 . The shaded vertical areas denote the MODIS
channel bandwidths in the solar region. The red crosses indicate the wavelengths that can be used for
cloud detection and retrieval of its optical depth and mean effective radius (after King et al. 1997).
(2) Second, construct numerous correlation look-up tables in the domain of visible and
near-IR wavelengths for a set of optical depths and particle sizes and for a prescribed
sun–satellite geometry (solar zenith angle, viewing zenith angle, and relative azimuthal
angle) and surface albedo, based on a line-by-line radiative transfer model, including
multiple scattering.
(3) Third, minimize the sum of the square of “residuals,” defined as the difference between
the natural logarithm of computed and measured reflectances. In mathematical terms,
this is expressed as follows:
χi2 = [ln |R1i,mea − R1i,com (τm , Den )|]2 + [ln |R2i,mea − R2i,com (τm , Den )|]2 , (5.2.8)
where i, m, and n are indices for a pixel and a priori optical depth and particle
size, respectively, and mea and com denote the measured and computed reflectances,
respectively.
(4) Fourth, use a systematic search method coupled with a linear interpolation scheme
to find the solutions for optical depth and particle size. Validation of this retrieval
algorithm is critical because of the non-uniqueness of the solutions determined from
the statistical optimization method.
MODIS contains solar channels that can be used for the retrieval of cloud optical depth
and particle mean effective size. For illustration purposes, we present Figure 5.7 to define
MODIS solar bands in terms of cloud spherical albedo, which is an integrated reflection
over the solar zenith angle, as a function of wavelength for a number of cloud droplet
mean effective radii with a cloud optical depth of 16 at 0.75 µm. The shaded vertical areas
denote the MODIS channel bandwidth in the solar region, while the red crosses indicate the
wavelengths that can be used for cloud detection and retrieval of its optical depth and mean
effective radius, including 0.681, 1.38, 1.617, 2.139, and 3.7 µm. The 3.7 µm wavelength
involves both solar and thermal IR components and is most useful at nighttime.
In conjunction with the retrieval of the optical depth and mean effective droplet radius
of water clouds from MODIS, King et al. (1997) performed radiative transfer calculations
for a number of solar zenith angles and a set of selective zenith and azimuthal angles at two
pairs of wavelengths: 0.664 µm1.621 µm and 0.664 µm2.142 µm. These wavelengths are
outside of the H2 O and O2 absorption bands and, at the same time, have substantially
different water droplet absorption properties. As discussed previously, at a non-absorbing
wavelength (0.664 µm), cloud optical depth is largely determined by reflectance, with little
dependence on droplet radius. The reflectances at 2.142 µm and 1.621 µm, in contrast, are
largely sensitive to mean effective radius, with the largest value occurring for the smallest
droplet radius (4 µm).
Application of the bi-spectral correlation technique developed by Nakajima and King
(1990) to clouds containing ice crystals to infer their optical depths and mean effective ice
crystal sizes has been conducted by Ou et al. (1999), Rolland et al. (2000), Rolland and Liou
(2001), and King et al. (2004). As an example, reflectance calculations have been carried
out by means of the adding/doubling method for a set of incoming and outgoing directions
corresponding to specific ER-2 aircraft operations for six ice crystal size distributions
with mean effective ice crystal sizes ranging from 24 µm (cold cirrus) to 124 µm (cirrus
uncinus). The single-scattering inputs to radiative transfer include the extinction coefficient,
single-scattering albedo, and asymmetry factor, which were computed from the geometric-
optics approach developed by Takano and Liou (1989a, 1989b) for columns and plates
(see Subsection 3.3.2). The results are displayed in 2-D reflectance diagrams in terms of
0.6811.617 µm and 0.6812.139 µm for a set of optical depths ranging from 0.5 to 16, as
displayed in Figure 5.8. The 0.681 µm reflectance mainly depends on optical depth, whereas
the 1.617 and 2.139 µm reflectances are primarily a function of mean effective particle size.
Also shown are the MAS data obtained from FIRE-II-IFO field experiments (see Subsection
1.6.2) over the northern Gulf of Mexico on December 5, 1991. For each MAS observation
point, a set of optical depths and mean effective ice crystal sizes can be determined via
procedure (4) above. It should be pointed out that the bi-spectral correlation technique
developed by Nakajima and King (1990) has, after improvements and refinements, played
a significant role in operational retrievals of cloud properties under daytime conditions
(Platnick et al. 2003, Minnis et al. 2011b).
Furthermore, Rolland et al. (2000) performed radiative transfer calculations using
MODIS channels of 0.65, 1.6, and 2.1 µm for six tropical ice crystal size distributions
with mean effective ice crystal sizes of 21, 23, 35, 41, 52, and 91 µm, as shown in
Figure 5.9. The shape distribution was based on in situ measurements covering a range
24 μm 93 μm
30 μm 124 μm
42 μm MAS data
(a) 75 μm (b)
q q q q
Figure 5.8 Correlation between the reflectances at (a) 0.681 µm1.617 µm and (b) 0.681 µm
2.139 µm. Radiative transfer and single-scattering calculations were performed for six ice-crystal
size distributions with mean effective sizes ranging from 24 to 124 µm. The optical depth used was
from 0.5 to 16. Overlaid on the curves are MAS data obtained from FIRE-II-IFO over the northern
Gulf of Mexico on December 5, 1991.
104 99.0 %
104 104
64.5 % 58.4 %
3 3 3
10 0.4 % 10 30.5 % 10 21.5 %
102 0.6 %
102 5.0 % 102
n(m-3 μm-1)
8.6 %
0.0 % 0.0 % 11.5 %

1 1 1
10 10 10
0
10 0 10 100
10-1 10-1 10-1

De = 21 μm De = 23 μm De = 35 μm
-2
10-2 10 10-2
0 600 1200 1800 0 600 1200 1800 0 600 1200 1800
4 4 4
10 10 10 58.6 %
50.7 % 57.9 %
3 3 3
10 16.3 % 10 26.7 % 10 8.4 %
2 2 20.0 %
10 22.9 % 102 6.3 % 10
n(m-3 μm-1)
9.1 % 13.0 %
1 10.1 %
10 10 1 101
100 100 100
10-1 10-1 10-1

De = 41 μm De = 52 μm De = 91 μm
10-2 10-2 10-2
0 600 1200 1800 0 600 1200 1800 0 600 1200 1800
Maximum Dimension (μm) Maximum Dimension (μm) Maximum Dimension (μm)
Figure 5.9 Six tropical cirrus size and shape distributions selected for single-scattering and radiative
transfer calculations to develop look-up tables for the two-channel retrieval technique to determine
cirrus optical depth and mean effective ice crystal size, which ranges from 21 to 91 µm (after
Rolland et al. 2000).
10 100
9 Optical Depth Ice Crystal Mean
8 Effective Size (μm)
75
MAS solar retrieval
MAS solar retrieval

7
6
5 50
4
3
SUCCESS 25 SUCCESS
2
CEPEX CEPEX
1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 25 50 75 100
In-situ analysis In-situ analysis
Figure 5.10 Validation of the retrieval technique based on collocated and coincident in situ ice
microphysics data derived from SUCCESS and CEPEX for (a) optical depth and (b) ice crystal
effective size (after Rolland et al. 2000 with modifications).
from hollow columns, plates, bullet rosettes, to aggregates (in terms of percentage; see
also Figure 1.19). The single-scattering calculations for complex ice crystals followed
the improved geometric-optics approaches presented in Subsection 3.3.3 (Yang and Liou
1995, 1996b, 1997). The results are displayed in 2-D reflectance diagrams similar to those
in Figure 5.8, in terms of 0.651.6 µm and 0.652.1 µm, for optical depths ranging from
0.5 to 8 and ice crystal sizes from 20 to 90 µm, and a set of sun–ER-2-aircraft geometries
was constructed for the inference of optical depth and ice crystal size.
The MAS data obtained from SUCCESS (1996) and CEPEX (1993), described in
Subsection 1.6.2, were employed for analysis. Additionally, the effective surface albedos
used in the calculations were determined from MAS reflectances over clear pixels. For each
data point, an optical depth and a mean effective ice crystal size can be determined using
the procedures outlined in steps (1)–(4) above.
Figure 5.10 illustrates a careful validation of retrieved optical depth and ice crystal size
on the basis of collocated and coincident in situ observations. In this figure, the retrieved
cirrus optical depth and mean effective ice crystal size are compared with those derived from
the analysis of concurrent in situ microphysical observations (6 s sampling at 200 m s−1
true air speed for a sampling path of 1200 m) in which clusters of 1200 m × 1200 m pixels
were used. The collocation process was carefully carried out using satellite visible images
overlaid with aircraft paths (NASA ER-2 and Learjet during CEPEX, and ER-2 and DC-8
during SUCCESS) and 2-DC data measurement locations. Radiation and microphysical
measurements carried out during CEPEX, specifically during the course of tracks 6 and
7 on April 4, 1993, were used for analysis. During track 6 the Learjet was flying within
the cloud mass at which time in situ observations were collected and used to validate the
retrieved optical depth. During track 7 the Learjet was flying near the cloud top. The in situ
observations collected during this time were used to validate the retrieved mean effective ice
crystal size. Cloud geometric depth was inferred from the Learjet pilot’s notes. Radiation
and microphysical measurements carried out during SUCCESS, specifically during track
7 on April 21, 1996, were also used for analysis. Cloud geometric thickness was obtained
from the lidar on board the NASA ER-2 aircraft. Note that for the SUCCESS case we found
only one instance of MAS and 2-DC measurements collocated in space and time.
The discrepancies displayed in Figure 5.10 could be the result of uncertainties in physi-
cal assumptions involving the scattering phase function, as discussed in Subsection 5.2.3.
Another major source of error concerning validation studies is the lack of vertical resolution
associated with in situ microphysical measurements. For example, the clouds sampled dur-
ing CEPEX were several kilometers thick. However, the available in-cloud microphysical
measurements were carried out along one single level from which extrapolation to other
levels must be relied upon in order to perform validation calculations (see Section 5.3).
5.2.3 The importance of ice crystal phase function in the two-channel approach
Errors in the retrieved parameters, both random and biased, may be attributed to several
factors, including limitations of instrument performance as well as the physical assumptions
inherent in retrieval methods. In conjunction with light scattering by ice crystals, we should
present the importance of ice crystal phase function in the construction of look-up tables
(see Figure 5.8) a priori for the determination of optical depth and mean effective ice crystal
size. Rolland et al. (2000) conducted comprehensive theoretical analyses using sets of ice
crystal size distributions, habit, and surface roughness in phase function calculations and
the resulting look-up table construction based on radiative transfer calculations.
The effect of size distribution was first investigated. Four different common size dis-
tributions – standard gamma, bimodal, log-normal, and power-law – were used in single-
scattering calculations assuming column shapes from 5 to 120 µm in size at a 5 µm resolu-
tion for 0.65, 1.6, and 2.2 µm wavelengths. The mean effective ice crystal size is defined in
Eq. (5.2.5), while the effective variance is defined by
,
1 ) 2 1/3 *2
Ve = 2 (D L) − De DL n(L)dL DL n(L)dL, (5.2.9)
De
where D is the width and L is the maximum dimension, the relation of which has been
determined by Auer and Veal (1970), Heymsfield (1972), and Mitchell and Arnott (1994)
based on laboratory and field measurements [see Eq. (1.6.2)].
Figure 5.11a shows phase functions for the four size distributions using Ve = 0.25, while
Figure 5.11b shows phase functions for bimodal distributions using Ve of 0, 0.1, and 0.25.
Both are for a De of 50 µm and a visible wavelength of 0.65 µm. From these results, it
is evident that differences in specific features such as forward scattering, backscattering,
3
(a) 10 (b) 103
l = 0.65 μm De = 50 μm l = 0.65 μm Bimodal distribution
Ve = 0.25
102 102 De = 50 μm
Phase Function
Phase Function
Ve = 0
Power
101 Gamma 101 Ve = 0.1
Log-normal Ve = 0.25
0 Bimodal 0
10 10
10-1 10-1
-2 -2
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
(c) 10
3 (d) 103
l = 1.6 μm Pristine habit model
l = 1.6 μm De = 52 μm
2 2
10 10 De = 52 μm
Pristine Phase Function
Phase Function
Aggregate Smooth
101 Mix 101 Rough
0 0
10 10
10-1 10-1
-2 -2
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Figure 5.11 Visible (0.65 µm) phase functions for (a) four analytic size distributions and (b) three
effective size distribution variances assuming columnar shape; near-IR (1.6 µm) phase functions for
(c) three ice crystals habit models and (d) two ice crystal surface conditions (after Rolland et al.
2000).
and 22° and 46° peaks are relatively small. Retrieval errors based on the look-up tables
constructed from these phase functions show that the errors are less than 1–3% and hence
are not significant.
To determine the sensitivity of the retrieved parameters to various assumptions about
ice crystal habit models, 12 representative size distributions with effective mean ice crystal
sizes ranging from 21 to 123 µm were used with habits comprised of (1) pristine (plates,
L < 100 µm; columns, L > 100 µm), (2) aggregate (bullet rosettes, L < 100 µm; column
aggregates, L > 100 µm), and (3) mixed (80% plates and 20% columns, L < 100 µm;
50% plates and 50% columns, L > 100 µm). Analysis of the single-scattering calculations
reveals that the assumed habit models can have a significant effect on phase functions.
Depicted in Figure 5.11c are 1.6 µm phase functions for three habit models. Consistent
differences in the phase functions are shown for scattering angles in satellite remote sensing
geometries. For instance, the pristine model has about twice the phase function values as
the aggregate model at a scattering angle of 50°. Inspection of the phase function at 0.65 µm
indicates that differences between the two models are primarily located at scattering angles
larger than 30°. Because of these phase function differences, it is anticipated that the effects
of ice crystal habit model assumption on the retrieval of optical depth and mean effective
ice crystal size are substantial [e.g., see Eq. (5.1.3)].
Ice particle surfaces could have substantial roughness associated with collision and
coalescence events (see Subsection 1.4.2). Surface roughness on laboratory-generated ice
crystals has also been observed by Cross (1968). To assess the impact of ice particle surface
roughness on retrieved cloud parameters, the pristine habit model (plate and column) was
used in light-scattering calculations following the procedure described in Subsection 3.1.3
(see also Subsection 3.4.3). In Figure 5.11d, we see that the backscattering, 22°, and 46°
peaks vanish for rough-surface ice particles. Moreover, substantial reduction in the phase
function can be seen throughout most of the scattering angle range that is pertinent for
satellite remote sensing applications. Similar results apply to the 0.65 µm wavelength.
We would expect that, employing the look-up table approach presented above, ice crystal
surface roughness assumptions could have a significant effect on both the retrieved optical
depth and mean effective ice crystal size. Yang et al. (2008) showed that neglecting ice
crystal surface roughness in generating look-up tables would systematically lead to under-
and overestimates of the retrieved cloud optical depth and mean effective ice crystal size,
respectively.
In summary, reliable phase functions for ice crystals are essential to the success of
remote sensing of optical and microphysical properties of cirrus clouds by means of sunlight
reflected from these clouds.
5.2.4 The current MODIS operation program for ice cloud studies
MODIS is a state-of-the-technology multi-disciplinary instrument flying on board two of
NASA’s polar-orbiting satellites: Aqua (see Figure 5.6) and Terra. Each MODIS instrument
provides unprecedented information on a global basis about processes occurring on land,
in the oceans, and in the atmosphere. The MODIS instrument makes observations in 36
discrete bands between 0.4 and 14.5 µm. In Subsection 5.2.2, we identified solar channels
that are pertinent to cloud and aerosol observations.
For data sampling, MODIS instrumentation employs a continuously rotating double-
sided scan mirror to perform cross-track, ±55° scans. Flying in a 705 km orbit, this scanning
pattern enables MODIS to achieve a 2330 km swath (cross-track) and provides global
coverage every one to two days. MODIS has a resolution of 250 m at nadir for two bands
(bands 1 and 2), with five bands (bands 3–7) at 500 m, and the remaining 29 bands (bands
8–36) at 1 km. The MODIS continuous observations are processed and interpreted on a
granule basis (a granule is simply a segment of MODIS orbital swath containing about
5 minutes of data).
As an example, Figure 5.12 shows a false-color RGB (red, green, blue) image of
a granule observed by the MODIS instrument on the Aqua satellite at 1845 UTC on
Figure 5.12 A false-color RGB (red, green, blue) image of a MODIS granule. The reflectance
observed by MODIS band 1 (band center near 0.65 µm) is selected for red; band 2 (band center is
about 0.86 µm) is used for the green channel. The intensity of blue corresponds to the brightness
temperature of band 31 (the band center is about 11 µm) after gray is flipped. In such a color scheme,
the ocean is dark, the land is green, low-level clouds appear yellowish, and high-level clouds are
white or light blue (courtesy of Zhibo Zhang).
July 22, 2007 over Central America. For the RGB display, the reflectance in MODIS band
1 (with the band center at approximately 0.65 µm) is selected for red; band 2 (with the band
center at approximately 0.86 µm) is used for green. The intensity of blue corresponds to
the brightness temperature of band 31 (with the band center at approximately 11 µm) after
being “gray flipped,” which simply means that objects with higher brightness temperature
are assigned lower intensity. In other words, colder objects appear bluer than warmer objects
in Figure 5.12. In such false-color schemes the ocean is dark and the land surface is green.
Note that the ocean, which appears warm in comparison with clouds, reflects little radiation
in the 0.65 and 0.86 µm bands. Therefore, the intensities of all three channels for the ocean
are small, which makes the ocean appear dark.
In comparison with clouds, land surfaces are warm and their reflectance in the 0.65 µm
band is small, but land surface reflectance in the 0.86 µm band is significant. These features
make the intensities of the red and blue channels for land surfaces small and the intensity of
the green channel moderate, which is the reason for the green land surfaces. In Figure 5.12,
high-level ice clouds generally have a whitish cast (although thin ice may appear bluish),
while low-level water clouds appear somewhat yellowish-green. This contrast is associated
with the difference in cloud-top temperature between low and high clouds, both of which
reflect 0.65 and 0.86 µm radiation effectively, so there is little difference in either the red
or the green channel between high and low clouds. However, because high clouds are
generally colder than low clouds, the intensity of the blue channel, which corresponds to
the gray-flipped 11 µm brightness temperature, is larger for high clouds than for low clouds.
The combination of the false red and green channels in the RGB composite makes low
clouds look yellow. In the case of high clouds, the intensities of all three channels are large,
so they appear bright white in Figure 5.12.
The MODIS operational retrieval of cloud optical and microphysical properties (specifi-
cally, optical depth and mean effective size) is based on the bi-spectral algorithm developed
by Nakajima and King (1990) mentioned previously. The physical principle of this retrieval
algorithm has been illustrated in Figure 5.8. For operational retrievals, a combination of
the 2.13 µm band and visible or near-IR band is used. Specifically, the spectral combina-
tion of 0.86 µm and 2.13 µm is used over land and ocean, whereas a combination of 1.24
and 2.13 µm bands is used over snow and sea ice surface. The latest MODIS operational
retrieval products are referred to as MODIS Collection 6 (C6) products. In the previous
MODIS collection (C5), the primary cloud effective radius was reported using the 2.13 µm
band. The 1.64 and 3.7 µm effective radius retrievals are reported in terms of deviations
from the 2.13 µm counterpart. Unlike C5 products, MODIS C6 products report all spectral
retrievals, including separate absolute cloud optical depth, effective radius, and water path
retrievals reported for combinations of 1.64, 2.13, and 3.7 µm bands. The detailed algorithm
theoretical basis document (ATBD) for MODIS operational cloud products (MOD06 and
MYD06) has been presented by Platnick et al. (2014), and will not be revisited in this text.
However, we briefly review ice cloud single-scattering property models used for operational
cloud property retrievals.
The earliest version of the MODIS ice cloud single-scattering property models was
developed by the authors of this text and was used by MODIS Collection 4 products
and MODIS/MAS retrieval (King et al. 2004). For ice crystals with maximum dimensions
smaller than 70 µm, 50% bullet rosettes, 25% plates, and 25% columns are employed, while
30% aggregates, 30% bullet rosettes, 20% hollow columns, and 20% plates are used for ice
crystals larger than 70 µm. These particles are assumed to be smooth. Furthermore, 12 in
situ particle size distributions, discretized into five bins, were used to obtain bulk single-
scattering properties, namely the ensemble-averaged phase function, single-scattering
albedo, and extinction efficient. Specifically, at each band the particle-ensemble-averaged

bulk single-scattering properties are defined in the form
λ2 Dmax N

fi (D)Qext,i (λ, D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin i=1
Qext = , (5.2.10a)
λ2 Dmax
N
fi (D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin
i=1
λ2 Dmax
N
fi (D)i (λ, D)Qext,i (λ, D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin i=1
= , (5.2.10b)
λ2 Dmax
N
fi (D)Qext,i (λ, D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin i=1

λ2 Dmax
N
fi (D)P ()i (λ, D)Qsca,i (λ, D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin i=1
P () = N ,
λ2 Dmax
fi (D)Qsca,i (λ, D)Ai (D) S(λ)n(D)dDdλ
λ1 Dmin i=1
(5.2.10c)
where Qext , , and P () are the broadband-averaged extinction efficiency, single-
scattering albedo, and phase function, respectively; n represents particle size distribution;
S is the band response function; λ1 and λ2 are specified wavelength limits; i is the index of
the mixture habit in ice clouds (total is N); and f is the ice cloud habit mixture such that
the summation of fi is equal to 1.
MODIS Collection 5 ice cloud single-scattering property models have been reported by
Baum et al. (2005a) in which a gamma distribution,
n(D) = N0 D μ e−γ D , (5.2.11a)
is used, where the slope (γ ) and dispersion (μ) parameters were determined from in situ
measurements. In this equation, N0 is the intercept, n(D) is the particle concentration per
unit volume, and D is the maximum dimension of an ice particle. Baum et al. (2005a) filtered
over 2000 particle size distributions in terms of cloud temperatures less than −25°C, which
ensured that the cloud particle phase is ice. After filtering, 1119 particle size distributions
remained.
The MODIS Collection 5 habit distribution consists of 100% droxtals when D < 60 µm;
15% bullet rosettes, 50% solid columns, and 35% plates when 60 µm < D < 1000 µm; 45%
hollow columns, 45% solid columns, and 10% aggregates when 1000 µm < D < 2500 µm;
and 97% bullet rosettes and 3% aggregates when D > 2500 µm. Except for aggregates,
which were assumed to be moderately rough, all other particle habits were assumed to have
smooth surfaces.
Contrary to the assumption of detailed particle habit distributions in the MODIS C5
model, the ice cloud single-scattering model in MODIS C6 is based on the assumption
(a) MODIS Collection 5 (b) MODIS Collection 6
Rough Surface Aggregate

(50 μm)
(c) (d)
4
1 10
MODIS C5
0.95 MODIS C6 3
10
Asymmetry Factor
Phase Function
0.9
2
10
0.85
1
10
0.8
0
10
0.75
-1
0.7 10
0 50 100 150 200 0 30 60 90 120 150 180
Effective Diameter ( μm) Scattering Angle (deg.)
Figure 5.13 Comparison between (a) MODIS C5 and (b) C6 ice cloud habit models; in (b), only one
ice aggregate with rough surface is shown. Panels (c) and (d) are the corresponding asymmetry factors
and phase functions for MODIS band 1 (at 0.67 µm) for the two models, using a mean effective ice
crystal size of 50 µm.
that all ice crystals have only one habit, namely an aggregate with a rough surface having
a gamma size distribution (Platnick et al. 2014). Figures 5.13a and 5.13b schematically
illustrate the ice crystal habit assumptions for MODIS C5 and C6 models, respectively.
The resulting asymmetry factor as a function of mean effective ice crystal size for C6 has a
constant value of 0.75, whereas for C5, its values are between 0.78 and 0.88. The phase
function as a function of the scattering angle for C6 is close to a smooth curve, due to
the ice crystal rough surface, whereas in C5, the phase function exhibits halo and other
features.
(a) (b)
320 320
8.5 μm 11 μm
Simulated BT (K)
Simulated BT (K)
280 280
240 240
200 200
160 160
200 240 280 320 200 240 280 320
Observed BT (K) Observed BT (K)
(c) (d)
320 12
12 μm Simulated BTD (K) 8.5-11 μm
Simulated BT (K)
9
280
6
240
3
200
0
160 -3
200 240 280 320 0 3 6 9 12
Observed BT (K) Observed BTD (K)
Figure 5.14 Scatter plots of the MODIS observed brightness temperatures (BTs) at (a) 8.5, (b)11,
and (c)12 µm, and (d) brightness temperature difference (BTD) (8.5−11 µm) versus simulated BTs
and BTD using the optimal values of ice cloud optical and ice crystal mean effective size in the
calculations (after Wang et al. 2011).
Spectral consistency is one advantage of the MODIS C6 ice cloud optical property
model over its MODIS C5 counterpart. According to physical principles, cloud properties
retrieved from satellite observations must be independent of the specific spectral bands used
for retrievals. Thus, cloud property retrievals from IR bands should be consistent with those
based on the bi-spectral Nakajima–King algorithm. Thermal IR bands centered at 8.5, 11,
and 12 µm are usually used to retrieve thin ice cloud optical depth and ice crystal mean
effective size (see Subsection 5.2.6). The basic principle is to select optimal cloud properties
(optical depth and mean effective size) so that simulated brightness temperatures (BTs) at
8.5, 11, and 12 µm channels and the brightness temperature difference (BTD) between
8.5 and 11 µm channels match observed values. Figure 5.14 displays scatter plots of the
Retrieved Cloud Optical Depth

(a) (b)
2.0 2.0
MODIS C6 IR-Based
MODIS C5 IR-Based
1.5 1.5
1.0 1.0
0.5 0.5
0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
MODIS C5 VIS/NIR-Based MODIS C6 VIS/NIR-Based
Figure 5.15 Comparison between the retrieval of ice cloud optical depth based on an IR band and a
visible/near-IR band: (a) results from the MODIS C5 model; (b) results from the MODIS C6 model
(courtesy of Chao Liu).
MODIS observed BTs and BTD against the simulated BTs and BTD using the optimal
values of ice crystal mean effective size and optical depth as input in the simulations. As
shown, observed and simulated values are well correlated. Note that Iwabuchi et al. (2014)
and Heidinger et al. (2015) have also developed retrieval approaches for thin cirrus clouds
based on thermal IR window bands.
Figure 5.15 shows a comparison between IR-band based and visible/near-IR-band based
retrievals of ice cloud optical depth. MODIS C5 and C6 ice cloud single-scattering property
models (Figure 5.13) were used for the retrieval, and the results are displayed in the left and
right panels, respectively. With the use of the MODIS C5 model, results derived from the
IR-band based retrieval are shown to underestimate ice cloud optical depth (Figure 5.15a).
Using the MODIS C6 model, results show that a spectral consistency can be achieved
(Figure 5.15b).
Determination of the robustness of ice cloud property models from MODIS C6 can
be checked by comparing polarimetric observations and simulations. In Section 5.4, we
will introduce the subject of remote sensing using reflected polarization. As a unique
instrument in the NASA A-Train satellite constellation (see Figure 5.6), the POLDER on
board PARASOL is a passive optical imaging radiometer and polarimeter that measures I,
Q, and U components of the Stokes parameter at three wavelengths with up to 16 viewing
angles for each pixel (Deschamps et al. 1994). To investigate the propriety of using ice
cloud optical property models for cloud retrieval, the Level-2 POLDER cloud product was
used, based on Level-1B radiances averaged over a larger spatial area (18 km× 18 km).
Included in this product is the percentage of cloud cover, surface type, cloud thermodynamic
6 10
Observation Density Distribution (x 10 -4)

5
Polarized Reflectivity (x 10 -2) 4 8
3
6
2
1
4
0
-1 2
-2
-3 0
100 110 120 130 140 150 160 170 180
Figure 5.16 Comparison between the POLDER observed polarized reflectivities and theoretical
simulations as a function of the scattering angle using MODIS C5 and C6 ice cloud single-scattering
properties depicted in Figure 5.13 (courtesy of Souichiro Hioki).
phase (ice, water, or mixed), and the normalized polarized radiance at 0.865 µm (Lnmp ),
which is defined as follows (C.-Labonnote et al. 2001):
πLp cos θs + cos θv
Lnmp = , (5.2.11b)
Es cos θs
where the subscripts s and v represent the solar zenith angle and the viewing angle,
respectively; Es is the TOA solar irradiance; Lnmp is normalized by the incident solar
irradiance so that it is a dimensionless quantity; and Lp is the linearly polarized radiance,
defined by

Lp = ± Q 2 + U 2 , (5.2.11c)
where the sign is determined by the angle between the polarization vector and the normal
to the scattering plane.
On the basis of the preceding discussions, Figure 5.16 shows a comparison between
the POLDER observed polarized reflectivity and the computed theoretical simulation as
a function of the scattering angle in the range 100–180°, the latter employing MODIS
C5 and C6 single-scattering models in the calculations. MODIS C6 produces statistical
consistency between observations and simulations in the domain of polarized reflectivity
shown by the red area (observations) and the red line (simulations), while the green area
and line correspond to C5.
In view of the preceding MODIS C5 and C6 model comparison, differences in the
retrieved optical depth appear to imply that the complexity and intricacy of the evolution
of ice crystal morphology which undergoes diffusion and accretion processes in the atmo-
sphere could be largely represented by a single ice crystal shape (or habit or morphology),
namely rough aggregates for global application. However, it seems that despite the consis-
tency in ice cloud optical depth retrieval between the use of different spectral bands substan-
tial variation in global retrieval of ice cloud optical depth and mean effective ice crystal size
could occur, considering that the MODIS retrieval algorithm involves complex decision-
making processes in order to obtain final cloud products. Of course, further research and
development are required to clarify the role of MODIS C6 products. At the time of this
writing (August 2015), these have yet to be finalized for public release, in the development
and presentation of global cloud climatology in conjunction with advances in contemporary
climate models for projection and prediction of climate and anthropogenic climate change.
5.2.5 Global mapping of cirrus clouds

In the following subsections, we present three cirrus cloud climatologies determined from
MODIS (King et al. 2013), AIRS (Kahn et al. 2014), and CERES-2 (Minnis et al. 2011b)
as cited in Subsection 1.2.2.
5.2.5.1 MODIS
The global distribution of clouds and their physical and optical properties have been derived
from MODIS since February 24, 2000, when the first analysis of Terra/MODIS data became
available. The MODIS atmosphere team has developed algorithms for cloud-top pressure,
cloud-top temperature, thermodynamic phase, as well as cloud optical thickness, effective
radius, and water path of liquid water and ice clouds (Platnick et al. 2003, Menzel et al.
2008, Kokhanovsky et al. 2011). Cloud optical thickness (or depth) and mean effective
radius are the two major cloud parameters determined in the cloud processing chain. The
basic physical principle behind the simultaneous retrieval of these two cloud parameters as
discussed previously is the bi-spectral solar reflectance method described by Nakajima and
King (1990). Implementing this method on a global basis and under all cloud conditions
has resulted in a number of practical enhancements and significant features.
Among more significant updates, the operational Collection 5 algorithm, in comparison
with previous versions (Collection 4, King et al. 2003), adopted detailed ice crystal
size/habit distribution models based on in situ measurements and corresponding ice optical
property library calculations (Baum et al. 2005a). This ice optical property library has
been developed for a set of ice particles that includes droxtals, hexagonal plates, hollow
columns, solid columns, three-dimensional bullet rosettes, and aggregates. The methods
generally involve a combination of the improved geometric-optics approach and the finite-
difference time domain method (Yang and Liou 1996a, 1996b). The geometric details for
various ice-particle habits can be found in Yang and Liou (1998a). Droxtals, representing
quasi-spherical ice particles, are described further in Yang et al. (2003a) and Zhang et al.
(2004). Details of the ice crystal size and shape distributions and associated phase function
and single-scattering properties have been presented in Subsection 5.2.4 and Figure 5.13.
Cloud Optical Depth Cloud Effective Radius
(a) December-February
(b) March-May
(c) June-August
(d) September-November
Cloud Optical Depth Cloud Effective Radius (μm)
Figure 5.17 Global ice cloud optical depth (0–32) maps (left-hand column) and ice crystal mean
effective radius (De 2, 16–32) (right-hand column) for winter, spring, summer, and autumn months,
derived from MODIS/Aqua datasets averaged over the period from 2002 to 2011 (after King et al.
2013, a rearrangement of their Figures 11 and 13).
Figure 5.17 shows the seasonal mean ice cloud optical depth (left-hand column) and
ice crystal mean effective radius (De 2, right-hand column) from MODIS/Aqua datasets
averaged over the period from 2002 to 2011. These optical depth results clearly show
that clouds over land, regardless of whether they are composed of liquid water or ice, are
generally much more optically thick than clouds over the ocean. Ice clouds over the tropical
ocean associated with the ITCZ have large optical depth associated with their convective
core and convergence, as do deep convective clouds over land in the Congo basin and the
Amazon. Ice clouds in the southern ocean surrounding Antarctica also exhibit large values
of cloud optical depth, particularly between March and August. The western tropical Pacific
has a large cloud fraction composed primarily of ice clouds with relatively low cloud optical
depth due to the preponderance of cirrus anvil clouds in this region. The mean effective
radius of ice clouds shows far less spatial variability when compared with the mean effective
radius of water clouds. Also, ice clouds appear to have a greater tendency to contain larger
particles over ocean than over land, particularly in tropical latitudes. Ice crystals also tend
to be smaller at the top of deep convective clouds. The range of ice crystal sizes in global
cloud maps is generally much smaller than that of water droplet sizes in water clouds.
5.2.5.2 AIRS
Numerical simulations (Chung et al. 2000, Huang et al. 2004) show that cirrus cloudy
radiances in the 800–1130 cm−1 thermal IR window are sufficiently sensitive to variations
in cirrus optical depth (τ c ), small ice crystal mean effective diameter (De < 50 µm), and ice
crystal shape, if appropriate shape distribution models are selected a priori for analysis. A
number of cirrus cloud remote sensing programs based on the Atmospheric Infrared Sounder
(AIRS; see Subsection 5.1.5) thermal IR window data have been developed by Wei et al.
(2004), Li et al. (2005a), Yue and Liou (2009), and Clarisse et al. (2010). Using AIRS data,
a thermal IR radiative transfer model has been constructed for specific application to the
remote sensing of thin cirrus clouds (Yue et al. 2007). The thin cirrus thermal IR radiative
transfer model was based on the delta-four-stream (D4S) approximation (Liou et al. 1988,
Fu and Liou 1993), which accounts for ice crystal scattering to facilitate high-spectral-
resolution remote sensing of cirrus cloud optical depth in AIRS data. It is combined with
the operational Optical Path Transmittance (OPTRAN) model (McMillin et al. 1995) and
a thin cirrus cloud parameterization, using a number of observed ice crystal size and shape
distributions.
All the above AIRS cirrus cloud retrieval programs are computationally time-consuming.
A new approach has been developed by Ou et al. (2013) to facilitate more rapid cirrus
cloud retrieval, based on a k-coefficient approach (note that this differs from the correlated
k-distribution method for sorting absorption lines, see Section 6.2). It uses AIRS cloud-
cleared radiances at the AIRS/Advanced Microwave Sounding Unit (AMSU) resolution as
input. The k-coefficient approach follows the heritage of the two-channel thermal IR window
(TIRW, 3.7 and 10.9 µm) correlation method developed by Ou et al. (1993), which was
applied to AVHRR data and validated using collocated in situ and ground-based lidar and
radar measurements (Ou et al. 1995, 1999; Liou et al. 2002b). Kahn et al. (2005) presented
an approach to cirrus detection at nighttime which they applied to AIRS and AMSU.
Their approach uses two thermal IR 3.8 and 10.4 µm brightness temperature differences
and total column perceptible water measurements. The two-channel method has also been
implemented in the S-NPP cloud retrieval chain for inferring both daytime and nighttime
cirrus cloud parameters (Ou et al. 2003, 2005b; Hutchison et al. 2006; Wong et al. 2007).
In this approach, the cirrus cloudy radiance in the thermal IR window region is expressed
as the sum of cloud emission and the transmitted below-cloud upwelling radiance, which
can be approximated by cloud-cleared radiance due to negligible above-cloud water vapor
emission/absorption.
The k-coefficient approach has been tested against a more accurate retrieval program, the
Stand-Alone clear-radiance Radiative Transfer Algorithm (SARTA) (Strow et al. 2003a)
that combines the D4S program with an accurate and computationally efficient radia-
tive transfer model. The SARTA is a narrow-band radiative transfer program, which also
includes surface and atmospheric emission/absorption, as well as surface reflected thermal
and solar radiation terms. This program was specifically designed to simulate clear-sky
upwelling radiance at TOA for any given AIRS channel and atmospheric profile and effec-
tively parameterized atmospheric transmittances in 100 pressure layers using AIRS spectral
response functions measured during pre-launch tests (Strow et al. 2003b).
The D4S model accounts for interaction between scattering and absorption/emission
within each cirrus cloud layer based on prescribed ice crystal single-scattering properties.
The most updated high-resolution ice crystal microphysics model was used to execute
D4S, along with thermal IR optical properties developed by the MODIS algorithm group
(Baum et al. 2007). The TOA AIRS spectral radiances for skies containing cirrus were
evaluated by convolving clear-sky transmittances from SARTA and cloudy layer transmit-
tances/reflectances from D4S. The combined SARTA and D4S radiative transfer program is
an accurate method to compute AIRS cirrus cloudy spectra for remote sensing application
and has been incorporated into the AIRS cloud retrieval algorithm to produce the version 6
cloud product.
In what follows we shall introduce the k-coefficient cirrus cloud retrieval program
developed by Ou et al. (1993), which was modified for application to multiple channels.
The upwelling radiance at TOA can be expressed for the six selected “clean” (with minimal
water vapor absorption/emission) channels in terms of the cirrus cloud-top temperature Tc ,
and emissivity εn , as follows:
Rn = (1 − εn )Ran + εn Bn (Tc ) , n = 1 − 6, (5.2.12)
where Ran denotes upwelling radiance reaching the cloud base for the six spectral bands
and Bn (Tc ) is the respective Planck function at Tc . The first term on the right-hand side of
Eq. (5.2.12) represents the contribution of transmitted radiance from below the cloud, while
the second term denotes the emission contribution from the cloud itself. The emission
by water vapor above the cirrus cloud was neglected. The effect of cloud reflectivity,
which is generally less than 3% of the incident radiance based on exact radiative transfer
calculations, was also neglected. Equation (5.2.12) is best applied to “clean” thermal IR
window wavelengths, where the cloud-base upwelling radiance can be approximated by
the TOA clear radiance because of negligible above-cloud absorption/emission. For each
selected channel n, we have
εn = 1 − exp(−kn τc ). (5.2.13)
The exponential term represents the effective transmissivity. The parameter kn represents
the effective extinction coefficients for the six channels, accounting for the effects of
multiple scattering within cirrus clouds and for ratios between visible and IR extinction
coefficients. This parameter is a function of both τ c and De and, generally, kn < 1 because of
multiple scattering effects (Ou et al. 2003). Thus, the products kn τ c can be considered as the
effective optical depth that would yield the same emissivity values under pure absorption
conditions.
By combining Eqs. (5.2.12) and (5.2.13), we obtain

Rn − Bn (Tc )
− ln = − ln (1 − εn ) = kn τc . (5.2.14)
Ran − Bn (Tc )
We then define the residual (cost function) which is to be minimized in order to solve for
De and τ c as follows:

6
) *2
χij = − ln (1 − εn ) − knij τci , (5.2.15)
n=1
where subscripts i and j denote the indices for reference τ c and De , respectively. Ou et al.
(2003) have performed a number of sensitivity studies on the accuracy of retrieved cirrus
cloud parameters.
The radiative transfer simulations of ice cloud brightness temperature spectra using the
D4S model have been incorporated in the new AIRS version 6 cloud product retrieval
algorithms (Kahn et al. 2014). The global ice cloud properties for January 2007 determined
from the AIRS version 6 cloud products are shown in Figure 5.18. The optical depth
distribution in the tropical western Pacific Ocean comprises a narrow band of high values
compared with ice cloud frequency. The highest optical depth values are associated with
the convective band closest to the ITCZ, which is surrounded by a wide latitudinal extent
of cirrus with lower optical depth values. Furthermore, large optical depths are shown in
the boreal winter oceanic storm track, but are greatly reduced in the Arctic region as well as
over the cold East Asian and North American continental regions. Very low optical depth
values dominate subtropical subsidence regions.
The De distributions are complex and have a broad minimum in the Arctic region,
with low values extending southward over East Asia and North America, corresponding
closely to the optical depth. A minimum is seen over the tropical western and central
Pacific Ocean, especially on either side of the ITCZ where thin cirrus is most common. A
small maximum in De along the ITCZ is consistent with large MODIS-derived re values
within deep convective activities in the tropics (Yuan and Li 2010). The retrieval of re from
Figure 5.18 Ice phase frequency (0–1), optical depth τ c (0–4), ice crystal mean effective size, De (10–70 µm), and ice cloud-top temperature Tc
(190–310 K) are presented for January 2007 (after Kahn et al. 2014).
surface-based radar and lidar observations also shows large values in profiles associated
with more convectively active regions (Protat et al. 2011). Maximum values of De are
obtained on the subtropical side of midlatitude jets in both hemispheres over oceans. There
is a decrease in De poleward across both storm tracks. Also, some structure is seen in
ice cloud frequency and De associated with orographic effects over coastal Antarctica, the
Andes, and other regions of elevated topography. A much larger dataset spanning several
years and all seasons is necessary to quantify the behavior of orographic clouds and relate
their properties to static stability, vertical velocity, and flow regime (Joos et al. 2010).
5.2.5.3 CERES-2
The NASA-CERES (Clouds and the Earth Radiant Energy System) project was designed to
improve current understanding of the relationship between clouds and solar and longwave
radiation. This is achieved by using space-borne broadband radiometric instruments to map
TOA radiation fields with coincident data from space-borne narrow-band imagers employed
to retrieve the properties of clouds associated with those fields. This subsection describes
the CERES Edition-2 cloud property retrieval system used to analyze data from the TRMM
Visible and Infrared Scanner (VIRS) and by the MODIS instruments on board both Terra and
Aqua satellites covering the period from 1998 through 2007. The daytime Visible Infrared
Shortwave-infrared Split-window Technique (VISST) for snow-free surfaces, along with
ancillary data and parameterizations of the cloud optical properties, are used to derive cloud
boundaries, phase, optical depth, effective particle size, and condensed/frozen water path
at both pixel and CERES footprint levels (Minnis et al. 2011a, 2011b).
The smooth-solid hexagonal column ice crystal distributions described by Minnis et al.
(1998b) were used to compute reflectance look-up tables for VIRS channels 1–3 (0.63, 1.6,
and 3.75 µm bands) and coefficients used in the emittance parameterizations for channels
3–5 (3.75, 10.8, 12.0 µm bands). For solar channels, the same optical properties listed in
Minnis et al. (1998b) were used in the adding/doubling radiative transfer model to develop
higher angular resolution look-up tables: 21 regularly spaced (0.05 intervals) µo and µ
nodes between 0 and 1, and 24 φ nodes with higher resolution near the extreme. The
adding/doubling radiative transfer model used 350 Legendre polynomials and 120 Fourier
terms to de-convolve the water droplet and ice crystal scattering phase functions. The
reflectance look-up tables cover the solar optical depth τ range from 0.25 to 128 for droplet
effective radii re between 2 and 32 µm and ice crystal effective diameters De between 6 and
135 µm.
The same angular resolutions and de-convolutions for solar bands were also used for
near-IR and shortwave IR calculations, where the optical properties of ice were com-
puted using geometric-optics ray-tracing results as in Takano and Liou (1989a), covering
the same optical depth range as solar models. In addition, cloud absorptance was com-
puted as a function of particle size, optical depth, and the solar zenith angle. Integration
over the solar zenith angle was performed to obtain diffuse solar absorptance. The scat-
tering phase functions were based on geometric-optics scattering computations for the
Ice Cloud Optical Depth
(a) Terra (b) Aqua
Ice Cloud Effective Diameter
(c) Terra (d) Aqua
Figure 5.19 Global distribution of mean daytime ice cloud optical depth: (a) Terra (2000–2007) and (b) Aqua (2002–2007); and ice cloud
mean effective diameter: (c) Terra (2000–2007) and (d) Aqua (2002–2007) (after Minnis et al. 2011b, a rearrangement of their Figures 3
and 7).
spectral-response-weighted indices of refraction. The sub-band values were integrated to

obtain a band average that was used to compute a normalization factor relative to the band
average for original ice crystal calculations. The ice sphere sub-band values were then
adjusted with a normalization factor to obtain the sub-band ice crystal optical properties,
while reflectances were computed for each sub-band using the adding/doubling model.
The reflectances were computed separately for each 0.1 µm sub-band between 3.5 and
4.0 µm. These sub-band reflectances were then integrated over the spectral response func-
tion and weighted by the TOA incoming solar radiances (Kurucz 1995) to obtain a single
reflectance for the band. The shortwave IR reflectance look-up tables are limited to optical
depths smaller than 32, since the reflectances asymptotically approach their maximum val-
ues at smaller optical depths for all considered particle sizes. The VIRS channel-3 effective
emittance was determined in a similar manner using the same set of sub-bands. In this
instance, the blackbody curve of cloud temperature substitutes for the solar spectrum to
weight the absorption and emission computed for each sub-band in the final integration over
the response function. The results were used to compute the model effective emittance for
each particle size, optical depth, and pairs of surface and cloud temperatures and to develop
emittance parameterizations as shown in Minnis et al. (1998b), along with a number of
parameterizations for solar reflectance and IR brightness temperature.
Figure 5.19 shows global distributions of mean daytime ice cloud optical depth and ice
crystal mean effective diameter for Terra and Aqua. The distribution patterns of each param-
eter are largely similar between Terra and Aqua. Ice cloud optical depths are particularly
prominent over the ITCZ and midlatitude storm tracks over northeastern Asia and America
coastal areas and north of Antarctic areas. Lower values of ice cloud optical depths are
noted over subtropical subsidence zones, particularly over southwestern Africa and western
South America. For Terra, ice cloud mean effective diameters are between 60 and 70 µm
over midlatitude storm tracks, northeastern Asia and America coastal areas and north of
Antarctic areas; between 40 and 50 µm over the ITCZ and between 30 and 40 µm over
spotty pocket areas in desert regions. Ice cloud mean effective diameters are smaller for
Aqua than for Terra over midlatitude storm areas, but larger over continental ITCZ areas,
namely the South American Amazon and Central African tropical forests. Such differences
can be produced by different local overpass times. Terra overflies in local late morning,
while Aqua overflies in local afternoon. Over storm track areas, solar heating may cause
ice crystals to sublime, resulting in a reduction in ice crystal sizes. Over the ITCZ, where
thunderstorm activity is stronger in the afternoon, ice crystal sizes are shown to increase.
5.2.6 Detection of thin cirrus

The presence of thin cirrus clouds with optical depths less than 0.1 is an important factor
which could be a critical parameter in global and regional thermal IR radiation balance,
with consequences for climate and climate change. In Subsection 5.2.5, we introduced three
global cirrus cloud climatologies presented by King et al. (2013), Minnis et al. (2011b),
and Kahn et al. (2014). However, the detection of visual and subvisual thin cirrus clouds is
a subject of great uncertainty. For this reason, it appears that past cirrus cloud climatologies
may have underestimated thin cirrus cloud cover. Lee et al. (2009b) pointed out that the
net radiative forcing of subvisual cirrus is on the order of 1–2 W m−2 , which has not been
accounted for in contemporary climate models.
Using a low-resolution CO2 slicing method, Wylie et al. (1994) estimated that thin
cirrus (τ < 0.7) covers about 20% of the midlatitude region and over 50% of the trop-
ics. Perhaps these values may overestimate cirrus cloud climatology due to low resolution
datasets. Nevertheless, cirrus clouds occur at all latitudes and their global fractional cover-
age remains in question. To investigate the possibility of detecting thin cirrus clouds using
high-resolution images from spectroradiometer channels, Gao et al. (1993) showed that
the reflectance of the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) 1.38 µm
channel, located in a strong water vapor absorption band, is particularly sensitive to the
presence of high-level cirrus clouds. When a significant amount of water vapor is present
in the atmosphere, the radiation at this wavelength is completely absorbed within the
lower levels. Consequently, only reflection off mid-level and high-level clouds can reach
satellite sensors, and strong reflectance in this channel can be entirely attributed to upper
tropospheric clouds. The use of the 1.38 µm band has been identified as a viable cir-
rus cloud detection tool (Gao and Kaufman 1995, Gao et al. 2003). In fact, Gao et al.
(2002a) developed an operational retrieval product for global cirrus reflectance. Hutchi-
son and Choe (1996) also showed that the use of 1.38 µm reflectance thresholds from
AVIRIS data significantly improved the accuracy of identification of thin cirrus clouds over
methods based on the Advanced Very High Resolution Radiometer (AVHRR) data for all
types of ground surfaces. Lastly, King et al. (1996) pointed out that the reflectance of the
MODIS airborne simulator (MAS) 1.9 µm water vapor absorption channel, a surrogate for
the 1.38 µm channel, is also sensitive to the presence of thin cirrus clouds in the same
manner.
On a related subject, high cloud contamination remains a major problem and a
leading source of error when retrieving aerosol and surface properties from satellites
(Mishchenko et al. 1999). Accurate cloud screening to determine cloud-free pixels is
essential. Techniques using spatial variance tests differentiate clouds from aerosols over
the oceans (Coakley and Bretherton 1982, Martins et al. 2002). However, these techniques
often fail to detect thin homogeneous cirrus and cannot be employed safely over land sur-
faces due to large variation in albedo. In view of the above, the use of the 1.38 µm band
has been recognized as an effective cirrus cloud detection approach. Moreover, the 1.38 µm
reflectance can be used in a ratio with a non-absorbing band at 1.24 µm to discriminate thin
cirrus from aerosols (Gao et al. 2002b).
Roskovensky and Liou (2003a) showed that the approach developed by Gao et al.
(2002a) may not be sufficiently robust to detect thin cirrus with optical depth less than
0.5 and may produce ambiguity in differentiating clear ocean pixels from those of thin
cirrus. To complement the preceding approach, they developed a new thin cirrus detection
scheme using the 1.38 µm0.65 µm reflectance ratio (RR) and the 8.6 and 11 µm brightness
temperature difference (BTD). Note that MODIS contains both 8.6 and 11 µm thermal IR
channels not displayed in Figure 5.7. Since each test is sensitive to a particular cirrus
trait – 1.38 µm reflectance to the altitude and thickness and the BTD to the ice phase – the
two tests work together well to produce a more powerful cirrus detection parameter. At
the same time, both tests can also be sensitive to dust in a dry atmosphere. The 0.65 µm
band was selected over the 1.24 µm band because, when used in ratio with the 1.38 µm
band, it can differentiate cirrus from aerosols more clearly over the ocean, where 1.24 µm
reflectance values can be exceedingly small. Roskovensky and Liou (2003a) performed
theoretical calculations to investigate the sensitivity of this new approach and used MODIS
data over both a land and an ocean ARM site where mm-wave radar data were available
to verify satellite detection results. A high cloud screening parameter P which combined
the 1.38 µm0.65 µm ratio (RR) and the 8.6–11 µm difference (BTD) was developed in the
form P = exp[RR × A + (BTD − B)], where the coefficient A is a scaling factor and the
coefficient B is the BTD offset. Employing this technique with A and B values determined
for thin-cirrus–dust conditions based on radiative transfer simulations, thin cirrus clouds
with optical depths smaller than 0.2 can be detected. In Figure 1.7, we presented a case
study on the detection of cirrus clouds with validation from mm-wave radar reflectivity
data.
Roskovensky and Liou (2003b, 2005, 2006) and Roskovensky et al. (2004) have
further developed simultaneous detection schemes to improve separation of thin cirrus
and aerosols by using the MODIS 1.38 µm reflectance as the primary tool in detecting
high ice clouds and BTD (8.6 and 11 µm) as a second, independent test to reinforce
or negate the 1.38 µm reflection results. These developments encompassed ground-based
validation datasets primarily available from ARM observation sites, including mm-wave
radar measurements at the time. As an example, we re-plotted the results presented by
Roskovensky and Liou (2003b, Figures 7, 8, and 10) using RR (1.380.65 µm) to replace
1.38 µm as a parameter for detection purposes, as originally proposed by Roskovensky
and Liou (2003a). Figure 5.20 displays scatterplots of RR as a function of (a) BTD and
(b) 0.65 µm reflectance on the basis of cloudy pixels inside a rectangular region defined
near the mm-wave radar for the three ARM cases obtained from the tropical western Pacific
(TWP) site and three scenes over the southern Great Plains (SGP) site, which represent
three different cloud regimes: clear-low (green), mid-level cloud (dark red), and thin cirrus
(blue). Points from each scene appear grouped together and are well separated from each
other so that cloud types are easily distinguished.
In Figure 5.20a, the horizontal dashed line at 0.12 represents the separation between
thin cirrus and mid-level clouds and clear/low clouds, while the vertical dashed line at
−1 denotes the separation between mid-level clouds and clear/low clouds. These points
are re-plotted in the domain of RR versus 0.65 µm reflectance in Figure 5.20b. The hor-
izontal line remains unchanged, but the vertical line at 20 separates mid-level clouds
and clear/low clouds more distinctly. Of course, additional research and development are
(a) (b)
0.4 0.4
+ Thin Cirrus
+ Mid-Level Cloud
+ Clear/Low
1.38/0.65 μm Reflectance Ratio
0.3 0.3
0.2 0.2
0.1 0.1
0.0 0.0
-2 0 2 4 6 0 20 40 60 80
8.6–11 μm BTD (K) 0.65 μm Reflectance (%)
Figure 5.20 The 1.38 µm0.65 µm reflectance ratio is plotted against both (a) the 8.6–11 µm bright-
ness temperature difference and (b) the visible reflectance for three clear and cloudy scenes over the
ARM TWP and SGP sites in March, April, May, and August 2001. Blue crosses represent thin cirrus
pixels with estimated optical depths between 0.1 and 0.9. Mid-level cloud (cloud top 4 km) pixels
are shown as dark red crosses, and clear sky and low cloud pixels as green crosses.
required to use this detection for global application. Moreover, in Figures 1.8 and 1.9, we
presented the capability of CALIOP/CALIPSO to determine cirrus high-resolution verti-
cal profiles, although the special coverage is limited due to its push-bloom satellite track
mode.
5.3 Remote sensing of cirrus cloud vertical size profile

5.3.1 The importance of cirrus cloud vertical profiles
Cirrus clouds have frequently been observed to be finite in nature and to display substantial
horizontal variability (Minnis et al. 1993, Mishchenko et al. 1996). On the basis of the
time series of backscattering coefficients derived from lidar and radar returns, Spinhirne
and Hart (1990), Sassen (1991), and Mace et al. (2002) also illustrated significant vertical
inhomogeneity in terms of extinction coefficient and ice water content, largely related to
vertical temperature distribution (Ou and Liou 1995). Miloshevich and Heymsfield (1997)
and Heymsfield and Iaquinta (2000) analyzed in situ balloon-borne ice crystal replicator
data and showed that the variation of ice crystal size and shape distributions within midlat-
itude cirrus clouds tend to be stratified: small non-spherical or quasi-spherical ice crystals
are predominant in the upper layer, pristine single ice crystals and bullet rosettes with
5.3 Cirrus cloud vertical size profile 289
well-defined hexagonal shapes are prominent in the middle layer, and the bottom layer
mainly contains larger but irregularly shaped aggregates formed by collision and subli-
mation processes of ice crystals (Figure 5.21a; see also Figure 1.18). Moreover, the mean
cloud particle size increases from cloud top downward, but sharply decreases to near-zero
values at cloud base.
The inhomogeneous nature of cirrus cloud composition has been shown to cause spatial
variation in atmospheric radiative heating/cooling rates. By conducting numerical experi-
ments using the UCLA (University of California, Los Angeles) general circulation model,
Gu and Liou (2001, 2006) and Gu et al. (2003) showed that the vertical profile of mean
effective ice crystal size (De ) determined radiative heating/cooling rate profiles within cirrus
clouds and surrounding clear areas, which would have a substantial impact on tempera-
ture horizontal and vertical gradients and consequent general circulation. A comprehensive
knowledge of the spatial distribution of De would have the potential to improve the repre-
sentation of cirrus clouds in weather and climate models.
The current MODIS cloud retrieval algorithm simultaneously retrieves the total cloud
optical thickness (τ c ) and a vertically averaged mean effective radius (re ) or size (De ) using
observed solar reflectance/radiance at one visible and a number of near-IR window bands, as
demonstrated in Figure 5.17, assuming a vertically homogeneous composition. This is based
on the physical principle that reflectances at non-absorbing (for cloud particles) visible and
absorbing near-IR wavelengths depend primarily on τ c and re , respectively. Although
this approach is straightforward, it cannot resolve the vertical structure of cirrus clouds.
Consequently, for any cloudy pixel, re values inferred from individual near-IR bands are
different from one another (Baum et al. 2000a). Platnick (2000) explained these differences
in retrieved water cloud re by theorizing that differences in maximum photon penetration
levels for individual near-IR bands cause the reflectance of stronger-absorbing bands to be
primarily sensitive to cloud-top properties, and that of weaker-absorbing bands to carry
more information about the cloud’s deeper structure. Based on this theory, Chang and Li
(2002) used three MODIS near-IR band reflectances to infer the vertically linear profile
of re in low-level non-precipitating stratus clouds. These authors assessed the performance
of the retrieval method by comparison with in situ observations of microphysical vertical
profiles and showed that the retrieved linear re profiles capture the vertical trend of re
variation. The feasibility of the vertical profiling of ice crystal sizes using MODIS channels
has also been explored by Yang et al. (2001). It appears that we can use the information
content of MODIS visible and near-IR band reflectances to infer the vertical profile of ice
crystal De from pixel levels based on the unique differences in cloud absorption strength
among these bands.
Subsection 5.3.2 presents a cirrus cloud vertical sizing algorithm for inferring cloud-
top and cloud-base effective sizes (Dt , Db ) and τ c from MODIS 0.645, 1.64, 2.13, and
3.75 µm band reflectances, using a successive minimization method based on a look-up
library of pre-computed reflectances from the adding/doubling radiative transfer program,
subject to corrections for Rayleigh scattering at the 0.645 µm band, above-cloud water
vapor absorption, and 3.75 µm thermal emission. The algorithmic accuracy and limitations
(a) (b)
14 0
0.1
De1 = De2 = 42 μm
13 0.2
0.3
12 -60°C
0.4
11 0.5
Optical depth within cloud

0 0.1 0.2 0.3 0.4 0.5
0
10
Height (km)
0.5
-40°C l = 1.64 μm
9 1.0 l = 2.13 μm
l = 3.75 μm
1.5
8
2.0
0 0.1 0.2 0.3 0.4 0.5
7 0
-20°C
1
2
6
3
200 μm 4
5 5
6
0°C 7
4
0 10 20 30 40 50 60 70 80 90 100 8
0 0.1 0.2 0.3 0.4 0.5
Relative Humidity (%) Vertical weighting functions
Figure 5.21 (a) Ice crystal size and shape as a function of height and relative humidity, captured by a replicator balloon sounding system over Marshall,
Colorado on November 10, 1994 (see also Figure 1.18). (b) Weighting functions for the MODIS 1.64 µm (blue), 2.13 µm (green), and 3.75 µm (red)
channels, corresponding to optically thin (upper panel, τ c = 0.5), moderate (middle panel, τ c = 2.0), and thick (lower panel τ c = 8.0) cirrus clouds.
of the retrieval method have been investigated by synthetic retrievals conditional on instru-
ment noise and perturbation of input parameters (Wang et al. 2009).
5.3.2 Development of a retrieval algorithm for vertical sizing

We first define the cloud-top and cloud-base effective sizes (Dt , Db ) as follows:

Lmax ,M Lmax
M
Dt,b = (3/2) Vh (L) nt,b (h, L) dL Ah (L) nt,b (h, L) dL ,
h=1 Lmin h=1 Lmin
(5.3.1)
where Ah (L) and Vh (L) are the projected area and volume of a particle with habit index h
and maximum dimension L, respectively, and nt (h, L) and nb (h, L) are ice crystal size and
shape distributions at cloud top and base, respectively.
To illustrate the relative contributions from different layers within the cloud to the total
reflectance/radiance at the top of the atmosphere, Figure 5.21b shows weighting functions
for all three near-IR bands of a model cirrus cloud with τ c = 0.5, 2.0, and 8.0, and De
= 42 µm. These weighting functions are defined as [dR(τ )/dτ ]/R(τc ), and were obtained
based on the maximum-photon penetration principle described by Platnick (2000). Due to
the stronger ice absorption in the 3.75 µm band, its weighting function peaks at a higher
level than the other two weighting functions, implying that the 3.75 µm reflectance is more
sensitive to cloud-top properties. On the other hand, the shapes of the weighting functions
for the 1.64 and 2.13 µm bands are similar, implying that only one piece of information
is available. Therefore, in the retrieval algorithm, the vertical profile of De is constrained
as a linear function of optical depth τ , because there are just two independent pieces of
information available, which can only be used to retrieve the two sizes Dt and Db .
The retrieval algorithm follows a numerical iterative approach to search for a converged
solution for Dt , Db , and τ c , subject to minimal differences between simulated and observed
reflectances for all four bands (including 0.645 µm). First, the adding/doubling radiative
transfer model is used to build a library of cloud-only radiative parameters for all combina-
tions of reference values of Dt , Db , and τ c . For each pixel identified as cirrus, reflectances
for the four bands and the viewing-geometry data are extracted from MODIS L1B products.
Required input parameters, including cloud-top temperature (Tc ), cloud-top pressure (Pc ),
surface albedo (Ag ), and surface temperature (Tg ), are extracted from MODIS data products.
A comprehensive set of cloud-only reflectances for a combination of reference Dt , Db , and
τ c are then computed and tabulated. Various pixel-dependent correction schemes are then
applied and a look-up library of total reflectance for this particular pixel is built for the
retrieval of cirrus parameters.
The numerical iterative search program used a successive minimization method [see
Eq. (5.2.8)] as follows: (1) prescribe initial values for Dt and Db and retrieve and update τ c
based on minimization of the 0.645 µm reflectance residual; (2) retrieve and update Dt by
minimizing the 3.75 µm radiance residual using updated values for τ c and Db : (3) retrieve
Db by minimizing 1.64 and/or 2.13 µm reflectance residual employing updated τ c and Dt

values. If all of the most recently updated Dt , Db , and τ c values differ from their respective
values from the previous iteration by less than prescribed thresholds, or the number of
iterative steps reaches a prescribed limit, the numerical procedure terminates and the last
set of Dt , Db , and τ c values is the retrieval solution. Otherwise, the iteration continues until
convergence is reached, usually within 3–4 steps.
For any given cirrus cloud pixel identified by the MODIS Cloud/Phase Mask programs,
the retrieval algorithm uses a look-up library of reflectance at the four spectral bands to
retrieve Dt , Db , and τ c . The generation of the look-up library is based on the ice crystal
size and habit distribution models developed by Baum et al. (2005a). These models were
based on 1100 in situ measurements of ice crystal habit distributions obtained during
several cirrus cloud field campaigns, including FIRE-I, FIRE-II, ARM Cloud IOP, CEPEX,
and CRYSTAL-FACE (see Sections 1.6 and 1.7), which were the most comprehensive
set of size and habit distribution models available for application to the satellite remote
sensing of cirrus clouds. This set of observations should represent ice crystal size and habit
distributions for a variety of cirrus clouds observed by MODIS and other A-Train satellites.
Each ice crystal size distribution was fitted to a gamma function whose parameters were
obtained by minimizing the difference between computed and measured IWC. Reference
Dt and Db both vary from 10 to 180 µm with an interval of 10 µm. A composite ice crystal
size distribution was obtained for each size bin, based on fitted gamma function parameters.
The prescribed ice crystal habit distribution model varies with the maximum dimension
(L), taking into account the stratified structure of in situ cirrus particle shapes and sizes
observations. For L < 60 µm, these small particles are composed of 100% droxtals. For
60 µm < L < 1000 µm, a mixture of 15% 3-D bullet rosettes, 50% solid columns, and
35% plates is prescribed. For 1000 µm < L < 2500 µm, a mixture of 45% hollow col-
umns, 45% solid columns, and 10% aggregates is prescribed. Finally, for 2500 µm < L <
9500 µm, these large particles are composed of 97% bullet rosettes and 3% aggregates (see
Figure 5.13a).
The bulk single-scattering properties associated with each reference Dt and Db were
based on archived single-scattering properties for individual wavelength and ice crystal
maximum dimension and habit, which were computed using the finite-difference time
domain method and the modified geometric-optics method (Yang et al. 2000a, 2000b;
Baum et al. 2005b). The prescribed reference τ c values are 0, 0.5, 1, 2, 4, 8, 16, and
32. Thus, for each spectral band and for a particular set of sun–sensor geometry, the
dimension of the look-up library is 18(Dt ) × 18(Db ) × 8(τ c ). For global application, the
look-up library will have to be expanded to include the functional dependence of reflectances
on a combination of reference solar and viewing zenith angles and relative azimuthal
angles.
The adding/doubling code specifically designed for simulating radiative transfer in cirrus
clouds was used to assist in the retrieval. For each combination of Dt , Db , and τ c , this code
was used to calculate cloud-only radiative parameters, which include bidirectional reflection
functions (R, R ∗ ) and transmission functions (T, T∗ ) in units of W m−2 sr−1 normalized by
the incoming solar flux; solar-zenith-angle-dependent reflections (r, r∗ ) and transmissions (t,
t∗ ) in units of W m−2 normalized by the incoming solar flux; and non-dimensional spherical
albedo (r̄ ∗ ), where ∗ denotes radiation from below. For a prescribed sun–sensor geometry,
simulated cloud-only properties (R, R∗ ; T, T∗ ; r, r∗ ; t, t∗ ; r̄ ∗ ) were calculated by linear
interpolation between reference directions. Spherical albedo (r̄ ∗ ) is a parameter integrated
over r∗ with respect to the solar zenith angle and is not affected by the viewing geometry.
The observed total reflectance consists of cloud-only reflectance and the multiple reflec-
tion between surface and clouds (Nakajima and Nakajima 1995). In this study, surface
albedo (Ag ) data for the 0.645, 1.64, and 2.13 µm bands were compiled from the MODIS
16-day L3 1 km global albedo data product MOD43B3, and the surface albedo at 3.75 µm
was estimated from the MODIS 8-day global surface temperature/emissivity data product
MOD11C2. Thus, variation in surface albedo has been accounted for, and retrieval errors
due to uncertainties in surface albedo have been minimized.
Water vapor absorption above the cloud is accounted for by assuming the atmosphere
above the cloud is purely absorptive (Platnick et al. 2003). The water-vapor-corrected
reflectance Rw is approximated by the product of cloudy reflectance and transmittances in
the solar incident and satellite viewing directions, where the water vapor absorption optical
thickness for the air mass above the cloud top is calculated from the correlated k-distribution
method by assuming the water vapor profile in the U.S. 1976 Standard Atmosphere.
The Rayleigh scattering effect is more prominent in the 0.645 µm band than in the other
three near-IR bands. Wang and King (1997) showed that neglecting the Rayleigh scattering
correction can cause a significant error in the retrieved τ c for large solar and/or viewing
zenith angles, and developed a parameterization, which was followed in the vertical sensing
program. The corrected bidirectional reflectance R (0.645 µm) is expressed as the sum of
the amount of direct Rayleigh scattering without cloud reflection, the reflection of direct
solar radiation by the cloud, the reflection of direct solar radiation by the cloud subject to
single scattering in the air, and the amount of single scattering in the air subject to cloud
reflection.
For the 3.75 µm band, the thermal emission from both the Earth’s surface and clouds
contributes significantly to the total measured radiance. The total radiance R (3.75 µm)
can be approximated by the sum of the reflected solar radiance, corrected for multiple
reflection and water vapor absorption, and the contribution from surface and cloud emissions
attenuated by above-cloud water vapor absorption. For each cirrus cloudy pixel, the above
corrections are applied to all combinations of reference Dt , Db , and τ c , to create a look-up
library of reflectance/radiance.
To assess the accuracy and limitations of the retrieval approach, we performed synthetic
retrievals, using simulated reflectance/radiance in lieu of measurement data. For each
combination of reference values of the three parameters Dt , Db , and τ c , the simulated
reflectance/radiance was determined from the adding/doubling radiative transfer program,
using interpolated single-scattering properties. Retrieval errors were obtained as differences
between retrieved and input (truth) cloud parameters. Since all reflectance/radiance values
were from simulations based on unperturbed input parameters with zero instrument noise,
0.5
Db =10 μm
tc = 8
30 μm
3.75 μm Radiance [W (m-2 μm-1 sr -1)]
0.4
50 μm
80 μm
110 μm
0.3
140 μm
180 μm
Dt =10 μm
0.2
30 μm
0.1
50 μm
80 μm
110 μm
140 μm
180 μm
0
0 0.1 0.2 0.3 0.4 0.5
1.64 μm Reflectance
Figure 5.22 Correlation between 1.64 µm reflectance and 3.75 µm radiance for selected combina-
tions of Dt and Db depicted in the figure, using a cloud optical depth τ c of 8 (after Wang et al.
2009).
the retrieval errors that were produced indicate the accuracy of the numerical iteration
process. The conditions under which the algorithm can perform well are thus identified.
For 2 < τ c < 15 and Dt < 80 µm, the retrieval results were found to be reasonably accurate.
The root-mean-square (rms) error of retrieved Dt , Db , and τ c are about 2.4 µm, 2.8 µm,
and 0.004, respectively. These error magnitudes for particle sizes and optical depths are
comparable with the error estimates for inferred τ c and De from the MODIS cloud retrieval
algorithm (King et al. 1997). For Dt > 80 µm, the accuracy of the retrieval algorithm is
limited by the low sensitivity of near-IR reflectance to Dt and Db .
Figure 5.22 displays the correlation of 1.64 µm reflectance and 3.75 µm radiance for
τ c = 8, selected combination of Dt and Db , and a prescribed viewing geometry (µ = 1,
µ0 = 0.82, φ = 180°). Solid and dashed curves denote reflectance/radiance for constant
Dt and Db , respectively. The constant-Dt and constant-Db curves are distinctly separated
for Dt < 80 µm, indicating a good sensitivity of reflectance/radiance to Dt and Db , and
small retrieval errors due to uncertainties in observed reflectance/radiance (Dt 2.6 µm,
Db 3.5 µm). However, for Dt > 80 µm, the constant Dt and Db curves are close
to each other, implying that a small perturbation of reflectance/radiance can cause an

unacceptable error in both retrieved Dt and Db . We estimate that a 5% uncertainty in the
1.64 µm reflectance can lead to errors of 5–20 µm and 20 µm in the retrieved Dt and Db ,
respectively.
Retrieval accuracy also depends on that of the input parameters Ag , Tc , Tg , and Pc , and
the instrument noise of measured reflectance/radiance. Synthetic retrievals with perturbed
input parameters and simulated reflectance/radiance were carried out to investigate the error
budget due to uncertainties in input parameters along with noise in reflectance/radiance data.
The perturbation of input parameters is based on the error specifications for MODIS data
products: Tg = ±1 K, Pc = ±50 hPa, and Tc = ±5 K. For each band, Ag = ±10%,
and the perturbation of simulated reflectance/radiance is based on the design-specified
signal-to-noise ratio. For each perturbation run, only a single input parameter or
reflectance/radiance for a band is perturbed, while all other input parameters and
reflectances/radiances are unperturbed. In this manner, we can identify major error sources
for each retrieved cloud parameter. Based on results of the algorithmic error analysis, syn-
thetic retrievals were carried out for Dt < 80 µm and 2 < τ c < 15. The restriction on τ c is
because retrieval performance deteriorates for either τ c < 2 or τ c > 15.
Most of the rms errors are comparable with error budget estimates of τ c and De for
the cloud optical property retrieval program using solar channels of the Visible-Infrared
Imaging-Radiometric Suites (VIIRS) on board the next-generation Suomi – National Polar-
orbiting Operational Environmental Satellite System (NPOESS) Preparatory Project (S-
NPP) (Ou et al. 2003). Retrievals of τ c are fairly accurate; the maximum rms percentage
errors around 3% are associated with Ag (0.645 µm) = ±10%, indicating that the accuracy
of the visible band surface albedo drives the retrieval performance involving τ c . Retrievals
of Dt and Db are also reasonably accurate, with the largest errors around 10 µm and 14 µm
associated with Tc = ±5 K, respectively. According to the design specification of the
currently operating VIIRS/S-NPP, the uncertainty threshold requirement for Tc is about
3 K, which is less than the MODIS specification of Tc = ±5 K (Menzel et al. 2006). This
requirement has been met on the pixel level by the VIIRS cloud-top temperature retrieval
program (Wong et al. 2007). Given the smaller uncertainty in Tc based on VIIRS/S-NPP
retrieval, errors in retrieved Dt and Db can be reduced by half.
5.3.3 Validation of the vertical sizing algorithm

The retrieval algorithm has been applied to MODIS cirrus scenes over the Atmospheric
Radiation Measurement Program’s Southern Great Plain (ARM-SGP) site, North Central
China, and Northeast Asia. The accuracy of retrieved cloud optical depths and mean
effective sizes was examined by comparison with MODIS cloud products. The quality of
retrieved cloud-top and cloud-base ice crystal effective sizes was evaluated by comparison
with those derived from observations using the ground-based Millimeter Wave Cloud Radar
(MMCR) and the Cloud Profiling Radar (CPR) aboard CloudSat. These assessments show
that retrieved cloud optical depths and mean effective sizes correlate well with MODIS
cloud products. Retrieved mean effective sizes are also well correlated with CPR/CloudSat
derived mean sizes, but are generally smaller in magnitude, because the CPR tends to miss
small cloud particles. A case study is presented below.
On March 6, 2001 at 1736 UTC, MODIS/Terra passed over the ARM-SGP Central
Facility (CF) site at the city of Lamont, Oklahoma (36.62°N, 97.5°W) with an excellent
viewing zenith angle of 1.3°. The MODIS true-color composite image (Figure 5.23a) shows
a large area of stratiform cirrus clouds over the south central United States. The operational
MMCR at CF also observed a continuous layer of cirrus clouds located between 6 and
11 km during the period from 0600 to 2200 UTC. For the purpose of assessing retrieval
results by comparison with MODIS and MMCR cloud products, we selected a rectangular
domain, denoted by the red box in Figure 5.23a, of 51 × 51 pixels around CF, which
is located approximately in the center of the box. The cloud mask product indicates that
pixels within the box are mostly cloudy, while the cloud phase product (Menzel et al.
2006) shows that these cloudy pixels are cirrus clouds in view of the fact that cloud-top
temperatures were lower than 240 K.
The retrieval algorithm was applied to each cirrus cloud pixel within the domain, and
mappings of retrieved τ c , Dt , and Db from Collection 005 algorithms are shown in Figures
5.23b, 5.23c, and 5.23d, respectively. The τ c values (6) over the northwest quadrant are
relatively large, while small τ c values (<3) over the southeast quadrant indicate optically
thin cirrus clouds. Except for the northwest quadrant, retrieved Dt values (50 µm) are
generally smaller than Db (>150 µm), consistent with the in situ observed stratified pattern
of ice crystal vertical size distribution. The vertically mean D̄e for each pixel from the
retrieval is calculated as the average of Dt and Db . Comparison with MODIS cloud products
reveals that retrieved (Figure 5.23b) and MODIS τ c maps are almost identical, but the
retrieved D̄e [= (Dt + Db )2] are systematically larger than MODIS D̄e (= 2re ), retrieved
using 0.645 µm (band 1) and 2.13 µm (band 7) reflectances.
Scatter plots in Figures 5.23e and 5.23f, respectively, show the correlation between
retrieved and MODIS τ c and D̄e for the selected domain. Figure 5.23e shows that the
retrieved τ c and MODIS τ c are highly correlated, with a correlation coefficient of 0.983.
This good correlation demonstrates that the retrieved τ c is highly accurate. The root-mean-
square (rms) difference is 0.219, which is about 5% of the mean τ c (4), and is comparable
to the objective requirement for the VIIRS/S-NPP algorithm (Ou et al. 2003). Figure 5.23f
shows that the retrieved mean D̄e also correlates well with MODIS D̄e , with a correlation
coefficient of 0.65. Differences between the retrieved and MODIS D̄e are due primarily to
differences in the assumed vertical De profiles. In the present radiative transfer program
used to generate the look-up library, the vertical linear profile of De , which increases from
the cloud top toward the cloud base, has been included in the calculations, whereas in the
MODIS forward radiative transfer modeling, the cloud effective particle size is assumed
to be vertically constant. Thus, retrieved D̄e are mostly larger than MODIS D̄e , which
tends to be biased toward the smaller cloud-top De as indicated in Figure 5.21a, while the
retrieved D̄e tends to be biased toward the larger mid-cloud De . Since both the present and
MODIS Collection 5 Cloud Optical Property 06_OD algorithms have employed the same
(a) MODIS Image (b) τc

7
48’
6
42’
5
36oN
36.00’ 4
30’
3
24’
2
50’ 40’ 97oW 20’ 10’
30.00’
(c) Dt (μm) (d) Db (μm)
48’ 48’ 150

150
42’ 42’
36oN 100 36oN 100
36.00’ 36.00’
30’ 50
30’ 50
24’ 24’
50’ 40’ 97oW 20’ 10’ 50’ 40’ 97oW 20’ 10’
30.00’ 30.00’
(e) (f)
7
6 150
De (μm)
5
τc
100
4
3 50
2
2 4 6 50 100 150
MODIS τc MODIS De (μm)
Figure 5.23 (a) MODIS true-color composite images for March 6, 2001 at 1736 UTC; (b) retrieved
τ c ; (c) retrieved Dt for a selected domain; (d) retrieved Db for a selected domain; (e) scatter plot for
retrieved τ c versus MODIS τ c for a selected domain; (f) scatter plot for retrieved De versus MODIS
De for a selected domain. The black rectangle in (b) denotes a 2 km × 16.8 km strip of cloud that is
supposed to have drifted over the MMCR site during the 10-minute period, and has been scanned by
the radar beam (after Wang et al. 2009).
ice crystal size and habit distribution models, the difference between the mean effective
particle sizes inferred from the present and MODIS retrieval programs should not be due
to the difference in the assumption of ice crystal habits.
At the time of Terra’s overpass, the MMCR at CF also observed a single layer of cirrus
clouds between 6 and 10 km. Figure 5.24a shows the MMCR time series of vertically
(a) MMCR De (μm)
(b)
MMCR
Dt (μm)
Retrieved
(c)
Db (μm)
Time (UTC)
(d)
0
Normalized Optical Depth tn
Retrieved average
MMCR average
0.2
0.4
0.6
0.8
1
40 50 60 70 80 90 100 110
De (μm)
Figure 5.24 (a) Time series of vertically resolved De derived from MMCR in the 10-min period
centered around the satellite overpass at 1736 UTC, March 6, 2001, which is denoted by the vertical
line; (b) collocated time series of retrieved and MMCR Dt ; (c) same as (b) except for Db ; and
(d) strip-averaged retrieved (dashed curve) and MMCR (solid curve) De profiles as functions of τ n .
Horizontal bars denote standard deviations (after Wang et al. 2009).
resolved De (Mace et al. 2002). Over a 10-minute period (5 minutes before and after the
MODIS overpass), MMCR observed 16 De profiles. The MMCR De varies between 50 and
150 µm, with a few spots larger than 150 µm.
Furthermore, to assess the validity of retrieved Dt and Db , MMCR De and IWC profiles
were processed and applied to MMCR, while a collocation scheme was applied to MMCR
and retrieved cloud parameters. For each of the 16 MMCR profiles, the vertical profile of
extinction coefficient β e (z) is given by [see Eq. (5.2.7)]
βe (z) ≈ IWC(z)[c + d/De (z)], (5.3.2)
where β e (z) has units of m−1 , IWC of g m−3 , and De of µm, and c and d are fitting coefficients.
τ (z) is then calculated by integrating β e (z) from the cloud top to the level z as follows:
zt
τ (z) = βe (z)dz. (5.3.3)
z
Finally, the MMCR De (z) is converted to De (τ ) using τ (z). A linear fitting procedure is
applied to each De (τ ) profile to generate Dt and Db . The MMCR time series of Dt and Db
are plotted as crosses in Figures 5.24b and 5.24c, respectively.
A space-to-time collocation scheme was subsequently applied to match the geographical
locations of the retrieved Dt and Db corresponding to the observation time of MMCR De
profiles. Since the MODIS data represented nearly instantaneous observation on the order
of only a few seconds, while the MMCR data provided a time series of observations from
a single location, the space-to-time correlation between the two datasets can be established
by using wind speed and direction data at the cirrus cloud level from atmospheric soundings
(Ou et al. 1995). As observed by collocated daytime soundings, which were launched at
CF (collocated with the MMCR site), and at 1130 and 2029 UTC, the wind speed and
direction at the cirrus cloud level based on interpolation to the MODIS overpass time
were estimated to be around 28 m s−1 and 276° clockwise from north, respectively. Within
the selected domain, a 2 km × 16.8 km strip of cloud (denoted by the black rectangle in
Figure 5.23b) was further identified, which was supposed to have drifted over the MMCR
site during the 10-minute period, and has been scanned by the radar beam. The strip length is
the distance traveled by the cloud within a 10-minute period (600 s × 28 m s−1 = 16 800 m).
The geographical location of each pixel within the strip was then matched with the time
of one of the 16 MMCR profiles. Because there were altogether 37 pixels within the strip,
each MMCR profile might correspond to two or three pixels. Mean values of the retrieved
Dt and Db for collocated pixels associated with each MMCR profile were then calculated.
The converted and MMCR time series of retrieved Dt and Db (plotted as circles and crosses)
are shown in Figures 5.24b and 5.24c, respectively.
Figure 5.24b shows that the retrieved and MMCR-determined Dt vary between 40 and
55 µm, and between 60 and 70 µm, respectively, and that both time series display similar
trends. The MMCR Dt is systematically 10–30 µm larger than the corresponding retrieved
values. This difference near the cloud top could be explained by the fact that MMCR were
mainly sensitive to large particles and missed smaller particles (Comstock et al. 2002).
Since cloud top is generally composed of smaller particles, the MMCR-derived Dt may
be positively biased. Figure 5.24c displays the retrieved and MMCR Db , which vary
between 70–100 µm and between 90–100 µm, respectively. The retrieved Db shows more
variability than the value from MMCR. The two Db series generally agree with one another
better than the two Dt series, with differences between 0 and 30 µm. This better agreement
near the cloud base could be because the MMCR determined Db might be less biased than
Dt , given the fact that particle sizes at the cloud base are larger than at the cloud top.
Yue et al. (2007) has reported similar magnitudes and differences between AIRS-retrieved
and MMCR-derived ice crystal sizes.
Figure 5.24d shows the averaged De profiles derived from MMCR observations (solid
lines) and retrievals (dashed lines) as functions of normalized cloud optical depth, defined
as τn = τ/τc . Horizontal bars at the six selected τ n levels represent the standard deviations
of De . Both the averaged profiles show that De generally increases from the cloud top
towards the cloud base, consistent with the general behavior of midlatitude cirrus clouds.
The slope of the retrieved profile is about 45 µm per unit τ n , while that of the MMCR
profile is around 53 µm per unit τ n . Figure 5.24d demonstrates a qualitative agreement in
the vertical trend of retrieved and MMCR De profiles. The standard deviations vary between
10 and 20 µm for retrieved profiles, and increase from the upper cloud (10 µm) to the
lower cloud (30 µm).
In addition to the remote sensing characteristics of MMCR, differences between MODIS
and MMCR retrievals can also be attributed to other factors involving spatial coverage and
microphysical processes. The spatial coverage for MODIS is considerably different from
MMCR, which has a nominal spatial resolution of 1 km near nadir. However, MMCR
deployed at the ARM-SGP site has a beam diameter of 3 m at ground level and beam width
of 0.19°. The beam diameter increases to 22 m at 6 km and 36 m at 10 km. The MMCR
beam diameters are much smaller than MODIS pixel sizes. Differences in MODIS-retrieved
and MMCR-derived cloud particle size parameters can result from different observation
scales. Moreover, in our collocation scheme, cirrus clouds were assumed to drift over the
MMCR site with nearly constant bulk microphysics properties during a 10-minute period.
However, the temporal scale of cloud microphysical processes is on the order of seconds, so
the evolution of clouds can be another source of differences between retrieved and MMCR
cloud parameters.
In summary, the cirrus cloud vertical sizing retrieval algorithm presented above repre-
sents a step forward in the development of satellite cloud retrieval programs in terms of
its ability to determine vertical cloud particle size profiles, and its potential applicability to
the processing of remote sensing data from next-generation radiometric sensors. If cirrus
cloud climatology could produce two vertical ice crystal size parameters, the results would
be extremely helpful in understanding vertical size structure in GCM and climate simula-
tions involving cirrus–H2 O dynamic feedback in the climate system, particularly over the
tropical upper troposphere, which appears to control sea surface temperatures. However, to
make the preceding algorithm operationally viable on a global scale, much more research
is required to improve retrieval algorithm performance.
5.4 Remote sensing using reflected polarization 301
5.4 Remote sensing of cirrus clouds using reflected polarization

In Subsection 2.4.2, we discussed the representation of a light beam by the Stokes param-
eters (I, Q, U, V), which are expressed in terms of amplitudes and phases [Eqs. (2.4.11a)–
(2.4.11d)] and the geometry defining an ellipse [Eqs. (2.4.22a)–(2.4.22d)], and from the per-
spective of measurements [Eqs. (2.4.27a)–(2.4.27c)]. Additionally, we defined the degree
of polarization in terms of these parameters as follows: PO = (Q2 + U2 + V2 )12 I. The
determination of V requires the measurement of ellipticity, which needs a quarter-wave
plate. In the laboratory (e.g., Kuik et al. 1991) or at a stationary ground-based facility (e.g.,
Coulson 1988), ellipticity measurements can be made. However, over a moving platform
(aircraft or satellite), such measurements appear to be extremely difficult. For this reason,
the conventional practice has been to measure the degree of linear polarization (or simply
the linear polarization), defined previously as: LP = −QI = −(Il − Ir )(Il + Ir ). It follows
that linear polarization provides a new dimension in addition to intensity in the remote
sensing of atmospheric clouds and aerosols as well as underlying surfaces. Moreover, tak-
ing ratios of two polarization intensities eliminates certain coefficients/parameters and the
uncertainties associated with them. Thus, high accuracies can be achieved in linear polar-
ization measurements from space (e.g., Cairns et al. 1999). For near-IR wavelengths, e.g.,
0.864 µm used by POLDER, as discussed in the following subsection, Rayleigh scattering
(molecular contributions) can be neglected for analysis of the reflected polarized intensity
in cirrus cloudy atmospheres. Also, land surfaces are usually considered to be Lambertian,
reflecting the same light intensity in all directions without polarization.
5.4.1 An historical review of polarization measurements

Polarization of the sunlight reflected from clouds appears to show a strong imprint of their
thermodynamic phase – ice or water, as well as ice crystal shape. Reflected polarization
observations from sunlight in the visible as a remote sensing tool for planetary exploration
were used by the French astronomer Lyot (1929) for the study of Venus. Although Venus is
the nearest planet, it is the most mysterious as it is surrounded by a veil of clouds. Hansen
and Hovenier (1974) performed an extensive investigation by comparing linear polarization
observations with comprehensive multiple scattering computations for spherical particles
by varying size parameter and refractive index based on Lorenz–Mie calculations. It was
shown that the Venus cloud layer was composed of spherical particles having a mean radius
of about 1.05 µm, an effective size dispersion of 0.07, and a refractive index of about
1.44 at a wavelength of 0.55 µm. Interpretations of the reflected polarization from sunlight
have been carried out to understand the physical, optical, and chemical properties of the
clouds and hazes of other planets (e.g., Kemp et al. 1971, Travis et al. 1979). Application
of polarimetric observations to remote sensing of the atmosphere and oceans has been
reported by Gehrels (1974), Mishchenko et al. (2010b), and Kattawar et al. (2016).
In the 1970s, NOAA explored the polarization of reflected sunlight as a means to
infer aerosols and clouds in the Earth’s atmosphere from space. Stowe (1977) reported a
high-altitude balloon experiment which carried spacecraft scanning radiometers with the
capability of measuring the linear polarization of reflected sunlight in broad intervals of the
visible and near-infrared spectrum. The radiometers scanned from horizon to horizon in
the principal plane of the sun as the balloon ascended to float at an altitude of 37 km. The
intensity and the plane and degree of linear polarization were used along with coincident
photographs to examine the effects of the Earth’s surface and of the atmosphere with
reference to polarization features. At that time, space polarization measurement of reflected
sunlight was a new concept in the United States. Since then, NOAA has not attempted
another such mission due to insufficient instrumentation heritage and associated risks.
More recently, NASA has equipped its Glory satellite with a polarimeter system to conduct
aerosol studies, as discussed below.
Meanwhile, the French space agency has developed a polarimeter for space remote sens-
ing called POLDER (see Subsection 5.1.5), which was launched as a passenger instrument
aboard the Advanced Earth Observing Satellite (ADEOS)-I in August 1996. ADEOS was
developed by the Japanese space agency, but the mission ended in June 1997 due to com-
munication failure. POLDER-2 was launched in December 2002 aboard ADEOS-II. The
second mission ended prematurely after 10 months when the satellite’s solar panel malfunc-
tioned. Although the operation of ADEOS-I was short-lived, polarization data collected by
POLDER has been analyzed by Chepfer et al. (2001) to study the global distribution of ice
crystal shapes in cirrus clouds.
A third-generation POLDER instrument was installed on the French space agency’s
PARASOL satellite. It was launched aboard an Ariane 5 carrier on December 18, 2004 and
joined the A-Train constellation, flying in close formation in low polar orbits. Zhang et al.
(2009) investigated differences in ice cloud optical depth derived from MODIS and PARA-
SOL/POLDER and found that POLDER values were substantially smaller than MODIS
values. Riédi et al. (2010) merged MODIS and PARASOL/POLDER data to provide more
reliable information on cloud thermodynamic phase to improve the retrieval of other cloud
properties. After five years of operation, PARASOL was maneuvered out of the A-Train
on December 2, 2009. However, polarization data were continuously recorded until its
termination date on December 18, 2013, exactly nine years post-launch.
While PARASOL/POLDER was operating, NASA authorized the development of a
polarimeter for aerosol and cloud studies through its Goddard Institute for Space Studies.
The Glory satellite was launched on March 4, 2011, carrying an instrument called the
Aerosol Polarimetry Sensor (APS) – a continuous scanning sensor that has the capability to
collect multi-angle observations of spectral polarization and intensity at visible and near-IR
wavelengths. Unfortunately, Glory was lost 30 minutes after launch when its Taurus XL
carrier rocket malfunctioned.
The APS is a heritage instrument based on the design of the Research Scanning Polarime-
ter (RSP, Cairns et al. 1999). The RSP has nine spectral channels, divided into two groups
based on the type of detector used: visible/near-IR bands at 0.41, 0.47, 0.55, 0.67, 0.865,
and 0.96 µm and shortwave infrared bands at 1.59, 1.88, and 2.25 µm. The RSP has pro-
vided high-accuracy polarization data enabling simultaneous retrieval of aerosol and cloud
microphysics parameters (Chowdhary et al. 2005). It has been used to collect extensive
data during a number of coordinated field campaigns (Mishchenko et al. 2007) for aerosol
and cloud studies. It is worth noting that the APS was originally scheduled to be included
in the instrument suite of the National Polar-orbiting Operational Environmental Satellite
System (NPOESS) (Ou et al. 2002); however, NPOESS failed to materialize.
5.4.2 Reflected polarization of sunlight from ice crystal clouds

Takano and Liou (1989b) and Liou and Takano (1994) have used single-scattering results for
hexagonal plates and columns to interpret polarization measurements. Further to these stud-
ies, Liou and Takano (2002) conducted extensive trial-and-error analyses and computations
to interpret the linear polarization pattern of sunlight reflected from optically thick cirrus
clouds as measured by Coffeen (1979), using an airborne infrared polarimeter at 2.2 µm
and shown in Figure 5.25a. The optical depth was estimated to be 32 and randomly oriented
hollow columns, plates, and aggregates were used, along with spheres, in the interpretation
of the observed polarization patterns. The results for spheres deviated significantly, espe-
cially at the rainbow angle of 140°. The results for randomly oriented aggregates best
matched the observed polarization feature, as compared with hollow columns and plates.
Chepfer et al. (1998) presented results of the polarization of sunlight reflected from
cirrus clouds, obtained from the 1994 EUCREX campaign using airborne polarimeter
measurements for two POLDER channels at 0.443 and 0.864 µm. Figure 5.25b shows
results for 0.864 µm over a scattering angle range of 40–180°. The results employing an
optical depth of about 3 most closely match the observed data. As displayed in the diagram,
the plate case (L2a = 880) fits the observed points in all scattering angles, except in the
backscattering range where the irregular ice particle case fits better.
Figure 5.25c shows observed polarization results from an airborne POLDER instrument
at 0.864 µm, covering scattering angles from 90° to 180°. Results in Figure 5.25d were
obtained from POLDER on the ADEOS-1 satellite platform, covering scattering angles
from 90° to 150° (Chepfer et al. 1999). Both involved extended cirrus cloud systems. The
polarization data illustrate weak (Figure 5.25c) and strong (Figure 5.25d) peaks at the
104° subsun feature associated with the horizontal orientation of ice plates and columns.
In order to provide appropriate interpretations, all 16 phase matrix elements are needed
in radiative transfer calculations. Following the approach presented in Subsection 2.5.2
for radiative transfer in horizontally oriented ice particles, 0.1% and 1.2% respectively of
hollow Parry columns (horizontal random orientation in a fixed direction) were added to a
combination of 60% rough-surface aggregate ice particles, 30% hollow columns, and 10%
plates randomly oriented in three-dimensional space to provide the best interpretation of
observed data.
Ou et al. (2005a) devised an approach for the remote sensing of cirrus particle size and
optical depth using the degree of polarization and polarized reflectances associated with the
first three Stokes parameters I, Q, and U for 0.865 and 2.25 µm wavelengths. The retrieval
algorithm employed the steepest-descent method in the form of a series of numerical
iteration procedures to search for the simulated polarization parameters that best matched
the measured parameters. Sensitivity studies were performed to investigate the behavior
10 0.10
(a) (b) POLDER
l = 2.22 μ m l = 0.864 μm Measurement
Linear Polarization (%)
Polarized Reflectance
q 0 = 70o 0.08 q 0 = 55.3o
5
t = 32 t =3
0.06
Aircraft 0.04
Measurement
-5
0.02
-10 0.00
0 60 120 180 0 60 120 180
0.05 0.10
(c) (d)
l = 0.864 μm
0.04 POLDER Measurement 0.08

l = 0.864 μm
q 0 = 38.0o q 0 = 43.6o
t = 2.5 60%
..
.. ..
.
. .. .
. . 30% 10% t =4 .. ..
..
. .. .
0.03 0.06 .. .
.
plus hollow Parry

columns (0.1%)
POLDER Measurement
0.02 0.04
.. ..
60%
..
. .. .
.. .
. 30% 10%
0.01 0.02 plus hollow Parry
columns (1.2%)
0.00 0.00
0 60 120 180 0 60 120 180

Figure 5.25 (a) Linear polarization of sunlight reflected from cirrus clouds, measured with an airborne infrared polarimeter at 2.2 µm (Coffeen
1979). (b) and (c) Full polarization observed by an airborne POLDER instrument at 0.864 µm over an extended cirrus cloud system (Chepfer et al.
1998). (d) Same as (b) and (c) but from the ADEOS-1 satellite platform (Chepfer et al. 1999). The theoretical results are computed for randomly
oriented hollow columns, plates, and irregular aggregates, and ice spheres. In (c) and (d), a small percentage of hollow Parry columns is added to
produce the observed peaks. The shaded areas denote scattering angle regions where the reflected sunlight is out of the observation range for the solar
zenith angles given in the figures (after Liou and Takano 2002).
0.3 0.3
l = 0.865 μm l = 2.25 μm
1/4
1 POLDER data
1/4
1
Polarization Ratio P
4
0.2 4 0.2 De = 124 μm
t = 16
De = 124 μm
t = 16
0.1 0.1
De = 42 μm
De = 42 μm
De = 24 μm De = 24 μm
0.0 0.0
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
Bidirectional Reflectance rI
Figure 5.26 Two-dimensional polarization/reflectance correlation diagram in terms of the mean

effective size De and the optical depth τ for two near-IR wavelengths of 0.865 and 2.25 µm, using
ice plates in the calculation for a solar zenith angle of 55.3°, a viewing zenith angle of −30°, and a
surface albedo of 0.0. The red crosses are POLDER data obtained on April 17, 1994 over Brittany,
France during EUCREX (after Ou et al. 2005a).
of the phase matrix elements as a function of the scattering angle for three ice crystal
size/shape combinations. Synthetic retrievals revealed that the retrieval algorithm itself is
highly accurate, and polarimetric and radiometric error sources caused small retrieval errors
that were less than S-NPP accuracy threshold requirements for effective particle size and
optical depth.
To illustrate the effectiveness and feasibility of this retrieval approach, a two-dimensional
correlation diagram was constructed between the polarization ratio P and the bidirectional
reflectance rI for these two wavelengths, and carried out for the defined shapes and sizes.
In Figure 5.26, we present results for the plate case involving a solar zenith angle of 55.3°
and a viewing zenith angle of −30°, corresponding to a scattering angle of 95° and a
surface albedo of 0.0. The solid lines are for constant ice crystal size in terms of the mean
effective width (size) De , while the dashed lines represent constant optical depth. We find
that there is a strong sensitivity of the polarization ratio to the effective ice crystal size.
Takano and Liou (1989a) show that P increases with decreasing length-to-width ratio LD
(L is the length and D is the width) for a range of scattering angles from 60° to 120°. The
mean effective size De is inversely proportional to LD, so if De increases, the polarization
ratio increases. Also displayed in these diagrams is the correlation of 0.865 µm P and rI
derived from the POLDER data obtained during EUCREX on the morning of April 17,
1994 over Brittany, France. These measurements were provided by Chepfer et al. (1998)
and are presented in Figure 5.25 (Liou and Takano 2002). The data were taken over the
thicker portion of a single layer of cirrus located between 8 and 10 km. The viewing angles
for the data were between −25° and −35°, which corresponded to scattering angles of
100° and 90°, respectively. Application of these data points to the retrieval scheme leads
to an optical depth of about 4 and a mean effective width of about 45 µm. Similar 2-D
correlation diagrams for polarization ratio and bidirectional reflectance can be constructed
for solid/hollow columns or irregular ice particles for shape and size studies.
Furthermore, the retrieval of ice cloud properties using RSP data gathered from the
CRYSTAL-FACE field campaign has been carried out by van Diedenhoven et al. (2013).
On July 29, 2002, a large storm developed on Florida’s west coast at about 1700–1800 UTC,
which was sampled by all aircraft involved in the CRYSTAL-FACE campaign. RSP mea-
surements obtained at about 2030 UTC over outflow of this storm aged by approximately
2 hours were analyzed. The mean solar zenith and relative azimuth angles for RSP data in
this case were 41° and 10°, respectively. Figure 5.27 shows retrieved cloud-top heights from
both RSP and MAS data, reflectivity profiles derived from the cloud radar system (CRS),
cloud optical depth, asymmetry parameter, roughness parameter, and aspect ratio from RSP
measurements as a function of time. The Proteus (RSP) and the ER-2 (CRS, MAS) were
well collocated for these legs. The CRS shows cloud-top heights around 12–14 km, with
slightly lower cloud-top heights in the trailing anvil than in the leading anvil (Figure 5.27a).
An apparently overshooting top occurs around 73.75 ks UTC Proteus time (i.e., the time
corresponding to the RSP data). High reflectivities occur around 73.5 ks UTC towards the
leading part of the storm. For the bulk of the storm, RSP retrieved cloud-top heights of
12–15 km.
Figure 5.27b shows the optical depth retrieved from RSP measurements, which is about
40–100 at the core of the system and strongly decreases in the trailing anvil. Asymmetry
parameters retrieved by the RSP are shown in Figure 5.27c and generally range from
about 0.76 to 0.8 with a median value of 0.78. The smallest asymmetry parameters are
generally found at the location with greatest optical depth and radar reflectivity in the
cloud top (around 73.5 ks UTC). Retrievals were not performed in the optically thin part
of the leading anvil because of the substantial influence of ocean sun glint. The retrieved
asymmetry parameters correspond to hexagonal columns with a median distortion of 0.45
(Figure 5.27d) and a median aspect ratio of 1.6 (Figure 5.27e).
5.5 Principle of backscattering depolarization for differentiation of ice

crystals and water drops
Differentiation of non-spherical ice crystals and spherical water droplets using backscat-
tering depolarization was first developed by Liou and Lahore (1974) on the basis of the
principles of geometric optics and controlled laboratory measurements. The subsequent
development of lidar backscattering depolarization for detecting ice crystals and non-
spherical aerosols has been presented by Sassen (1976, 1991). The use of backscattering
polarization for cirrus cloud study is evident from the CALIOP/CALIPSO results pre-
sented in Figures 1.8 and 1.9. In the following subsections, we capture the essence of the
fundamental development of backscattering depolarization based on the principles of geo-
metric optics.
5.5 Differentiation of ice crystals and water drops 307
Figure 5.27 RSP, CRS, and MAS data for July 29, 2002 as a function of time corresponding to
RSP data on the Proteus aircraft. CRS and MAS data are shifted and stretched/squeezed in time
to best collocate with the RSP data. (a) Cloud radar reflectivity (color contours); RSP retrieved
cloud-top height (red) with minimum and maximum retrieved heights (dotted lines) and cloud-top
height derived from MAS 11 µm brightness temperature (green). Also shown are RSP retrieved ice
cloud properties: (b) cloud optical depth, (c) asymmetry parameter, (d) distortion parameter, and
(e) aspect ratio (after van Diedenhoven et al. 2013).
5.5.1 Theoretical basis

From Eq. (2.4.30), the incident and scattered electric vectors can be expressed as

Es exp(−ikr + ikz) S2 S3 Ei A2 A3 Ei
= = , (5.5.1)
E⊥s ikr S4 S1 E⊥i A4 A1 E⊥i
where the matrix A is defined as the transformation matrix and we have changed Er and El
to E⊥ and E , respectively, where these denote the propagating electric vectors respectively
perpendicular and parallel to a reference plane. After a scattering event, the reference plane
for the scattered electric vector normally differs from the incident electric vector through
the transformation matrix, whose elements Aj ( j = 1, 2, 3, 4) are quantities associated with
the scattering medium and are functions of the shape, size, and orientation of ice particles,
along with their indices of refraction. For spherical water droplets, A3 = A4 = 0.
The principle of depolarization due to backscattering from large cloud particles can be
described by the laws of geometric optics in terms of light rays reflected and refracted by the
scatterer, and surface waves, as presented in Subsection 3.3.6 for spheres and in Subsection
3.3.7 for non-spherical particles. In this matter, the scattered light beam can be divided
into separate localized rays so that the backscattered electric vector can be evaluated from
geometry. For a sphere, which is symmetrical with respect to the incident electrical vector,
there are three possibilities to produce 180° backscattering: a central ray externally reflected
from the sphere, a central ray undergoing two transmissions and one internal reflection, and
surface waves (or edge rays) that travel around the sphere and produce backscattering, as
illustrated in Figure 5.28a. For the first two cases, the direction of vibration of the incident
electric vector, representing the polarization state, remains the same after backscattering.
Since edge rays do not touch the surface, the direction of vibration of the electric vector
again retains its polarization state.
In the case of columns and plates, it is clear that in order to evaluate the scattered electric
vector due to two refractions and two reflections, the initial incident electric vector must be
transferred first to the plane containing the incident and refracted rays – the scattering plane.
It follows that the transmitted and reflected fractions of the energy for the two components
of the electric field must be evaluated with respect to the scattering plane. The scattered
electric field must be rotated again to the initial coordinate so that the depolarization of the
incident radiation can be evaluated in this coordinate system. Thus, if the angle between
the reference plane for the incident electric vector and the scattering plane is denoted γ ,
the transformation matrix of a hexagonal crystal for the backscattered light beam can be
expressed in the form
A = Lγ TRR∗ TL−γ , (5.5.2)
where the rotational matrix, which defines the transformation of the scattered electric vector
to the initial plane of reference, is given by

cos γ sin γ
Lγ = , (5.5.3)
− sin γ cos γ
and the term L−γ , which rotates the initial plane of reference to the scattering plane, is the
inverse (denoted by the symbol −1) of Lγ :
L−γ = L−1
γ . (5.5.4)
The transmission matrix for rays transmitted in and out of the crystal can be written as

t 0
T= , (5.5.5)
0 t⊥
where t and t⊥ are the two components of the transmitted portion of the energy. Because
of geometrical similarity, the transmission matrix is the same for rays transmitted in or out
(a)
E
E E
Light beam E
(b)
E
Light beam E
(c)
E
Light beam E
Figure 5.28 Principle of backscattering depolarization from a spherical droplet, a plate, and a
solid column. The incident electric field is assumed to be vertically polarized. The electric vector
backscattered from a sphere retains the polarization state in view of the geometric symmetry. However,
the backscattered electric vector produces a cross-polarization due to internal rotations of the incident
electric vector in order to achieve backscattering.
of hexagonal crystals, as shown in Figure 5.28. The reflection matrix R∗ for rays reflected
at the edge of the crystal and R for rays reflected at the bottom of the crystal, as shown in
Eq. (5.5.2), can be expressed as

∗

∗ r 0 r 0
R = , R= . (5.5.6)
0 r⊥∗ 0 r⊥
From Fresnel’s formulas [see Eqs. (3.1.9a) and (3.1.9b)], the two components of the
reflected portion of the energy are defined by
cos θi − m cos θt
r⊥ = , (5.5.7)
cos θi + m cos θt
m cos θi − cos θt
r = , (5.5.8)
m cos θi + cos θt
where the incident angle θ i differs for the cases with and without ∗ . The transmitted parts
are given by
) *1/2 ∗
t,⊥ = 1 − |r,⊥ |2 , and t,⊥ = t,⊥ . (5.5.9)
Because of the symmetry between the ray going into the crystal and the ray emerging
from the crystal, the two transmitted energies are the same. In the above equations, θ i and
θ t denote the incident and transmitted (or refracted) angles, respectively, with reference to
the normal (θ i,t = 0 for normal incidence) and m is the index of refraction of ice crystals.
Finally, the transmitted angle θ t can be determined from Snell’s law, sin θi = m sin θt
[Eq. (3.1.1)]. Subsequently, all the remaining parameters can be determined.
The terms on the right-hand side of Eq. (5.5.2) can be expanded explicitly as follows:
A2 = r⊥ r⊥∗ t⊥2 cos2 γ + r r∗ t2 sin2 γ , (5.5.10)
A1 = r⊥ r⊥∗ t⊥2
sin γ + 2
r r∗ t2
cos γ , 2
(5.5.11)
∗ 2
A3 = −A4 = r⊥ r⊥ t⊥ − r r∗ t2 sin γ cos γ . (5.5.12)
For normal incidence, only one internal reflection is required to produce backscattering.
If we imagine that rays had undergone reflection at the edge of an ice crystal with a
refracted angle θ t of 90°, then from Eqs. (5.5.7) and (5.5.8), r⊥∗ = r∗ = 1. On the basis of
the preceding analysis, it is clear that the backscattered polarized beam would be partially
depolarized due to transformation of the electric vector’s vibration plane within the crystal.
We may define the depolarization ratio for a vertically polarized incident beam as
follows:
s 2
E |A3 |2

δ⊥ = 2 = , (5.5.13)
E s |A1 |2
⊥
where Es is the electric field vector received in the cross component. Similarly, the depo-
larization ratio for a horizontal polarized beam can be expressed as
s 2
E |A4 |2
δ = ⊥2 = . (5.5.14)
E s |A2 |2

5.5.2 Issues of orientation and theoretical results

The above discussion is concerned only with a single crystal. For a sample of ice crystals,
the orientation properties must be accounted for. In Subsection 1.6.3, we pointed out that
it is very likely that naturally occurring hexagonal columns and plates orient randomly in
a preferred plane, although a completely random orientation may occasionally be possible,
especially in the initial stages of nucleation.
The relative geometry of an incident polarized beam and an ice crystal is given in
Figure 5.29; γ is the azimuthal angle between the Poynting vector and a crystal, while
θ i is the angle of incidence with respect to the normal plane of the surfaces of hexagonal
(a) (b)
Z NORMAL OBLIQUE
INCIDENCE INCIDENCE
S
g Y
E
X
Figure 5.29 (a) Definition of the orientation angle γ and the inclination angle α in Cartesian
coordinates in which the major axis of a column crystal is aligned with the y-axis. The directions
of the Poynting vector S and the electric (E) and magnetic (H) vectors are also shown. (b) The
configurations of normal and oblique incidence with reference to a sample of columns and plates.
crystals. From this simple geometry, we have sin θ i = cos γ sin α, where α denotes the
angle between the Poynting vector and the zenith. The two angles γ and α completely
describe the orientation of an ice crystal relative to the backscattering geometry. Hence, the
elements in the transformation matrix for a single crystal can be expressed as
Aj = Aj (γ , α), j = 1, 2, 3, 4. (5.5.15)
If the crystals are randomly oriented in a horizontal plane and in the context of backscat-
tering, then these elements can be expressed as

2 π/2
Aj = Aj (γ , α) dγ . (5.5.16)
π 0
For convenience of discussion, γ and α are referred to as the orientation and inclination
angles, respectively. Based on this analysis, it is clear that the depolarization ratio defined in
Eqs. (5.5.13) and (5.5.14) would depend on the inclination angle as well as the polarization
state of the incident beam. Referring to Figure 5.29b, if the incident beam is normal
to a sample of plates whose orientations are in a preferred plane, then depolarization
should not be expected. Also, no depolarization would be produced by an incident light
beam perpendicular to the preferred plane of randomly oriented columns. If the major
axes of plates and columns have the same dimensions, the backscattering produced by
oblique incidence would be equivalent despite the difference in their shapes, because of the
geometrical similarity of the hexagons. However, this conclusion may not be applicable to
other complex ice crystal shapes.
Finally, for a completely random orientation such that every position of an ice crystal is
equally probable, we have
π/2 π/2
4
Aj = 2 Aj (γ , α) dγ dα. (5.5.17)
π 0 0
(a) Theory
50 NORMAL INCIDENCE OBLIQUE INCIDENCE

a = 0o d⊥ g = RANDOM
d||
40
PERCENT
30
dR
20
10
0
0 30 60 0 30 60 90
ORIENTATION ANGLE g INCLINATION ANGLE a
(b) Experiment
ADD H2O ADD H2O CO2 SEEDING

WATER WATER ICE
1
d⊥
0
P⊥
i
P⊥
P||
10 SECONDS
Figure 5.30 (a) Theoretical results for backscattering depolarization from hexagonal plates and
columns for vertically and horizontally polarized beams with normal and oblique configurations. (b)
A typical backscattering return from a water cloud and its transitional stage transformation to an ice
crystal cloud for a vertically polarized laser beam (after Liou and Lahore 1974).
Because the coordinate systems of randomly oriented ice crystals are symmetric to any inci-
dent light beam, it follows that the cloud would have no preferred selection with reference
to incident electric vectors. The foregoing discussions reveal that we may learn orientation
characteristics from the backscattering of polarized light if a preferred orientation for ice
crystals occurs.
Figure 5.30a shows theoretical results for backscattered depolarization from hexagonal
crystals. The curves on the left-hand side of the figure denote the values of depolarization
ratios for the light beam with normal incidence onto a single column as a function of
its orientation angle γ . The depolarization shows a constant shift between vertically and
horizontally polarized incident beams with a maximum value located at γ = 45°. The
right-hand side of the figure gives the depolarization ratio for randomly oriented hexagonal
crystals. These results are plotted as a function of the inclination angle with respect to
the zenith. Backscattered depolarization depends on the incident polarization state and the
geometry defining the incident beam and the positions of the ice crystals. Depolarization
ratios of the vertically polarized beam are smaller than those of the horizontally polarized
beam because of horizontally oriented properties of the ice particles. However, if the
ice crystals are in completely random orientations, a depolarization ratio δ R of 29%
is obtained for both vertically and horizontally polarized incident beams. Takano and
Jayaweera (1985) obtained a higher value of 39% using a more extended geometric ray-
tracing program to account for direct reflection and crystal aspect ratio for an ensemble of
randomly oriented ice crystals.
In summary, based on the preceding theoretical formulation and analysis we find, first,
that the light beam backscattered from spherical water drops retains the polarization state of
the incident energy. Second, the light beam backscattered from non-spherical ice crystals is
partially depolarized as a result of internal reflections and the transformation of coordinate
systems governing the electric vector. Third, backscattered depolarization from randomly
oriented ice crystals in 3-D space is about 29% for both vertically and horizontally polarized
beams. Lastly, a state of preferred orientation of the ice crystals results in a dependence of
depolarization on the polarization state of the incident beam and on the inclination angle
of the ice crystal orientation with respect to the incident beam, as well as on ice crystal
shape.
5.5.3 Experimental arrangements and results

Unpolarized light from a 0.6328 µm helium–neon (He-Ne) laser was employed as a radia-
tion source. A polarizer in front of the laser was used to produce a vertically or horizontally
polarized beam as desired. A small 45° mirror was mounted directly in front of the receiver
so that an exact 180° backscattering from clouds can be observed. This arrangement is
very important because the polarization properties change drastically for spherical water
drops within a few degrees of exact backscattering. The receiver used in this experiment
was a photomultiplier along with a 10 Å He-Ne bandpass filter to reduce background light.
An automatic rotating polarizer followed by the photomultiplier housing analyzed the
backscattered signals into two orthogonal components. The rotating polarizer consists of
two orthogonal polarizer sheets which rotate back and forth at a rate of 1–5 rotations per sec-
ond. Since cloud compositions change rapidly, it is extremely important to record two cross-
components simultaneously. The photomultiplier was followed by an amplifier with voltage
gains of up to 32 volts per microampere of input current. Signals were recorded on a data
processor so that the depolarization ratio, evaluated by dividing the cross-polarized compo-
nents by the two neighboring parallel components, could be obtained. The data processor
was then connected to a chart recorder which plotted the depolarization ratio and the two
almost-simultaneously observed orthogonal backscattered components in relative values.
The optical system was aimed at an angle of about 45° to an open-top cold chamber
whose temperature was about −20°C. Water clouds were produced by introducing water
vapor from a moist air vaporizer-humidifier. Seeding by suddenly scratching solid CO2
with a sharp edge produces an ice crystal cloud when required. The inside walls of the cold
chamber were covered by black chiffon to eliminate possible reflections of the scattered
light beam from the walls. In addition, a light trap composed of polished black glass plates
was employed to absorb the direct light beam. Slides coated with a Formvar solution were
used to sample the water droplets and ice crystals. Sample slides were examined with a
microscope to determine the phase, shape, size, and number density of cloud particles.
Figure 5.30b shows a typical backscattered return from a water cloud and its transitional
stage to an ice crystal cloud. The incident beam is vertically polarized. The backscattering
history on the left-hand side of this figure shows that the water cloud is dissipating after
about 2 min while the returned depolarization remains at about 2–4%. The concentration
of water drops is about 1000 cm−3 with a mean radius of about 5 µm. The small amount of
depolarization could have been caused by the effects of multiple scattering. A dense water
cloud is again produced by adding water vapor to the chamber.
The cloud was seeded by scratching solid CO2 when the supercooled water cloud was
relatively quiet. This is indicated on the right-hand side of Figure 5.30b. A few seconds after
seeding, the depolarization ratio immediately increases, as can be clearly seen in the figure;
it remains at about 35% during the next 30 s, with instantaneous variations of 30 to 40%.
Since the signals are presented essentially without time averaging, it is their mean trend that
is important in physical interpretations. Ice crystals were found to be predominantly plates,
the sizes of which were about 20 µm in the initial stage and about 50 µm when they fell
out due to sedimentation. The signal decrease after seeding revealed the low concentration
of ice crystals. From this illustration, it is clear that the backscattered radiation from water
drops retains nearly the polarization state of the incident laser beam, whereas radiation
backscattered from non-spherical ice crystals is strongly depolarized. Although ice crystals
produced in the laboratory normally do not have the ideal shapes assumed in the theoretical
analysis, the 35% depolarization ratio appears to agree with the theoretically predicted
value of 29% if multiple scattering contributions due to an ensemble of cloud particles are
accounted for.
5.5.4 Application to lidar (laser radar)

The lidar equation for the average backscattered power P̄r (r) as a function of distance r
can be expressed by the following equation:
r
Pt CAr βπ (r)h
P̄r (r) = exp −2 βe (r )dr , (5.5.18)
8πr 2 0
Figure 5.31 A compilation of depolarization (δ) values for a number of ice-related particles
(snowflakes, graupels, rimed ice particles, and others), ice clouds, water clouds, and laboratory
raindrops, and their dependence on temperature, from a variety of laboratory (laser) and field (lidar)
studies (after Sassen 1976, 1991).
where Pt is the transmitted power, C is a known instrument factor, Ar is the collecting

aperture, β π = P(π) β s , P is the phase function at π(180°), β s is the volume scattering
coefficient, h is the pulse length, β e is the extinction coefficient, and the exponential
term represents two-way attenuation. The energy gains corresponding to the transmitter
and receiver have been neglected in this lidar (or radar) equation.
In Eq. (5.5.18), we have neglected polarization information in the development of the
lidar equation. The transmitted power Pt can be generated to be vertically (Pt ⊥ ) or hori-
zontally (Pt ) polarized, while the detector can be set so that both polarization components
(Pr , Pr ⊥ ) are measured. This provides additional information about particle characteristics
and is referred to as the depolarization technique. Consider a vertically polarized incident
laser beam. A cross-polarization component can be produced after internal reflection. The
ratio of this component to the component that retains the same polarization as the incident
light beam is referred to as the backscattering depolarization ratio, defined as
δ = Pr /Pr ⊥ . (5.5.19)
Likewise, we can also define the backscattering depolarization ratio for a horizontally
polarized incident laser beam.
Sassen (1976, 1991) has presented depolarization ratios for various types of ice crystals
and water droplets derived from laboratory and field studies (Figure 5.31). For water
droplets, this ratio is close to zero, as predicted by theory. For ice crystals, the depolarization
ratio varies from 50 to 70%, depending on their shape and size. Figure 5.31 displays
snowflakes of various conditions, graupels, and other shapes. Furthermore, Takano and
Jayaweera (1985) showed that the depolarization ratio for columns (50%) is larger than
for plates (30%), in line with experimental results presented by Sassen (1991). The
depolarization ratio provides a powerful means of distinguishing between water and ice
clouds in the atmosphere.
The backscattering depolarization technique can also be used to probe the orientation
properties of ice particles. In the case of a vertically pointing lidar involving horizontally
oriented plate crystals, the depolarization ratio would be zero because of the symmetry of
the two parallel surfaces with respect to the laser beam. Lidar experiments have shown that
the backscattering depolarization ratio in this case is close to zero, whereas it increases
significantly as the lidar scans a few degrees off the vertical (e.g., Platt et al. 1978).
Ground-based high-resolution 0.532 µm1.06 µm polarization lidar has been used to
measure the depolarization ratio for contrail cirrus. Sassen and Hsueh (1998) presented
depolarization ratios for contrail cirrus and showed that this ratio in persisting contrails
ranged from about 0.3 to 0.7. Freudenthaler et al. (1996) observed ratios of 0.1 to 0.5 for
contrails with temperatures ranging from −60 to −50°C, depending on the stage of their
growth.
Figure 5.32 depicts backscattering depolarization ratios for various ice crystal shapes and
sizes in terms of their aspect ratio, L2a, where L and 2a are the vertical length and horizontal
maximum dimension, respectively. The results were computed from the unified theory of
light scattering by ice crystals presented in Section 3.4 for shapes including bullet rosettes,
solid and hollow columns, Koch snowflakes, and plates, which are related to temperatures
from −15 to −40°C (see Figure 1.14). The horizontal bars denote the depolarization range
as a function of aspect ratio for each ice crystal type. The top panel shows depolarization
ratios determined from high-resolution polarization lidar for contrail cirrus, as presented
by Sassen and Hsueh (1998) and Freudenthaler et al. (1996) for temperatures from −50
to −60°C with three measurement points. Note that the depolarization ratio values for ice
clouds presented in Figure 5.31 are 0.5, consistent with the theoretical results presented
in Figure 5.32.
In the discussion of the CALIPSO satellite associated with the A-Train constellation
in Subsection 5.1.5, we pointed out that this satellite carries Cloud-Aerosol Lidar with
Orthogonal Polarization (CALIOP), a space lidar which provides high-resolution vertical
profiles of optically thin clouds and aerosols. The measurement of both vertically and
horizontally polarized backscattered power provides a new dimension for cirrus cloud
studies vis-à-vis Eqs. (5.5.18) and (5.5.19). Hu (2007) and Hu et al. (2007, 2009) have
undertaken a comprehensive analysis of CALIOP backscattering depolarization data for
the identification of ice and water clouds and ice particle orientation properties. These
data contain contributions from multiple backscattering (Liou 1972c). In the following, we
present a depolarization study of ice clouds by satellite-based polarized lidar.
Figure 5.33 displays backscattering depolarization as a function of temperature for
clouds with layer-integrated backscatter larger than 0.01 per solid angle (sr), corresponding
to an ice cloud optical depth greater than 0.3. The data presented in Figure 5.33a are for
January 2007, during which the lidar in CALIPSO/CALIOP was pointed 0.3° off the nadir.
−60
Contrail Observations
−50
−40
Theoretical Results
(Single Scattering)
−35
Temperature (oC)
L/2a = 2 - 3
Ice Crystal Shape
−30
L/2a = 2 - 4.7
−25
L/2a = 1.4 - 2.5
−20
L/2a = 0.1 - 0.4
−15
L/2a = 0.05 - 0.4
0.1 0.2 0.3 0.4 0.5 0.6 0.7

Depolarization Ratio
Figure 5.32 Theoretical depolarization ratios computed from a unified theory of light scattering by
ice crystals for shapes including bullet rosettes, solid and hollow columns, Koch snowflakes, and
plates, related to temperatures from −15 to −40°C (see Figure 1.14). The horizontal bars denote
the depolarization range as a function of aspect ratio for each ice crystal type. The top panel shows
depolarization ratios determined from high-resolution polarization lidar for contrail cirrus [upper
curve and data point from Sassen and Hsueh (1998), lower curve from Freudenthaler et al. (1996)],
for temperatures from −50 to −60°C.
Figure 5.33b shows data for January 2008; at that time the lidar was pointed 3° off the
nadir. Globally, the depolarization ratios of ice clouds with cloud-top temperatures colder
than 240 K are between 0.25 and 0.45. The depolarization ratio is negatively correlated
with cloud-top temperature, i.e., the colder the temperature, the larger the ratio, which
is consistent with depolarization statistics based on ground measurements and theoretical
calculations (Platt et al. 1987, 1998; Sassen and Benson 2001; You et al. 2006).
For cloud-top temperatures between 240 and 270 K, depolarization ratios are mostly
between 0.05 and 0.3. In contrast to the signals from colder clouds, the depolarization
(a) (b)
° °
Figure 5.33 Frequency of occurrence of clouds as a function of cloud temperature and depolarization
ratio. Only those clouds with a layer-integrated backscatter larger than 0.01 sr−1 , corresponding to an
optical depth greater than about 0.3, are included. The data were obtained (a) in January 2007 when
the lidar was pointed 0.3° off the nadir and (b) in January 2008 when the lidar pointed 3° off the nadir
(after Hu et al. 2009).
ratios increase with cloud-top temperature. At these temperatures, water, ice, and mixed-
phase clouds can coexist, making the thermodynamic phases difficult to define based on
cloud temperature and depolarization. For optically thin clouds with extinction coefficients
less than 0.2 km−1 , molecular backscatter accounts for more than 10% of the 0.532 µm
lidar return. For this reason, the depolarization ratio measured directly by CALIOP for thin
cirrus can be significantly lower than the corresponding ice crystal depolarization ratio.
In summary, the backscattering depolarization ratio from an incident polarized laser
beam provides a new dimension for the discrimination of ice and water clouds and the
orientation properties of ice particles.
5.6 Remote sensing of cirrus using reflected line spectra in 1.38 and 0.76 µm bands
5.6.1 A brief review of the use of line spectra for Earth remote sensing
In Figure 5.4, we presented spectral bidirectional reflectances in water vapor bands at 0.94,
1.1, 1.38, and 1.87 µm wavelength, produced by vibrational transitions associated with
combination and overtone states in the presence of cirrus clouds comprised of three mean
effective ice crystal sizes. The 1.38 µm band is particularly useful in view of its line strength
and associated amounts of solar energy. We also pointed out in Subsection 5.2.6 that the
AVIRIS 1.38 µm channel is specifically sensitive to the presence of high-level clouds.
When a substantial amount of water vapor is present in the atmosphere, the radiation in
5.6 Remote sensing using reflected line spectra 319
this spectral range is almost completely absorbed within lower levels. As a result, only
the spectral reflection from high-cloud levels could reach a satellite sensor for detection
(Gao et al. 1993). The spectral lines of sunlight reflected from other planets have been
observed and used to determine the composition of planetary atmospheres (Goody and
Yung 1989). Such observations have not, however, been made from Earth satellites (but
see Subsection 5.6.3). In the following, we capture a number of important developments
pertinent to the use of emission spectra for cirrus identification.
In analysis of satellite Infrared Interferometer Spectrometer (IRIS) data, Prabhakara et al.
(1993) showed that thin cirrus in the tropics can be identified in the 8–12 µm window region.
The IRIS instrument was on board Nimbus III (Conrath et al. 1970) and IV (Kunde et al.
1974), and for the first time provided data on infrared emission by the atmosphere from
about 400 to 1600 cm−1 with a 5 cm−1 spectral resolution. Unfortunately, the interferom-
eter spectrometer experiment was not followed up in the United States satellite program.
However, three Fourier spectrometers in the spectral range 400–1600 cm−1 were included
in European METEOR satellites in the late 1970s. In the analysis of emission spectra
measured in clear and cirrus cloudy atmospheres, Spänkuch and Döhler (1985) found that
the presence of cirrus clouds significantly reduced the upwelling radiances in the entire
spectral region, except in the center of the CO2 15 µm band. Smith et al. (1995) developed
a high spectral resolution infrared spectrometer, referred to as HIS, intended for satellite
application. The HIS is a Michelson interferometer covering the spectral region from 3.5 to
19 µm with a high spectral resolution of λλ > 2000, which depends on the line position.
The spatial resolution is about 2 km at the Earth’s surface below the ER-2 aircraft at 20 km.
The authors showed that sufficient information on cirrus clouds existed in the HIS spectrum
and suggested that the cirrus cloud structure and composition could be inferred. Further,
Smith et al. (1998) displayed an interesting spectrum for a case involving a cold cirrus that
was particularly evident in the 800–1000 cm−1 window region.
5.6.2 Reflected line spectra in the 1.38 µm band

On the basis of computational results from a line-by-line equivalent radiative transfer
program, we have investigated the information content of bidirectional reflectances in the
1.38 µm H2 O band covering 6600–7500 cm−1 . The calculations employed an ice crystal size
distribution with a mean effective ice crystal size of 42 µm, corresponding to cirrostratus
with a shape composition of 50% aggregates/bullet rosettes, 30% hollow columns, and
20% plates. Using the adding/doubling program for multiple scattering combined with
the correlated k-distribution method that integrates water vapor absorption, the scattering
and absorption properties of ice crystals, Rayleigh and background aerosol scattering,
and surface reflection, we have carried out a number of sensitivity experiments within
the 1.38 µm band covering 6600–7500 cm−1 with a 1 cm−1 resolution. The correlated k-
coefficients for water vapor were determined using the method developed by Fu and Liou
(1992a) using 10 cumulative probability functions g. The non-homogeneous atmosphere
(a)
0.14
tc = 1.0
q 0 = 60o, q = 0o
tc = 0.0 (clear)
0.12
0.1
0.08
0.06
0.04
0.02
0
6600 6700 6800 6900 7000 7100 7200 7300 7400 7500
(b)
Wavenumber, n (cm ) -1
0.14
Cs, Zb = 8 km, Δ Z = 2 km, As = 0.1
0.12
0.1
0.08
tc = 1.0
0.06
0.04
tc = 0.5
0.02
tc = 0.1 tc = 0.0 (clear)

0
0 150 300 450 600 750 900

Rearranged Wavenumber, n * (cm-1 )
Figure 5.34 (a) Bidirectional reflectance in the 1.38 µm water vapor band as a function of wavenum-
ber from 6600 to 7500 cm−1 for clear (τ c = 0) and cirrus (τ c = 1) cloudy atmospheres. A cirrostratus
located at 8 km with a thickness of 2 km in the midlatitude summer atmosphere is used in the calcula-
tion. The surface albedo and the solar and emergent angles used are 0.1 and 60° and 0°, respectively.
(b) Bidirectional reflectance for optical depths τ c = 0.1, 0.5, and 1.0 in the wavenumber domain
rearranged according to the order of clear reflectance values, a monotonic increasing curve (after
Liou et al. 2002a).
was divided into a number of layers, each of which was assumed to be homogeneous. In
each layer, water vapor optical depth τ w (g) = k(g, p, T) ρz was added to total optical
depth, where k(g, p, T) is the correlated k-coefficient in the domain of the g function as well
as pressure and temperature, ρ is water vapor density, and z is the geometric thickness
of the layer. Once the single-scattering properties have been defined for a given g for
each level, monochromatic radiative transfer calculations can be carried out so that the
reflectance for each spectral subinterval (ν = 1 cm−1 ) can be obtained as summation of
the reflectance component in the g-domain.
Figure 5.34a shows the results for clear and cirrus cloudy conditions in which the
midlatitude summer atmosphere water vapor profile is used. The cirrus cloud base is placed
at 8 km with an optical depth of 1. The line structure of water vapor absorption exhibits
significant fluctuations. At about 7100–7400 cm−1 , reflectances in clear conditions are
extremely small because of strong water vapor absorption. Multiple scattering due to ice
crystals contributes to the strength of reflectances in the line-wing regions. It is difficult to
examine the full information content of cirrus clouds because of the complexity of the line
structure. For this reason, bidirectional spectral lines for the clear condition were ordered
according to their magnitudes in such a manner that a monotonic increasing function is
displayed in the rearranged wavenumber domain (Figure 5.34b).
Subsequently, the bidirectional spectral lines for cirrus cloudy conditions were also
ordered in accordance with this rearranged wavenumber domain. Low values represent
reflectances associated with line centers, whereas high values correspond to line wings.
In Figure 5.34b, the largest reflectance in the line wing is about 0.1 for the clear case
corresponding to the surface albedo used in the calculation. For a small optical depth of
0.1, reflectance increases in the center of the water vapor absorption lines, produced by
the scattering contribution of ice crystals. Pronounced scattering events occur for optical
depths of 0.5 and 1, as demonstrated in the rearranged spectra. Larger fluctuations shown
at an optical depth of 1 arise because the scattering and absorption coefficients for ice
particles do not align with water vapor absorption coefficients. In addition to optical depth,
it appears that sufficient and useful information is also available in the rearranged spectra
with reference to ice crystal size and cloud height.
To inspect the information content of cirrus optical depth τ c , mean effective ice crystal
size De , cirrus cloud base height Zb , and solar zenith angle θ 0 , we have carried out a number
of analyses in terms of cloud radiative forcing (CRF), i.e., differences between cloudy and
clear reflectances as defined by R = R(cloudy) − R(clear). Figure 5.35a displays CRF as
a function of rearranged wavenumber in the 1.38 µm band according to the order of clear
reflectance values for τ c = 0.1, 0.5, and 1.0 using De = 42 µm, Zb = 8 km, and θ 0 = 60°.
The cirrus optical depth information is clearly shown by the separation of the curves in the
rearranged wavenumber domain.
In the following investigations, we used an optical depth of 0.1. First, we examined the
effect of ice crystal size on the CRF spectrum assuming other pertinent variables remain
unchanged. Mean effective ice crystal sizes of 24, 42, and 124 µm were used (Figure 5.35b).
The largest CRF difference appears in the line wings unaffected by water vapor absorption
0.006
(a) (b)
0.07 tc = 1.0
tc = 0.5 0.004
tc = 0.1
0.05 0.002
0 De = 24 μm
De = 42 μm
0.03
De = 124 μm
−0.002
Cloud Radiative Forcing
0.01
−0.004
De = 42 μm, q 0 = 60o, Zb = 8 km Zb = 8 km, q 0 = 60o, tc = 0.1

−0.01 −0.006
0 300 600 900 0 300 600 900
0.006 0.016
(c) (d)
z = 6 km q 0 = 30o
z = 10 km q 0 = 70o
0.004 0.012
0.002 0.008
0 0.004
De = 42 μm, q 0 = 60o, tc = 0.1 De = 42 μm, Zb = 8 km, tc = 0.1

−0.002 0
0 300 600 900 0 300 600 900
Rearranged Wavenumber, n * (cm −1 )

Figure 5.35 Differences of cloud and clear reflectances defined by R = R(cloudy) − R(clear),
referred to as cloud radiative forcing (CRF), as a function of rearranged wavenumber for a number of
sensitivity studies: (a) optical depth, (b) ice crystal mean effective size, (c) cirrus cloud height, and
(d) solar zenith angle.
at the rearranged wavenumber of 900 cm−1 . Scattering and absorption by ice particles
contribute to large deviations, and in these cases the reflectances are less than the surface
albedo because of the addition of a small cloud optical depth.
We also inspected the effect on the CRF spectrum of cloud-base height using two
values, 10 and 6 km. Significant differences occurred for rearranged wavenumbers smaller
than 450 cm−1 associated with the center of a water vapor absorption line, as shown in
Figure 5.35c. The reflectance signals in this case exclusively arise from scattering by ice
crystals. For the lower cloud, absorption of water vapor above the cloud substantially
reduces its reflectance. Lastly, we examined the effect of the position of the sun on the
CRF spectrum using solar zenith angles of 70° and 30° (Figure 5.35d). The larger angle,
representing a lower sun position, produces stronger reflectances in the entire rearranged
wavenumber domain. In this case, the effective optical depth for the thin cloud is enhanced
so that ice crystals can undergo more multiple scattering.
5.6.3 Reflected line spectra in the 0.76 µm band

In addition to the preceding presentation using the 1.38 µm band, we wish to point out that
the oxygen A-band (0.76 µm) has a similar absorption feature in the Earth’s atmosphere.
It is produced by transitions from ground electronic states accompanied by vibrational–
rotational transitions and is centered at 13 121 cm−1 (red band). The optical depth of the
A-band can be determined by molecular oxygen (or air) density and therefore is a known
quantity. The absorption of solar irradiance by this band is similar to but not exactly the
same as that by the 1.38 µm water vapor bands. The A-band radiometer sees the Earth’s
atmosphere principally in the upper troposphere and low stratosphere and can therefore be
used to infer the position of high-level cirrus clouds.
In Subsection 5.1.5, we presented the Orbiting Carbon Observatory (OCO), a NASA
Earth System Science Pathfinder Project (ESSP) mission designed to make precise, time-
dependent global measurements of atmospheric CO2 from an Earth-orbiting satellite. OCO
unfortunately failed to reach orbit. The replacement OCO2 was successfully launched in
July 2014. It comprises spectrometers to measure reflected sunlight intensities at three
wavelength bands: the weak 1.6 µm and relatively stronger 2.0 µm CO2 bands, and the
0.76 µm O2 A-band.
Finally, in conjunction with the 1.38 µm H2 O band and the 0.76 µm O2 band, we wish to
present as an historical note the following development. Under the auspices of the NASA
small satellite program, Liou and Goody (1996) proposed a low-cost cloud/climate satellite,
referred to as CIRRUS, making use of the reflected solar line spectra in these two bands.
While this proposal was not selected at that time, it showed that water vapor and oxygen
band measurements made at the same time on the same cloud could yield complementary
information that would enhance the discrimination of the analysis. The line structure
corresponds to photons with levels of line formation at different depths in the cloud and
provides a sounding technique to probe the cloud interior, yielding information that is
not available from conventional broadband solar sensors (see also Section 5.3 for comple-
mentary discussions). Analysis of reflected line spectra has provided much knowledge and
understanding of planetary atmospheres (e.g., West et al. 1986, Goody and Yung 1989).
The proposed CIRRUS would record reflected spectra from all clouds, with an emphasis
on high ice clouds, since these were the least understood (Liou 1992) and the most important
for climate studies (Liou 1986). The variety of ice crystal shapes and sizes and their vertical
Figure 5.36 A theoretical 0.76 µm O2 band spectrum, as a function of wavenumber from 13 000
to 13 180 cm−1 and with a line resolution of 0.5 cm−1 , of a cirrus cloud at 8–10 km above an ocean
surface where the bidirectional reflectance is a function of the solar zenith angle of 60° and the
viewing angle of 0° in the principal plane (φ = 0°). Cirrus optical depths of 1, 0.5, 0.1, and 0.05,
along with a clear atmosphere, were used in the calculation (after Liou and Goody 1996).
profiles presented in Chapter 1 offers an unusual challenge for remote sensing. Reflected
spectra (0.5 cm−1 ) sufficient to resolve the vibration–rotation fine structure of both bands
would be recorded by a small nadir-pointing Fourier transform spectrometer (FTS) once
every 10–12 seconds with a footprint of 50 km and would be calibrated by viewing a
sintered scattering surface illuminated by direct sunlight. It was noted that planetary studies
have succeeded in usefully characterizing the optical properties of clouds on substantially
large scales.
Reflected line spectra in the 1.38 µm band have been presented in Subsection 5.6.2. In
Figure 5.36 we present a theoretical 0.76 µm O2 band spectrum, with a line resolution of
0.5 cm−1 , from Liou and Goody’s proposal (Liou and Goody 1996), involving a cirrus cloud
at 8–10 km above an ocean surface (dark background) where the bidirectional reflectance
is a function of the solar zenith angle of 60° (in) and the viewing angle of 0° (out) in
the principal plane (φ = 0°). Cirrus optical depths of 1, 0.5, 0.1, and 0.05, along with
a clear atmosphere, were used in the calculation. The optical depth is clearly shown in
the spectra. Additionally, the preceding analysis for 1.38 µm spectra could be carried out
for the 0.76 µm spectra to demonstrate the information content of ice crystal size and ice
water content (determined from optical depth) and their vertical profiles (see below). The
challenging subject of vertical ice crystal shape distribution requires further analysis.
5.6.4 Retrieval of cirrus cloud parameters using reflected line spectra

It appears that the relative strength of water vapor lines in the 1.38 µm band or of oxygen
lines in the 0.76 µm, coupled with ice crystal scattering and absorption, could offer vertical
profile information on cirrus clouds on the basis of spectral bidirectional reflectance mea-
surements. Consider a mixture of ice crystals and water vapor in a given cirrus layer with
a thickness of z. The total extinction coefficient and single-scattering albedo for a given
wavenumber for this layer can be expressed as
σe (total) = σe + k(ν), (5.6.1)
σs ∼ c
(ν) = = , (5.6.2)
σe + k(ν) 1 + 2k(ν)/πDe2
where σ e and σ s denote, respectively, the extinction and scattering cross-sections for ice
particles; k(ν) is the absorption coefficient in units of cross-section; and c = σs /σe repre-
sents the single-scattering albedo for ice particles, where the limit of conventional geometric
optics has been used to obtain σe ∼= 2 · π(De /2)2 . We may further use the parameterization
equation developed for optical depth denoted in Eq. (5.2.7) for ice particles. When this
equation is coupled with the absorption optical depth for water vapor, the total optical
depth for the combined system can be written in the form
τ (ν) ∼
= [IWC(c + b/De ) + ρw k(ν)] · z, (5.6.3)
where ρw denotes the number density for water vapor. In the preceding equations, all
variables are dependent on the selected layer.
Consider further an optically thin layer such that single-scattering events prevail. The
spectral reflectance (reflection function) [see Eq. (5.1.3)] and transmittance (transmission
function) are given by
1 P c (De )
R(ν) = [IWC(c + b/De ) + ρw k(ν)] · z, (5.6.4)
4μμ0 1 + 2k(ν)/πDe2
1 P ∗ c (De )
T (ν) = [IWC(c + b/De ) + ρw k(ν)] · z, (5.6.5)
4μμ0 1 + 2k(ν)/πDe2
where reflectance and transmittance have the same form except that their scattering phase
functions are given by P (μ, φ; −μ0 , φ0 ) and P ∗ (−μ, φ; −μ0 , φ0 ), where −μ0 denotes the
cosine of the solar zenith angle (incoming light beam) and the minus sign represents the
downward direction; μ denotes the cosine of the zenith angle (outgoing light beam); and
φ and φ0 are associated azimuthal angles.
In view of the fact that IWC and De are explicitly included in these equations, it appears
feasible to determine these parameters from reflectance measurements. However, the term
P c must be expressed in terms of De or predetermined. We may now consider a two-layer
system such that the combined spectral reflectance from the adding principle of radiative
transfer is given by (Liou 2002)
) *
R12 (ν) = R1 (ν) + T̃1∗ (ν)R2 (ν)T̃1 (ν)/ 1 − R1∗ (ν)R2 (ν) . (5.6.6)
For an optically thin layer, the reflectance and transmittance for light beams from above and
below (with ∗ ) may be assumed to be the same, i.e., T̃1∗ = T̃1 and R1∗ = R1 . Equation (5.6.6)
contains four unknowns: De (1, 2) and IWC(1, 2). Thus, four or more spectral reflectance
measurements may be selected to perform retrievals. A direct inversion is obviously not
feasible, but numerical solutions could be developed for inversion purposes.
Further, the adding principle for reflectance may be extended to a number of layers
such that a suitable vertical profile for De and IWC may be inferred from a set of carefully
selected spectral measurements in the 1.38 µm or 0.76 µm bands. The preceding analysis
illustrates the rich information inherent in the 1.38 µm water vapor and 0.76 µm O2 line
spectra reflected from cirrus cloudy atmospheres, a subject that can be explored for the
remote sensing of cirrus cloud vertical structure. In Section 5.2, we have demonstrated the
retrieval of cirrus optical depth and two ice crystal sizes from available MODIS channels
with validation from independent collocated and coincident observations.
6
Application of light scattering by ice crystals
to climate studies
In Chapter 5, we presented the application of light scattering by ice crystals to remote

sensing from various perspectives. We demonstrated the importance of an ice crystal’s
single-scattering properties, including phase function and polarization, in the retrieval of
optical and microphysical properties of ice clouds using bidirectional reflectance in visible
and near-IR wavelengths and emission in thermal IR window regions. This chapter discusses
the application of light scattering by ice crystals to climate studies.
Climate is usually defined as the average state of the atmosphere observed as weather
in terms of the mean and its statistical deviations that measure the variability over a period
of time. Thus, we may speak of the climate of a season, a year, or longer periods. The
terms climatological temperature, precipitation, and other atmospheric parameters, such as
data averaged over a 30-year period or longer, are frequently used to represent the mean
atmospheric state. Changes in climate involve interactions of the atmosphere with other
parts of the climate system, including the oceans, ice, snow, and land, associated with
the coupling of natural variability and anthropogenic perturbations. Because the impact of
radiative transfer in the atmosphere on dynamic processes takes time to be fully effectual, it
has generally been assumed that its significance is more on climate, a time scale of perhaps
at least a month, rather than on weather forecasting involving a few days. Through radiative
transfer, clouds and cloud feedback play critical roles in regulating climate and climate
change in the Earth–atmosphere system.
An understanding of the Earth’s climate and climate change must begin with a com-
prehensive understanding of the radiation emitted from the sun and the absorption of solar
energy by the atmosphere and the surface, as well as of the emission of thermal infrared
energy from the Earth–atmosphere system. Radiation equilibrium at the top of the atmo-
sphere represents the fundamental mode of the climate system. The incoming solar energy
that is absorbed by the Earth–atmosphere system must be balanced by an equal amount
of emitted thermal infrared energy so as to achieve climate equilibrium. Otherwise, the
equilibrium temperature of the Earth, its brightness as viewed from space, would undergo a
continuous change until the restoration of the energy balance, in which global distributions
of ice clouds (see Subsections 1.2.1 and 5.2.5) play a key role. Global radiation budgets
have been observed from space since the beginning of the meteorological satellite era. The
climate system is also affected by the internal heating and cooling of the atmosphere, as
well as by surface radiative and heat budgets.
327
328 Application of light scattering to climate studies
Cirrus clouds are globally distributed, being present at all latitudes and without respect
to land or sea or season of the year. As presented in Chapters 1 and 5, they regularly
cover about 20–30% of the globe and strongly influence weather and climate processes
through their effects on the radiation budget of the Earth and the atmosphere. These clouds
possess a number of unique features: in addition to being global and located high in the
troposphere, as well as extending to the lower stratosphere on some occasions, they contain
almost exclusively non-spherical ice crystals of various shapes and sizes, as shown in
Sections 1.5 and 1.6. In view of these features, there are significant observational difficulties
in determining the radiative properties of cirrus clouds. From a computational perspective,
a reliable and efficient determination of the radiative properties of cirrus clouds requires
fundamental scattering and absorption data involving complex and intricate ice crystals.
In Chapters 2–4, we provided comprehensive discussions on the scattering and absorption
processes involving ice crystals.
To quantify the effects of reflection and emission by ice clouds in the solar (0.2–5 µm)
and thermal IR (5–100 µm) spectral regions, respectively, the general cloud radiative
forcing concept may be applied (Dickinson 1977, Hartmann et al. 1986). We may define
the flux density (W m−2 ) or simply the flux at the top of the atmosphere (TOA), and
define cirrus solar (shortwave) and thermal IR (longwave) radiative forcings as net flux
differences between clear and cirrus cloudy atmospheres, where the net flux is defined as
the difference between upward and downward fluxes. To understand radiative forcing of
ice crystal clouds, we shall begin with a discussion of the spectral parameterization of the
single-scattering properties of ice crystals for use in cirrus cloud radiative flux calculations
and, where possible, compare these with observations.
6.1 Physical basis for the spectral single-scattering parameterization

of ice crystals
Liou and Wittman (1979) developed a parameterization of the broadband radiative proper-
ties of ice clouds in terms of IWC using the scattering and absorption properties of circular
cylinders. Ebert and Curry (1992) developed a single-scattering parameterization of ice
clouds at solar wavelengths in terms of IWC and De using the results for randomly ori-
ented hexagonal columns and plates provided in Takano and Liou (1989a). Fu and Liou
(1993) presented a parameterization of the single-scattering properties of ice columns and
plates covering both solar and thermal IR regions, accounting for pertinent gaseous absorp-
tion lines, and providing a physical rationale for parameterization that includes extinction,
absorption, and phase function expansion coefficients. Subsequent updates were given by
Fu (1996, 2007) and Gu et al. (2003). Further to these studies, Liou et al. (2008) used a
mixed ice crystal habit (see Figure 6.4) representative of aircraft measured ice crystal data
in radiation parameterization. Edwards et al. (2007) presented a radiation parameterization
scheme employing an ice aggregate model developed by Baran and Francis (2004) which
considered rough surface textures of ice particles. A general review of parameterization
of the single-scattering properties of ice clouds has been given in Baran (2012) and Yang
6.1 Single-scattering parameterization of ice crystals 329
et al. (2015). In Section 6.1.1, we present the physical foundations of the radiation param-
eterization of the scattering and absorption of ice crystals, largely following Fu and Liou
(1993), Takano and Liou (1989a), Gu et al. (2003), and Liou et al. (2008).
6.1.1 Physical foundations

Calculations of single-scattering properties, including phase function, single-scattering
albedo, and extinction coefficient, require a light-scattering program and detailed ice
crystal size and shape distributions. The calculations are usually time-consuming, as shown
in Chapters 3 and 4. If radiation calculations are to interact with an evolving cloud where
particle size distribution varies as a function of time and/or space, the computer time
needed to investigate just this aspect of the radiation program would be formidable, even
with a supercomputer. Thus, there is a practical need to simplify the computational pro-
cedure for calculation of the single-scattering properties of cloud particles. Since spheres
scatter an amount of light proportional to their cross-sectional area, a mean effective
radius, defined as the mean radius weighted by the cross-sectional area of the spheres, has
been used in conjunction with radiation calculations (Hansen and Travis 1974; see also
Subsection 5.2.1).
Ice crystals have complex shapes and ice crystal size distributions are usually expressed
in terms of the maximum dimension. Representation of the size distribution for ice crystals
is much more involved than that for spherical water droplets. To the extent that the scattering
of light is proportional to the cross-sectional area of ice particles, we may use the mean
effective ice crystal size defined in Eq. (1.7.2) as follows:
Lmax Lmax
De = V (L, D) n(L) dL A(L, D) n(L) dL. (6.1.1)
Lmin Lmin
In this equation, we have neglected the factor 32 (see Subsection 5.2.1), because all
parameterization coefficients are to be obtained on the basis of numerical fitting to results
calculated from “exact” light-scattering programs. The volume and geometric cross-section
for randomly oriented columns and plates are given, respectively, by
√
V (L, D) = 3 3LD 2 /8, (6.1.2)
√
A(L, D) = 3D( 3D/4 + L)/4. (6.1.3)
Based on aircraft observations by Ono (1969), Auer and Veal (1970), and Heymsfield
(1972) and the work of Mitchell and Arnott (1994) noted in Subsection 1.6.3, the width D
can be expressed in terms of the length L. It follows that the mean effective ice crystal width
(or size), volume, and geometric cross-section can be defined solely in terms of ice crystal
size distribution, with the ice crystal maximum as the defining parameter. The numerator
in Eq. (6.1.1) is related to the ice water content (IWC, g m−3 ) in the domain of ice crystal
size D in the form
Lmax
IWC = ρi V (L)n(L) dL, (6.1.4)
Lmin
where ρ i is the density of ice. We shall now confine our attention to the mean effective size
to represent ice crystal size distribution in single-scattering analysis for ice crystals. The
extinction coefficient is defined by
Lmax
β= σ (L) n(L)dL, (6.1.5)
Lmin
where σ is the extinction cross-section (area, cm2 ) for a single crystal. In the limits of
conventional geometric optics and using randomly oriented hexagonal columns and plates,
the extinction cross-section is twice the geometric cross-section and is given by [see also
Eq. (2.4.49)]
√
σ = 2A(L, D) = 3( 3D 2 /4 + DL)/2. (6.1.6)
Using the definitions of De , IWC, and β, we have

IWC β
De = i.e., β = constant × IWC/De . (6.1.7a)
ρi 2
On the basis of this analysis, it is clear that the extinction coefficient β and the term IWCDe
are linearly correlated. We may add an intercept to define the extinction coefficient in the
following general form [see also Eq. (5.2.7)]:
β ≈ IWC(a + b/De ), (6.1.7b)
where a and b are certain coefficients. Using observed ice crystal size distributions, it can
be shown that β/IWC and 1/De are linearly correlated in the solar spectral region.
Because cloud absorption is critically dependent on variation of the single-scattering
albedo, it must be accurately parameterized. For a given ice crystal size distribution, the
single-scattering albedo, , is defined by
Lmax Lmax
1− = σa n(L) dL σ n(L) dL , (6.1.8)
Lmin Lmin
where σa denotes the absorption cross-section of a single crystal. When absorption is small,
σa ≈ kλ V , where kλ = 4πmi /λ, mi is the imaginary refractive index, λ is the wavelength
[see Eq. (2.3.18)], and V is the volume of a hexagonal crystal in Eq. (6.1.2). Thus, we have
√
3 3πmi (λ) 2
σa = LD . (6.1.9)
2λ
In the case of strong absorption and under the equal partition principle, the scattering
cross-section is produced by diffraction, which is proportional to the projected geometric
area. At the same time, light rays undergoing geometric reflection and refraction are largely
absorbed and the absorption cross-section is also proportional to the projected geometric
area. Thus, the single-scattering albedo ≈ 1/2 (Fu 1996). Using the extinction cross-
section defined in Eq. (6.1.6) and the absorption cross-section discussed above, and noting
that D is related to L based on observations, we may express the single-scattering co-albedo

in the form
1 − ≈ c + dDe , (6.1.10)
where c and d are again certain coefficients. In the preceding discussion, we have used
the conventional geometric-optics limit to derive expressions for extinction coefficient
and single-scattering albedo. In view of the observed ice crystal sizes in cirrus clouds
(10–2000 µm), the simple linear relationships denoted in Eqs. (6.1.7b) and (6.1.10) should
be valid for solar wavelengths (0.2–5 µm). For thermal IR wavelengths (e.g., 10 µm),
the limits of the conventional geometric-optics approximation without improvement (see
Subsection 3.3.5) may not be sufficiently accurate for small ice crystals. From aircraft
observations, however, it is noted that there is a good linear relationship between the
extinction coefficient in the infrared spectrum and the extinction coefficient based on the
large-particle approximation (Foot 1988).
In the context of randomly oriented ice particles, the phase function is a function of the
scattering angle only [see Eq. (2.4.59a)], and can be expanded in a series of Legendre
polynomials Pl for radiative transfer calculations in the form

M
P (cos ) = l Pl (cos ), (6.1.11)
l=0
where we may set 0 = 1. The first-order expansion coefficient is related to the asymmetry
parameter, first introduced in Eq. (2.4.60b), in the form g = < cos > = 1 /3. Note that
the phase function notation P is the same as the first element of the scattering phase matrix
P11 . Also, in the discussion of radiative flux transfer, the effect of polarization can be
practically neglected. To the extent that the phase function is dependent on De , defined in
Eq. (6.1.1), which accounts for the width, aspect ratio, and shape (roughness) illustrated in
Figures 3.25 and 3.26 (see also Fu 2007), we may postulate that the expansion coefficients
l (l
= 0), like the extinction coefficient and single-scattering albedo, may be expressed
by the following linear equations:
l ≈ el + fl De , l = 1, 2, . . . N, (6.1.12a)
where el and fl are certain fitting coefficients. In radiative transfer, in order to speed up
computations, phase function expansion coefficients may be represented by the Henyey–
Greenstein function (Henyey and Greenstein 1941) in the form
l = (2l + 1)g l , l = 1, 2, . . . N, and g = 1 /3. (6.1.12b)
Equations (6.1.12a), (6.1.12b), (6.1.10), and (6.1.7b) constitute the first-order parameteri-
zation of the extinction coefficient, single-scattering albedo, and phase function in terms of
IWC and De .
6.1.2 A generalized single-scattering parameterization for ice crystals

Fu and Liou (1993) pointed out that higher-order terms in De may be necessary in order to
increase the accuracy of the fitting. For application to general circulation models (GCMs),
the following generalized parameterization equations have been developed (Gu et al. 2003):

N
β(λ; x, y, z) ≈ IWC(x, y, z) [an (λ)/Den (x, y, z)], (6.1.13)
n=0

N
(λ; x, y, z) ≈ 1 − [bn (λ)Den (x, y, z)], (6.1.14)
n=0

N
g(λ; x, y, z) ≈ [cn (λ)Den (x, y, z)], (6.1.15a)
n=0
where (x, y, z) denotes the position in space, and an , bn , and cn are coefficients determined
from numerical fitting based on detailed light scattering and absorption calculations for a
range of ice crystal size distributions and shapes (column and plate types). Fu and Liou
(1993) and Gu et al. (2003) have demonstrated that the first-order polynomial expansion
(N = 1) is sufficient in solar bands to achieve an accuracy within 1% in the fitting. However,
for thermal IR bands, the second-order polynomial fitting (N = 2) is required to achieve
this level of accuracy.
Moreover, Liou et al. (2008) have incorporated polynomial coefficients l (l =
1, 2, 3, 4) into the transfer of solar flux by means of the delta-four-stream approach and
the similarity principle in radiative transfer (Liou 1974, Liou et al. 1988, Liou 2002) to
enhance accuracy in the case of the strong forward scattering nature of cloud and aerosol
particles in solar bands. The parameterization equations can then be written as follows:

N
l (λ; x, y, z) ≈ [cnl (λ)Den (x, y, z)], l = 1 − 4. (6.1.15b)
n=0
In Eq. (6.1.15b), l is to be replaced by the similarity equation l∗ = [l − f (2l +

1)]/(1 − f ), where the forward peak coefficient f = 4 /9, to account for strong forward
scattering by cloud and aerosol particles. Coefficients in the fitting of these equations
have been obtained from the basic scattering and absorption database provided by Yang
et al. (2000a) for solar bands and Yang et al. (2005) for thermal IR bands. In radiation
calculations, it is convenient to use the non-dimensional optical depth τ = βz, where
z is the cloud thickness, and the ice water path (IWP, g m−2 ) = IWCz for a cloud
represented by IWP and De . IWC is a prognostic variable in current GCM and climate
models. However, De is generally prescribed. Liou et al. (2008) have provided a simple
correlation between De and IWC based on the physical definitions of these parameters and
ice crystal size and shape data collected in various field experiments (see Section 1.7). It
follows that, in an evolving cloud field, parameterization of the single-scattering properties
of ice crystals will involve the predicted IWC (t, x, y, z) as a function of time, along with ice
crystal size and shape distribution represented by De (t, x, y, z), from which a corresponding
radiative field will follow. In this manner, cirrus–radiation interactions and feedback can
be fully accounted for in GCM and climate simulations.
In a discussion of the radiative forcing of contrail cirrus contaminated by black carbon,
Liou et al. (2013) further developed the parameterization of the extinction coefficient, the
single-scattering albedo, and four expansion coefficients in the phase function. This work
used the spectrally consistent database developed by Yang et al. (2013), which contains the
scattering, absorption, and polarization properties of 11 randomly oriented ice crystal habits
ranging in size from 2 to 10 000 µm for wavelengths of 0.2–100 µm. This comprehensive
database represents a substantial improvement on the previous databases published by
Yang et al. (2000a) for the solar spectrum (0.2–5 µm) and Yang et al. (2005) for the
thermal IR spectrum (5–100 µm). It also bridges gaps in spectral resolution and provides
improved calculations for small ice crystal size parameters, as well as information relating
to the volume and projected area of each habit. We have adopted this database in the solar
spectrum to obtain single-scattering properties in terms of the extinction coefficient, single-
scattering albedo, and asymmetry factor for the condition of pure ice. We have selected a
simple and representative shape – namely, hexagonal plate – for this study.
Figure 6.1 shows the spectral extinction coefficient, single-scattering co-albedo, and
asymmetry factor as a function of wavelength from 0.2 to 5 µm for mean effective sizes
De of 5, 10, and 20 µm, representing contrail ice crystal sizes (Iwabuchi et al. 2012). In
order to achieve an accuracy within 0.1% for presentation purposes, we have set N = 6.
The spectral extinction coefficient is in units of km−1 and when multiplied by the cloud
thickness results in the optical depth. For De of 10 and 20 µm, the average size parameter is
sufficiently large at visible and near-IR wavelengths that the averaged extinction efficiency
is approximately 2 on the basis of the conventional geometric-optics approach, except in
the vicinity of 3 µm where a significant dispersion of the real part of the refractive index
for ice occurs (Warren and Brandt 2008; see also Figure 2.8). The minimum at 2.85 µm is
referred to as the Christiansen effect (Christiansen 1884, 1885), which occurs when the real
part of the refractive index approaches 1 while the imaginary part is substantially larger,
resulting in domination by absorption. The single-scattering co-albedo pattern, which uses
the logarithmic parameterization to represent extremely small values, mimics the variability
of the imaginary refractive index for ice and generally increases with increasing wavelength.
The maximum asymmetry factor values shown in the vicinity of 2.85 µm for the three mean
effective sizes are related to the Christiansen effect, leading to sharp diffraction peaks in
the phase functions.
6.1.3 Absorption line formation in the scattering of light by ice crystals

In order to connect light scattering by ice crystals to radiative and climate impacts in
the Earth’s atmosphere, we must consider atmospheric line absorption, as presented
in Subsection 5.1.2. Absorption line parameters for various atmospheric gases have
been computed from fundamental quantum mechanics theory. Laboratory data have also
0.10
0.08
De = 5 μm
0.06
0.3 0.04
0.02
De = 10 μm
0.2
0.00
Extinction Coefficient (km -1)
10 0
Single-Scattering Co-Albedo
1.2 0.1 1.0
Asymmetry Factor
De = 20 μm
10 −2
0.8 0.0 0.9
10 −4
0.4 parameterization 0.8
original database
10 −6
0.0 0.7
0.2 0.8 1.4 2.0 2.6 3.2 3.8 4.4 5.0 0.2 0.8 1.4 2.0 2.6 3.2 3.8 4.4 5.0 0.2 0.8 1.4 2.0 2.6 3.2 3.8 4.4 5.0
Wavelength (μm)
Figure 6.1 Spectral extinction coefficient, single-scattering co-albedo (on a logarithmic scale), and asymmetry factor as a function of solar wavelength
from 0.2 to 5 µm for mean effective ice crystal sizes of 5, 10, and 20 µm, for plate-type crystals. The results are derived from the database developed by
Yang et al. (2013). Also shown are parameterization results based on Eqs. (6.1.13)–(6.1.15b) (after Liou et al. 2013).
been available for a limited number of spectral intervals. Based on theory and measure-
ments, line parameters have been compiled over the range 0–17 900 cm−1 , in which data for
more than 1 million lines have been presented (e.g., Rothman et al. 1998). The absorption
line parameters are listed in data form in terms of line position (in cm−1 ), line intensity
[in cm−1 (mol cm−1 ) at 296 K], air-broadened half-width (in cm−1 atm−1 at 296 K), and
lower-state energy (in cm−1 ). The molecular species include H2 O, CO2 , O3 , CH4 , N2 O,
O2 , CO, and other trace gases. Over the years, these line parameters have been refined and
improved (Rothman et al. 2003, Rothman et al. 2013).
The absorption coefficient (in cm2 g−1 ) for a given wavenumber ν is defined as

N
kν (p, T ) = Sj (T )fν,j (p, T ), (6.1.16)
j =1
where j (1, 2, . . . , N) is the index of the absorption coefficient for the jth line, the line
strength S is a function of temperature, the line shape fν is a function of both pressure
and temperature, and the summation signifies all relevant lines making contributions at
the wavenumber ν. In order to resolve individual lines, the absorption coefficient must be
computed at wavenumber intervals smaller than line half-width.
Moreover, the basic radiation parameter in infrared flux calculations, termed spectral
transmittance, covers a small spectral interval ν such that variation of the Planck function
can be neglected. It is defined by
% &

−τ dν dν
Tν̄ (u) = e = exp − kν,j (u)du , (6.1.17)
ν ν ν u j ν
where the optical depth is given by

N
N
τ= τj = kν,j (u)du, (6.1.18)
j =1 u j =1
and u denotes the path length, which is an integration of the density of a specific absorber
over height or pressure coordinates. Thus, in order to calculate the spectral transmittance
exactly, adequate and reliable summations must be performed to cover all absorption lines,
spectral intervals, and non-homogeneous paths.
In the upper stratosphere, absorption and emission processes are dominated by CO2
and O3 . Broadening of the absorption lines is primarily due to the Doppler effect. The
Doppler half-width in the 15 µm CO2 and 9.6 µm O3 bands is 0.0005–0.001 cm−1 . The
spectral interval in these two bands covers about 400 cm−1 . Thus, the absorption must be
calculated at more than half a million points if individual lines are to be resolved. In the
troposphere, absorption due to H2 O predominates. H2 O lines cover essentially the entire
infrared spectrum with a spectral region of about 15 000 cm−1 . These lines are broadened
by collisions, and their half-widths are >0.01 cm−1 . Computations must be performed at
about one million points to resolve H2 O lines. For each wavenumber, there are numerous
lines and atmospheric conditions that must be considered for application to atmospheric
flux radiative transfer, which involves vertical and angular integrations. Exact line-by-line
calculations require formidable computer time, even with a supercomputer. For this reason,
all infrared radiative transfer theories that have been developed are aimed at simplifying
and economizing the computation of spectral transmittance by circumventing integration
over both the spectral interval and the non-homogeneous path length.
A traditional approach referred to as the band model (Elsasser 1938, Goody 1952) covers
a suitably small wavenumber interval in which semi-analytical solutions can be mathemat-
ically derived under homogeneous conditions (a reference temperature and pressure) so
that the tedious integration in the wavenumber domain required in spectral transmittance
calculations can be accomplished a priori. The band model approach led to the develop-
ment of the broadband emissivity concept (Elsasser 1942, Möller 1943, Yamamoto 1952),
according to which atmospheric temperature is used instead of Planck’s function through
the Stefan–Boltzmann law under the condition of Kirchhoff’s thermodynamic equilibrium.
Application to non-homogeneous atmospheres requires suitable adjustment or scaling of the
absorber’s path length and atmospheric pressure referred to as the two-parameter Curtis–
Godson approximation (Curtis 1952, Godson 1953, van de Hulst 1945), or a simplified
adjustment of path length referred to as the one-parameter scaling approximation (Chou
and Arking 1980). A less well-known three-parameter adjustment for ozone cooling rate
calculations has also been developed by Fu and Liou (1992a).
At about the same time as Elsasser’s band model work, an approach based on grouping
of absorption lines within a spectral interval – scrambling and ranking them, so to speak –
for spectral transmittance calculations, referred to as the k-distribution method, was pre-
sented in the work of Ambartzumian (1936) on the equilibrium temperature of stellar
atmospheres. Arking and Grossman (1972) followed this approach to understand the line
shape effect on the temperature of planetary atmospheres. In a homogeneous atmosphere,
spectral transmittance is independent of the ordering of absorption coefficients for a given
spectral interval ν. Consequently, the intricate wavenumber integration may be replaced
by an integration over g(k) space, referred to as the cumulative probability, such that
1
−kν u dν
Tν̄ (u) = e = e−ku dg(k), (6.1.19a)
ν ν 0
where g(0) = 0 and g(k → ) = 1. By definition, g(k) is a monotonically increasing

and smooth function in k space. As a result, the integration in g space, which replaces the
tedious wavenumber integration, can be evaluated as a finite sum of exponential terms. For
example, Figure 6.2a depicts the absorption coefficient kν as a function of ν in a part of
the CO2 15 µm band at a pressure of 600 hPa and a temperature of 260 K, and Figure 6.2b
illustrates the absorption coefficient k(g) as a function of the cumulative probability g. Since
g is a smooth monotonic function, a few integration points will suffice to achieve a high
degree of accuracy in spectral transmittance computations.
The absorption coefficient varies significantly with pressure and temperature in terms
of its half-width and line strength noted above. It follows that in order to apply the k-
distribution method to realistic non-homogeneous atmospheres, variation of the absorption
3 3
(a) (b)
2 2
Absorption Coefficient (log k(g))

Absorption Coefficient (log k)
1 1
0 0
−1 −1
−2 −2
−3 −3
−4 −4
−5 −5
−6 −6
−7 −7
600 620 640 660 680 700 0 0.2 0.4 0.6 0.8 1
Wavenumber (cm-1) Cumulative Probability (g)
Figure 6.2 (a) Absorption coefficient kν in units of (cm atm)−1 as a function of wavenumber with
a resolution of 0.01 cm−1 in the CO2 15 µm band with p = 600 hPa and T = 260 K. (b) Absorption
coefficient on a logarithmic scale as a function of the cumulative probability g in the interval 0–1.
coefficient in the vertical must be accounted for. Therefore, it is necessary to determine

whether the spectral transmittances are equivalent in terms of the following two integrations:
1

dν
Tν̄ (u) = exp − kν du = exp − k(g)du dg. (6.1.19b)
ν u ν 0 u
Because the line profile varies significantly from one pressure–temperature level to
another, the rearrangement of all the lines so that the ν- and g-integrations are the same
is not obvious. The method that assumes the two are equivalent is referred to as the
correlated k-distribution (CKD) method (Lacis et al. 1979, Goody et al. 1989, Lacis and
Oinas 1991, Fu and Liou 1992a). The last two authors proved that the two integrations
are exactly the same for a single line and periodic lines, as well as in the limits of the
weak- and strong-line approximations. Because the CKD approach allows the use of k-
distributions at each altitude, an appropriate Voigt profile can be accounted for throughout
the atmosphere. Numerous improvements for CKD in the limits of the weak- and strong-line
approximations have been made (Mlawer et al. 1997, Jacobson 2005, Li and Baker 2005,
Hogan 2010). Lastly, and most important in the context of light scattering by ice crystals,
the CKD approach can be applied to absorption bands in both solar and thermal infrared
spectra and, at the same time, results from this method can be directly incorporated into
multiple-scattering processes associated with cloud and aerosol particles.
6.1.4 Single-scattering properties of combined ice crystals and absorbing gases

Consider a non-homogeneous atmosphere that is divided into N homogeneous layers. For
the case when both Rayleigh scattering and extinction by cloud particles are mixed with
gaseous absorption, the total optical depth for each layer is given by
τ (g) = τ R + τ M + τ G (g), (6.1.20)
where τ M represents the optical depth due to ice crystals and τ G (g) is the optical depth
contributed by the gaseous absorption for a given g (cumulative probability), which can be
expressed as τ G (g) = k(g, p, T )ρz, where k (g, p, T) is the equivalent k-function, ρ is
the density of absorber, and z is the geometric thickness of the layer. The total ice crystal
optical depth is τ M = τsM + τaM , where τsM and τaM are cloud scattering and
absorption optical depths, respectively. The optical depth produced by Rayleigh molecules
is τ R = βz ≈ (Rp/T ) · z, where p is the pressure (hPa), T is the temperature (K),
and the coefficient R is a function of wavelength. For Rayleigh scattering, the single-
scattering albedo is 1 and the phase function P (cos ) = 3(1 + cos2 )/4, so that expansion
coefficients 1 = 0, 2 = 1/2, and l = 0, where l ≥ 3. Note that contributions from
scattering and absorption by water clouds and various types of aerosols can be added into
Eq. (6.1.20).
Thus, the combined single-scattering albedo is given by
-
(g) = τ R + τsM [τ R + τ M + τ G (g)], (6.1.21)
and the expansion coefficients for the phase function can be obtained from
- R
l = τ R lR + τsM lM τ + τsM , l = 1, 2, 3, 4, (6.1.22)
where lM and lR denote the expansion coefficients for the phase function for ice particles
and Rayleigh molecules, respectively. Since the phase function is independent of gaseous
absorption, the combined l is constant over a given spectral absorption band. Once the
single-scattering properties have been defined for a given g for each level, monochromatic
radiative transfer calculations may be carried out, and the total flux over each spectral band
can be obtained by integrating the flux solution in g space.
After a careful inspection of the location of absorption line spectra in both solar and
thermal IR regions and overlaps as well as consideration of computational efficiency, Fu
and Liou (1993) divided solar and IR spectra into 6 and 12 bands, respectively. These bands
are 0.2–0.69, 0.69–1.3, 1.3–1.9, 1.9–2.5, 2.5–3.5, and 3.5–5.0 µm for the solar spectrum,
and 2200–1900, 1900–1700, 1700–1400, 1400–1250, 1250–1100, 1100–980, 980–800,
800–670, 670–540, 540–400, 400–280, and 280–10 cm−1 for the IR spectrum. In the
former, absorption due to H2 O (2500–14 500 cm−1 ), O3 (50 000–14 500 cm−1 ), CO2 (2850–
5200 cm−1 ), and O2 (A, B, and γ bands) is taken into account. In the latter, absorption due to
H2 O (0–2200 cm−1 ), CO2 (540–800 cm−1 ), O3 (980–1100 cm−1 ), CH4 (1100–1400 cm−1 ),
and N2 O (1100–1400 cm−1 ) is included. The continuum absorption of H2 O is included
in the spectral region 280–1250 cm−1 . The use of CKD leads to an immense numerical
simplification, in which about 10 000 frequency intervals can be replaced by a few g
intervals. Subject to the required accuracy of 1%, the minimum number of quadrature
points in g space ranges from one to about ten for different absorbing gases in different
spectral regions. Further, in order to treat the overlapping of CO2 and H2 O absorption, CKD
has been formulated for a single gas mixture to economize the computational time (Fu and
Liou 1992a). Using a minimum number of g intervals to represent gaseous absorption and
to treat overlap within each spectral interval, 121 spectral calculations are required for each
vertical profile.
In addition to the preceding absorbers, absorption contributions to the 10 µm window
region from chlorofluorocarbons (CFCs), including CFCl3 (CFC-11), CF2 Cl2 (CFC-12),
and CF3 Cl (CFC-13), are accounted for. Also, absorption by the H2 O continuum and a
number of minor absorbers in the solar spectrum, including CH4 , N2 O, NO2 , O3 , CO, SO2 ,
O2 -O2 , and N2 -O2 , are incorporated in Fu and Liou’s parameterization programs (Zhang
et al. 2005, Liou et al. 2008). The latter inclusions lead to an additional absorption of solar
flux in a clear atmosphere on the order of 1–3 W m−2 .
6.2 Radiative transfer in cirrus cloudy atmospheres

6.2.1 The subject of radiative transfer for application to climate studies
It suffices to use the basic scalar equation for the transfer of diffuse intensity in plane-
parallel atmospheres denoted in Eq. (2.5.8) to seek an exact solution for flux density
(W m−2 ) distributions at a given level. For practical reasons, the term “flux density” has
also been called flux. For solar radiation, however, this equation only describes the diffuse
component in which light beams are scattered more than once. For this reason, the direct
component must be included to account for the attenuation of unscattered solar radiation
in atmospheres for extinction processes. Moreover, the azimuthal component of diffuse
radiation can be integrated out of the equation. Consequently, for a given level denoted by
the optical depth τ , the upward and downward flux densities are defined as
1
F ↑ (τ ) = 2π I (τ, μ)dμ, (6.2.1)
0
−1
F ↓ (τ ) = 2π I (τ, μ)dμ + μ0 F e−τ/μ0 , (6.2.2)
0
where μ = cos θ, μ0 = cos θ0 , θ and θ0 denote the zenith angles defining the diffuse and
direct solar radiation, and the term μ0 F represents the downward solar flux density at
TOA. For thermal IR radiation (5–50 µm), the second term in the right-hand side of Eq.
(6.2.2) can be neglected. The net flux density is then given by
F (τ ) = F ↓ (τ ) − F ↑ (τ ). (6.2.3)
After the azimuthal component is decoupled, Eq. (2.5.8) can be exactly solved by
the discrete-ordinates method developed by Chandrasekhar (1950), as presented by Liou
(1973a) for atmospheric flux density calculations. The other exact method uses the adding
principle (van de Hulst 1980, Hansen 1971). Takano and Liou (1989a) and Liou and
Takano (1994) applied the adding method for radiative transfer to oriented and complex
ice crystals. The “exact” solutions for radiative transfer require substantial computational
efforts, particularly so when atmospheric vertical inhomogeneity is taken into consideration.

In Subsection 6.1.4, we presented the single-scattering properties of a mixture of ice crystals,
absorbing gases, and air for use in a given atmospheric layer. Because of the computational
consideration for radiation processes in GCM and climate models, it is practical to develop
simplified radiative transfer methods with accuracy known.
Schuster (1905) and Schwarzschild (1906), two prominent astrophysicists, originated the
two-beam (upward and downward) approximation for radiative transfer. Liou (1973b) and
Meador and Weaver (1980) developed the modern two-stream approximation for application
to scattering by clouds and aerosols. To understand the physical structure of the interior of
a star, Eddington (1916) developed an expansion of the intensity using two-term Legendre
polynomials, now referred to as Eddington’s approximation for radiative transfer. These
two approximations are similar, yielding reasonable results for optically thick layers in
comparison with more exact calculations, and have been used for application to radiative
flux calculations in association with climate studies. However, they produce inaccurate
results for thin layers and when significant absorption is involved.
Further to these studies, the four-stream approximation concept for radiative transfer,
first derived by Liou (1974), has been employed to enhance flux calculations without
substantially increasing computational efforts. As with the two-stream approximation,
analytical solutions in the form of more involved mathematical coefficients can be derived
for two upward and downward streams in this approximation, with enhanced computational
accuracy. Scattering by atmospheric cloud and aerosol particles is highly peaked in the
forward direction, particularly for ice crystals for which the forward scattered energy
within 5° scattering angles produced by diffraction is 5–6 orders of magnitude greater
than that in the side- and backscattering directions (see, e.g., Figures 3.25 and 3.26). In
the case of cloud and aerosol particles, we may carry out the so-called delta-function
adjustment, using the similarity principle for radiative transfer (Sobolev 1975, Joseph et al.
1976, Wiscombe 1977, van de Hulst 1980, Liou 2002) such that expansion coefficients in
the phase function l are replaced by the similarity equation as follows:
l∗ = [l − f (2l + 1)]/(1 − f ), l = 1, 2, . . . , N, (6.2.4)
where the forward peak coefficient is f = 2 /5 (l = 1) for two-stream and f = 4 /9 (l =
4) for four-stream, to account for strong forward scattering by cloud and aerosol particles.
A systematic development of the solution for these two approximations has been presented
by Liou et al. (1988).
Fu et al. (1997) demonstrated that the delta-two-stream method is the most computa-
tionally efficient but produces significant errors in fluxes and heating rates under cloudy
conditions. High accuracy can be obtained by using the delta-four-stream method, but sub-
stantial computer time is required for the calculation of thermal infrared radiative transfer.
The delta-two/four-stream combination method is sufficiently economical for IR calcu-
lations; it is four times faster than the delta-four-stream method but only 50% slower
than the two-stream method, and at the same time produces acceptable accuracy under
most atmospheric conditions. For application of the delta-four-stream scheme, we may
divide the atmosphere into N homogeneous layers with respect to extinction coefficient,
single-scattering albedo, and phase function. The bottom layer can be considered as the
Lambertian surface albedo, such that it reflects the incoming solar flux without transmis-
sion. The 4 × N unknown coefficients in the analytic solution of the transfer equation are
determined by matching internal continuity boundary conditions, following the procedure
described in Liou (1975b). A numerically stable program has been developed to solve
the system of linear equations using the property that the coefficient matrix is a sparse
matrix.
Gu et al. (2003) implemented in the University of California, Los Angeles (UCLA)
atmospheric GCM (AGCM) a combination of the delta-four-stream approximation for solar
flux calculations and the delta-two/four-stream approximation for IR flux calculations. This
combination has proved to be computationally efficient and at the same time to produce
a high degree of accuracy. In addition to water and ice clouds, Gu et al. (2006) have
incorporated various aerosol species in the UCLA AGCM for investigation of the impact
of black carbon and dust on cloud configurations and dynamic and circulation patterns. Gu
et al. (2011) further incorporated Fu and Liou’s radiation parameterization into the Weather
Research Forecast (WRF) program. The Fu–Liou–Gu code, which is unique in treating the
light scattering properties of ice particles in line absorbing atmospheres, is now available in
the physics package of the WRF program managed by the National Center for Atmospheric
Research (NCAR).
6.2.2 Radiative transfer model results versus aircraft observations for cirrus clouds
On the basis of the preceding discussion, we may perform radiative flux transfer calculations
in cirrus cloudy atmospheres once the location and thickness of cirrus are defined and
atmospheric absorbing gas and temperature profiles and surface albedo are given as input.
Flux measurements of stratocumulus clouds were reported by Stephens et al. (1978).
For cirrus clouds, which require high-flying aircraft, a cirrus intensive field observation
(IFO) was conducted in Wisconsin from October 13 to November 2, 1986 as a major
component of the First ISCCP Regional Experiment (FIRE) to improve our understanding
of cirrus clouds and their influence on weather and climate processes (Starr 1987). In
the cirrus IFO, simultaneous microphysical and radiation measurements were made using
high-altitude research aircraft, allowing comparisons between model calculations and in situ
observations. In this section we present a comparison of model-derived infrared emissivity
and solar albedo calculations with measurements from the NCAR Sabreliner aircraft on
October 28, 1986.
The Sabreliner was flown near the western shore of Lake Michigan in the vicinity of
Green Bay, Wisconsin, in a thin banded cirrus layer, between 1530 and 1630 UTC. Cloud-
top and base heights were 11.1 and 8.9 km, respectively. A racetrack pattern was flown
at six different levels throughout the cirrus. The mean solar zenith angle during the flight
was 61.3°. Since radiative and microphysical characteristics of the cloud sampled on the
south side of the racetrack pattern (referred to as cloud 1) are different from those sampled
on the north side (referred to as cloud 2), data from each side were analyzed separately.
Furthermore, each side of the racetrack pattern was divided into thinner, mean, and thicker
cases to minimize sampling (Smith et al. 1990, Heymsfield et al. 1990).
From an observational perspective, scattering and absorption of ice particles take place
simultaneously along with pure H2 O absorption. Thus, for clouds imbedded in the atmo-
sphere, the conventional Kirchhoff thermodynamic equilibrium for emission and absorp-
tion may not be adequately applicable. Following Paltridge and Platt (1981), the “effective”
upward and downward emissivities involving a non-black cloud may be defined in the
forms
↑ ↑
FL (zt ) − FL (zb )
ε↑ = ↑
, (6.2.5a)
σ T̄ 4 − FL (zb )
and
↓ ↓
FL (zb ) − FL (zt )
ε↓ = ↓
, (6.2.5b)
σ T̄ 4 − FL (zt )
↑ ↓
where FL and FL are thermal infrared (longwave) broadband upward and downward fluxes,
respectively; zt and zb are cloud-top and base heights, respectively; T̄ is the mean cloud
temperature; and σ T̄ 4 is the flux emitted by a blackbody at T̄ .
Moreover, to determine the solar albedo for clouds, which differs from surface albedo,
we must consider cloud absorption. Let α be the solar albedo with reference to the incoming
solar radiation at cloud top, α ∗ be its counterpart for radiation incident from cloud bottom,
and A be cloud absorptance. We may then relate the downward and upward solar fluxes at
the cloud top and bottom as follows:
Fs↓ (zb ) = (1 − A − α)Fs↓ (zt ) + α ∗ Fs↑ (zb ), (6.2.6a)
Fs↑ (zt ) = (1 − A − α ∗ )Fs↑ (zb ) + αFs↓ (zt ), (6.2.6b)

↑ ↓
where Fs and Fs denote solar (shortwave) broadband upward and downward fluxes,
respectively. Adding Eqs. (6.2.6a) and (6.2.6b) yields
) *
Fs↓ (zb ) + Fs↑ (zt ) = (1 − A) Fs↓ (zt ) + Fs↑ (zb ) . (6.2.7)
We can then solve for A to obtain

) ↓ ↑ * ) ↓ ↑ *
Fs (zt ) − Fs (zt ) − Fs (zb ) − Fs (zb )
A= ↓ ↑
, (6.2.8)
Fs (zt ) + Fs (zb )
where the numerator represents the net solar flux divergence for cloud. In this manner, the
downward fluxes at cloud top and bottom, and the upward flux at cloud bottom resulting
from reflection from the surface and from the atmosphere below can be accounted for. To
a good approximation for a single layer, we may consider α α ∗ , to obtain
Fs↓ (zb ) = (1 − A − α)Fs↓ (zt ) + αFs↑ (zb ). (6.2.9)

1.2 1.2
(a) (b)
Effective Downward Emissivity

De = 25 μm De = 25 μm
Effective Upward Emissivity
1.0 1.0
50
50 75
75 100
0.8 100 0.8 125
125
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
Ice Water Path (g m –2) Ice Water Path (g m –2)
1.0 0.8
(c) (d)
De = 25 μm
0.8 Observation (Stackhouse and Stephens 1991)
0.6 Cloud Sample
1 2
Solar Albedo
50
Solar Albedo
thinner thinner
0.6 mean mean
De = 25 μm
thicker thicker 75
0.4
50 Present Model
100
0.4
75 125
100
0.2
0.2
125
0.0 0.0
0 20 40 60 80 100 0 0.2 0.4 0.6 0.8 1 1.2
Ice Water Path (g m –2) Effective Downward Emissivity
Figure 6.3 (a) Effective upward emissivity, (b) effective downward emissivity, and (c) solar albedo
as functions of ice water path (IWP). Model results representing mean effective ice crystal sizes of
25, 50, 75, 100, and 125 µm are compared with observed values obtained from the cirrus FIRE IFO.
(d) Solar albedo as a function of effective downward emissivity (after Fu and Liou 1993).
Solving for α, we obtain the equation for cloud solar albedo in the form
↓ ↓
Fs (zt )(1 − A) − Fs (zb )
α= ↓ ↑
. (6.2.10)
Fs (zt ) − Fs (zb )
Figures 6.3a–c show computed and observed effective upward and downward emissivi-
ties and solar albedo as a function of IWP. The broadband flux observations of Stackhouse
and Stephens (1991) for the October 28, 1986 FIRE case are included for comparison, along
with error bars to show measurement uncertainties. In theoretical calculations, atmospheric
temperature and moisture profiles were averaged over the two soundings from Green Bay at
1400 and 1730 UTC. For CO2 , CH4 , and N2 O, uniform mixings were assumed throughout
the atmosphere with concentrations of 330, 1.6, and 0.28 ppm, respectively. The ozone,
temperature, and moisture profiles above the sounding level were the same as the U.S.
1976 Standard Atmosphere. The cirrus cloud was located between 8.9 and 11.1 km, while
IWP was varied by changing the assumed IWC but keeping the cloud position fixed. Ice
saturation mixing ratio was assumed in the cloud for water vapor absorption calculations.
Mean effective sizes of 25, 50, 75, 100, and 125 µm, as defined in Eq. (6.1.1), were
used, corresponding to the five solid lines for effective upward and downward emissivities
and solar albedo in these figures. The solar zenith angle and surface albedo used were 61.3°
and 0.072, respectively, following observed values.
Comparing theoretical results to observed data, mean effective ice crystal sizes for these
cases lie in the range of 50–100 µm for emissivities and 50–75 µm for solar albedo. The
mean effective ice crystal size for effective upward and downward emissivities in each case
appears to be consistent in view of observational uncertainties; however, the size associated
with solar albedo is smaller. This difference can be explained by the fact that solar albedo
is more sensitive to ice crystal size spectra at cloud top. Based on observations (Heymsfield
et al. 1990), ice crystals progressively range from smaller sizes near cloud top to larger
sizes at cloud base (see Figure 1.18). For cloud 1, these figures show that De increases
with increasing IWP, because larger IWPs are usually correlated with larger ice crystal
sizes. As shown in Eqs. (6.2.5a) and (6.2.5b), effective emissivities are defined in terms of
fluxes that contain components due to the scattering and absorption of cloud particles and
gaseous absorption, principally by H2 O. For this reason, computed effective emissivities
may exceed 1 for large IWPs and may deviate from 0 when IWP → 0. Figure 6.3d shows the
solar albedo versus effective downward emissivity for a number of mean effective ice crystal
sizes. The observed data derived from the October 28, 1986 FIRE case are also displayed.
In the domain of albedo–emissivity, the effect of ice crystal size distribution is fairly small.
Also, reasonable agreement is shown between theoretical results and observations.
In light of the preceding discussion, the observed cloud emissivity and cloud albedo
can be reasonably interpreted from theoretical results using the scattering and absorption
properties of column and plate ice crystals. The actual ice crystals that were collected by the
optical probe aboard the Sabreliner during the FIRE cirrus experiment were largely bullet
rosettes, columns, plates, and compacted spatial ice particles with occasional extensions
(Heymsfield et al. 1990). Although exact ice crystal shapes were not specified in the
scattering and absorption calculations, theoretical results using a combination of columns
and plates appear to account for the general hexagonal structure of ice particles, and
yield comparable flux results. This perhaps explains some of the success of the Fu–Liou
parameterization scheme in the context of interpreting aircraft observed data.
Figure 6.4 shows broadband solar albedo as a function of IWP. The solid lines are the-
oretical results computed from a line-by-line equivalent model (higher spectral resolutions
than Fu and Liou’s parameterization) using a set of observed ice crystal size distributions for
a mixture of bullet rosettes (50%), hollow columns (30%), and plates (20%), proposed as a
global ice crystal shape factor for use in radiative forcing analysis. Also presented are results
derived from aircraft observations for cirrus clouds during the FIRE experiment in Wis-
consin in October–November, 1986, depicted in Figure 6.3c. The shaded area in Figure 6.4
shows results based on these size distributions converted to equivalent spheres. Regardless
of the input parameters for spheres, the theoretical results significantly underestimate the
observed values, primarily because of the stronger forward scattering for spherical particles
and stronger absorption for spheres at near-infrared wavelengths (Kinne and Liou 1989).
Using the mean effective ice crystal size defined in Figure 1.23, the ice crystal size that
1
Observation
50% bullet rosettes, 30% hollow columns, and

20% plates
Equivalent Ice Spheres
0.8
Cloud Solar Albedo
De = 24 μm (Cold Ci)
0.6
42 μm (Cs)
0.4 75 μm (Nov 1)
124 μm (Ci Uncinus)
0.2
0
0 20 40 60 80 100
Ice Water Path (g m–2)
Figure 6.4 Solar albedo as a function of ice water path, determined from broadband flux observations
of cirrus clouds from aircraft during the FIRE experiment, Wisconsin, October–November, 1986
(Stackhouse and Stephens 1991). The solid lines represent theoretical results computed from a line-
by-line equivalent solar model using observed ice crystal sizes and shapes for a range of mean
effective ice crystal diameters, as shown in Figure 1.23. The shaded area shows corresponding results
for equivalent spheres.
best fits the observed data lies between 50 and 75 µm, typical values at the top portion of
midlatitude cirrus cloud systems and in line with the conclusion presented in Fu and Liou
(1993).
6.2.3 Radiative transfer model results compared with satellite observations

The Earth radiation budget (ERB) at TOA has been derived from satellite observations since
the beginning of the satellite era, spanning over 50 years (Suomi 1958). The advancement
of scanning radiometers allows the measurement of reflected shortwave (solar) and long-
wave (IR) radiation. However, the determination of outgoing broadband flux density (or
simply flux) from measured spectral radiances (per solid angle) has required conversion of
measured filtered radiances to radiances covering the entire solar and thermal IR spectrum,
integration over all angles of measurements using bidirectional models, as well as appro-
priate time averaging procedures. Moreover, the transformation of narrow spectral interval
data from NOAA polar-orbiting and geostationary satellites to the broadband estimation of
fluxes has required a number of assumptions and models. Lastly, the development of the
self-calibrating cavity radiometer has led to high-accuracy, high-stability measurements of
incoming solar irradiance. The ERB is usually presented in terms of the emitted longwave
flux, referred to as the outgoing longwave radiation (OLR); the planetary albedo, which is
the ratio of the reflected solar flux to the incoming solar flux at TOA; and the net radiative
flux, which is the difference between the reflected solar flux and OLR. Global distributions
of the mean annual planetary albedo, OLR, and net radiative flux have been presented by
Vonder Haar and Suomi (1971), Stephens et al. (1981), and Hartmann (1994).
Satellite observations of incoming and reflected solar fluxes and emitted thermal IR flux
over the last 40 years have provided the radiation equilibrium of the Earth–atmosphere sys-
tem, representing the fundamental mode of the climate system. However, reliable observed
solar and thermal IR fluxes within the atmosphere and at the surface have been extremely
limited in time and space and insufficient for use in climate study. Increasing concentra-
tions of greenhouse gases, chiefly CO2 , would force secular changes in the vertical profiles
of fluxes, resulting in a warmer troposphere and cooler stratosphere. In climate models,
uncertainties in the simulation of the surface and atmospheric radiation budget (SARB)
would translate directly into uncertainties in climate simulations. In an attempt to address
these and related issues, Charlock and Alberta (1996) developed a conceptual framework by
designing an experiment, referred to as the cooperative CERES/ARM/GEWEX experiment
(CAGEX), at the Department of Energy’s Atmospheric Radiation Measurement (ARM)
Southern Great Plains (SGP) Cloud and Radiation Testbed (CART) site in Oklahoma. The
experiment, described in Figure 6.5, was carried out on a 3 × 3 grid, 0.3° × 0.3° on each
side, every 30 min from 1409 UTC to 2239 UTC for 26 days, beginning on April 5, 1994. It
was intended to provide (1) satellite-based cloud properties and atmospheric sounding data
that fulfill input requirements for radiative transfer model calculations, (2) vertical profiles
of radiative fluxes calculated with those data as input, and (3) measurements of radiative
fluxes and cloud properties for comparison with some of the flux calculations and input data.
Cloud properties for input to the radiative transfer code (Fu and Liou 1993), modified
by the inclusion of scattering and absorption by aerosols, were obtained from Minnis et al.
(1995), which also provided empirical estimates of the TOA planetary albedo (albedo) and
OLR, along with temperature and humidity soundings obtained by interpolating data from
standard radiosondes. Satellite-derived temperatures above 100 hPa and O3 were used. For
each grid-box, CAGEX provided a sounding from the Mesoscale Atmospheric Predic-
tion System (MAPS). Because of uncertainties relating to upper-tropospheric humidity in
soundings, a climatological humidity above 300 hPa was employed. With respect to surface
(a) (c)
(b) (d)
Figure 6.5 The CERES/ARM/GEWEX Experiment (CAGEX) for comparison between satellite
ERB observations and simulations by the radiation parameterization model using observed atmo-
spheric and surface parameters as input. (a) Map of the CAGEX horizontal 3 × 3 grid surrounded by
ARM surface facilities. (b) 3-D depiction of 48 vertical levels with 3 × 3 horizontal grids. (c) GOES-7
pixels (Minnis et al. 1995) in 3 × 3 horizontal grids. (d) Time series of the TOA albedo calculated
from the Fu–Liou code using MAPS and core data and the GOES-inferred value; calculated and
observed (pyrgeometer) surface downward longwave flux; and satellite-retrieved cloud-top and base
heights, along with the surface lidar-measured cloud base (after Charlock and Alberta 1996).
data, the optical properties of the surface and of aerosols in the forward flux calculations
were taken partly from measurements at the ARM SGP CART site. MicroPulse lidar mea-
sured the altitude of the lowest cloud base above the central facilities. The total aerosol
optical depth in solar bands was obtained from the Multi Filter Rotating Shadowband
Radiometer (MFRSR) spectral measurement of the direct and diffuse solar beam at the
surface.
For the Fu–Liou radiative transfer calculation, aerosol optical depths in the near and
thermal infrared were determined by scaling the MFRSR values at shorter wavelengths
using the humidity-dependent tables of d’Almeida et al. (1991). The d’Almeida et al.
tables for continental aerosol were used to estimate the aerosol single-scattering albedo and
asymmetry parameter at all wavelengths. Aerosols were apportioned by altitude using a
distribution function. The core soundings, which were based on interpolations from 12-hour
radiosonde data, were used for calculations over the 3 × 3 horizontal array of CAGEX. Both
the core and MAPS soundings produce mean TOA albedos with the Fu–Liou code that are
quite close to the observed albedos of Minnis et al. (1995) for clear skies (Figure 6.5d,
top graph). Comparison of the calculated downward fluxes at the surface based on the core
and MAPS soundings (Figure 6.5d, middle graph) with pyrgeometer observations displays
substantial differences. The satellite-retrieved cloud base height also substantially differs
from the value determined from surface lidar observations (Figure 6.5d, bottom graph).
Figures 6.6a–6.6d compare radiative fluxes calculated from the Fu–Liou code, using the
preceding cloud, surface, and atmospheric data as input, with satellite-observed shortwave
(SW) albedo, surface insolation, OLR, and surface downward longwave (LW) fluxes. The
term “Bias” in this figure indicates the differences between the two. Figure 6.6a, comparing
observed and calculated total-sky albedo, displays significant scatter throughout its range;
clear-sky results are much better. Calculated surface insolation for clear-sky conditions is
too large (Figure 6.6b), in agreement with Wild et al. (1995). Calculated surface insolation
for total sky conditions is again too large, in agreement with the GEWEX SRB Project
(Alberta et al. 1994, Whitlock et al. 1995). Both are shown in Figure 6.6b. The bias in
total-sky surface downward flux using MAPS is 42 W m−2 for a mean TOA insolation of
985 W m−2 . The daylight CAGEX (local scale) bias would be approximately equivalent to
an annual global, 24-hour bias of 15 W m−2 . If more accurate aerosol composition, size
distribution, and spectral indices of refraction were known a priori, it appears that the bias
in comparison could be significantly reduced.
For LW radiation, TOA observations were considered to be the empirical conversion of
narrowband GOES-7 to OLR. Surface LW observations were obtained from the pyrgeome-
ters at the ARM central facility. As with shortwave, Fu–Liou calculations using the MAPS
soundings were generally close to observations. MAPS results are featured in the scatter
plot (Figures 6.6c and 6.6d). The total-sky bias becomes slightly negative at low values of
OLR (Figure 6.6c). Conditions of intermediate OLR are associated with intermediate cloud
cover conditions, such as partial sky cover and cloud overlap, which challenge satellite-
based retrievals of cloud properties; the scatter in Figure 6.6c is large for intermediate OLR.
The small bias and scatter in the observed versus computed surface downward LW flux
Fu–Liou vs. Measurements (MAPS sounding)

April 1994 ARM IOP Center CAGEX Grid-box
(a) TOA Shortwave Albedo (c) Outgoing Longwave Radiation
1.0
Bias Bias
Total-Sky 0.014 300
Total-Sky 5
GOES-7/LBTM OLR (W m–2)

0.8 Clear-Sky 0.004 Clear-Sky 12
GOES-7/LBTM Albedo
250
0.6
200
0.4
0.2 150
Total-Sky Total-Sky
Clear-Sky Clear-Sky
0.0 100
0.0 0.2 0.4 0.6 0.8 1.0 100 150 200 250 300
Fu–Liou Albedo Fu–Liou OLR (W m–2)
(b) Shortwave Surface Insolation (d) Surface Downward Longwave

Bias Bias
1000 400
Total-Sky 43 Total-Sky 1
Measured SFC Down (W m–2)
Clear-Sky 32
Pyrgeometer DLF (W m–2)
Clear-Sky -6
800
350
600
300
400
250
200
Total-Sky Total-Sky
Clear-Sky Clear-Sky
0 200
0 200 400 600 800 1000 200 250 300 350 400
Fu–Liou SFL Down (W m–2) Fu–Liou DLF (W m–2)
Figure 6.6 Comparisons between calculations using Fu–Liou code, with input from satellite-
retrieved clouds, and MAPS soundings, with observations of clear-sky and total-sky conditions:
(a) TOA shortwave albedo, (b) SW surface insolation, (c) outgoing longwave radiation, and
(d) surface downward longwave (after Charlock and Alberta 1996).
attest to the surprising fidelity of the surface LW cloud forcing inferred from the retrievals
of cloud areal coverage and altitude of cloud base from Minnis et al. (1995) (Figure 6.6d).
The CAGEX biases for LW fluxes appear to be more straightforward than for SW. In LW,
a shift to the more time-intensive MAPS soundings as input for the radiative transfer code
generally improved the match between calculations and observations. At both TOA and
the surface, comparison of clear-sky calculations with observations suggests that the real
atmosphere is more opaque than the modeled atmosphere; the clear-sky OLR calculated by
the Fu–Liou code exceeds observations by 12 W m−2 ; the corresponding clear-sky surface
downward LW is 6 W m−2 smaller than observations. More opaque H2 O continuum data
would be more likely to affect the surface flux than that at TOA. The OLR error could
be caused by an error in the sounding at high altitude, by the radiative transfer code, by
a misrepresentation of surface temperature or emissivity at the central facility, which may
not represent the whole grid box adequately, or by the OLR observation itself.
6.3 Radiative forcing of cirrus clouds

6.3.1 A discussion of the radiative forcing of cirrus clouds
The amount of sunlight that cirrus clouds reflect, absorb, and transmit depends on their
coverage, position, thickness, and ice crystal size and shape distributions. Cirrus clouds can
also reflect and transmit the thermal infrared emitted from the surface and the atmosphere
and, at the same time, emit infrared radiation according to the temperature structure within
them. Ice crystal size and shape distributions and cloud thickness are fundamental cirrus
parameters that determine the relative strength of the solar albedo (reflecting of sunlight)
and infrared greenhouse (trapping of thermal infrared radiation) effects, which are essential
components of the discussion of cirrus clouds and climate. These radiative effects are deter-
mined by the basic scattering and absorption properties of ice crystals. Unlike the scattering
of light by spherical water droplets (which can be solved by the Lorenz–Mie theory), an
exact solution for the scattering of light by non-spherical ice crystals, covering all sizes and
shapes that occur in the Earth’s atmosphere, does not exist in practical terms. In Chapter
3, we advanced this area and demonstrated that the scattering and absorption properties
of ice crystals of all sizes and shapes, which commonly occur in the atmosphere, can be
calculated with high precision by a unified theory of light scattering which incorporates
a surface-wave (edging effect) component into an improved geometric-optics approach,
specifically for computations of the extinction coefficient, the single-scattering albedo, and
the asymmetry factor for ice crystal size parameters larger than 0.1–1. It is particularly
useful in the application of radiation parameterization to GCM and climate models. To
comprehend the impact of cirrus clouds on the radiation field of the Earth and the atmo-
sphere and thus on climate, the term “cloud radiative forcing” has been used to quantify
the relative significance of the solar albedo and infrared greenhouse effects. Cloud radia-
tive forcing is the difference between the radiative fluxes at TOA under clear and cloudy
conditions. The addition of a cloud layer in a clear sky would lead to more sunlight being
reflected back into space, reducing the amount of solar energy available to the atmosphere
and the surface. In contrast, the trapping of atmospheric thermal IR emission by non-black
cirrus clouds enhances the radiative energy, or heat, available in the atmosphere and at the
surface. Accordingly, cloud radiative forcing is defined as follows:
CL = FLcl − FLov and CS = FScl − FSov , (6.3.1)

where cl and ov denote clear and overcast conditions, respectively, and FL and FS are the
upward IR (longwave) and solar (shortwave) fluxes, respectively, at TOA. The net cloud
radiative forcing is the sum of shortwave and longwave radiative forcings, as follows:
C = CL + CS . (6.3.2)
For partly cloudy conditions, if the cloud cover is denoted as η, the cloud radiative forcing
is given by ηC.
6.3.2 Theoretical analysis for cirrus radiative forcing

In this subsection, we present cirrus radiative forcing results from the perspective of radiative
transfer parameterization discussed above, under the condition that cirrus cloud cover is
100% and its position is located between 8.9 and 11.1 km, a typical value in midlatitude.
Figures 6.7a–6.7c present CL , CS , and C as a function of IWP for different mean effective
ice crystal sizes. IR cloud forcing CL (Figure 6.7a) is always positive, corresponding to
the heating of the Earth–atmosphere system due to the greenhouse effect of clouds, while
solar cloud forcing CS (Figure 6.7b) is always negative, showing the cooling of the system
by the cloud albedo effect. The terms CL and C are strongly dependent on both IWP and
De . For an IWP of 30 g m−2 , CL decreases from 131 to 54 W m−2 as De increases from 25
to 125 µm. The solar cloud forcing counterpart increases from −114 to −24 W m−2 . Net
cloud forcing (Figure 6.7c) shows that cirrus clouds have a net heating effect on the system
except for IWP > 50 g m−2 when De = 25 µm. These results are similar to those derived
from a climate model presented by Stephens et al. (1990).
In general, small De is associated with cold cirrus clouds that have small IWP. We may
safely assume that midlatitude cirrus clouds would have positive cloud forcing; that is, the
greenhouse effect produced by the presence of these clouds is more pronounced than the
albedo effect. The net greenhouse effect is a function of both ice crystal size and IWP.
The data points are computed from ten observed ice crystal size distributions using a cloud
thickness of 2 km. The visible optical depth of these clouds ranges from about 0.15 to 1.5.
Assuming a 20% cloud cover, the cloud forcing values generally range from about 4 to
8 W m−2 . When coupled with positive radiative forcings, such as those due to the increase
of greenhouse gases, feedback produced by an increase or decrease of cirrus parameters
could be significant.
The preceding theoretical calculations show that the infrared greenhouse effect for cirrus
clouds generally outweighs their solar albedo counterpart, except when the clouds contain
very small ice crystals on the order of a few micrometers, which exert a strong solar albedo
effect. The relative significance of the solar albedo versus infrared greenhouse effects is
clearly dependent on the ice crystal size and the amount of ice in the cloud. Because of the
complexity of sorting cirrus signatures from satellite observations, actual data to calculate
the global cirrus cloud radiative forcing are yet to become available.
We further looked into cloud radiative forcing at the surface. In the solar case, negative
cloud forcing values produced by cloud reflection are substantially similar to those presented
200
Top IR Cloud Forcing (W m–2)

(a)
150
100
50
0
0 20 40 60 80 100
0
Top Solar Cloud Forcing (W m–2)
(b)
-50
-100
De = 25 μm
-150 50
75
100
125
-200
0 20 40 60 80 100
100
(c)
Top Cloud Forcing (W m–2)
50
-50
-100
0 20 40 60 80 100
Figure 6.7 Cloud radiative forcing at TOA as a function of ice water path for mean effective ice
crystal sizes of 25, 50, 75, 100, and 125 µm: (a) thermal IR (longwave), (b) solar (shortwave), and
(c) net. For solar radiation, the solar zenith angle and surface albedo used were 60° and 0.1, respec-
tively, along with a 12-hour solar day (after Fu and Liou 1993).
in Figure 6.7b, in view of the fact that, without clouds, the atmosphere is largely transparent
with respect to solar radiation. For the IR case, the insertion of cirrus clouds produces
small effects at the surface (0–30 W m−2 ) because of strong absorption by the atmosphere
at thermal IR wavelengths between the surface and cloud. Cirrus cloud-induced IR heating
within the atmosphere and solar cooling at the surface could have a significant impact on
atmospheric circulation and temperature distribution, as shown by Ramanathan (1987) and
Slingo and Slingo (1988).
In Section 1.7, we demonstrated that the mean effective ice crystal size is statistically
correlated to the ice water content (IWC) in the form
De ≈ exp{aj + bj ln (IWC) + cj [ln (IWC)]2 }, j = 1, 2, 3, (6.3.3)
where aj , bj , and cj are fitting coefficients (see Subsection 1.7.2) and the indices 1, 2, and
3 divide the Earth into tropical (0–30° latitude), midlatitude (30–60° latitude), and Arctic
(60–90° latitude) regions, respectively, in view of their unique ice microphysical properties
associated with temperature and dynamic structures. For the tropics, a1 = 5.4199, b1 =
0.352 11, and c1 = 0.012 680. For warm cirrus in midlatitude, a2 = 5.2375, b2 = 0.131 42,
and c2 = 0, while for cold cirrus, a∗ 2 = 4.3257, b∗ 2 = 0.265 35, and c∗ 2 = 0.021 864. For
the Arctic region, a3 = 4.8510, b3 = 0.331 59, and c3 = 0.026 189.
In Eqs. (6.1.13)–(6.1.15b), we have illustrated that the extinction coefficient can be
expressed in terms of IWC and De , whereas the single-scattering albedo and the expansion
coefficients for the phase function are functions only of De . Consequently, these single-
scattering properties are driven by a linear combination of IWC subject to the relationship
defined in Eq. (6.3.3). Also note that the ice water path IWP (g m−2 ) = IWC (g m−3 )z,
where z is the cirrus cloud thickness, and IWC is a prognostic variable in GCM and
climate models. From Eq. (6.3.3), the mean effective ice crystal size De is now defined as
a function of IWC.
Using the detailed ice crystal sizes and shapes described in Subsection 6.1.2, we have
conducted radiative forcing analyses for the three regions. Figure 6.8 illustrates solar,
thermal IR, and net radiative forcings as a function of IWP (and De ). The scales for the
solar (negative) and IR (positive) forcings range from 0 to 140 W m−2 . In the calculations,
the solar constant, solar zenith angle, surface albedo, cloud base height, and cloud thickness
used are 1366 W m−2 , 60°, 0.1, 9 km, and 2 km, respectively. Also, cirrus clouds cover 100%
of the sky. For a given IWP, a cirrus cloud has the smallest De for the Arctic region and
the largest De for warm midlatitude conditions. As a result, the Arctic case displays the
largest solar and largest IR forcing, but the smallest net forcing because the negative solar
radiative forcing increases more than its IR counterpart as IWP increases. Midlatitude
results show the smallest solar forcing associated with the largest De and the smallest IR
radiative forcing, resulting in the largest net forcing. Cirrus solar and IR radiative forcings
in the tropical condition are in between these.
One last point: For a given IWC, ice crystals can have different size distributions leading
to different De . To account for De variation due to uncertainty in small ice crystal mea-
surements, we have used randomly selected values of De that are within the possible range
(a) 140
120 Arctic
IR Forcing (W m–2)
100 Tropics
80
Midlatitude
60
40
20
0
0 20 40 60 80 100
(b) 0
-20
Solar Forcing (W m–2)
-40
-60 Midlatitude
-80 Tropics
Arctic
-100
-120
-140
0 20 40 60 80 100
(c) 50
Midlatitude
Net Forcing (W m–2)
40
Tropics
30
Arctic
20
10
0
0 20 40 60 80 100
Figure 6.8 Cirrus cloud (a) IR, (b) solar, and (c) net radiative forcings at TOA as a function of IWP
for Arctic, midlatitude, and tropical regions on the basis of parameterization of the mean effective ice
crystal size De in terms of IWC, with observed ice crystal sizes and shapes accounted for in correlation
analysis.
of the parameterized De related to this uncertainty corresponding to a specific IWC value

as determined from observations. Differences in solar and IR forcings and the net radia-
tive forcing produced using the mean De values within the uncertainty range associated
with smaller ice crystal concentrations are less than 1–2 W m−2 . This appears to suggest
that using the IWC–De correlation in terms of the mean value presented in Eq. (6.3.3)
would be an excellent approach to radiative forcing calculations. Finally, assuming a cirrus
cloud cover of 20% and an IWP range of 20–80 g m−2 , we obtain positive cirrus radiative
forcings of 4–8 W m−2 , a conclusion similar to that presented in Figure 6.7. Of course, ice
cloud cover and position vary significantly over tropical, midlatitude, and Arctic regions,
which could yield substantial variability in radiative forcing magnitude, especially over the
tropics.
In addition to cloud radiative forcing, the vertical variability of atmospheric radiative
heating and cooling rates produced by the presence of cirrus clouds is an important param-
eter that affects temperature fields and consequently circulation and precipitation. We shall
first define the solar heating rate. For a vertical layer of thickness z, the net flux divergence
is given by [see Eq. (6.2.3)]
F (z) = F (z) − F (z + z). (6.3.4)
The absorbed solar energy must be used to heat the layer and the resulting rate of temperature
change can be written in the form
∂T ∂T 1 F (z)
F (z) = −ρCp z , i.e., =− , (6.3.5)
∂t ∂t ρCp z
where ρ is the air density, Cp is the specific heat at constant pressure, and t represents time.
The heating rate can also be expressed in pressure coordinates. The thermal IR cooling rate
(or negative heating rate) can be expressed likewise, except that the net flux divergence is
given by F (z) = F ↑ (z) − F ↓ (z).
As an example, Figure 6.9 shows solar heating rate profiles for (1) a clear atmosphere,
(2) an atmosphere containing a contrail cirrus with pure ice located between 9.5 and 10.5 km,
(3) ice contaminated by a black carbon (BC) particle of 0.1 or 0.5 µm radius with external
mixing, and (4) same as (3) except with internal mixing. The height scale used was from 0
to 20 km with a vertical resolution of 1 km, except in the cloud layer (9.5–10.5 km), which
is divided into four layers for illustrative purposes. The heating rate is proportional to the
net flux divergence at two levels divided by the associated thickness in cirrus and is hence
dependent on the vertical layer used. We adopted the U.S. 1976 Standard Atmosphere; the
solar constant, solar zenith angle, surface albedo, cloud base height, and cloud thickness of
1366 W m−2 , 60°, 0.1, 9.5 km, and 1 km, respectively; and a visible optical depth of 0.5 for
contrail cirrus in heating rate calculations. Also see Figure 6.1 and associated discussions
as well as Subsection 3.4.3 for a presentation on BC and its mixing state.
The instantaneous heating rate in a clear troposphere is on the order of 0.5–1 K day−1 ,
primarily produced by water vapor. Above 15 km, the stratospheric heating rate is due
mainly to ozone absorption. The case of internal mixing produces the largest heating rate.
rBC = 0.1 μm, De = 5 μm rBC = 0.5 μm, De = 5 μm

20 20
15 15
Height (km)
10 10
5 5
0 0
0.0 0.5 1.0 1.5 0 2 4 6 8
rBC = 0.1 μm, De = 10 μm rBC = 0.5 μm, De = 10 μm

20 20
Clear
Pure Ice
15 15 External Mixing
Internal Mixing
Height (km)
10 10
5 5
0 0
0.0 0.5 1.0 1.5 0 2 4 6 8
Heating Rate (K Day–1) Heating Rate (K Day–1)
Figure 6.9 Instantaneous atmospheric and cloud heating rates produced by the absorption of solar
radiation in clear and contrail cirrus atmospheres. Contrail cirrus contains pure and contaminated ice
particles of 5 and 10 µm mean effective size, externally and internally mixed with BC particles of 0.1
and 0.5 µm radii. Contrail cirrus is located between 9.5 and 10.5 km (after Liou et al. 2013).
However, the effect is largely confined within the thin cloud layer; heating rates above and
below the cloud do not show noticeable differences between pure and contaminated cases.
For the most realistic case, involving De = 5 µm and rBC = 0.1 µm, the cloud heating
rate increases from 1.1 K day−1 for pure ice to 1.35 K day−1 for external mixing to
1.4 K day−1 for internal mixing. A thin contrail cirrus cloud slightly increases heating
above the cloud, but reduces the heating by about 0.2 K day−1 below the cloud. For a larger
De of 10 µm, ice absorbs more solar radiation, the BC contamination effect reduces, and
the condition of external or internal mixing appears to have little consequence. For the case
of rBC = 0.5 µm, substantial cloud heating can be generated and the mixing state effect
becomes significant, as shown in the right-hand diagrams of Figure 6.9, especially for
De = 5 µm, for which an instantaneous heating rate of 8 K day−1 is shown.
Solar heating and IR cooling profiles involving a thin cirrus over the tropical tropopause
are also shown in Figure 6.19.
6.4 Climatic effects of cirrus clouds

6.4.1 Investigation using one-dimensional climate models
In the introduction to this chapter, we stated that the action of natural variability as well as
anthropogenic perturbations would lead to climate change. The former involves changes
in solar constant associated with sunspot activities, solar insolation distributions due to
the Earth’s orbit around the sun, and atmospheric composition resulting from volcanic
eruptions. Anthropogenic perturbations are associated with human activities starting around
the era of the Industrial Revolution in 1860, specifically increases in greenhouse gases and
anthropogenic aerosols. Under the scenario of greenhouse perturbations and based on the
principle of radiative energy transfer, the surface and the troposphere will trap more thermal
IR radiation and lead to greenhouse warming.
As a result of tropospheric warming, clouds would form higher in the atmosphere. If a
formation of high clouds were to rise even higher in the atmosphere, there would be a positive
feedback due to the enhanced downward IR flux, while the reflection and transmission of
sunlight would remain largely unchanged. While optically thick low and middle clouds with
optical depths greater than 30 would reflect sunlight, their capacity to trap thermal IR
flux from below would remain largely unchanged due to the small temperature differential
between these clouds and the surface, as well as the small amount of water vapor in between
them. Therefore, from the viewpoint of radiative energy exchange, the issue of the role of
high clouds comprised of ice crystals in greenhouse perturbations is more challenging than
that of low–middle clouds, which largely contain water droplets.
The second issue that determines the role cirrus clouds play in global climate and green-
house warming is related to ice microphysics in terms of their IWC and ice crystal size. The
temperature dependence of IWC is evident from aircraft measurements of midlatitude cirrus
clouds. Shown in Figure 6.10 is the observed IWC as a function of temperature. Although
observed values show large standard deviations, they do reveal a temperature-dependent
trend in terms of mean value. Liou (1986) developed the following parameterization equa-
tion to fit the observed data points:
[ln (IWC) + 7.6]/4 = exp[2.443 × 10−4 (|Tc | − 20)2.455 ], Tc < −20◦ C, (6.4.1)
where Tc denotes cloud temperature and the coefficients are the best fit to data points. Addi-
tionally, Heymsfield and Platt (1984) showed that there is a distinct correlation between ice
10-1
Aircraft Observations
Parameterization
Ice Water Content (g m–3)
10-2
10-3
10-4
-60 -55 -50 -45 -40 -35 -30 -25 -20
Cirrus Temperature (oC)
Figure 6.10 Averaged IWC and its standard deviation as a function of cirrus cloud temperature,
determined from aircraft observations presented by Heymsfield and Platt (1984). The red curve
represents the best-fit polynomial parameterization denoted in Eq. (6.4.1).
crystal size and cloud temperature. Ice crystals are smaller (larger) at colder (warmer)
temperatures. Observations show that the colder cloud top primarily contains smaller
pristine ice crystals, whereas the warmer cloud bottom is composed of larger irregu-
lar ice particles [see Figure 1.18 and Eq. (6.3.3)]. Under greenhouse warming scenarios
accompanied by observational evidence, the IWC and ice crystal size would both become
larger.
A climate model is required to understand the translation from greenhouse radiative
perturbations of cloud configuration and radiative forcing to temperature fields in the
complex and multidimensional atmosphere. Globally averaged one-dimensional (in the
vertical) climate models have been developed for this purpose. The incoming solar flux at
TOA must be balanced by reflected solar and emitted infrared fluxes over a climatological
time scale, since the only energy exchange with space is by means of radiative processes. At
the surface, however, equilibrium must be achieved by a balance between net radiative fluxes
and vertical fluxes of sensible and latent heat, which are governed by motions involving
various scales, including conduction and diffusion, eddies, and cumulus convection. These
scales of vertical motion associated with the transport of sensible and latent heat are
generally referred to as convection.
The simplest way to account for convection in a one-dimensional model for climate stud-
ies is the convective adjustment scheme based on the concept of static stability, introduced
by Manabe and Wetherald (1967). In this scheme, the atmosphere is divided into layers
that include non-convection, convection without contact with the surface, and convection
in contact with the surface. An iterative procedure may be constructed for computation of
the local rate of temperature change with time. It begins at the surface and progressively
scans the layers above until all layers of the supercritical lapse rate have been eliminated.
This process is repeated at each time step. In the computation, atmospheric temperatures
are first constructed based on the balance between radiative heating and cooling rates.
Subsequently, the surface temperature is calculated from the balance of solar and IR net
fluxes at the surface. Since solar fluxes heat the surface, there will be net upward fluxes,
which are distributed in the layer above the surface. Adjusting the temperature in steps
upward whenever the layer is convectively unstable eliminates all the supercritical lapse
rates. For application to one-dimensional climate models, the critical lapse rate is usually
assumed to be 6.5 K km−1 for the globally averaged condition, based on the fact that the
climatological atmospheric temperature profile in the troposphere has a lapse rate close to
this value. Using globally averaged cloudiness, the temperature profile obtained is close to
that of the U.S. 1976 Standard Atmosphere. Without the inclusion of clouds, temperatures
are much higher than those with clouds; a difference of as much as 20 K is shown for surface
temperature. Manabe and Wetherald (1967) performed the first sensitivity experiment of
the radiative–convective climate model to examine radiative forcings of carbon dioxide and
solar input.
If the temperature increases in the region of low clouds, which are composed entirely
of water droplets, these clouds will reflect more solar radiation because of greater liquid
water content (the solar albedo effect), leading to a negative feedback (e.g., see Charlock
1982, Somerville and Remer 1984). Further, both theory and experiments reveal that the
cloud droplet concentration is approximately proportional to the level of cloud condensation
nuclei (CCN), and that the size of droplets decreases as CCN in clouds increases. A number
of observations over land have shown that pollution increases the number of CCN leading
to more small cloud droplets per unit volume, which, in turn, increases the reflection of
sunlight (or solar albedo; Twomey et al. 1984); this is referred to as the Twomey effect. This
effect has also been observed in the modification by ship stack effluents of the radiative
properties of shallow stratus clouds over oceans (Coakley et al. 1987, Radke et al. 1989).
6.4.1.1 A one-dimensional cloud–precipitation–climate model

Liou and Ou (1989) developed a 1-D radiative–convective climate model to investigate the
impact of temperature increases induced by doubling CO2 (300 to 600 ppm) on the forma-
tion of cloud droplet size distribution and precipitation. The model includes the vertical
transport of water vapor, liquid water, and precipitation flux, which are governed by the rates
of condensation, evaporation, and precipitation generation. The rate of condensation for a
given particle size distribution is directly proportional to a mean effective size r̄1 (the first
power) through the 1-D diffusion theory for water vapor and latent heat transport. The rate
of precipitation generation is produced by autoconversion, representing initial conversion of
cloud droplets to raindrops, and accretion, denoting collision and coalescence. Autoconver-
sion, within the limits of Stokes law, is proportional to a mean radius, r̄4 , to the fourth power
(see also Liu et al. 2004 and Liu and Daum 2004) and liquid water content (LWC), which is
a function of a mean effective radius to the third power, r̄3 [see Eq. (5.2.2)]. Accretion is a
function of precipitation flux, a variable in the governing equation set, and of LWC. The rate
of evaporation for raindrops can be obtained from an equation analogous to condensation
and by including a ventilation factor. Thus, the rates of condensation, evaporation, and
precipitation generation can be parameterized in terms of specific humidity, cloud liquid
water mixing ratio, and precipitation flux, wherein the rate coefficients are functions of
temperature and therefore interactive with greenhouse perturbations by means of radiative
forcing. It is convenient in radiative forcing calculations to use cloud optical depth, which
drives the transfer of radiation and is proportional to a mean radius to the second power, r̄2
[see Eq. (5.2.3)]. This radius is referred to here as the “mean radius r̄ = r̄2 .” Using available
droplet size distributions for a number of low and middle clouds and cumulonimbus, the
following relationships have been derived: r̄4 ≈ 1.247 r̄, r̄3 ≈ 1.132 r̄, and r̄1 ≈ 0.852 r̄.
The precipitation generation rate is directly dependent on r̄ 4 .
Based on the interactive cloud and precipitation program within the context of the
preceding 1-D radiative–convective climate model, the link between cloud microphysical
processes and CO2 climatic perturbations can be hypothesized with reference to mean state.
Two such hypotheses are as follows: (1) If the perturbed mean radius r̄ is less than the
climatological mean value r̄0 , precipitation decreases, leading to an increase in cloud LWC;
thus, the solar albedo effect outweighs the IR greenhouse effect, resulting in a negative
perturbed temperature change due to a positive CO2 radiative forcing. (2) If the perturbed
mean radii r̄ are larger than the climatological mean value r̄0 , precipitation increases,
leading to a decrease in cloud LWC; thus, the IR greenhouse effect outweighs the solar
albedo effect, resulting in a positive temperature-change amplification due to a positive
CO2 radiative forcing.
With respect to the first hypothesis, Charlson et al. (1987) pointed out that the major
source of cloud condensation nuclei (CCN) over the oceans is non–sea-salt sulfate produced
from the emission of dimethyl sulfide (DMS) by marine organisms. They suggested that
the highest rate of DMS emission to the atmosphere is associated with the warmest, most
saline, and most intensely illuminated regions of the oceans. Thus, an increase in surface
temperature as a result of CO2 doubling could cause an increase in DMS emission, and
hence CCN. Moreover, theory and experiments indicate that cloud droplet concentration
is approximately proportional to CCN and that droplet size decreases as CCNs in water
clouds increase (Warner and Twomey 1967, Hudson 1983, Pruppacher and Klett 1996).
With respect to the second hypothesis, the competing mechanism for the potential
existence of larger cloud droplets (with reference to the climatological mean) in the atmo-
sphere under the global warming scenario may be associated with precipitation. Using the
climatological mean radius in perturbation runs, we find that precipitation computed from
the present 1-D cloud–precipitation–climate model increases by l.4% if the CO2 con-
centration is doubled. Additionally, a number of GCM experiments have showed that
precipitation would be increased in almost all latitudes with the CO2 doubling experiment
(e.g., Manabe and Wetherald 1980, Washington and Meehl l984). Precipitation is consid-
ered to be the primary mechanism for the removal of atmospheric aerosols, including CCN.
This process is referred to as wet removal (Prospero et al. 1983). Wet removal includes the
absorption of CCN in precipitation droplets within clouds (rainout) and the capture of CCN
in precipitation occurring below the cloud base (washout). With increased precipitation,
the CCN number could be reduced. In accordance with the second hypothesis above, larger
cloud droplets could be formed from a low concentration of efficient CCN. This process
provides a means of positive feedback involving precipitation as well as temperature.
An important note is in order at this point. Simulation results based on the 1-D cloud–
precipitation–climate model have demonstrated that the precipitation rate is a function of
the fourth power of a mean radius. A decrease in this droplet radius, which is inversely
proportional to cloud optical depth if LWP remains unchanged [see Eq. (5.2.4)], would
reflect more sunlight (cloud albedo effect), referred to as the first indirect effect of aerosols
(Twomey et al. 1984), as cited above. At the same time, this decrease would also result
in precipitation reduction, referred to as the second indirect effect of aerosols. The link
with precipitation vis-à-vis aerosol–cloud interactions in climate change was first reported
by Liou and Ou (1989) in the June issue of the Journal of Geophysical Research and by
Albrecht (1989) in the September issue of Science (see footnote 30 in the reference section
of this article).
The foregoing discussions have pointed out the importance of cloud particle size on
both solar albedo (negative feedback) and precipitation (positive feedback) under global
warming scenarios. It is not known from this 1-D model study which of the two competing
mechanisms is more efficient.
To further complicate the aerosol (CCN)–cloud–precipitation–climate issue under global
warming, numerous observations over land have revealed that anthropogenic pollution
increases CCN numbers (Warner and Twomey 1967, Hobbs et al. 1974, Braham 1974),
and leads to more small cloud droplets (Twomey et al. 1984) as mentioned above. This
increase is caused by regional anthropogenic sources and is not directly related to global
warming under the first hypothesis described above. If the perturbed mean cloud droplet
radius produced by man-made air pollution is smaller than the climatological mean value,
precipitation (proportional to the fourth power of the cloud particle radius) could decrease,
leading to an increase in cloud LWC. The solar albedo effect would then outweigh the
IR greenhouse effect. A reduction of the mean droplet radius by 0.5 µm could cool the
atmosphere to such a degree as to offset the warming produced by CO2 doubling. Leaitch
et al. (1992) have observed a reduction in droplet radii of about 1 µm in eastern North
America as a result of anthropogenic pollution.
The issue of man-made aerosols on cloud formation and precipitation is complicated
by the fact that these particles consist of various compositions defined by different indices
of refraction, with radii ranging from 0.01 to 10 µm, while some of them exhibit
complex morphology. For instance, sulfate, nitrate, and organic carbon aerosols are non-
absorbing spherical particles which only scatter sunlight. Black carbon (BC or soot) particles
significantly absorb sunlight in view of their imaginary indices of refraction. During aging
processes in the atmosphere, these can evolve from initial aggregate shapes into complex
coated particles in association with other aerosols and later with water vapor. Non-spherical
mineral dust particles, having the largest sizes (1–20 µm) of all aerosols, can scatter
and absorb solar (shortwave) radiation as effectively as BC, and can also trap thermal
IR (longwave) radiation. The preceding absorption and scattering processes involving
aerosols can modify an atmospheric radiation field and are referred to as the direct effect.
The modification of vertical and horizontal radiation patterns directly from aerosols leads
to changes in temperature, circulation, and precipitation distributions in the atmosphere,
resulting in a redistribution of cloud radiative forcing fields, now known as the semi-
direct effect. The aerosol first (solar albedo) and second (precipitation) indirect effects
are associated with cloud and precipitation formation and are a consequence of radiative
forcing, as elaborated above.
While the preceding discussion focuses primarily on clouds containing water droplets,
the same arguments can be applied to clouds comprised of ice crystals, particularly those
involving BC/dust and cirrus cloud formation.
6.4.1.2 A one-dimensional climate–cirrus-cloud model

For high cirrus clouds containing primarily ice crystals, a one-dimensional radiative–
convective model study illustrates that the balance of solar albedo versus greenhouse
effects depends on both IWC and ice crystal size (Ou and Liou 1995). We may relate cloud
optical depth to IWP and De as follows: τ ≈ IWP(−6.656 × 10−3 + 3.686/De ), where
IWP = IWCz. See Eq. (6.1.7b). Thus, increasing IWP enhances cloud optical depth. By
contrast, increasing De , reduces cloud optical depth, subject to the condition that IWP
remains unchanged.
In the following, we develop a model similar to the one discussed in Subsection 6.4.1.1
for high clouds consisting of ice crystals whose size distribution is represented by De ,
first introduced in Section 1.7 and in numerous subjects including remote sensing and its
correlation with IWC given in Eq. (6.3.3). The 1-D radiative–convective model coupled
with a 1-D cirrus model is based on the work of Liou and Ou (1983), Liou et al. (1985),
Liou and Ou (1989), and the discussions of ice microphysics in Sections 1.4 and 1.7. The
model includes water vapor and cloud ice water equations, as well as parameterizations of
ice generation and growth. The basic 1-D radiative–convective–ice equilibrium equations
may be written as follows:

∂T
ρCp kh + γs − γ = FS (z) − FIR (z, T ), (6.4.2)
∂z
1 ∂(ρw q )
= −Cdif (T ), (6.4.3)
ρ ∂z
1 ∂ [ρwc qc (T )]
= Cdif (T ), (6.4.4)
ρ ∂z
Cdif (T ) = 4π[De (T )/2][q/qi (T ) − 1]/[fK (T ) + fD (T )], (6.4.5)
De (T ) ≈ exp(a + b ln [IWC(T )] + c{ln [IWC(T )]}2 ), (6.4.6)
where ρ is the density of air, Cp is the specific heat of air at a constant pressure, kh is the
thermal eddy diffusion coefficient, γ s is the moist adiabatic lapse rate, and γ is the counter-
gradient lapse rate (Deardorff 1972) for lapse rate adjustment associated with the upward
sensible heat flux in the planetary boundary layer which is slightly less than adiabatic, FS
is the solar flux, and FIR is the thermal infrared flux, which is a function of temperature
T. In Eq. (6.4.3), q is the specific humidity, w is the air velocity, and ρw q denotes the
vertical eddy fluxes of specific humidity. The 1-D model is assumed to be comprised of
uniform clear and cloudy regions with known cloud cover and thickness, on the basis of
earlier cloud climatology compiled by London (1957), and only cirrus cloud IWP and De in
addition to water vapor are subject to greenhouse temperature increases due to a doubling
of CO2 .
Let w c and qc (= IWCρ) denote the vertical velocity and specific humidity, respectively,
in the cirrus cloudy region. For the clear region, these two variables are denoted by w and
q. Then the eddy flux of specific humidity for clear plus cirrus cloud regions may be
written as w q = ηwc qc + (1 − η)wq, where η is the cirrus cloud fraction (a constant)
and w c can be computed from Richardson’s equation (Liou et al. 1985). The horizontal
averages of temperature T, specific humidity q, and vertical velocity w may be expressed
as a linear combination of the two regions: χ = ηχc + (1 − η)χ0 , where the term χ can
be T, q, or w. For application to 1-D radiative–convective models, we can safely assume
that T(cloud) T(clear) = T. In the context of a 1-D global climate model, the average
vertical velocity must be equal to 0, so w = −ηwc /(1 − η). Finally, the specific humidity
may be expressed as q = qi [η + (1 − η)h0 ], where qi is the saturation specific humidity
with respect to ice, which is a function of temperature through the Clausius–Clapeyron
equation [see Eqs. (1.4.1a) and (1.4.1b)], and h0 denotes the threshold relative humidity
parameterized in terms of its surface value in the form h0 = h0 (p∗ )(p/p∗ − 0.02)/0.98
(Manabe and Wetherald 1967), where p is the pressure and p∗ is its surface value and
h0 (p∗ ) 0.8 based on general circulation statistics (Oort 1983).
In Eqs. (6.4.3)–(6.4.5), Cdif denotes the rate of diffusional growth for a given ice crystal
size distribution with a mean effective size De . Within the context of a 1-D equilibrium
climate model, we postulate that instantaneous ice nucleations by means of homogeneous
and heterogeneous processes have already been initiated and exist in equilibrium with
temperature. In Eq. (6.4.5), fK (T ) = L2I /KRv T 2 and fD (T ) = Rv T /Dqi (T ), where LI is
the latent heat associated with heat transformation between ice and water vapor, Rv is the gas
constant for water vapor, K is the thermal conductivity coefficient, and D is the molecular
diffusivity coefficient. Lastly, the parameterization of De as a function of IWC is given in
Eq. (6.4.6) where a, b, and c are coefficients for global conditions. See also Eq. (6.3.3).
The 1-D climate–cirrus cloud model was first run to produce a temperature profile
corresponding to the standard atmospheric condition under the 1 × CO2 condition (control
run), which produces a surface temperature Ts of 288 K, a mean effective ice crystal size De
of 20 µm, and a mean IWP (= IWC × z), where IWC can be computed from Eq. (6.4.6)
and z is a constant, from the 1-D model. This action was then followed by 2 × CO2
perturbation simulations to investigate the relative importance of IWP and De feedbacks, as
well as the relative strength of solar albedo (denoted by α) and IR greenhouse (denoted by
ε) effects in these two parameters. We may express perturbed values for these two effects,
in terms of their present climate counterparts (1 × CO2 ) with a subscript 0, in the forms

dε ∂ε dIWP ∂ε dDe
ε = ε0 + Ts = ε0 + + Ts , (6.4.7a)
dTs ∂IWP dTs ∂De dTs

dα ∂α dIWP ∂α dDe
α = α0 + Ts = α0 + + Ts , (6.4.7b)
dTs ∂IWP dTs ∂De dTs
where the terms dε/dTs and dα/dTs are further expressed as functions of differential
changes in IWP and De with respect to surface temperature changes Ts . In this manner,
the perturbed surface temperature is fully interactive with IWP, De , α, and ε.
Figure 6.11 illustrates simulation results fitted into a 3-D diagram in terms of Ts ,
IWP, and De , where the planes corresponding to IWP = 0 and De = 0 can be
clearly visualized. The symbol • represents the case with only water vapor feedback with
a surface temperature increase Ts = 2.41 K, and corresponds to the intersection of two
planes defined by IWP = De = 0. On the plane for De = 0, the symbols and
denote cases with only the IR greenhouse effect (Ts = 2.93 K, positive feedback) or only
the solar albedo effect (Ts = 1.76 K, negative feedback), respectively, while indicates
that both effects are included (Ts = 2.14 K, negative feedback). These values reveal that
the solar albedo effect generated by the enhanced reflection of sunlight by a large IWP
results in a decrease in surface temperature. On the plane for IWP = 0, the symbols
and again denote cases with only the IR greenhouse effect (Ts = 2.30 K, negative
feedback) or only the solar albedo effect (Ts = 3.20 K, positive feedback), respectively,
while indicates that both effects are included (Ts = 3.03 K, positive feedback). These
results show the domination of the solar albedo effect caused by the reduced reflection for
a large De , leading to substantial increases in surface temperature.
In Figure 6.11, the symbol × denotes the case when both IWP and De , as well as IR
greenhouse and solar albedo feedback, are all accounted for in the calculations, resulting
in a temperature increase of Ts = 2.68 K. The surrounding points and correspond,
respectively, to cases with only the IR greenhouse effect (positive feedback) and only the
solar albedo effect (negative feedback). Under the CO2 doubling scenario and compared
to the H2 O feedback of Ts = 2.41 K, cirrus cloud microphysics, through IWP and De
interactions, lead to a positive feedback of more than 10% due to the prevalence of an
IR greenhouse effect, based on a 1-D global climate model simulation without regional
considerations. Lastly, the valley depicted in Figure 6.11 is the result of numerical fitting,
and in this context represents minimum-Ts values on the IWP–De horizontal plane.
3.5
3.0
2.5
ΔTs (K)
2.0
1.5
1.0
6
ΔIW
6
3
5
P(
4
2
gm
3
–2
1
2
m)
ΔD e (μ
)
1
0
0
Figure 6.11 A 3-D depiction of surface temperature deviation Ts due to 2 × CO2 as a function
of De and IWP, defined as deviations of the cirrus cloud mean effective size and ice water path
from climatological mean values (1 × CO2 ), respectively. The results were obtained by running the
radiative–convective model with interactive water vapor, IWP, and De . Separation of IR greenhouse
and solar albedo effects was also included in the calculations. The symbols • and × represent,
respectively, without and with IWP and De feedback, while the symbols and denote separate
or combined IR greenhouse and solar albedo effects in the model calculations (see text for further
explanation).
In view of the above and to summarize, the role of clouds in climate and climatic
perturbations due to external radiative forcing is indeed intricate and multidimensional.
The results derived from 1-D climate models shed some light on potential interactions and
feedback involving cloud height, cloud water content, and cloud particle size associated
with temperature increases produced by greenhouse warming. However, 1-D models do not
account for the horizontal variation in cloud parameters. Since clouds are 3-D in nature,
their role in climate and climate change must be resolved using computer models that
can simulate both vertical and horizontal cloud structure and composition derived from
fundamental physical principles.
6.4.2 Ice microphysics sensitivity investigation using GCMs

6.4.2.1 The effect of interactive ice crystal size on precipitation simulations
Including prognostic equations capable of predicting IWC for high-level clouds formed
in GCMs and climate models is a milestone in incorporating a physically based cloud
microphysics scheme into these models. This advance is also significant from the perspec-
tive of studying cloud–radiation interactions. However, the interaction of cloud particle size,
an independent parameter that also affects cloud radiation transfer, has not been appropri-
ately accounted for in GCMs. For example, for a given IWC in clouds, smaller particles
would reflect more sunlight than larger ones because of larger cloud optical depths [see
Eqs. (5.2.7) and (6.1.7b)], an effect that has been recognized by Twomey et al. (1984) and
Liou and Ou (1989) in conjunction with aerosol–cloud indirect effects.
In Chapter 1, we pointed out that ice crystal size and shape in the Earth’s atmosphere are
complex and intricate. After initial homogeneous and/or heterogeneous nucleation involv-
ing suitable aerosol particles and atmospheric conditions, ice crystal growth is governed
by diffusion processes and subsequent actions resulting from collision and coalescence.
These physical processes are complicated by the nature of the ice crystal’s hexagonal and
irregular shape. Incorporating a fully interactive ice microphysics simulation based on first
principles into a GCM appears to be a challenging and an extremely difficult computational
task. It has been common practice to prescribe a mean effective ice crystal size in GCMs
(e.g., Gu et al. 2003). A number of GCMs have also used temperature to determine De
(Kristjánsson et al. 2005, Gu and Liou 2006). In Section 1.7, we demonstrated that the
mean effective ice crystal size De correlates statistically with IWC based on ice crystal size
and shape observations from intensive aircraft campaigns over tropical, midlatitude, and
polar regions. A summary of the IWC–De parameterization was given in Eq. (6.3.3).
We incorporated the IWC–De correlation into the UCLA atmospheric GCM to investi-
gate its usefulness and importance in climate simulations (Liou et al. 2008). The UCLA
Department of Atmospheric and Oceanic Sciences, evolved from Meteorology and later
Atmospheric Sciences, has been one of two primary pioneering research centers in the
development of AGCM (see Randall 2000), and more recently of coupled atmosphere–
ocean models for a number of climate studies, including El Ninõ/Southern Oscillation and
the Asian Monsoon (e.g., see Arakawa 2000, Mechoso et al. 2000).
The UCLA AGCM is a state-of-the-art grid point model of the global atmosphere
extending from the Earth’s surface to a height of 50 km. The physically based spectral
radiation parameterization presented above for cloud and aerosol studies and conceived by
Fu and Liou (1993) is built into the UCLA AGCM to enhance its capability to study a variety
of climate problems involving cloud and aerosol interactions (Gu et al. 2003, 2006, 2010,
2012; Gu and Liou 2006; Liou et al. 2008). Prognostic variables involved in the UCLA
AGCM are: horizontal wind, potential temperature, water vapor mixing ratio, cloud liquid
water and cloud ice water, planetary boundary layer (PBL) depth, surface pressure, land
surface temperature, and snow depth over land. The PBL is parameterized as a well-mixed
layer of variable depth (Li et al. 2002). Parameterization of cumulus convection and its
interaction with the PBL follows Pan and Randall (1998). The geographical distribution of
sea surface temperature is based on a 31-year (1960–1990) climatology corresponding to
the global sea ice and sea surface temperature (SST) (GISST) version 2.2 data set (Rayner
et al. 1995). Sea ice thickness and extent as well as surface albedo and roughness length are
prescribed (Alexander and Mobley 1975, Dorman and Sellers 1989). Ozone mixing ratios
are prescribed as a function of latitude, height, and time based on a 1985–1990 climatology
(Li and Shine 1995). In this study, we used a low-resolution model version with a grid size
covering 4-latitude by 5-longitude and with 15 layers from the Earth’s surface to the top at
1 hPa.
Two numerical experiments were carried out. In the control run, De for ice clouds was
fixed at 80 µm, representing a global mean. In the perturbation run, De was parameterized
in terms of IWC in accordance with the correlation equations presented above. It is a
function of the model-predicted IWC at each time-step and interacts with cloud, radiation,
and dynamic processes in the model. Ice crystal sizes computed from parameterization
are generally smaller in the tropics than in midlatitude, ranging from 10 to 60 µm. In
midlatitude, De ranges from 50 to about 80 µm for colder temperatures (−65°C < T <
−40°C), and from 80 to more than 100 µm for warmer cirrus. In the Arctic regions, De is
about 70–80 µm.
Geographical differences in OLR and precipitation are illustrated in Figure 6.12. Since
smaller ice particles reflect more solar radiation and trap more IR, differences in OLR
closely follow variations in ice crystal size. Negative values are mostly located in the lower
latitudes where the parameterized ice crystal sizes are generally smaller than the prescribed
value in the control run, resulting in more reflected solar fluxes, not shown in this figure, and
more trapped IR fluxes. Increased OLRs are displayed in the midlatitude region due to larger
De than the prescribed 80 µm. Differences between the two simulations are not particularly
significant in high latitudes since the parameterized De is close to the fixed value of 80 µm
in the control run. Changes in the precipitation pattern do not directly result from the ice
crystal size effect; rather, these changes are related to intricate interactions among cloud,
radiation, and dynamic processes through modified vertical heating profiles associated with
the interactive De used in the simulation. Increases in precipitation are found in tropical
regions where ice particles are smaller and OLR associated with colder temperatures is
reduced, an indication of stronger convective activities. Reduced precipitation is seen in
areas corresponding to enhanced OLRs in subtropical and midlatitude regions. This brief
illustration suffices to demonstrate the importance of using correct ice crystal sizes in GCM
simulations.
6.4.2.2 The effect of ice crystal shape on cloud radiative forcing simulations
In Figures 3.26 and 3.27, we showed that ice crystal surface roughness has a substantial
impact on the phase function, especially at forward directions, which would reduce the
asymmetry factor and hence the reflection of sunlight from ice clouds. There is some
evidence that the growth of ice crystals may lead to rough surfaces caused by evapora-
tion, sublimation, or riming processes (Cross 1969; Ono 1969; Davy and Branton 1970;
Ulanowski et al. 2006, 2010). Fu (2007) and Yang et al. (2013) considered the effect of ice
surface roughness on the asymmetry factor, which resulted in a decrease in forward scatter-
ing, and illustrated the subsequent reduction on reflected solar fluxes. Edwards et al. (2007)
implemented a single-scattering parameterization, taking into account rough surfaces in an
ice aggregate model in the Third Hadley Centre Atmosphere GCM.
July Mean Difference

OLR
Precipitation
Figure 6.12 Differences in OLR and precipitation patterns simulated using the UCLA GCM, based
on values of De determined from IWC–De correlations for Arctic (60–90° latitude), midlatitude (30–
60° latitude), and tropical (0–30° latitude) regions and a control run using a fixed ice crystal size of
80 µm (after Liou et al. 2008).
Figure 6.13 CAM5.1-simulated 10-year mean annual total cloud radiative effect (W m−2 ) and roughening effects (CRE for the severely roughened
case minus that for the smooth case) at TOA: (a) SW cloud radiative effect (SWCRE), (b) LW cloud radiative effect (LWCRE), (c) roughening
effect on SWCRE, and (d) roughening effect on LWCRE. Hatched areas correspond to a 95% confidence level (after Yi et al. 2013).
The issue of ice crystal roughness was revisited by Yi et al. (2013) using NCAR’s
Community Atmosphere Model (CAM, version 5.1). Figure 6.13 shows the simulated 10-
year mean annual total cloud radiative effect at TOA associated with ice particle surface
roughness resulting from completely smooth (CS) and severely roughened (SR) conditions.
Hatched areas indicate a 95% confidence level. The total cloud radiative effect is shown
because the CAM5.1 radiation scheme does not treat different cloud types individually, thus
making it impossible to single out ice cloud effects. The results reveal that the shortwave
cloud radiative effect induced by ice surface roughness is substantial in terms of spatial and
seasonal variations. Strong shortwave surface-roughness effects are found in the tropics
and in both hemisphere storm-track regions; longwave differences are limited mostly to
the tropics. This could indicate either feedback processes or internal variability in the
AGCM that require further exploration. Perhaps the perturbation of the shortwave radiation
by ice particle roughening induces changes in regional thermodynamic structures that
subsequently lead to persistent local circulation anomalies. In turn, these circulation changes
could lead to further changes in ice cloud properties and, thus, notable changes in longwave
cloud radiative effects. This appears to be in line with the fact that ice surface roughness
leads to both positive and negative anomalies in the model’s shortwave cloud radiative
effect, as shown in Figure 6.13.
AGCM simulation results appear to agree in general with offline radiative transfer
simulation results, but the simulated results show more variation in both sign and magnitude,
as well as changes with respect to time and geographical location, indicating potential
feedback effects. The NCAR CAM5.1 results also provide a way of estimating the effect
of ice particle surface roughening (SR – CS) on global cloud radiative forcing, with global
annual average effects on shortwave and longwave due to ice surface roughness being
1.83 and 0.37 W m−2 , respectively. The net cloud radiative impact of ice crystal roughness
is −1.46 W m−2 , a sizeable value compared to other forcing values caused by various
scenarios of increased CO2 concentration.
6.5 Climate issues associated with cirrus clouds

6.5.1 Contrails and contrail cirrus
In Section 1.1, we introduced upper-level ice crystal clouds produced by high-flying aircraft
as well as subsequent climate issues over the last five decades. These clouds are known
as condensation trails or simply contrails. Contrails are visible line clouds resulting from
water vapor emissions that form behind aircraft flying in sufficiently cold air where water
droplets form on soot (or black carbon) and sulfuric acid particles emitted from the aircraft
or on background particles, and subsequently freeze to become ice particles. Persistent
contrails often develop into more extensive contrail cirrus in ice-supersaturated air masses.
During the Subsonic Aircraft: Contrail and Cloud Effect Special Study (SUCCESS)
conducted in 1996 and several other cirrus cloud field campaigns (see Subsection 1.6.2),
useful ice crystal and microphysical datasets were obtained for contrails, contrail cirrus,
and cirrus. Contrails were found to consist predominantly of bullet rosettes, columns,
and plates, with sizes ranging from about 1 to 50 µm (Liou et al. 1998). Minnis et al.
(1998a) conducted a comprehensive analysis of jet aircraft contrails over the United States
and Europe using satellite infrared imagery. Their findings, along with potential climatic
impacts, were included in the Intergovernmental Panel on Climate Change (IPCC) report
Aviation and the Global Atmosphere (IPCC Report 1999).
Among the IPCC findings is the expectation that aircraft emissions worldwide will
increase in the future following rapid growth in aviation operations due to economic
improvements and population increase. The formation of contrails from high-flying aircraft
and the subsequent induced contrail cirrus would impact the three-dimensional radiation
field in the atmosphere through the reflection of sunlight and the trapping of thermal infrared
radiation. Radiation and climatic impacts from these clouds also include their interaction
with the water vapor budget in the upper troposphere and lower stratosphere (UTLS; see
Subsection 6.5.1.1), which is important in controlling thermal infrared radiation exchange.
An historical review on the subject of contrail and contrail cirrus follows.
6.5.1.1 Early studies

Reinking (1968) discussed the reduction in solar insolation at the surface due to the presence
of contrails, while Kuhn (1970) reported the trapping of thermal infrared radiation by
contrails, based on aircraft observations. Machta and Carpenter (1971) reported secular
increases in high-cloud cover in the absence of low or middle clouds at a number of
midlatitude stations in the United States between 1948 and 1970. It has been suggested
that this increase in cloudiness could be linked to the expansion of jet aircraft flights in
the upper troposphere and lower stratosphere, reported in the Study of Man’s Impact on
Climate (SMIC 1971). The exhaust plumes of jet aircraft, consisting primarily of water
vapor, carbon dioxide, and other hydrocarbons, produce contrails. The water vapor within
jet plumes may undergo homogeneous and/or heterogeneous nucleation processes, creating
centers of attraction upon which ice particles form and grow. Contrails persist for only a
short time if the ambient air is very dry. In humid conditions, however, they may persist
for minutes or hours and spread into linear formations a few kilometers in width and tens
of kilometers in length. Contrails tend to cluster in groups. Changnon (1981) conducted a
detailed analysis of cloud observations, sunshine measurements, surface temperatures, and
air traffic data over the northern midwest United States for the period 1901–1977. A general
downward trend in the annual number of clear days had been observed, implying that cloud
formation frequency was increasing. In particular, high cloudiness increased over north-
central Illinois and Indiana during the period 1951–1976, corresponding approximately
to the rapid expansion of air traffic. Moreover, the number of months with moderated
temperatures, defined as differences between monthly average maximum and minimum
temperatures that are smaller than the normal differences, has been increasing. This trend
is consistent with the increase in cloudiness.
In a paper published in Air Traffic and the Environment: Background, Tendencies and
Potential Global Atmospheric Effects (Schumann 1990), Liou et al. (1990) investigated
30
Cons. (10 gallons)

Domestic Jet Fuel
8
(a) Salt Lake City
Annual High Cloud

6
20
Cover (%)
9
4
2
10
0
0
50 55 60 65 70 75 80
13
(b)
Surface Temperature
12
( o C)
11
10
9
50 55 60 65 70 75 80
Year
Figure 6.14 (a) Mean annual high-cloud cover in Salt Lake City from 1949 to 1982 (×) and domestic
jet fuel consumption (). The two red lines are statistical fitting curves for high-cloud cover during
the periods 1949–1964 and 1965–1982. The fitting results for the entire period are indicated by the
blue line. (b) Corresponding mean annual surface temperatures (o), where two red lines are statistical
fitting curves for surface temperature during the periods 1949–1964 and 1965–1982. The statistical
fitting curve for the entire period is shown by the blue line (after Liou et al. 1990).
the climatic impact of contrail cirrus and analyzed high cloudiness for Salt Lake City,
Utah during the period 1949–1982 using cloud data obtained from surface observations.
The monthly average high cloudiness was computed using only those cases in which the
sum of low and middle cloud amounts was 0.5 tenths or less. The annual high cloudiness
was then computed from the monthly average values. Figure 6.14a shows the amounts of
mean high cloud with no low or middle intervening cloudiness for the period 1949–1982.
Based on statistical analysis, the high-cloud record may be considered as two separate
cloud populations corresponding to the periods 1949–1964 and 1965–1982. The mean
high-cloud amounts for these two periods are 11.8% and 19.6%, respectively. Domestic jet
fuel consumption from 1956 to 1982 is also displayed in Figure 6.14a, where a significant
increase in the mid 1960s is clearly noted. The sharp anomalous increase in annual high
cloudiness appears to coincide with a significant increase in domestic jet fuel consumption
in the mid 1960s. The sharp increase appears to suggest a possible causative relationship
between increases in jet aircraft traffic and annual high-cloud amount since 1965.
Annual surface temperature values for Salt Lake City during the period 1949–1982 were
also evaluated using surface temperature data published by the National Climate Center,
as shown in Figure 6.14b. Temperature data were divided into two time periods coinciding
with those for high-cloud amount. The slope of the linear fit of temperature for the first time
period is negative with a very small coefficient of determination. However, for the second
time period, the slope is positive and the coefficient of determination is approximately five
times greater than for the first time period. An increase in regional surface temperatures
could be caused by a number of factors, ranging from natural climate variability to man-
made perturbations. Nevertheless, the increase in surface temperature in Salt Lake City
during the period 1965–1982 appears to be directly related to the increase in high-cloud
amount, considering the high statistical correlation between these two parameters.
In terms of model studies, Manabe (1975) discussed the effect of contrails on surface
temperature and pointed out the importance of cirrus blackness on temperature sensitivity
experiments. Freeman and Liou (1979) investigated the effect of increased contrail cirrus
cover in midlatitude on the radiative budget of the Earth–atmosphere system. Liou and
Gebhart (1982) conducted extensive 1-D numerical experiments on the effect of cirrus
clouds on equilibrium temperatures and showed that the effect is a function of cirrus optical
depth (or emissivity). Ou and Liou (1984) constructed a 2-D climate model based on the
energy balance approach to study the effect of the radiative properties of cirrus clouds
on global temperature perturbations and to illustrate the importance of cirrus emissivity
and position. Liou (1986) further defined the role of cirrus clouds in weather and climate
processes.
Liou et al. (1990) assessed the climatic effect of contrail cirrus from the viewpoint of a
2-D cloud–climate model (latitude and height). This model was a combination of a 2-D
energy balance climate model and an interactive cloud formation model that generated
cloud cover and water content based on thermodynamic principles. Temperatures and cloud
models were interactive through the radiation program. The cloud–climate model was first
adjusted to the current climate state in terms of temperature, cloud cover, and TOA radiation
budget. Humidity and surface albedo feedback were included in the model. The effect of
contrail cirrus cover on cloud formation and temperature fields was investigated by increas-
ing the cloud cover between 20 and 70°N, roughly corresponding to the region where most
jet aircraft traffic occurs. A 5% increase in high-cloud cover in a perturbation run resulted
in a substantial amplification in high-cloud cover increase (15%) at 20–40°N, caused by
specific humidity enhancement. Low and middle clouds also increased slightly because
of additional moisture supply (solar albedo effect). Overall, enhanced downward thermal
IR emission from additional high clouds led to temperature increases in the troposphere
at lower latitudes (IR greenhouse effect) and latitudinal surface temperature increased by
1 K. A 10% increase in high-cloud cover in a perturbation run resulted in tropospheric
temperature increases of more than a factor of two and a surface temperature increase of
2.5 K. All perturbation experiments involving an increase in high-cloud cover revealed
increases in atmospheric and surface temperatures induced by a positive greenhouse
feedback.
In Subsection 6.3.2, we showed that the infrared greenhouse effect for cirrus clouds
generally outweighs the solar albedo effect, resulting in a radiative flux gain at the surface,
a conclusion based on radiative transfer principles that are also applicable to contrails
and contrail cirrus. Of course, the degree of surface warming induced by the presence of
high-level ice clouds is dependent on their radiative properties and positions, as well as
on feedback associated with ice cloud formation involving IWC and De , representing ice
crystal size and shape, as articulated in Section 6.4.
6.5.1.2 Contemporary programs and investigations

Schumann (1990) first presented the state of knowledge and understanding of contrails
and reported results on radiative transfer which provided an estimate of the order of
magnitude of the effect of cirrus clouds and water vapor on temperature perturbations in
comparison to CO2 doubling from a global perspective. This paper also presented contrail
analysis from NOAA-AVHRR data and found that contrails covered a few percent of certain
regions including southern Germany and the Alps. A pattern recognition method was also
developed to quantify contrail vertical cross-sections, which reported values of several
kilometers in width and 700 m in thickness, based on which the contrail optical depth
could reach a value of 1 (Schumann and Wendling 1990). Schumann (2002) provided
a comprehensive review of the physical and microphysical properties of contrails and
contrail cirrus and their importance to climate change. Additionally, Schumann (2005)
discussed the formation, occurrence, properties, and climatic effects of contrails. Line-
shaped contrails could trigger the formation of contrail cirrus, which have much larger
coverage; however, the magnitude of this effect has yet to be assessed. Further, Schumann
(2012) has developed a numerical model for the prediction of contrails for the purpose of
simulating and predicting a large ensemble of contrails as a function of given air traffic and
meteorology.
Meanwhile, in view of the rapid expansion of aviation activities worldwide, with
unknown environmental and climate impacts, Friedl et al. (1997) issued an interim report
assessing the impact of subsonic fleets on NOx emission and the formation of tropospheric
O3 , as well as the addition of contrails and contrail cirrus from NASA’s viewpoint (see also
Subsection 1.6.2, on the SUCCESS field experiments of 1996). Furthermore, the Federal
Aviation Administration (FAA) has, together with NASA and other agencies, developed and
implemented a dynamic, flexible, and scalable Next Generation Air Transportation System
(NextGen) that is safe, secure, efficient, and environmentally sound, with the objective
of environmental protection that allows sustained aviation growth. One of the NextGen
environmental goals is to reduce the impact of aviation emissions on global climate. A
report to the U.S. Congress in 2004 on “Aviation and the Environment,” prepared by the
FAA-NASA-Transport Canada sponsored Center of Excellence PARTNER (the Partnership
for AiR Transportation Noise and Emissions Reduction), stated that, among the environ-
mental impacts of aviation, the greatest uncertainty and contention were associated with
climate and climate change. There is a need to better characterize non-CO2 aviation climate
impacts and their associated uncertainties in order to formulate realistic goals and actions
to mitigate these impacts.
Both the IPCC report, Aviation and the Global Atmosphere (IPCC Report 1999), men-
tioned above, and the IPCC Fourth Assessment Report, Climate Change 2007 (IPCC
Report 2007, a draft of which was available in 2006), stated that the level of scientific
understanding of the non-CO2 climate impacts of aviation ranges from medium to very
low. In response, FAA and NASA jointly sponsored a workshop in 2006 on “Impacts of
Aviation on Climate Change” and concluded that there is a need for focused research
efforts in the United States to address uncertainties and gaps in our understanding of this
subject. Similar recommendations were also voiced during an International Civil Aviation
Organization (ICAO)-sponsored international workshop on “Assessing current scientific
knowledge, uncertainty and gaps quantifying climate change, noise and air quality aviation
impacts” in 2007, echoing the need for more research in this area by the international
community.
Further to the preceding studies, FAA developed the Aviation Climate Change Research
Initiative (ACCRI) with participation from NASA, NOAA, and the Environmental Protec-
tion Agency (EPA). The main objective of ACCRI is to identify and address key scientific
gaps and uncertainties regarding climate impacts while providing timely scientific input to
inform optimal mitigation actions and policies. To develop its research plan, ACCRI openly
solicited and supported national and international science teams to develop subject-specific
white papers dealing with various aspects of the climate impacts of aviation. These reports
were designed to provide a review of the state of knowledge, uncertainties, analysis capa-
bilities, and gaps as well as to develop key research recommendations for goals that can be
achieved within long-term time horizons consistent with environmental and climate needs.
ACCRI convened an international science meeting in February 2008 and produced a report
summarizing findings and recommendations; in particular, Brasseur and Gupta (2010) have
presented a way forward.
The ACCRI program phase II was continued in 2010 as a result of efforts by the
FAA to reduce uncertainties in their earlier estimates of aviation-related radiative forc-
ing and climate response while accounting for projected aviation growth. Drivers for the
ACCRI program include computation of aviation emissions under various growth scenarios;
computation of climate change parameters including atmospheric composition, contrails,
contrail-induced cirrus clouds, the effects of particles on contrails and contrail-induced
cirrus clouds (as well as natural cirrus clouds); and the identification of appropriate metrics
to define climatic impacts. Brasseur et al. (2016) have presented the state of knowledge,
understanding, and quantification of current and future climate impacts of aviation emis-
sions using atmospheric models, surface and satellite observations, and laboratory experi-
ments. In what follows, we present a number of study highlights pertaining to contrails and
contrail-induced cirrus in terms of satellite observations and model simulations.
On the subject of satellite observations, Duda et al. (2013) produced a comprehensive
observational dataset for linear contrails by analyzing all Northern Hemisphere MODIS
data taken during 2006. Their best estimate of 0.135% mean hemispheric linear contrail
coverage yields a 0.07% global average when extrapolated to the Southern Hemisphere. The
Aqua afternoon contrail coverage is greatest over the North Pacific, the North Atlantic, and
the South China Sea (where its coverage reaches 1.2%). Just after midnight, linear contrail
coverage decreases everywhere, except over the mid-north Atlantic where coverage is
1.25%. A different pattern emerges from the Terra data, which yield more contrails over the
United States and northern Europe, where air traffic density suggests maximum coverage.
Iwabuchi et al. (2012) analyzed coincident and collocated MODIS and CALIPSO lidar data
corresponding to several thousand linear contrails visually detected in MODIS imagery
(see also Subsection 1.2.4) and found that their altitudes and temperatures were 10.9 ±
1.0 km and −54.6 ± 5.3°C, respectively. The mean contrail optical depth determined from
CALIPSO data was 0.19, along with a mean effective ice crystal size De of 20 µm. Xie
et al. (2012) also employed MODIS and CALIPSO linear contrail observations to develop
an ice crystal habit parameterization that best matched observed lidar parameters. Linear
contrail properties determined from CALIPSO and MODIS appeared to be remarkably
similar. Minnis et al. (2013) tracked selected contrails over the United States and found
that the combined linear and contrail cirrus coverage was 3.5 times the linear contrail value
and that contrail cirrus optical depth and De values were greater than linear contrail values.
Using MODIS results from Duda et al. (2013), Bedka et al. (2013) obtained an average
contrail temperature of −51.8°C with daytime and nighttime modes at −55 and −47°C,
respectively, and derived a mean contrail optical depth and De of 0.208 and 34.4 µm,
respectively. These values are consistent with those presented by Iwabuchi et al. (2012).
Spangenberg et al. (2013) computed the radiative forcing (RF, defined in Section 6.3)
of linear contrails over the Northern Hemisphere for the year 2006 based on concurrent
observations of their cover, temperature, optical depth, and ice crystal size determined by
Duda et al. (2013) and Bedka et al. (2013), as well as coincident cloud analyses provided by
Minnis et al. (2011a). The net RF at night is generally larger than in daytime, during which
longwave and shortwave forcings are largely canceled. Overall, the Northern Hemisphere
and global net contrail RFs are about 10.6 and 5.7 mW m−2 (10−3 W m−2 ), respectively, for
2006 Aqua results. The global mean net RF is slightly greater than the minimum estimate
obtained by Sausen et al. (2005) and close to the lower end of the RF range reported by Lee
et al. (2009a). The global net RF corresponding to total aviation-induced clouds would be
much larger because of additional contrail cirrus cover and would increase by a factor of
9, yielding a value of 51.3 mW m−2 .
With reference to model simulations, Chen et al. (2012) and Chen and Gettelman (2013)
assessed the RF from linear contrails and persistent contrail cirrus using 2006 air traffic
data and a temperature and humidity dependent contrail parameterization in the NCAR
Community Earth System Model (CESM) version 1. They found that the RF from linear
contrails is sensitive to the diurnal cycle. They obtained an estimated global value for linear
contrails ranging from 0 to about 6 mW m−2 assuming a 30-min contrail lifetime, consistent
with satellite-based observations. When the spreading of contrails is accounted for in model
simulations, the estimated global RF increases to about 4–20 mW m−2 . Over regions where
the highest air traffic occurs, the local RF can be as large as 1 W m−2 .
Yi et al. (2012) estimated the 2006 global mean linear contrail net RF using the NCAR
Community Atmospheric Model (CAM) version 5, augmented by a specified linear contrail
distribution provided by Rap et al. (2010), and obtained a value of about 11.3 mW m−2 .
Figure 6.15 depicts (a) regional shortwave (solar), (b) longwave (thermal IR), and (c) net
RF distributions over the globe. Due to the reflection of sunlight, the contrail solar albedo
effect is always negative, whereas its IR greenhouse counterpart is always positive due
to the trapping of thermal infrared fluxes emitted from the surface and the atmosphere
below them. It must be pointed out that these two opposing effects involving contrails are
not canceled out (see also Section 6.3 regarding naturally occurring cirrus) even for the
globally averaged condition associated with the domination of the IR greenhouse effect,
albeit the net RF is small. However, over dense air traffic areas, net RFs are up to 10 times
greater than global average values, particularly in the three regions displayed in this figure:
154.9 mW m−2 over North America, 85.8 mW m−2 over Europe, and 25.0 mW m−2 over
East Asia. The contrail cover, optical depth, ice crystal size, and model cloud overlap, as
well as input atmospheric and surface parameters, all contribute to computed contrail RFs
and produce RF differences among models. However, it is quite clear that there is broad
consistency between the preceding estimates, with their overlapping uncertainty ranges.
Burkhardt and Kärcher (2011) developed a global contrail cirrus model that produced
global RFs of 38 and 31 mW m−2 , respectively without and with the feedback of contrail
water vapor reductions in conjunction with cirrus formation. This model estimate was
about twice that estimated by Chen and Gettelman (2013). The difference between the
two probably lies within the uncertainty range associated with the performance of different
GCMs. Moreover, Schumann and Graf (2013) found a larger aviation-induced cirrus impact
using combined observations and a contrail cirrus model developed by Schumann (2012).
An “aviation fingerprint” due to a daily cycle of air traffic was found in the diurnal cycle
of cirrus properties in the North Atlantic region (Graf et al. 2012). The same fingerprint
was found in annual mean diurnal cycles of cirrus cover and OLR derived from Meteosat
data. This fingerprint analysis suggests a contrail cirrus coverage of about 1–2% with
optical depths exceeding 0.1, a longwave RF of 600–900 mW m−2 , and mean contrail cirrus
lifetimes of about 3–4 hours. Also, linear contrails represent a small fraction with respect
to the contrail cirrus cover at shorter time scales. The large RF-to-cover ratio suggests that
man-made aviation induces additional cirrus cover as well as thickening existing cirrus.
Various ACCRI analyses suggest that the impact of aviation-induced cirrus on “global” RF is
about 16–100 mW m−2 , which is about 1–6% of the total estimated anthropogenic radiative
forcing of 1.6 W m−2 (Solomon et al. 2007) without regional considerations. However,
as pointed out in the preceding paragraph, aviation emissions are mostly concentrated in
flight corridor regions. Thus, regional studies are needed to better understand geographical
disparities in the regional climate impacts of aviation emissions.
Lee et al. (2009a), cited previously, conducted an investigation on aviation and global
climate change in the twenty-first century and presented in a rather comprehensive fashion
the RFs resulting from aviation-produced contrails and subsequent induced cirrus clouds. In
addition to the subject of contrails and contrail cirrus, the IPCC Fourth Assessment Report
(a) SW Global average is -0.00624478

90N
60N
30N
30S
60S
90S
180 150W120W90W60W 30W 0 30E 60E 90E 120E150E 180
(b) LW Global average is 0.0175618

90N
60N
30N
30S
60S
90S
180 150W120W90W60W 30W 0 30E 60E 90E 120E150E 180
(c) NET Global average is 0.011317

90N
60N
30N
30S
60S
90S
180 150W120W90W60W 30W 0 30E 60E 90E 120E150E 180
-0.5 -0.1 -0.01 -0.005-0.001 0 0.001 0.005 0.01 0.1 0.5
Figure 6.15 Simulated 2006 global annual averaged (a) shortwave, (b) longwave, and (c) net contrail
radiative forcings (W m−2 ) using the NCAR Community Atmospheric Model (CAM) version 5, aug-
mented by a specified linear contrail distribution. Large averaged regional net RFs of 154.9 mW m−2 ,
85.8 mW m−2 , and 25.0 mW m−2 are seen over North America, Europe, and East Asia, respectively
(after Yi et al. 2012).
Aviation Radiative Forcing Components in 2005

Spatial LO
RF Terms (W m-2) scale SU
0.0280
Carbon dioxide (0.0253)
Global High
{
Ozone 0.0263 Continental Med
to
production (0.0219) hemispheric -Low
Methane -0.0125 Med
Global -Low
NOx reduction (-0.0104)
0.0138 Med
Total NOx (0.0115) Global -Low
0.0028 Hemispheric
Water vapor (0.0020) to global Low
Best estimate
-0.0048 Local to
Sulfate aerosol Estimate
global
Low
(-0.0035)
(IPCC AR4 values)
90% confidence 0.0034 Local
Soot aerosol (0.0025) to global
Low
0.0118 Local to
Linear contrails continental Low
(0.01)
Induced cirrus 0.033

Local to Very
cloudiness hemispheric Low
Total aviation 0.055 Low

Global
(Excl. induced cirrus) (0.0478)
Total aviation 0.078 Global Low

(Incl. induced cirrus)
-0.08 -0.04 0 0.04 0.08 0.12

Radiative Forcing (W m-2)
Figure 6.16 Radiative forcing (RF) components from global aviation as evaluated from pre-
industrial times until 2005. Bars represent updated best estimates, or an estimate in the case of
aviation-induced cloudiness. Error bars represent the 90% likelihood range for each estimate. IPCC
AR4 values are indicated by white lines within the bars, as reported by Forster et al. (2007). The
induced-cloudiness estimate includes linear contrails. Numerical values are given on the right for
both IPCC AR4 (in parentheses) and updated values. The median value of total RF from aviation is
displayed, with and without aviation-induced cloudiness, in the last two rows. The geographic spatial
scale of RF from each component and the level of scientific understanding are also shown on the right
(after Lee et al. 2009a).
(AR4, Forster et al. 2007), Lee et al. (2009a), and Brasseur et al. (2016) have discussed
RFs produced by CO2 , NOx , and its related CH4 and O3 reductions, and H2 O, direct RFs
involving soot and sulfate aerosols, as well as policy issues and technological options for
mitigation and adaptation. These discussions are largely beyond the purview of this text,
but see the comments on aerosol indirect effects in the context of soot and sulfate, in line
with the discussion in Subsection 6.4.1, later in this subsection.
Figure 6.16 depicts RF components from global aviation evaluated from pre-industrial
times to 2005, as presented in Lee et al. (2009a). RF items include individual components
mentioned above. The red and blue boxes indicate the best estimates for RFs corresponding
to individual elements. Bars represent updated best estimates, or an estimate in the case
of aviation-induced cloudiness, which has the largest uncertainty; error bars represent the
90% likelihood range. IPCC AR4 values are indicated by white lines within the bars, as
reported by Forster et al. (2007). Numerical values are given on the right for both IPCC
AR4 (in parentheses) and updated values provided by Lee et al. (2009a). The median value
of total RFs from aviation, with and without aviation-induced cloudiness, is displayed in
the last two rows, while the geographic spatial scale of RF from each component and the
level of scientific understanding (LOSU) are also shown on the right.
The successful ACCRI program, initiated primarily by the FAA in the United States,
has made substantial contributions in terms of satellite observations of contrail and con-
trail cirrus cover, and in accounting for feedback and interactions of the climate system
using modern GCMs with advanced parameterizations of the radiative properties of ice
particles of various sizes and morphologies representative of contrails and contrail cirrus.
ACCRI further identified additional aviation-related RFs with reference to long-term O3
changes, H2 O–NOx feedback, nitrate–aerosol effects, and direct and indirect effects of
sulfate aerosols. Lastly, the ACCRI team pointed out that the largest uncertainties in the
indirect effects of aircraft aerosols are associated with the impacts of aviation soot on
cirrus clouds (Penner et al. 2009, Liou et al. 2013, Gettelman and Chen 2013, Jacobson
et al. 2013), leading to substantial uncertainties in climate and climate change assessments
(Brasseur et al. 2016).
In view of the preceding discussion and in summary, it is clear that the impacts of
contrail-induced cirrus clouds on RF and the potential consequences for temperature (and
precipitation) perturbations have been and continue to be largely unknown, particularly at
the regional scale. Indeed, further concerted research endeavors are required.
6.5.2 The role of cirrus in upper-troposphere and lower-stratosphere exchange

6.5.2.1 Introductory notes
Exchanges of natural and anthropogenic trace chemical species and water vapor between
upper troposphere and lower stratosphere (UTLS) as well as cirrus cloud formation are
among the many important processes that must be understood for prediction of climate
and climate change. Holton et al. (1995) have provided a comprehensive review on the
subject from the viewpoint of the global-scale aspects of exchange in addition to the
small- and synoptic-scale mechanisms of exchange. Stratospheric water vapor plays a
critical role in the region’s chemistry and the radiative budget of the atmosphere (Forster
and Shine 2002, Solomon et al. 2010). Moreover, climate change would influence the
tropopause layer, the key in UTLS exchanges, through enhanced large-scale upwelling
of air and potential changes in tropical convection, temperature, chemical composition,
and cirrus clouds (Randel and Jensen 2013). Lastly, man-made contrails and contrail-
induced cirrus, as presented in Subsection 6.5.1, could also play an important role in UTLS
exchanges.
101 30
(a)
25
Pressure (hPa)
Altitude (km)
Stratosphere
20
Cold Point
102 Tropopause 15
Troposphere 10
Cold Point
Temperature 5
103
180 200 220 240 260 280 300
Temperature (K)
30 N 0.24
(b)
15 N
0.16
Latitude
0
0.08
15 S
0
30 S 20 m/s
0 60 E 120 E 180 120 W 60 W 0
Longitude
193 (c) 12
Cloud Fraction (%)

2
194 10
BDC Index
1
T100 hPa (K)
195 8
196 0
6
197 -1
4
198 -2
2
199
2007 2008 2009 2010 2011 2012 2013
Year
Figure 6.17 Cirrus clouds and temperatures near the tropical tropopause. (a) Vertical profile of
temperature. (b) Horizontal distributions of temperature (contours) and winds (vectors) at 100 hPa
from reanalysis data, and cirrus fractions with bases higher than 15 km obtained from CALIPSO
lidar observations (color scale). (c) Temporal variation in temperature at 100 hPa (blue), cirrus cloud
fraction with cloud bases higher than 15 km (averaged over the tropics, red), and the BDC index (after
Fu 2013).
Fu (2013) presented the following figures to illustrate 2-D structures of key parameters
in UTLS exchanges. Figure 6.17a depicts an illustrative temperature profile for the tro-
posphere and stratosphere as a function of altitude (in km) and pressure (in hPa), where
the minimum (cold point) is the tropopause where thin cirrus clouds frequently form.
Cirrus clouds and temperatures near the tropical tropopause are determined by a com-
plex combination of stratospheric (top-down) and tropospheric (bottom-up) processes.
Figure 6.17b shows the horizontal distribution of temperatures (contours) and winds (vec-
tors) at 100 hPa from reanalysis data, and cirrus cloud fractions with bases higher than 15 km
obtained from CALIPSO lidar observations (color scale). The zonal structures in both cir-
rus and temperature at 100 hPa closely resemble the mean pattern of equatorial planetary
waves driven by massive convection over the Indo-Pacific warm pool area. Figure 6.17c
presents the temporal variations of cirrus and temperature near the tropical tropopause,
which are strongly modulated by extratropical stratospheric waves and drive a large-scale
latitudinal circulation in the stratosphere, with air rising across the tropical tropopause,
moving poleward, and sinking towards the extratropical troposphere. This is referred to as
the Brewer–Dobson circulation (BDC), and is particularly evident in seasonal cycle. The
response of both equatorial planetary waves and upward motion associated with the BDC
to global warming would have an important impact on tropical thin cirrus clouds, which,
in turn, could have a significant feedback to the climate system.
6.5.2.2 Cirrus clouds and the mechanism of stratosphere dehydration

In Subsection 1.8.1, we pointed out that outflow cirrus clouds from tropical towers could
be maintained in a convectively active state by radiative flux gradients within clouds,
which are warmed and lifted into the lower stratosphere by net radiative heating. Danielsen
(1982) postulated that extended anvils would become radiatively destabilized by top cooling
coupled with base warming, leading to lapse rate increases within anvils and providing an
upward flux of water vapor into the low stratosphere. The additional moisture would
promote rapid ice crystal growth and fallout and, hence, serve as a dehydration mechanism
for the tropical stratosphere.
Ackerman et al. (1988) carried out broadband radiative transfer calculations for typical
tropical anvils and found that heating rate differences between cloud bottom and top ranged
from about 30 to 200 K day−1 . Lilly (1988) developed a mixed-layer model to analyze
the dynamic mechanism for cirrus anvil formation and showed that destabilization of the
layer could be produced by a strong radiative heating gradient, in support of Danielsen’s
hypothesis. The dehydration mechanism was discussed in an early paper by Brewer (1949),
who explained that the stratosphere dryness over the western Pacific Ocean is caused by air
parcels crossing the tropopause in the tropics, undergoing freezing/drying due to very low
temperatures, and returning at high latitudes; this is referred to as the “fountain theory.”
The stratospheric circulation pattern is now called the Brewer–Dobson circulation, as noted
above.
In an early study of stratosphere dehydration, Robinson (1980) suggested that the flux of
air across the tropical tropopause might be dominated by sporadic mass injections associated
with overshooting convective cumulus turrets that penetrate into the lower stratosphere.
To investigate and evaluate this mechanism, aircraft direct observations of the tropical
convective exchange were conducted over Panama in September 1980 and subsequently
over Darwin, Australia in January–February 1987 during the Tropical Experiment of the
Stratosphere-Troposphere Exchange Project (STEP). ER-2 flights near the cold western
Pacific tropopause region during the STEP experiment showed upward transport of air
into the stratosphere, clearly linked with local dehydration for some convective systems
(Danielsen 1993). Selkirk (1993) carried out radiosonde data analyses and demonstrated
that this convective transport occurred on a regional scale. Russell et al. (1993) showed that
a variety of convective cloud types were observed with disparate effects on water vapor
transport into the stratosphere. It appears to be inconclusive as to whether the convective
exchange and dehydration mechanism can account for stratospheric water vapor budget.
Prabhakara et al. (1988), Wang et al. (1994), and Jensen et al. (1996a) showed from
satellite, lidar, and aircraft observations that a persistent, thin, frequently subvisible cirrus
cloud layer, apparently not directly associated with deep convective systems, is usually
present just below the tropical tropopause. This thin cirrus layer occurs most frequently
over the tropical western Pacific and may be present as much as 80% of the time in
this region. The fact that this cloud layer occurs in tropopause regions having the lowest
temperatures suggests that it could play an important role in the dehydration of air entering
the stratosphere. Using a detailed cloud model, Jensen et al. (1996b) investigated the
potential impact on water vapor of ice clouds formed in situ near the tropical tropopause.
The potential was noted for dehydration when a few ice particles are formed and collect
water vapor to become sufficiently large to fall from the cloud.
The stratospheric fountain theory was challenged by Dessler (1998), who stated that
the tropopause temperature is cold enough to allow the observed low moisture value.
Sherwood (2000) analyzed radiosonde wind data and indicated that net subsidence rather
than upwelling occurred in the vicinity of the tropopause in the western Pacific region.
Gettelman et al. (2000) also showed a downward water vapor flux in this region, based on
a global chemical tracer modeling study.
Hartmann et al. (2001) argued that tropical tropopause altitudes could reach 17 km.
Cirrus clouds at tropical tropopause could be strongly heated if there are no clouds below
them. However, if tropopause cirrus clouds lie above convective cumulonimbus anvil top at
above 15 km, then radiative cooling of that tropopause cirrus could result, and its magnitude
is large enough to offset the heating due to subsidence, even at the lowest tropopause
temperature observed. For this reason, radiative cooling of tropopause cirrus clouds may
be important in UTLS dehydration processes. If thin layers of cirrus clouds occur in the
upper troposphere without other clouds below, thin cirrus will be heated by emission
from the underlying surface and will undergo evaporation. The authors postulated that,
with sufficiently cold convective cloud tops, the upper atmosphere could be cooled to
form cirrus clouds more easily. Shown in Figure 6.18 is a cartoon depiction of a two-
cloud radiation system imbedded in the standard temperature profile. For simplification of
argument, consider that atmospheric absorption is absent, solar absorption is neglected, and
the convective anvil cloud top (CA) is a blackbody. If the emissivity of optically thin cirrus
above is denoted by ε, then the radiative balance equation is given by
4
σ TCA − εσ Tcirrus
4
= εσ Tcirrus
4
, (6.5.1)
where T is the temperature and σ is the Stefan–Boltzmann constant. It follows that Tcirrus =
4
(TCA /2ε)1/4 . To obtain a radiative equilibrium temperature that is less than the mean
temperature of about 195 K of the tropical tropopause, the emission temperature of the
4
εσTcirrus
Cirrus
4
σTCA
4
εσTcirrus
Convective
Anvil
Temperature
Profile
Figure 6.18 A schematic diagram illustrating the thermal IR emissions of a thin cirrus above an
optically thick convective anvil cloud, including a typical tropical temperature profile where thin
cirrus occurs around the coldest temperature area (after Hartmann et al. 2001).
underlying convective anvil must be less than about 232 K, which corresponds to the
temperature at an altitude of about 10.5 km.
Hartmann et al. (2001) carried out radiative heating rate calculations for thin cirrus above
thick convective anvil clouds in the tropics, based on the broadband solar and thermal IR
radiative transfer program developed by Fu and Liou (1993) with further modifications and
improvements. The calculations were performed using atmospheric profiles – including
temperature, water vapor mixing ratio, and ozone – typical of the western Pacific in the
winter season, where the temperature has a cold point of 188 K at 17 km (note that the
standard tropical atmosphere has its coldest temperature of 195 K at this height). Also used
were a vertical resolution of 125 m, a surface albedo of 0.05, and a solar zenith angle of 54°
with a daytime fractional length of 0.496, corresponding to January at 5°N. Additionally,
the calculations employed a 0.5 km-thick thin cirrus with a top height of 16.5 km and a mean
effective ice crystal size of 13 µm, based on in situ aircraft observations for subvisual cirrus.
Figure 6.19 shows heating rate profiles for thin cirrus between 15 and 18 km with different
optical depths, with and without cooling contributions from an underlying thick anvil cloud
assumed to be a blackbody. Without the presence of a convective anvil cloud below, the
resulting heating rates for thin cirrus are 0.3, 1.3, 3.4, and 5.4 K day−1 , all positive values,
for thin cirrus optical depths of 0 (clear atmosphere), 0.01, 0.03, and 0.05, respectively.
However, when convective anvil clouds are added, the respective heating rate counterparts
become 0.1 (clear + anvil), −0.1, −0.4, and −0.7 K day−1 .
18
Cirrus Optical Depth

0
0.01
0.03
17 Cloud below with 0.05
Height (km)
the top at 14 km
16
No cloud below
15
-2 -1 0 1 2 3 4 5 6
Radiative Heating Rate (K day –1)
Figure 6.19 Radiative heating rates near the tropical tropopause with cirrus clouds present between
16 and 16.5 km with various cloud optical depths including no cloud below, and with and without
convective cloud top at 14 km below (after Hartmann et al. 2001).
Based on the preceding computational results, contributions from thermal IR emis-

sions dominate in cases involving the thin-cirrus and convective-anvil-cloud system, i.e.,
tropopause cirrus clouds in colder regions above convective anvils can lead to net radiative
cooling accompanied by subsidence and contribute to a mechanism for drying the air to the
3.5 ppmv observed in the lower stratosphere. Thus, if thin cirrus overlap convective anvils
for more than 1000 km or so, this will allow time for radiative cooling to take place and
for ice crystals to grow and fall significantly relative to air, as postulated by Jensen et al.
(1996b). Extended cirrus cloud decks of this scale can often be seen in satellite images over
the Maritime Continent region. Once the parcel leaves the shelter of the underlying anvil
cloud, it would again be heated by radiation from below and could rise into the stratosphere
with its moisture depleted. The stratospheric dehydration mechanism is one of cycling of
air by horizontal advection through the cold region above tropical convection and the slow
ascent of dehydrated air in non-convective regions, in line with the view of the horizon-
tal transport associated with stratosphere dehydration discussed in Holton and Gettelman
(2001).
Corti et al. (2006) further investigated the mechanism involving air parcels entering the
stratosphere from the troposphere through upwelling in the tropics. It was pointed out that
the transport is either too slow or insufficient in mass flux to act as a major supply for the
Brewer–Dobson circulation. These authors found that a combination of deep convection
and subsequent upwelling related to cirrus and clear sky via cirrus–radiation interactions
and feedback offer a pathway for sufficient troposphere–stratosphere transport to sustain
the Brewer–Dobson circulation. Using radiative heating calculations based on cirrus cloud
fields determined from CALIPSO and CloudSat, Lin et al. (2013) showed that upward
mass fluxes in the tropical lower stratosphere can be extended to the upper troposphere
with vertical velocities of 0.77 mm s−1 at 15.5 km. The paper by Fu (2013) cited above
and his latitude–longitude depiction of cirrus clouds, temperature, and winds near the
tropopause, as well as the seasonal variation of cirrus cover, temperature, and Brewer–
Dobson circulation index presented in Figure 6.17, clearly provide a unified perspective
for UTLS exchanges, including an explanation for stratosphere dehydration. It is quite
apparent that cirrus clouds, of various sizes and morphologies, play a key role in transport
processes through radiation interactions and feedback.
6.5.3 Thin cirrus clouds and climate

6.5.3.1 Thin cirrus detection and comparison with GCM output
The presence of thin cirrus clouds with optical depths smaller than 0.1 is an impor-
tant factor which could play a key role in global and regional radiative forcing assess-
ments and in the consequences for climate and climate change. In this connection, in
Subsection 5.2.5 we introduced three global cirrus cloud climatologies in terms of cloud
cover, optical depth, and mean effective ice crystal size, as presented by King et al. (2013),
Minnis et al. (2011b), and Kahn et al. (2014). These parameters were determined from
retrieval algorithms based on data gathered from nadir-looking radiometers at visible
and infrared wavelengths. However, nadir-looking radiometers have limitations, and large
uncertainties in detecting and determining thin and subvisible cirrus clouds because of the
nature of these optically thin clouds, especially over land surfaces. For this reason, it appears
that past cirrus cloud climatologies may have underestimated thin cirrus cloud parameters
and that the net radiative forcing of these clouds could be on the order of 1–2 W m−2 (Lee
et al. 2009a), an amount that has been unaccounted for in contemporary climate models,
cited previously. Clearly, the representation of thin ice clouds in the upper troposphere
in GCMs has shortcomings due to a lack of adequate, high-quality, global observations
to constrain and calibrate model simulation results for the purpose of minimizing errors
in weather and climate predictions, as well as narrowing down sources of uncertainty in
climate change projections.
Li et al. (2005b) introduced a set of IWC measurements from the Earth Observing
System’s Microwave Limb Sounder (MLS, Waters et al. 1999) to assess the status of global
climate models in simulating upper tropospheric IWC. The MLS, aboard the Aura satellite
launched in July 2004 as part of the A-Train constellation (see Subsection 5.1.5), measures
microwave emissions from atmospheric long paths to determine chemical composition,
water vapor, temperature, and cloud ice. Limb geometry allows the retrieval program to
derive vertical profiles of these parameters on fixed pressure surfaces, providing nearly
global coverage with a vertical resolution of 4 km and horizontal along- and cross-track
resolutions of 200 and 7 km, respectively. Cloud ice in terms of IWC has been derived
from MLS 240 GHz measurements. Comparisons have been made with European Centre for
Medium-Range Weather Forecasts (ECMWF) analyses and with simulations from several
GCMs, including the Geophysical Fluid Dynamics Laboratory (GFDL) model with the
relaxed Arakawa–Schubert scheme (Moorthi and Suarez 1992), the GFDL with Donner’s
convection scheme (Donner et al. 2001), UCLA with Liou’s cloud–radiation scheme (Gu
et al. 2003), the NCAR Community Atmosphere Model V.3 (CAM3), and the NASA
Goddard Institute for Space Studies (GISS) model (Del Genio et al. 2005).
For January 2005 monthly and daily mean IWC values, the general spatial agreement
between MLS and ECMWF is quite reasonable; however, MLS estimates are larger by a
factor of 2–3 over a number of areas. For GCMs, the agreement between model and data
is within a factor of 2–4. These differences were produced by sampling and uncertainties
in MLS observations as well as uncertainties in model simulated values. These results
demonstrate the potential usefulness of the MLS dataset for evaluating GCM performance
and guiding development efforts.
Further to the preceding study, Waliser et al. (2009) presented two types of satellite
ice cloud observations based on MLS and CloudSat measurements. The MLS IWCs were
derived from correlation studies between modeled cloud induced radiances and IWCs based
on 240 GHz measurements (Livesey et al. 2007, Wu et al. 2009). Here we note that the
detection of ice clouds by means of microwave emissions is not directly related to light
scattering by ice crystals. The Cloud Profiling Radar (CPR) on the CloudSat satellite, part
of the A-Train constellation, is a fixed 94 GHz nadir-viewing active radar, operational since
June 2006, that also measures backscattered power from the Earth’s surface and cloud
particles in the atmospheric column as a function of distance (Stephens et al. 2002). The
CloudSat retrieval for IWC employs an optimal estimation approach to determine ice cloud
parameters based on radar reflectivity measurements, and consists of vertical profiles along
the satellite ground track. Because of the along-track motion, the CPR footprint is oblong,
1.3 km across track and 1.7 km along track.
Figure 6.20 shows mean annual distribution of global ice-cloud IWC fields at 215 hPa (in
the vicinity of the tropopause) as a function of latitude and longitude, derived from (a) MLS
and (b) CloudSat under conditions of no precipitation at the surface and no convective clouds
(deep convection and cumulus). Comparisons were made with IWC fields simulated from
two GCMs: (c) the Goddard Earth Observing System Model, Version 5 (NASA/GEOS5)
and (d) NCAR/CAM3. The two GCMs perform relatively well within the context of MLS
and CloudSat ice cloud IWC fields, a significant improvement considering the substantial
disparity displayed in the results presented in Li et al. (2005b). As illustrated in these figures
based on satellite observations and model simulations the mean annual thin ice cloud IWC
distribution generally covers between 40°N and 40°S with a maximum along equatorial
areas and two gaps between 180° and 60°W over the Pacific Ocean.
6.5.3.2 Some notes on high thin clouds in global radiative and climate forcing
It is quite clear that satellite-retrieved data on cloud ice in the upper troposphere derived
from MLS and CloudSat (as well as CALIPSO) can, on the basis of remote sensing
30
28
24
20
15
10
8
4
3
2
1
Figure 6.20 Mean annual IWC (x 10−3 g m−3 ) at 215 hPa for (a) MLS covering 2007, (b) CloudSat
in cases of no precipitation (NP) and no convective clouds (NC) covering August 2006 to July 2007,
(c) NASA/GEOS5 simulations covering January 1999 to December 2002, and (d) NCAR/CAM3
simulations covering 1979 to 1999 (after Waliser et al. 2009, with modifications).
principles, be used to assist GCM diagnosis of ice in the atmosphere and, at the same
time, can offer some indications of possible model improvements. However, the question
of compatibility between satellite data and model output for cloud ice has not been well
addressed because the related observing systems and model parameterizations and frame-
works have developed essentially independently, leading to differences in approach to the
same problem. On the one hand, the retrieval of ice cloud fields is still in its embryonic stage.
For instance, see the MODIS product evolution (C6 versus C5) discussed in Subsection
5.2.4, which describes concerns with the a priori use of ice crystal size and morphology
and the consequences for the phase function and single-scattering albedo employed in
retrieval algorithms. Additionally, retrievals/sensors often have issues and uncertainties,
while GCMs exhibit considerable variation in their representations of ice clouds.
The fundamental question appears to be whether MODIS-, MLS-, or CloudSat-derived
IWC can be used to evaluate IWC values simulated by GCMs. Future inquiries should
consider what meaningful comparisons can be made between remote sensing data and
GCM output, and how the former can best be applied to evaluation, assessment, and
improvement of the latter. It is a great challenge to bridge significant gaps between the two
independent data sources. Of course, nothing is exact even in a perfect world – uncertainties
exist everywhere. It appears that the gaps could be reduced by the following approach. To the
extent that global satellite observations trump GCM simulations and that independent data
are needed to check “crystal ball” (computer model) performance in future predictions,
we may use the independent data continuously available from satellites to evaluate and
improve GCM physical parameterizations of ice cloud formation over a time span of years.
This principle may apply to other parameterizations in the model as well.
We now discuss a couple of cloud ice variables that have not been gathered from the
aforementioned satellites. First, a long-term trend must be established for the frequency
of satellite-based observations of thin cirrus occurrence globally in order to understand
whether global warming has impacted cloud ice cover in the upper troposphere. MLS,
CALIPSO, and CloudSat have demonstrated their ability to provide accurate and reliable
cloud ice data over more than 10 years. Perhaps an extrapolation to a few years back
could be accomplished in combination with other satellite datasets and/or model simulated
results.
Second, present ice cloud data consist of only IWC, but lack mean effective size De .
For radiative forcing analysis, however, both IWC and De are required in radiative transfer
calculations, as discussed in Subsection 6.3.1. In Subsection 6.4.1, we showed that IWC
has a tendency to increase with temperature and that De can be parameterized in terms of
IWC. Despite the lack of De data from the above-mentioned satellites, the parameterization
approach outlined in that subsection may be useful as a first approximation to investigate
radiative forcing of cloud ice in the upper troposphere, make correlations with tropospheric
temperature trends, and understand the relative role of IWC and De in these analyses.
Here we can also consider the role of optically thin cirrus in global cloud radiative
forcing from the separate perspectives of cloud cover and cloud microphysics (IWP and
De ). With respect to the former, thin cirrus clouds with optical depths less than 0.7 cover
approximately 20% of the midlatitude region and more than 50% over the tropics, based
on NOAAHIRS observations from 1979–2001 (Wylie et al. 2005). These values could
overestimate high-level cloud cover because of low-resolution datasets used in the anal-
ysis. For about 10 years the conventional passive spectral radiometers on board MODIS,
AIRS, and CERES have provided cirrus cloud climatology, including cloud cover, opti-
cal depth (proportional to IWP), and mean effective ice crystal size (De ), as presented in
Subsection 5.2.5. Limitations on and uncertainties in these nadir-looking spectral radiome-
ters for thin cirrus clouds have been discussed in Subsection 5.2.6. Moreover, based on
ground-based mm-wave radar observations (see e.g., Roskovensky and Liou 2003a), high
cirrus clouds frequently overlap with low clouds with top heights less than 2 km (see
Section 1.1) and temperatures close to those of the underlying surfaces. For this reason,
the greenhouse effect of cirrus clouds would not be affected. This high/low cloud system
must be differentiated from contiguous convective clouds, often referred to as high clouds,
whose net radiative forcings were estimated to be nearly zero (Kiehl and Ramanathan 1990,
Bretherton and Hartmann 2009).
In Subsection 6.2.3, we introduced analysis of the Earth radiation budget (ERB) at TOA
derived from satellite observations available since the onset of the meteorological satellite
era. Extracting cloud effects from ERB data requires the identification of the ERB in clear-
sky conditions. Harrison et al. (1990) provided the first estimate of global cloud radiative
forcings with values of −48 and 31 W m−2 for solar and IR radiation, respectively, and a net
value of −17 W m−2 . The partition of contributions from high, middle, low, and convective
clouds cannot, however, be uniquely derived from top-down satellite ERB observations.
In Subsection 6.2.3, we presented an indirect approach to determining TOA radiation
budgets by means of a radiative transfer model coupled with input data from satellite
observations, namely, 3-D climatological fields of clouds, temperature, humidity, ozone,
and other relevant atmospheric and surface parameters, as shown in Figures 6.5 and 6.6.
In this manner, a consistent check between direct satellite ERB observations and model
computed values may be made, despite uncertainties encountered in both. Uncertainties
in the indirect approach involve cloud retrievals as well as parameterizations developed
for radiative transfer models. On the basis of simulation results from climate models, the
recent IPCC Report (2013) provided an estimate of global cloud radiative forcings of −47
and 26 W m−2 for solar and IR radiation, respectively, and a net value of −21 W m−2 ,
a substantial difference of −4 W m−2 compared to earlier satellite observations due to a
reduction in the IR greenhouse effect. Of course, this difference is only a number (similar
to the 30% global albedo of the Earth–atmosphere system); however, its implications
for climate radiative forcing in response to anthropogenic greenhouse warming could be
overlooked. The reasons for this difference must be multidimensional, but is it possible that
cloud ice in terms of cover and microphysics has not been accurately represented in climate
models?
With respect to cirrus cloud microphysics, we presented in Subsection 6.4.1 an analysis
of the relative strengths of the solar albedo and IR greenhouse effects in the case of cirrus
clouds as a function of IWP and De , and showed that the IR effect dominates the solar
effect. There, we used a 1-D climate–cirrus model to illustrate intricate interactions and
feedback involving IR, solar, IWP, and De on surface temperature perturbations under the
greenhouse warming scenario of CO2 doubling. We showed that cirrus cloud microphysics
through IWP and De interactions led to an overall positive feedback, largely through the
prevalent IR greenhouse effect, and resulted in an increase in surface temperature of more
than 10%. With the capability of MLS, CALIPSO, and CloudSat and their potential global
datasets for ice cloud cover, IWC, and De , global and regional ice cloud radiative forcings
could be determined, providing a basis for improvements in ice cloud parameterization and
simulation in climate models.
In view of the presentation in this text and in summary, the authors’ fundamental work
on light scattering by ice crystals has provided valuable and definitive contributions to the
development of satellite remote sensing of ubiquitous cirrus clouds comprised of ice crystals
of various sizes and morphology in the Earth’s atmosphere, as well as to the development
of radiation parameterizations for these clouds in climate models for investigating the
role of ice clouds in temperature and precipitation responses to global warming and climate
change.
To close this chapter and the book, we offer the following lines (see also the Epigraph):
Let there be light.

Let there be beautiful ice crystals in the air and mountain ranges.
And here come the reindeers and Santa Claus carrying Maxwell’s equations, and light
rays are shining in the wonderlands.
Let the glory of Geometric Optics for ice crystals, Newton’s optics, and sun’s light rays
rise again from the horizon.
Let ice crystals’ old friends – black carbon and dust – be not forgot for Auld Lang Syne.
And ice crystals are carried by the ceaseless winds; and
After traveling thousands of miles up and down, the sky looks very blue.
Let there be space missions to tender ubiquitous light rays in the sky,
And all things considered, let light scattering by ice crystals in remote sensing and
climate change be a delight.
References
Ackerman, T. P., Liou, K. N., Valero, F., and Pfister, L. (1988). Heating rates in tropical
anvils. J. Atmos. Sci., 45, 1606–23.
Airy, G. B. (1838). On the intensity of light in the neighbourhood of a caustic. Trans.
Cambridge Philos. Soc., 6, 397–403.
Alberta, T. L., Charlock, T. P., Whitlock, C. H., et al. (1994). Climate observations with
GEWEX Surface Radiation Budget Project data. In Proceedings of the Eighth Con-
ference on Atmospheric Radiation. Nashville, TN: American Meteorological Society,
pp. 22–4.
Albrecht, B. A. (1989). Aerosols, cloud microphysics, and fractional cloudiness. Science,
245, 1227–30.
Alexander, R. C. and Mobley, R. L. (1975). Monthly average sea-surface temperatures and
ice-pack limits on a 1° global grid. Mon. Wea. Rev., 104, 143–8.
Ambartzumian, V. A. (1936). The effect of absorption lines on the radiative equilibrium of
the outer layers of the stars. Publ. Observ. Astronom. Univ. Leningrad, 6, 7–18.
Aoki, T., Aoki, T., Fukabori, M., et al. (2000). Effects of snow physical parameters on
spectral albedo and bidirectional reflectance of snow surface. J. Geophys. Res., 105,
10219–36.
Appleman, H. (1953). The formation of exhaust condensation trails by jet aircraft. Bull.
Am. Meteorol. Soc., 34, 14–20.
Arakawa, A. (2000). A personal perspective on the early years of general circulation
modeling at UCLA. In General Circulation Model Development: Past, Present, and
Future – Proceedings of a Symposium in Honor of Professor Akio Arakawa, ed. D. A.
Randall. New York: Elsevier, pp. 1–65.
Arfken, G. (1985). Mathematical Methods for Physicists, 3rd edn. Orlando: Academic
Press.
Arking, A. and Grossman, K. (1972). The influence of line shape and band structure on
temperatures in planetary atmospheres. J. Atmos. Sci., 29, 937–49.
Asano, S. and Sato, M. (1980). Light scattering by randomly oriented spheroidal particles.
Appl. Opt., 19, 962–74.
Asano, S. and Yamamoto, G. (1975). Light scattering by a spheroidal particle. Appl. Opt.,
14, 29–49.
Auer, A. H., Jr. and Veal, D. L. (1970). The dimension of ice crystals in natural clouds. J.
Atmos. Sci., 27, 919–26.
Auriol, F., Gayet, J.-F., Febvre, G., et al. (2001). In situ observation of cirrus scattering
phase functions with 22° and 46° halos: Cloud field study on 19 February 1998. J.
Atmos. Sci., 58, 3376–90.
392
References 393
Babinet, J. (1837). Mémoires d’optique météorologique. C. R. Acad. Sci., 4, 638–48.

Bailey, M. and Hallett, J. (2004). Growth rates and habits of ice crystals between −20° and
−70°C. J. Atmos. Sci., 61, 514–44.
Bailey, M. and Hallett, J. (2009). A comprehensive habit diagram for atmospheric ice
crystals: Confirmation from the laboratory, AIRS II, and other field studies. J. Atmos.
Sci., 66, 2888–99.
Baran, A. J. (2003). Simulation of infrared scattering from ice aggregates by use of a
size-shape distribution of circular ice cylinders. Appl. Opt., 42, 2811–18.
Baran, A. J. (2012). From the single-scattering properties of ice crystals to climate predic-
tion: A way forward. Atmos. Res., 112, 45–69.
Baran, A. J. and Francis, P. N. (2004). On the radiative properties of cirrus cloud at solar
and thermal wavelengths: A test of model consistency using high-resolution airborne
radiance measurements. Quart. J. Roy. Meteorol. Soc., 130, 763–78.
Barber, P. W. and Hill, S. C. (1990). Light Scattering by Particles: Computational Methods.
Singapore: World Scientific.
Barber, P. W. and Yeh, C. (1975). Scattering of Electromagnetic Waves by Arbitrarily
Shaped Dielectric Bodies. Appl. Opt., 14, 2864–72.
Barkey, B. and Liou, K. N. (2001). Polar nephelometer for light-scattering measurements
of ice crystals. Opt. Lett., 26, 232–4.
Barkey, B. and Liou, K. N. (2006). Laboratory measurements of spectral reflection from
ice clouds of various habits. Appl. Opt., 45, 5716–24.
Barkey, B. and Liou, K. N. (2008). Visible and near infrared reflectances measured from
laboratory ice clouds. Appl. Opt., 47, 2533–40.
Barkey, B., Liou, K. N., Takano, Y., Gellerman, W., and Sokolsky, P. (1999). An analog
light scattering experiment of hexagonal icelike particles. Part II: Experimental and
theoretical results. J. Atmos. Sci., 56, 613–25.
Barkey, B., Bailey, M., Liou, K. N., and Hallett, J. (2002). Light-scattering properties of
plate and column ice crystals generated in a laboratory cold chamber. Appl. Opt., 41,
5792–6.
Baum, B. A., Kratz, D. P., Yang, P., et al. (2000a). Remote sensing of cloud properties
using MODIS airborne simulator imagery during SUCCESS: 1. Data and models. J.
Geophys. Res., 105, 11767–80.
Baum, B. A., Soulen, P. F., Strabala, K. I., et al. (2000b). Remote sensing of cloud properties
using MODIS airborne simulator imagery during SUCCESS: 2. Cloud thermodynamic
phase. J. Geophys. Res., 105, 11781–92.
Baum, B. A., Heymsfield, A. J., Yang, P., and Bedka, S. T. (2005a). Bulk scattering
properties for the remote sensing of ice clouds. Part I: Microphysical data and models.
J. Appl. Meteorol., 44, 1885–95, doi:10.1175/JAM2308.1.
Baum, B. A., Yang, P., Heymsfield, A. J., et al. (2005b). Bulk scattering properties for
the remote sensing of ice clouds. Part II: Narrowband models. J. Appl. Meteorol., 44,
1896–911, doi:10.1175/JAM2309.1.
Baum, B. A., Yang, P., Nasiri, S. L., et al. (2007). Bulk scattering properties for the remote
sensing of ice clouds. Part III: High-resolution spectral models from 100 to 3250 cm−1 .
J. Appl. Meteorol. Clim., 46, 423–34.
Bedka, S. T., Minnis, P., Duda, D. P., Chee, T. L., and Palikonda, R. (2013). Properties of
linear contrails in the northern hemisphere derived from 2006 Aqua MODIS observa-
tions. Geophys. Res. Lett., 40, 772–7.
Bentley, W. A. and Humphreys, W. J. (1931). Snow Crystals. New York: Dover Publi-
cations.
394 References
Berenger, J.-P. (1994). A perfectly matched layer for the absorption of electromagnetic
waves. J. Comput. Phys., 114, 185–200.
Bergeron, T. (1935). On the physics of clouds and precipitation. Proc. Fifth Assembly
U.G.G.I. Lisbon, 2, 156–78.
Bertie, J. E., Labbé, H. J., and Whalley, E. (1969). Absorptivity of ice I in the range of
4000–30 cm−1 . J. Chem. Phys., 50, 4501–20.
Bi, L. and Yang, P. (2014). Accurate simulation of the optical properties of atmospheric ice
crystals with the invariant imbedding T-matrix method. J. Quant. Spectrosc. Radiat.
Transfer, 138, 17–35.
Bi, L. and Yang, P. (2015). Impact of calcification on the inherent optical properties
of Emiliania huxleyi coccoliths and coccolithophores. J. Quant. Spectrosc. Radiat.
Transfer, 155, 10–21.
Bi, L., Yang, P., Kattawar, G. W., Hu, Y., and Baum, B. A. (2011a). Diffraction and external
reflection by dielectric faceted particles. J. Quant. Spectrosc. Radiat. Transfer, 112,
163–73.
Bi, L., Yang, P., Kattawar, G. W., Hu,Y., and Baum, B. A. (2011b). Scattering and absorption
of light by ice particles: Solution by a new physical-geometric optics hybrid method.
J. Quant. Spectrosc. Radiat. Transfer, 112, 1492–508.
Bi, L., Yang, P., Kattawar, G. W., and Mishchenko, M. I. (2013). Efficient implementation
of the invariant imbedding T-matrix method and the separation of variables method
applied to large non-spherical inhomogeneous particles. J. Quant. Spectrosc. Radiat.
Transfer, 116, 169–83.
Bishop, S. E. (1884). The remarkable sunsets. Nature, 29, 259–60.
Blaschak, J. G. and Kriegsmann, G. A. (1988). A comparative study of absorbing boundary
conditions. J. Comput. Phys., 77, 109–39.
Blau, H. H., Jr., Espinola, R. P., and Reifenstein, E. C. (1966). Near infrared scattering by
sunlit terrestrial clouds. Appl. Opt., 5, 555–64.
Böhm, H. P. (1989). A general equation for the terminal fall speed of solid hydrometeors.
J. Atmos. Sci., 46, 2419–27.
Bohren, C. F. and Huffman, D. R. (1983). Absorption and Scattering of Light by Small
Particles. New York: John Wiley & Sons.
Born, M. and Wolf, M. (1975). Principles of Optics, 5th edn. Oxford: Pergamon Press.
Borovoi, A. G. and Grishin, I. A. (2003). Scattering matrices for large ice crystal particles.
J. Opt. Soc. Am. A, 20, 2071–80.
Borovoi, A., Grishin, I., Naats, E., and Oppel, U. (2002). Light backscattering by hexagonal
ice crystals. J. Quant. Spectrosc. Radiat. Transfer, 72, 403–17.
Boudala, F. S., Isaac, G. A., Fu, Q., and Cober, S. G. (2002). Parameterization of effective
ice particle size for high-latitude clouds. Int. J. Climatol., 22, 1267–84.
Braham, R. R. (1974). Cloud physics of urban weather modification: A preliminary report.
Bull. Am. Meteorol. Soc., 55, 100–6.
Brasseur, G. P. and Gupta, M. (2010). Impact of aviation on climate: Research priorities.
Brasseur, G. P., Gupta, M., Anderson, B. E., et al. (2016). Impact of aviation on climate:
FAA’s Aviation Climate Change Research Initiative (ACCRI) Phase II. Bull. Amer.
Meteorol. Soc., 97, 561–83.
Bretherton, C. S. and Hartmann, D. L. (2009). Large-scale controls on cloudiness. In Clouds
in the Perturbed Climate System: Their Relationship to Energy Balance, Atmospheric
Dynamics, and Precipitation, ed. J. Heintzenberg and R. J. Charlson. Strüngmann
Forum Reports, 2. Cambridge, MA: MIT Press, pp. 217–34.
References 395
Brewer, A. W. (1949). Evidence for a world circulation provided by the measurements of

helium and water vapor distribution in the stratosphere. Quart. J. Roy. Meteorol. Soc.,
75, 351–63.
Britt, C. L. (1989). Solution of electromagnetic scattering problems using time domain
techniques. IEEE Trans. Antennas Propag., 37, 1181–92.
Burkhardt, U. and Kärcher, B. (2011). Global radiative forcing from contrail cirrus. Nature
Clim. Change, 1, 54–8.
Byers, H. R. (1965). Elements of Cloud Physics. Chicago: University of Chicago Press.
C.-Labonnote, L., Brogniez, G., Doutriaux-Boucher, M., et al. (2000). Modeling of light
scattering in cirrus clouds with inhomogeneous hexagonal monocrystals: Comparison
with in-situ and ADEOS-POLDER measurements. Geophys. Res. Lett., 27, 113–16.
C.-Labonnote, L., Brogniez, G., Buriez, J.-C., and Doutriaux-Boucher, M. (2001). Polarized
light scattering by inhomogeneous hexagonal monocrystals: Validation with ADEOS-
POLDER measurements. J. Geophys. Res., 106, 139–53.
Cai, Q. and Liou, K. N. (1982). Polarized light scattering by hexagonal ice crystals: Theory.
Appl. Opt., 21, 3569–80.
Cairns, B., Russell, E. E., and Travis, L. D. (1999). The research scanning polarimeter:
Calibration and ground-based measurements. Proc. SPIE, 3754, 186–96.
Chamberlain, J. W. and Hunten, D. M. (1987). Theory of Planetary Atmospheres: An
Introduction to Their Physics and Chemistry. New York: Academic Press.
Chandrasekhar, S. (1950). Radiative Transfer. London: Oxford University Press.
Chang, F.-L. and Li, Z. (2002). Estimating the vertical variation of cloud droplet effective
radius using multispectral near-infrared satellite measurements. J. Geophys. Res., 107,
4257.
Chang, P. C., Walker, J. G., and Hopcraft, K. I. (2005). Ray tracing in absorbing media. J.
Quant. Spectrosc. Radiat. Transfer, 96, 327–41.
Changnon, S. A. (1981). Midwestern cloud, sunshine and temperature trends since 1901:
Possible evidence of jet contrail effects. J. Appl. Meteorol., 20, 496–508.
Charlock, T. P. (1982). Mid-latitude model analysis of solar radiation, the upper layers of
the sea, and seasonal climate. J. Geophy. Res., 87, 8923–30.
Charlock, T. P. and Alberta, T. L. (1996). The CERES/ARM/GEWEX Experiment
(CAGEX) for the retrieval of radiative fluxes with satellite data. Bull. Am. Meteo-
rol. Soc., 77, 2673–83.
Charlson, R. J., Lovelock, J. E., Andreae, M. O., and Warren, S. G. (1987). Oceanic
phytoplankton, atmospheric sulphur, cloud albedo and climate. Nature, 326, 655–61.
Chen, C.-C. and Gettelman, A. (2013). Simulated radiative forcing from contrails and
contrail cirrus. Atmos. Chem. Phys., 13, 12525–36.
Chen, C.-C., Gettelman, A., Craig, C., Minnis, P., and Duda, D. P. (2012). Global contrail
coverage simulated by CAM5 with the inventory of 2006 global aircraft emissions. J.
Adv. Model. Earth Syst., 4, M04003.
Chen, G., Yang, P., and Kattawar, G. W. (2008). Application of the pseudospectral time-
domain method to the scattering of light by non-spherical particles. J. Opt. Soc. Am.
A, 25, 785–90.
Chen, J. P. and Lamb, D. (1994). Simulation of cloud microphysics and chemical processes
using a multicomponent framework. Part I: Description of the microphysical model.
J. Atmos. Sci., 51, 2613–30.
Chen, J. P., McFarquhar, G. M., Heymsfield, A. J., and Ramanathan, V. (1997). A modeling
and observational study of the detailed microphysical structure of tropical cirrus anvils.
J. Geophys. Res., 102, 6637–53.
396 References
Chepfer, H., Brogniez, G., and Fouquart, Y. (1998). Cirrus clouds’ microphysical properties
deduced from POLDER observations. J. Quant. Spectrosc. Radiat. Transfer, 60, 375–
90.
Chepfer, H., Brogniez, G., Goloub, P., Bréon, F. M., and Flamant, P. H. (1999). Observations
of horizontally oriented ice crystals in cirrus clouds with POLDER-1/ADEOS-1. J.
Chepfer, H., Goloub, P., Riedi, J., et al. (2001). Ice crystal shapes in cirrus clouds derived
from POLDER/ADEOS-1. J. Geophys. Res., 106, 7955–66.
Chou, M.-D. and Arking, A. (1980). Computation of infrared cooling rates in the water
vapor bands. J. Atmos. Sci., 37, 855–67.
Chou, M.-D., Lee, K.-T., Tsay, S.-C., and Fu, Q. (1999). Parameterization for cloud long-
wave scattering for use in atmospheric models. J. Climate, 12, 159–69.
Chowdhary, J., Cairns, B., Mishchenko, M. I., et al. (2005). Retrieval of aerosol scattering
and absorption properties from photopolarimetric observations over the ocean during
the CLAMS experiment. J. Atmos. Sci., 62, 1093–117, doi:10.1175/JAS3389.1.
Christiansen, C. (1884). Untersuchungen über die optischen Eigenschaften von fein ver-
theilten Körpern. Ann. Physik, 259, 298–306.
Christiansen, C. (1885), Untersuchungen über die optischen Eigenschaften von fein ver-
theilten Körpern. Ann. Physik, 260, 439–46.
Chung, S., Ackerman, S., van Delst, P. F., and Menzel, W. P. (2000). Model calculations
and interferometer measurements of ice-cloud characteristics. J. Appl. Meteorol., 39,
634–44.
Chýlek, P., Ramaswamy, V., and Cheng, R. J. (1984). Effect of graphitic carbon on the
albedo of clouds. J. Atmos. Sci., 41, 3076–84.
Clapp, M. L., Miller, R. E., and Worsnop, D. R. (1995). Frequency-dependent optical
constants of water ice obtained directly from aerosol extinction spectra. J. Phys.
Chem., 99, 6317–26.
Clarisse, L., Hurtmans, D., Prata, A. J., et al. (2010). Retrieving radius, concentration,
optical depth, and mass of different types of aerosols from high-resolution infrared
nadir spectra. Appl. Opt., 49, 3713–22.
Coakley, J. A. and Bretherton, F. P. (1982). Cloud cover from high-resolution scanner
data: Detecting and allowing for partially filled fields of view. J. Geophys. Res., 87,
4917–32.
Coakley, J. A., Jr., Bernstein, R. L., and Durkee, P. A. (1987). Effect of ship-stack effluents
on cloud reflectivity. Science, 237, 1020–2.
Coffeen, D. L. (1979). Polarization and scattering characteristics in the atmospheres of
Earth, Venus, and Jupiter. J. Opt. Soc. Am., 69, 1051–64.
Coleman, R. F. and Liou, K. N. (1981). Light scattering by hexagonal ice crystals. J. Atmos.
Sci., 38, 1260–71.
Comstock, J. M., Ackerman, T. P., and Mace, G. G. (2002). Ground-based lidar and radar
remote sensing of tropical cirrus clouds at Nauru Island: Cloud statistics and radiative
impacts. J. Geophys. Res., 107, 4714.
Conrath, B. J., Hanel, R. A., Kunde, V. G., and Prabhakara, C. (1970). The infrared
interferometer experiment on Nimbus 3. J. Geophys. Res., 75, 5831–57.
Corti, T., Luo, B. P., Fu, Q., Vömel, H., and Peter, T. (2006). The impact of cirrus clouds
on tropical troposphere-to-stratosphere transport. Atmos. Chem. Phys., 6, 2539–47.
Coulson, K. L. (1988). Polarization and Intensity of Light in the Atmosphere. Hampton,
VA: A. Deepak Publishing.
Courant, R., Friedrichs, K., and Lewy, H. (1928). On the partial difference equations
of mathematical physics. In AEC Research and Development Report, NYO-7689.
References 397
New York: AEC Computing and Applied Mathematics Center, Courant Institute of
Mathematical Sciences, pp. 1–76.
Cowley, L. and Schroeder, M. (2009). HaloSim3 ray tracing simulator. http://www.atoptics
.co.uk/halosim.htm.
Cox, C. and Munk, W. (1954). Measurement of the roughness of the sea surface from
photographs of the Sun’s glitter. J. Opt. Soc. Am., 44, 838–50.
Cox, S. K. and Griffith, K. T. (1979). Estimates of radiative divergence during phase III
of the GARP Atlantic tropical experiment. Part I: Methodology. J. Atmos. Sci., 36,
576–85.
Cross, J. D. (1968). Study of the surface of ice with a scanning electron microscope.
In Proceedings of the International Symposium on Physics of Ice, ed. N. Riehl, B.
Bullemer and H. Engelhardt. Plattsburgh, NY: Plenum Press, pp. 81–94.
Cross, J. D. (1969). Scanning electron microscopy of evaporating ice. Science, 164, 174–5.
Curtis, A. R. (1952). Contribution to a discussion of “A statistical model for water vapor
absorption,” by R. M. Goody. Quart. J. Roy. Meteorol. Soc., 78, 638–40.
Curtis, D. B., Rajaram, B., Toon, O. B., and Tolbert, M. A. (2005). Measurement of the
temperature-dependent optical constants of water ice in the 15–200 µm range. Appl.
Opt., 44, 4102–18.
d’Almeida, G. A., Koepke, P., and Shettle, E. P. (1991). Atmospheric Aerosols:
Global Climatology and Radiative Characteristics. Hampton, VA: A. Deepak Pub-
lishing.
Danielsen, E. F. (1982). A dehydration mechanism for the stratosphere. Geophys. Res. Lett.,
9, 605–8.
Danielsen, E. F. (1993). In situ evidence of rapid, vertical, irreversible transport of lower
tropospheric air into the lower tropical stratosphere by convective cloud turrets and by
larger-scale upwelling in tropical cyclones. J. Geophys. Res., 98, 8665–81.
Davy, J. G. and Branton, D. (1970). Subliming ice surfaces: Freeze-etch electron
microscopy. Science, 168, 1216–18.
Deardorff, J. W. (1972). Theoretical expression for the counter gradient vertical heat flux.
J. Geophys. Res., 77, 5900–4.
Debye, P. (1909). Der Lichtdruck auf Kugeln von beliebigem Material. Ann. Physik., 335,
57–136.
Debye, P. (1915). Zerstreuung von Röntgenstrahlen. Ann. Phys., 351, 809–23.
Del Genio, A. D., Kovari, W., Yao, M.-S., and Jonas, J. (2005). Cumulus microphysics and
climate sensitivity. J. Climate, 18, 2376–87.
Descartes, R. (1637). Discours de la Méthode. Leiden (Discourse on Method, 3rd edn.
(1998), trans. D. A. Cress. Indianapolis: Hackett Publishing Company).
Deschamps, P.-Y., Breon, F.-M., Leroy, M., et al. (1994). The POLDER mission: Instrument
characteristics and scientific objectives. IEEE Trans. Geosci. Remote Sens., 32, 598–
615.
Dessler, A .E. (1998). A reexamination of the “stratospheric fountain” hypothesis. Geophys.
Res. Lett., 25, 4165–8.
Detwiler, A. and Ramaswamy, V. (1990) Radiative heating profiles in simple cirrus cloud
systems. J. Atmos. Sci., 47, 2167–76.
DeVoe, H. (1964). Optical properties of molecular aggregates. I: Classical model of elec-
tronic absorption and refraction. J. Chem. Phys., 41, 393–400.
Dickinson, R. E. (1977). Cloud radiative feedbacks and parameterizations in 3-D climate
models. Trans., Am. Geophys. Union, 58, 796.
Doi, T. (1832). Sekka Zusetsu (Illustrations of Snow Blossoms). [Reproduced in Nihon
Kagaku Koten Zenshu series, ed. Saigusa Hiroto, 6, Tokyo 1946.]
398 References
Doi, T. (1839). Zoku Sekka Zusetsu (Supplement to Illustrations of Snow Blossoms).

[Reproduced in Nihon Kagaku Koten Zenshu series, ed. Saigusa Hiroto, 6, Tokyo
1946.]
Doicu, A., Wriedt, T., and Eremin, Y. A. (2006). Light Scattering by Systems of Particles –
Null-Field Method with Discrete Sources: Theory and Programs. Berlin, Heidelberg:
Springer-Verlag.
Dominé, F., Lauzier, T., Cabanes, A., et al. (2003). Snow metamorphism as revealed by
scanning electron microscopy. Micros. Res. Tech., 62, 33–48.
Donner, L., Seman, C. J., Hemler, R. S., et al. (2001). A cumulus parameterization including
mass fluxes, convective vertical velocities, and mesoscale effects: Thermodynamic and
hydrological aspects in a general circulation model. J. Climate, 14, 3444–63.
Dorman, J. L. and Sellers, P. J. (1989). A global climatology of albedo, roughness length
and stomatal resistance for atmospheric general circulation models as represented by
the Simple Biosphere model (SiB). J. Appl. Meteorol., 28, 833–55.
Draine, B. T. (1988). The discrete-dipole approximation and its application to interstellar
graphite grains. Astrophys. J., 333, 848–72.
Draine, B. T. and Flatau, P. J. (1994). Discrete-dipole approximation for scattering calcu-
lations. J. Opt. Soc. Am. A, 11, 1491–9.
Duda, D. P., Minnis, P., Khlopenkov, K., Chee, T. L., and Boeke, R. (2013). Estimation of
2006 northern hemisphere contrail coverage using MODIS data. Geophys. Res. Lett.,
40, 612–7.
Dugin, V. P. and Mirumyants, S. O. (1976). The light scattering matrices of artificial
crystalline clouds. Izv., Acad. Sci., USSR, Atmos. Oceanic Phys., 12, 988–91.
Dupertuis, M. A., Proctor, M., and Acklin, B. (1994). Generalization of complex Snell–
Descartes and Fresnel laws. J. Opt. Soc. Am. A, 11, 1159–66.
Ebert, E. E. and Curry, J. A. (1992). A parameterization of ice cloud optical properties for
climate models. J. Geophys. Res., 97, 3831–6.
Eddington, A. S. (1916). On the radiative equilibrium of the stars. Mon. Not. Roy. Astron.
Soc., 77, 16–35.
Edwards, J. M., Havemann, S., Thelen, J.-C., and Baran, A. J. (2007). A new parametrization
for the radiative properties of ice crystals: Comparison with existing schemes and
impact in a GCM. Atmos. Res., 83, 19–35.
Elsasser, W. M. (1938). Mean absorption and equivalent absorption coefficient of a band
spectrum. Phys. Rev., 54, 126–9.
Elsasser, W. M. (1942). Heat Transfer by Infrared Radiation in the Atmosphere. Harvard
Meteorological Studies, 6. Cambridge, MA: Harvard University Press.
Findeisen, W. (1938). Die kolloidmeteorologischen Vorgänge bei der Niederschlagsbil-
dung. Meteorol. Zeitschr., 55, 121–33.
Flammer, C. (1957). Spheroidal Wave Functions. Stanford: Stanford University Press.
Fletcher, N. H. (1962). The Physics of Rainclouds. New York: Cambridge University Press.
Foot, J. S. (1988). Some observations of the optical properties of clouds. II: Cirrus. Quart.
J. Roy. Meteorol. Soc., 114, 145–64.
Forster, P. M. and Shine, K. P. (2002). Assessing the climate impact of trends in stratospheric
water vapor. Geophys. Res. Lett., 29, 1086.
Forster, P. M., Ramaswamy, V., Artaxo, P., et al. (2007). Changes in atmospheric con-
stituents and in radiative forcing. In Climate Change 2007: The Physical Science
Basis. Contribution of Working Group I to the Fourth Assessment Report of the Inter-
governmental Panel on Climate Change, ed. S. Solomon, D. Qin, M. Manning, Z.
Chen, M. Marquis, K. Averyt, M. M. B. Tignor, and H. L. Miller. Cambridge, UK:
Cambridge University Press, pp. 129–234.
References 399
Francis, P. N., Jones, A., Saunders, R. W., et al. (1994). An observational and theoretical
study of the radiative properties of cirrus: Some results from ICE’89. Quart. J. Roy.
Meteorol. Soc., 120, 809–48.
Fraunhofer, J. (1821). Neue Modifikation des Lichtes durch gegenseitige Einwirkung und
Beugung der Strahlen, und Gesetze derselben (New modification of light by the
mutual influence and the diffraction of rays, and the laws thereof). Denkschriften
der Königlichen Akademie der Wissenschaften zu München (Memoirs of the Royal
Academy of Science in Munich), 8, 1–76.
Freeman, K. P. and Liou, K. N. (1979). Climatic effects of cirrus clouds. Adv. Geophys.,
21, 231–87.
Fresnel, A. J. (1819). Mémoire sur la diffraction de la lumière. In Oeuvres Complétes
d’Augustin Fresnel, ed. H. Senarmont, E. Verdet, and L. Fresnel. Paris: Chrochard,
pp. 247–382.
Freudenthaler, V., Homburg, F., and Jäger, H. (1996). Optical parameters of contrails from
lidar measurements: Linear depolarization. Geophys. Res. Lett., 23, 3715–8.
Friedl, R. R., Baughcum, S. L., Anderson, B. E., et al. (1997). Atmospheric Effects of
Subsonic Aircraft: Interim Assessment Report of the Advanced Subsonic Technology
Program, NASA Reference Publication 1400. 1997–2005.
Friedlander, S. K. (2000). Smoke, Dust, and Haze: Fundamentals of Aerosol Dynamics,
2nd edn. New York: Oxford University Press.
Fritsch, K. (1853). On snowflake forms and temperature of precipitation. Sber. Akad. Wiss.
Wien (Math-Natur. Kl.), 11, 492–504.
Fu, Q. (1996). An accurate parameterization of the solar radiative properties of cirrus clouds
for climate models. J. Climate, 9, 2058–82.
Fu, Q. (2007). A new parameterization of an asymmetry factor of cirrus clouds for climate
models. J. Atmos. Sci., 64, 4140–50.
Fu, Q. (2013). Bottom up in the tropics. Nature Clim. Change, 3, 957–8.
Fu, Q. and Liou, K. N. (1992a). On the correlated k-distribution method for radiative transfer
in non-homogeneous atmospheres. J. Atmos. Sci., 49, 2139–56.
Fu, Q. and Liou, K. N. (1992b). A three-parameter approximation for radiative transfer in
non-homogeneous atmospheres: Application to the O3 9.6-µm band. J. Geophy. Res.,
97, 13051–8.
Fu, Q. and Liou, K. N. (1993). Parameterization of the radiative properties of cirrus clouds.
J. Atmos. Sci., 50, 2008–25.
Fu, Q., Liou, K. N., Cribb, M. C., Charlock, T. P., and Grossman, A. (1997). Multiple
scattering parameterization in thermal infrared radiative transfer. J. Atmos. Sci., 54,
2799–812.
Fu, Q., Yang, P., and Sun, W. B. (1998). An accurate parameterization of the infrared
radiative properties of cirrus clouds for climate models. J. Climate, 11, 2223–
37.
Gagin, A. (1972). The effects of supersaturation on the ice crystal production by natural
aerosols. J. Rech. Atmos., 6, 175–85.
Gans, R. (1925). Strahlungsdiagramme ultramikroskopischer Teilchen. Ann. Phys., 381,
29–38.
Gao, B.-C. and Kaufman, Y. J. (1995). Selection of the 1.375-µm MODIS channel for
remote sensing of cirrus clouds and stratospheric aerosols from space. J. Atmos. Sci.,
52, 4231–7.
Gao, B.-C., Goetz, A. F. H., and Wiscombe, W. J. (1993). Cirrus cloud detection from
airborne imaging spectrometer data using the 1.38 µm water vapor band. Geophys.
Res. Lett., 20, 301–4.
400 References
Gao, B.-C., Kaufman, Y. J., Tanre, D., and Li, R.-R. (2002a). Distinguishing tropospheric
aerosols from thin cirrus clouds for improved aerosol retrievals using the ratio of
1.38-µm and 1.24-µm channels. Geophys. Res. Lett., 29, 1890.
Gao, B.-C., Yang, P., Han, W., Li, R.-R., and Wiscombe, W. J. (2002b). An algorithm using
visible and 1.38-µm channels to retrieve cirrus cloud reflectances from aircraft and
satellite data. IEEE Trans. Geosci. Remote Sens., 40, 1659–68.
Gao, B.-C., Yang, P., and Li, R.-R. (2003). Detection of high clouds in polar regions during
the daytime using the MODIS 1.375-µm channel. IEEE Trans. Geosci. Remote Sens.,
41, 474–81.
Gao, R. S., Fahey, D. W., Popp, P. J., et al. (2006). Measurements of relative humidity in a
persistent contrail. Atmos. Envir., 40, 1590–600.
Gedney, S. D. (1996). An anisotropic perfectly matched layer-absorbing medium for the
truncation of FDTD lattices. IEEE Trans. Antennas Propag., 44, 1630–9.
Gehrels, T. (1974). Planets, Stars and Nebulae Studied with Photopolarimetry. Tucson,
AZ: University of Arizona Press.
Gettelman, A. and Chen, C. (2013). The climate impact of aviation aerosols. Geophys. Res.
Lett., 40, 2785–9.
Gettelman, A., Holton, J. R., and Douglass, A. R. (2000). Simulations of water vapor in the
lower stratosphere and upper troposphere. J. Geophys. Res. 105, 9003–23.
Glaisher, J. (1855). Snow crystals. Rep. Council Brit. Meteorol. Soc., 17. Abridged in Q. J.
Microsc. Sci., 3, 179–85.
Godson, W. L. (1953). The evaluation of infrared radiative fluxes due to atmospheric water
vapor. Quart. J. Roy. Meteorol. Soc., 79, 367–79.
Goody, R. M. (1952). A statistical model for water-vapour absorption. Quart. J. Roy.
Meteorol. Soc., 78, 165–9.
Goody, R. M. and Yung, Y. L. (1989). Atmospheric Radiation: Theoretical Basis, 2nd edn.
New York: Oxford University Press.
Goody, R. M., West, R., Chen, L., and Crisp, D. (1989). The correlated-k method for
radiation calculations in non-homogeneous atmospheres. J. Quant. Spectrosc. Radiat.
Transfer, 42, 539–50.
Gosse, S., Labrie, D., and Chylek, P. (1995). Refractive index of ice in the 1.4–7.8-µm
spectral range. Appl. Opt., 34, 6582–6.
Graf, K., Schumann, U., Mannstein, H., and Mayer, B. (2012). Aviation induced diurnal
north Atlantic cirrus cover cycle. Geophys. Res. Lett., 39, L16804.
Greenberg, J. M., Lind, A. C., Wang, R. T., and Libelo, L. F. (1967). Scattering by non-
spherical systems. In ICES II: Electromagnetic Scattering, ed. R. L. Rowell and R. S.
Stein. New York: Gordon and Breach, pp. 3–54.
Greenler, R. (1980). Rainbows, Halos, and Glories. New York: Cambridge University
Press.
Grenfell, T. C. and Warren, S. G. (1999). Representation of a non-spherical ice particle
by a collection of independent spheres for scattering and absorption of radiation. J.
Geophys. Res., 104, 31697–709.
Grundy, W. M. and Schmitt, B. (1998). The temperature-dependent near-infrared absorption
spectrum of hexagonal H2 O-ice. J. Geophys. Res., 103, 25809–22.
Grynko, Y. and Skhuratov, Y. (2003). Scattering matrix calculated in geometric optics
approximation for semitransparent particles faceted with various shapes. J. Quant.
Spectrosc. Radiat. Transfer, 78, 319–40.
Gu, Y. and Liou, K. N. (1997a): Interactions of radiation, microphysics, and turbulence in
a two dimensional cirrus cloud model. In IRS’96: Current Problems in Atmospheric
Radiation, ed. W. L. Smith and K. Stamnes. Hampton, VA: A. Deepak Publishing,
pp. 238–41.
References 401
Gu, Y. and Liou, K. N. (1997b): Numerical experiments on the interactions of radia-

tion, turbulence, and microphysics in cirrus clouds. In Proceedings of the Ninth
AMS Conference on Atmospheric Radiation, Long Beach, CA, February 2–7,
pp. 97–101.
Gu, Y. and Liou, K. N. (2000). Interactions of radiation, microphysics, and turbulence in
the evolution of cirrus clouds. J. Atmos. Sci. 57, 2463–79.
Gu, Y. and Liou, K. N. (2001). Radiation parameterization for three-dimensional inhomo-
geneous cirrus clouds: Application to climate models. J. Climate, 14, 2443–57.
Gu, Y. and Liou, K. N. (2006). Cirrus cloud horizontal and vertical inhomogeneity effects
in a GCM. Meteorol. Atmos. Phys., 91, 223–35.
Gu, Y., Farrara, J., Liou, K. N., and Mechoso, C. R. (2003). Parameterization of cloud–
radiation processes in the UCLA general circulation model. J. Climate, 16, 3357–
70.
Gu, Y., Liou, K. N., Xue, Y., et al. (2006). Climatic effects of different aerosol types in China
simulated by the UCLA general circulation model. J. Geophys. Res., 111, D15201.
Gu, Y., Liou, K. N., Chen, W., and Liao, H. (2010). Direct climate effect of black carbon
in China and its impact on dust storms. J. Geophys. Res., 115, D00K14.
Gu, Y., Liou, K. N., Ou, S. C. S., and Fovell, R. (2011). Cirrus cloud simulations using
WRF with improved radiation parameterization and increased vertical resolution. J.
Geophys. Res., 116, D06119.
Gu, Y., Liou, K. N., Jiang, J. H., Su, H., and Liu, X. (2012). Dust aerosol impact on North
Africa climate: A GCM investigation of aerosol–cloud–radiation interactions using
A-Train satellite data. Atmos. Chem. Phys., 12, 1667–79.
Guettard, M. (1762). Observations météorologiques, faites à Varsovie pendant les années
1760, 1761 & 1762. Mem. Acad. Sci. Paris, 402–30.
Gultepe, I. and Starr, D. O’C. (1995). Dynamical structure and turbulence in cirrus clouds:
Aircraft observations during FIRE. J. Atmos. Sci., 52, 4159–82.
Hadley, O. L., Corrigan, C. E., Kirchstetter, T. W., Cliff, S. S., and Ramanathan, V. (2010).
Measured black carbon deposition on the Sierra Nevada snow pack and implication
for snow pack retreat. Atmos. Chem. Phys., 10, 7505–13.
Hallett, J. and Mason, B. J. (1958). The influence of temperature and supersaturation on
the habit of ice crystals grown from the vapour. Proc. Roy. Soc., A27, 440–53.
Hammer, C. U., Clausen, H. B., Dansgaard, W., et al. (1985). Continuous impurity analysis
along the Dye 3 deep core. Geophys. Monogr. Ser., 33, 90–4.
Hansen, J. E. (1971). Multiple scattering of polarized light in planetary atmospheres.
Part II: Sunlight reflected by terrestrial water clouds. J. Atmos. Sci., 28, 1400–26.
Hansen, J. E. and Hovenier, J. W. (1974). Interpretation of the polarization of Venus. J.
Atmos. Sci., 31, 1137–60.
Hansen, J. E. and Pollack, J. B. (1970). Near-infrared light scattering by terrestrial clouds.
J. Atmos. Sci., 27, 265–81.
Hansen, J. E. and Travis, L. D. (1974). Light scattering in planetary atmospheres. Space
Sci. Rev., 16, 527–610.
Harrison, E. F., Minnis, P., Barkstrom, B. R., et al. (1990). Seasonal variation of cloud
radiative forcing derived from the Earth Radiation Budget Experiment. J. Geophys.
Res., 95, 18687–703.
Hartmann, D. L. (1994). Global Physical Climatology. San Diego: Academic Press.
Hartmann, D. L., Ramanathan, V., Berroir, A., and Hunt, G. E. (1986). Earth radiation
budget data and climate research. Rev. Geophys., 24, 439–68.
Hartmann, D. L., Holton, J. R., and Fu, Q. (2001). The heat balance of the tropical
tropopause, cirrus, and stratospheric dehydration. Geophys. Res. Lett., 10, 1969–
72.
402 References
Heidinger, A. K., Li, Y., Baum, B. A., et al. (2015). Retrieval of cirrus cloud optical depth
under day and night conditions from MODIS Collection 6 cloud property data. Remote
Sensing, 7, 7257–71.
Hellman, G. (1893). Schneekristalle. Berlin: J. Mückenberger, p. 66.
Henyey, L. G. and Greenstein, J. L. (1941). Diffuse radiation in the galaxy. Astrophys. J.,
93, 70–83.
Hess, M. and Wiegner, M. (1994). COP: A data library of optical properties of hexagonal
ice crystals. Appl. Opt., 33, 7740–6.
Hesse, E., Macke, A., Havemann, S., et al. (2012). Modelling diffraction by facetted
particles. J. Quant. Spectrosc. Radiat. Transfer, 113, 342–7.
Heymsfield, A. J. (1972). Ice crystal terminal velocities. J. Atmos. Sci., 29, 1348–57.
Heymsfield, A. J. (1975). Cirrus uncinus generating cells and the evolution of cirriform
clouds. Part I: Aircraft observations of the growth of the ice phase. J. Atmos. Sci., 32,
799–808.
Heymsfield, A. J. (1997). Precipitation development in stratiform ice clouds: A microphys-
ical and dynamical study. J. Atmos. Sci., 34, 367–81.
Heymsfield, A. J. and Iaquinta, J. (2000). Cirrus crystal terminal velocities. J. Atmos. Sci.,
57, 916–38.
Heymsfield, A. J. and McFarquhar, G. M. (1996). High albedos of cirrus in the tropical
Pacific warm pool: Microphysical interpretations from CEPEX and from Kwajalein,
Marshall Islands. J. Atmos. Sci., 53, 2424–51.
Heymsfield, A. J. and Platt, C. M. R. (1984). A parameterization of the particle-size
spectrum of ice clouds in terms of the ambient temperature and the ice water content.
J. Atmos. Sci., 41, 846–55.
Heymsfield, A. J. and Sabin, R. M. (1989). Cirrus crystal nucleation by homogeneous
freezing of solution droplets. J. Atmos. Sci., 46, 2252–64.
Heymsfield, A. J., Miller, K. M., and Spinhirne, J. D. (1990). The 27–28 October 1986
FIRE IFO cirrus case study: Cloud microstructure. Mon. Wea. Rev., 118, 2313–28.
Heymsfield, A. J., Lewis, S., Bansemer, A., et al. (2002). A general approach for deriving
the properties of cirrus and stratiform ice cloud particles. J. Atmos. Sci., 59, 3–29.
Heymsfield, A. J., Schmitt, C., Bansemer, A., et al. (2006). Effective radius of ice cloud
particle populations derived from aircraft probes. J. Atmos. Ocean. Technol., 23, 361–
80.
Ho, P.-Y. and Needham, J. (1959). Ancient Chinese observations of solar haloes and
parhelia. Weather, 14, 124–34.
Hobbs, P. V. (1974). Ice Physics. Oxford: Clarendon Press.
Hobbs, P. V., Harrison, H., and Robinson, E. (1974). Atmospheric effects of pollutants.
Science, 183, 909–15.
Hogan, R. J. (2010). The full-spectrum correlated-k method for longwave atmospheric
radiative transfer using an effective Planck function. J. Atmos. Sci., 67, 2086–100.
Holland, R. (1977). THREDE: A free-field EMP coupling and scattering code. IEEE Trans.
Nucl. Sci., 24, 2416–21.
Holton, J. R. and Gettelman, A. (2001). Horizontal transport and the dehydration of the
stratosphere. Geophys. Res. Lett., 28, 2799–802.
Holton, J. R., Haynes, P. H., McIntyre, M. E., et al. (1995). Stratosphere–troposphere
exchange. Rev. Geophys., 33, 403–39.
Houze, R. A., Jr., Geotis, S. G., Marks, F. D., Jr., and West, A. K. (1981). Winter monsoon
convection in the vicinity of North Borneo. Part I: Structure and time variation of the
clouds and precipitation. Mon. Wea. Rev., 109, 1595–614.
References 403
Hsu, N. C., Tsay, S. C., King, M. D., and Herman, J. R. (2004). Aerosol properties over
bright-reflecting source regions. IEEE Trans. Geosci. Remote Sens., 42, 557–69.
Hu, Y. (2007). Depolarization ratio–effective lidar ratio relation: Theoretical basis for space
lidar cloud phase discrimination. Geophys. Res. Lett., 34, L11812.
Hu, Y., Vaughan, M., Liu, Z., et al. (2007). The depolarization–attenuated backscatter
relation: CALIPSO lidar measurements vs. theory. Opt. Express, 15, 5327–32.
Hu, Y., Winker, D., Vaughan, M., et al. (2009). CALIPSO/CALIOP Cloud Phase Discrim-
ination Algorithm. J. Atmos. Oceanic Technol., 26, 2293–309.
Huang, H.-L., Yang, P., Wei, H., et al. (2004). Inference of ice cloud properties from
high spectral resolution infrared observations. IEEE Trans. Geosci. Remote. Sens., 42,
842–53.
Hudson, J. G. (1983). Effects of CCN concentrations on stratus clouds. J. Atmos. Sci., 40,
480–6.
Huffman, P. J. (1970). Polarization of light scattered by ice crystals. J. Atmos. Sci., 27,
1207–8.
Huffman, P. J. (1973). Supersaturation spectra of AgI and natural ice nuclei. J. Appl.
Meteorol., 12, 1080–2.
Huffman, P. J. and Thursby, W. R., Jr. (1969). Light scattering by ice crystals. J. Atmos.
Sci., 26, 1073–7.
Hutchison, K. D. and Choe, N. J. (1996). Application of 1.38-µm imagery for thin cirrus
detection in daytime imagery collected over land surfaces. Int. J. Remote Sens., 17,
3325–42.
Hutchison, K. D., Wong, E., and Ou, S. C. S. (2006). Cloud base heights retrieved during
night-time conditions with MODIS data. Int. J. Remote Sens., 27, 2847–62.
Huygens, C. (1662). Oeuvres complètes de Christiaan Huygens. Vol. XVII. The Hague:
Martinus Nijhoff, Société Hollandaise des Sciences.
Iaquinta, J., Isaka, H., and Personne, P. (1995). Scattering phase function of bullet rosette
ice crystals. J. Atmos. Sci., 52, 1401–13.
IPCC Report (1999). Aviation and the Global Atmosphere: A Special Report of Working
Groups I and III of the Intergovernmental Panel on Climate Change, ed. J. E. Penner,
D. H. Lister, D. J. Griggs, D. J. Dokken, and M. McFarland. Cambridge: Cambridge
University Press.
IPCC Report (2007). Climate Change 2007: Synthesis Report. Contribution of Working
Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel
on Climate Change, ed. Core Writing Team, R. K. Pachauri, and A. Reisinger. Geneva:
IPCC, 104 pp.
IPCC Report (2013). Climate Change 2013: The Physical Science Basis, Contribution of
Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on
Climate Change, ed. T. F., Stocker, D. Qin, G.-K. Plattner, M. Tignor, S. K. Allen,
J. Boschung, A. Nauels, Y. Xia, V. Bex, and P. M. Midgley. Cambridge: Cambridge
University Press, 1535 pp.
Irvine, W. M. and Pollack, J. B. (1968). Infrared optical properties of water and ice spheres.
Icarus, 8, 324–60.
Ishimoto, H., Masuda, K., Mano, Y., Orikasa, N., and Uchiyama, A. (2012). Irregularly
shaped ice aggregates in optical modeling of convectively generated ice clouds. J.
Iskander, M. F., Lakhtakia, A., and Durney, C. H. (1983). A new procedure for improving
the solution stability and extending the frequency range of the EBCM. IEEE Trans.
Antennas Propag. 31, 317–24.
404 References
Iwabuchi, H. and Yang, P. (2011). Temperature dependence of ice optical constants: Impli-
cations for simulating the single-scattering properties of cold ice clouds. J. Quant.
Spectrosc. Radiat. Transfer, 112, 2520–5.
Iwabuchi, H., Yang, P., Liou, K. N., and Minnis, P. (2012). Physical and optical properties
of persistent contrails: Climatology and interpretation. J. Geophys. Res., 117, D06215.
Iwabuchi, H., Yamada, S., Katagiri, S., Yang, P., and Okamoto, H. (2014). Radiative and
microphysical properties of cirrus cloud inferred from infrared measurements made
by the moderate resolution imaging spectroradiometer (MODIS). Part I: Retrieval
method. J. App. Meteorol. Clim., 53, 1297–316.
Jackson, J. D. (1975). Classical Electrodynamics. New York: John Wiley & Sons.
Jacobowitz, H. (1971). A method for computing the transfer of solar radiation through
clouds of hexagonal ice crystals. J. Quant. Spectrosc. Radiat. Transfer, 11, 691–5.
Jacobson, M. Z. (2005). A refined method of parameterizing absorption coefficients among
multiple gases simultaneously from line-by-line data. J. Atmos. Sci., 62, 506–17.
Jacobson, M. Z., Wilkerson, J. T., Naiman, A. D., and Lele, S. K. (2013). The effects
of aircraft on climate and pollution. Part II: 20-year impacts of exhaust from all
commercial aircraft worldwide treated individually at the subgrid scale. Faraday
Discuss., 165, 369–82.
Jayaweera, K. O. L. F. and Mason, B. J. (1965). The behavior of freely falling cylinders
and cones in a viscous fluid. J. Fluid Mech., 22, 709–20.
Jenk, T. M., Szidat, S., Schwikowski, M., et al. (2006). Radiocarbon analysis in an Alpine
ice core: Record of anthropogenic and biogenic contributions to carbonaceous aerosols
in the past (1650–1940). Atmos. Chem. Phys., 6, 5381–90.
Jensen, E. J., Toon, O. B., Westphal, D. L., Kinne, S., and Heymsfield, A. J. (1994a).
Microphysical modeling of cirrus: 1. Comparison with 1986 FIRE IFO measurements.
J. Geophys. Res., 99, 10421–42.
Jensen, E. J., Toon, O. B., Westphal, D. L., Kinne, S., and Heymsfield, A. J. (1994b).
Microphysical modeling of cirrus: 2. Sensitivity studies. J. Geophys. Res., 99, 10443–
54.
Jensen, E. J., Toon, O. B., Selkirk, H. B., Spinhirne, J. D., and Schoeberl, M. R. (1996a). On
the formation and persistence of subvisible cirrus clouds near the tropical tropopause.
J. Geophys. Res., 101, 21361–75.
Jensen, E. J., Toon, O. B., Pfister, L., and Selkirk, H. B. (1996b). Dehydration of the
upper troposphere and lower stratosphere by subvisible cirrus clouds near the tropical
tropopause. Geophys. Res. Lett., 23, 825–8.
Johnson, B. R. (1988). Invariant imbedding T matrix approach to electromagnetic scattering.
Appl. Opt., 27, 4861–73.
Johnson, B. R. and Secrest, D. (1968). Quantum-mechanical calculations of the inelastic
cross sections for rotational excitation of para and ortho H2 upon collision with He. J.
Chem. Phys., 48, 4682–93.
Joos, H., Spichtinger, P., and Lohmann, U. (2010). Influence of a future climate on the
microphysical and optical properties of orographic cirrus clouds in ECHAM5. J.
Geophys. Res., 115, D19129.
Joseph, J. H., Wiscombe, W. J., and Weinman, J. A. (1976). The delta-Eddington approxi-
mation for radiative flux transfer. J. Atmos. Sci., 33, 2453–9.
Jourdan, O., Oshchepkov, S., Shcherbakov, V., Gayet, J.-F., and Isaka, H. (2003). Assess-
ment of cloud optical parameters in the solar region: Retrievals from airborne mea-
surements of scattering phase functions. J. Geophys. Res., 108, 4572–84.
Kahn, B. H., Liou, K. N., Lee S.-Y., et al. (2005). Nighttime cirrus detection using Atmo-
spheric Infrared Sounder window channels and total column water vapor. J. Geophy.
Res., 110, D07203.
References 405
Kahn, B. H., Liang, C. K., Eldering, A., et al. (2008). Tropical thin cirrus and relative
humidity observed by the Atmospheric Infrared Sounder. Atmos. Chem. Phys., 8,
1501–18.
Kahn, B. H., Irion, F. W., Dang, V. T., et al. (2014). The Atmospheric Infrared Sounder
version 6 cloud products. Atmos. Chem. Phys., 14, 399–426.
Kahnert, M. (2003). Numerical methods in electromagnetic scattering theory. J. Quant.
Spectrosc. Radiat. Transfer, 79–80, 775–824.
Kang, S., Zhang, Y., Zhang, Y., et al. (2010). Variability of atmospheric dust loading over
the central Tibetan Plateau based on ice core glaciochemistry. Atmos. Environ., 44,
2980–9.
Kattawar, G. W., Yang, P., You, Y., et al. (2016). Polarization of light in the atmosphere
and ocean. In Light Scattering Reviews 10, ed. A. A. Kokhanovsky, pp. 3–29. Berlin,
Heidelberg: Springer-Verlag.
Katz, D. S., Thiele, E. T., and Taflove, A. (1994). Validation and extension to three dimen-
sions of the Berenger PML absorbing boundary condition for FD-TD meshes. IEEE
Microw. Guided W. Lett., 4, 268–70.
Kemp, J. C., Wolstencroft, R. D., and Swedlund, J. B. (1971). Circular polarization: Jupiter
and other planets. Nature, 232, 165–8.
Kepler, J. (1611). Strena seu de nive sexangula, ed. G. Tampach, Frankfurt (English trans-
lation by C. Hardie, The Six-Cornered Snowflake (2014). Oxford: Oxford University
Press).
Kerker, M. (1969). The Scattering of Light and Other Electromagnetic Radiation. New
York: Academic Press.
Khvorostyanov, V. I. and Sassen, K. (1998a). Cirrus cloud simulation using explicit micro-
physics and radiation. Part I: Model description. J. Atmos. Sci., 55, 1808–21.
Khvorostyanov, V. I. and Sassen, K. (1998b). Cirrus cloud simulation using explicit micro-
physics and radiation. Part II: Microphysics, vapor and ice mass budgets, and optical
and radiative properties. J. Atmos. Sci., 55, 1822–45.
Kiehl, J. T. and Ramanathan, V. (1990). Comparison of cloud forcing derived from the
Earth Radiation Budget Experiment with that simulated by the NCAR community
climate model. J. Geophys. Res., 95, 11679–98.
King, M. D., Menzel, W. P., Grant, P. S., et al. (1996). Airborne scanning spectrometer
for remote sensing of cloud, aerosol, water vapor, and surface properties. J. Atmos.
Oceanic Technol., 13, 777–94.
King, M. D., Tsay, S. C., Platnick, S. E., Wang, M., and Liou, K. N. (1997). Cloud
retrieval algorithms for MODIS: Optical thickness, effective particle radius, and
thermodynamic phase. MODIS Algorithm Theoretical Basis Document ATBDMOD-
05. Greenbelt, MD: NASA Goddard Space Flight Center, 1997-12-23.
King, M. D., Menzel, W. P., Kaufman, Y. J., et al. (2003). Cloud and aerosol properties,
precipitable water, and profiles of temperature and water vapor from MODIS. IEEE
Trans. Geosci. Remote Sens., 41, 442–58.
King, M. D., Platnick, S., Yang, P., et al. (2004). Remote sensing of liquid water and ice
cloud optical thickness and effective radius in the Arctic: Application of airborne
multispectral MAS data. J. Atmos. Oceanic Technol., 21, 857–75.
King, M. D., Platnick, S., Menzel, W. P., Ackerman, S. A., and Hubanks, P. A. (2013).
Spatial and temporal distribution of clouds observed by MODIS onboard the Terra
and Aqua Satellites. IEEE Trans. Geosci. Remote Sens., 51, 3826–52.
Kinne, S. and Liou, K. N. (1989). The effects of the nonsphericity and size distribution of
ice crystals on the radiative properties of cirrus clouds. Atmos. Res., 24, 273–84.
Kobayashi, T. (1961). The growth of snow crystals at low supersaturations. Phil. Mag., 6,
1363–70.
406 References
Kokhanovsky, A. (1999). Optics of Light Scattering Media: Problems and Solutions. Chich-
ester: John Wiley & Sons.
Kokhanovsky, A. A., Platnick, S., and King, M. D. (2011). Remote sensing of terrestrial
clouds from space using backscattering and thermal emission techniques. In The
Remote Sensing of Tropospheric Composition from Space, ed. J. P. Burrows, U. Platt,
and P. Borrell. Berlin, Heidelberg: Springer-Verlag, pp. 231–57.
Komrska, J. (1972). Fraunhofer diffraction at apertures in the form of regular polygons, I.
Opt. Acta., 19, 807–16.
Komrska, J. (1973). Fraunhofer diffraction at apertures in the form of regular polygons, II.
Opt. Acta., 20, 549–63.
Korolev, A. V., Isaac, G. A., and Hallett, J. (1999). Ice particle habits in Arctic clouds.
Geophys. Res. Lett., 26, 1299–302.
Kou, L., Labrie, D., and Chylek, P. (1993). Refractive indices of water and ice in the 0.65-
to 2.5-µm spectral range. Appl. Opt., 32, 3531–40.
Kramers, H. A. (1927). La diffusion de la lumiere par les atomes. Atti. Congr. Intern. Fisica
Como., 2, 545–57.
Kristjansson, J. E., Iversen, T., Kirkevag, A., Seland, O., and Debernard, J. (2005). Response
of the climate system to aerosol direct and indirect forcing: Role of cloud feedbacks.
J. Geophys. Res., 110, D24206.
Kronig, R. d. L. (1926). On the theory of dispersion of X-rays. J. Opt. Soc. Am., 12, 547–57.
Kuhn, P. M. (1970). Airborne observations of contrail effects on the thermal radiation
budget. J. Atmos. Sci., 27, 937–42.
Kuik, F., Stammes, P., and Hovenier, J. W. (1991). Experimental determination of scattering
matrices of water droplets and quartz particles. Appl. Opt., 30, 4872–81.
Kumai, M. (1977). Electron microscope analysis of aerosols in snow and deep ice cores
from Greenland. In Isotopes and Impurities in Snow and Ice. International Association
of Hydrological Sciences Publication No. 118, pp. 341–50.
Kunde, V. G., Conrath, B. J., Hanel, R. A., et al. (1974). The Nimbus 4 infrared spec-
troscopy experiment. 2: Comparison of observed and theoretical radiances from 425–
1450 cm−1 . J. Geophys. Res., 79, 777–84.
Kunz, K. S. and Lee, K.-M. (1978). A three-dimensional finite-difference solution of the
external response of an aircraft to a complex transient EM environment. Part I: The
method and its implementation. IEEE Trans. Electromagn. Compat., EMC-20, 328–
33.
Kunz, K. S. and Simpson, L. (1981). A technique for increasing the resolution of finite-
difference solutions of the Maxwell equation. IEEE Trans. Electromagn. Compat.,
EMC–23, 419–22.
Kurucz, R. L. (1995). The solar irradiance by computation. In Proceedings of the 17th
Annual Review Conference on Atmospheric Transmission Models, ed. G. P. Anderson,
R. H. Picard, and J. H. Chetwynd, PL-TR-95-2060(II), Special Reports, No. 274,
Hanscom AFB, MA: Phillips Laboratory, Directorate of Geophysics, 1995-05-24,
pp. 333–4.
Lacis, A. A. and Oinas, V. (1991). A description of the correlated k distribution method
for modeling nongray gaseous absorption, thermal emission, and multiple scattering
in vertically inhomogeneous atmospheres. J. Geophys. Res., 96, 9027–63.
Lacis, A. A., Wang, W. C., and Hansen, J. E. (1979). Correlated k-distribution method for
radiative transfer in climate models: Application to effect of cirrus clouds on climate,
4th NASA Weather and Climate Program Sci. Rev. 19790012513 Paper-51. Greenbelt,
MD: NASA Goddard Space Flight Center, 1979-01-01.
References 407
Lakhtakia, A. (1992). Strong and weak forms of the method of moments and the coupled
dipole method for scattering of time-harmonic electromagnetic fields. Int. J. Mod.
Phys. C, 3, 583–603.
Langleben, M. P. (1954). The terminal velocity of snowflakes. Quart. J. Roy. Meteorol.
Soc., 80, 174–81.
Laube, M. and Holler, H. (1988). Numerical data and functional relations in science and
technology. Cloud Phys., 4, 1–110.
Lawson, R. P., Baker, B. A., Zmarzly, P., et al. (2006). Microphysical and optical properties
of atmospheric ice crystals at South Pole Station. J. Appl. Meteorol. Clim., 45, 1505–
24.
Lazzi, G. and Gandhi, O. P. (1997). On the optimal design of the PML absorbing boundary
condition for the FDTD code. IEEE Trans. Antennas Propag., 45, 914–16.
Leaitch, W. R., Isaac, G. A., Strapp, J. W., Banic, C. M., and Wiebe, H. A. (1992).
The relationship between cloud droplet number concentrations and anthropogenic
pollution: Observations and climatic implications. J. Geophys. Res., 97, 2463–
74.
Lee, D. S., Fahey, D. W., Forster, P. M., et al. (2009a). Aviation and global climate change
in the 21st century. Atmos. Env., 43, 3520–37.
Lee, J., Yang, P., Dessler, A. E., Gao, B.-C., and Platnick, S. (2009b). Distribution and
radiative forcing of tropical thin cirrus clouds. J. Atmos. Sci., 66, 3721–31.
Li, D.-M. and Shine, K. P. (1995). A 4-dimensional ozone climatology for UGAMP mod-
els. UGAMP Internal Rep., 35. Reading, UK: Center for Global and Atmospheric
Modelling, Department of Meteorology, University of Reading.
Li, J. and Baker, H. W. (2005). A radiation algorithm with correlated-k distribution. Part I:
Local thermal equilibrium. J. Atmos. Sci., 62, 286–309.
Li, J., Huang, H.-L., Liu, C.-Y., et al. (2005a). Retrieval of cloud microphysical properties
from MODIS and AIRS. J. Appl. Meteorol., 44, 1526–43.
Li, J.-L. F., Köhler, M., Farrara, J. D., and Mechoso, C. R. (2002). The impact of stra-
tocumulus cloud radiative properties on surface heat fluxes simulated with a general
circulation model. Mon. Wea. Rev., 130, 1433–41.
Li, J.-L. F., Waliser, D. E., Jiang, J. H., et al. (2005b). Comparisons of EOS MLS cloud ice
measurements with ECMWF analyses and GCM simulations: Initial results. Geophys.
Res. Lett., 32, L18710.
Lilly, D. K. (1988). Cirrus outflow dynamics. J. Atmos. Sci., 45, 1594–605.
Lin, L., Fu, Q., Zhang, H., et al. (2013). Upward mass fluxes in tropical upper troposphere
and lower stratosphere derived from radiative transfer calculations. J. Quant. Spectrosc.
Radiat. Transfer, 117, 114–22.
Lin, R.-F. (1997). A numerical study of the evolution of nocturnal cirrus by a two-
dimensional model with explicit microphysics. Ph.D. dissertation, The Pennsylvania
State University, 198 pp.
Liou, K. N. (1972a). Electromagnetic scattering by arbitrarily oriented ice cylinders. Appl.
Opt., 11, 667–74.
Liou, K. N. (1972b). Light scattering by ice clouds in the visible and infrared: A theoretical
study. J. Atmos. Sci., 29, 524–36.
Liou, K. N. (1972c). On depolarization of visible light from water clouds for a monostatic
lidar. J. Atmos. Sci., 29, 1000–3.
Liou, K. N. (1973a). A numerical experiment on Chandrasekhar’s discrete-ordinate method
for radiative transfer: Applications to cloudy and hazy atmospheres. J. Atmos. Sci.,
30, 1303–26.
408 References
Liou, K. N. (1973b). Transfer of solar irradiance through cirrus cloud layers. J. Geophys.
Res., 78, 1409–18.
Liou, K. N. (1974). Analytic two-stream and four-stream solutions for radiative transfer. J.
Atmos. Sci., 31, 1473–5.
Liou, K. N. (1975a). Theory of the scattering-phase-matrix determination for ice crystals.
J. Opt. Soc. Am. 65, 159–62.
Liou, K. N. (1975b). Applications of the discrete-ordinate method for radiative transfer to
inhomogeneous aerosol atmospheres. J. Geophys. Res., 80, 3434–40.
Liou, K. N. (1977). A complementary theory of light scattering by homogeneous spheres.
Appl. Math. Comp., 3, 331–58.
Liou, K. N. (1986). Influence of cirrus clouds on weather and climate processes: A global
perspective. Mon. Wea. Rev., 114, 1167–99.
Liou, K. N. (1992). Radiation and Cloud Processes in the Atmosphere: Theory, Observation,
and Modeling. New York: Oxford University Press.
Liou, K. N. (2002). An Introduction to Atmospheric Radiation, 2nd edn. San Diego: Aca-
demic Press.
Liou, K. N. and Coleman, R. F. (1980). Light scattering by hexagonal columns and plates.
In Light Scattering by Irregularly Shaped Particles, ed. D. W. Schuerman. New York:
Plenum Publishing Corporation, pp. 207–18.
Liou, K. N. and Gebhart, K. L. (1982). Numerical experiments on the thermal equilibrium
temperature in cirrus cloudy atmospheres. J. Meteorol. Soc. Jpn., 60, 570–82.
Liou, K. N. and Goody, R. M. (1996). CIRRUS: A low cost cloud/climate mission. A
proposal submitted to the National Aeronautics and Space Administration in reference
to AO-96-MTPE-01.
Liou, K. N. and Hansen, J. E. (1971). Intensity and polarization for single scattering by
polydisperse spheres: A comparison of ray optics and Mie theory. J. Atmos. Sci., 28,
995–1004.
Liou, K. N. and Lahore, H. (1974). Laser sensing of cloud composition: A backscattered
depolarization technique. J. Appl. Meteorol., 13, 257–63.
Liou, K. N. and Ou, S. C. S. (1983). Theory of equilibrium temperatures in radiative-
turbulent atmospheres. J. Atmos. Sci., 40, 214–29.
Liou, K. N. and Ou, S. C. S. (1989). The role of cloud microphysical processes in climate:
An assessment from a one-dimensional perspective. J. Geophys. Res., 94, 8599–
607.
Liou, K. N. and Takano, Y. (1994). Light scattering by non-spherical particles: Remote
sensing and climatic implications. Atmos. Res., 31, 271–98.
Liou, K. N. and Takano, Y. (2002). Interpretation of cirrus cloud polarization measurements
from radiative transfer theory. Geophys. Res. Lett., 29, 1313.
Liou, K. N. and Wittman, G. D. (1979). Parameterization of the radiative properties of
clouds. J. Atmos. Sci., 36, 1261–73.
Liou, K. N., Baldwin, R., and Kaser, T. (1976). Preliminary experiments on the scattering
of polarized laser light by ice crystals. J. Atmos. Sci., 33, 553–7.
Liou, K. N., Cai, Q., Barber, P. W., and Hill, S. C. (1983a). Scattering phase matrix
comparison for randomly hexagonal cylinders and spheroids. Appl. Opt., 22, 1684–7.
Liou, K. N., Cai, Q., Pollack, J. B., and Cuzzi, J. N. (1983b). Light scattering by randomly
oriented cubes and parallelepipeds. Appl. Opt., 22, 3001–8.
Liou, K. N., Ou, S. C. S., and Lu, P. J. (1985). Interactive cloud formation and climatic
temperature perturbations. J. Atmos. Sci., 42, 1969–81.
References 409
Liou, K. N., Fu, Q., and Ackerman, T. P. (1988). A simple formulation of the delta-
four-stream approximation for radiative transfer parameterizations. J. Atmos. Sci., 45,
1940–8.
Liou, K. N., Ou, S. C. S., and Koenig, G. (1990). An investigation on the climatic effect
of contrail cirrus. In Air Traffic and the Environment: Background, Tendencies and
Potential Global Atmospheric Effects, ed. U. Schumann. Lecture Notes in Engineering,
60. Berlin, Heidelberg: Springer-Verlag, pp. 154–69.
Liou, K. N., Yang, P., Takano, Y., et al. (1998). On the radiative properties of contrail cirrus.
Liou, K. N., Takano, Y., and Yang, P. (2000). Light scattering and radiative transfer in ice
crystal clouds: Applications to climate research. In Light Scattering by Non-spherical
Particles: Theory, Measurements, and Applications 2000, ed. M. I. Mishchenko, J. W.
Hovenier, and L. D. Travis. San Diego: Academic Press, pp. 417–49.
Liou, K. N., Takano, Y., Yang, P., and Gu, Y. (2002a). Radiative transfer in cirrus clouds:
Light scattering and spectral information. In Cirrus, ed. D. K. Lynch, K. Sassen, D.
O. Starr, and G. Stephens. New York: Oxford University Press, pp. 265–96.
Liou, K. N., Ou, S. C. S., Takano, Y., et al. (2002b). Remote sensing of three-dimensional
inhomogeneous cirrus clouds using satellite and mm-wave cloud radar data. Geophys.
Res. Lett., 29, 1360.
Liou, K. N., Gu, Y., Yue, Q., and McFarguhar, G. (2008). On the correlation between ice
water content and ice crystal size and its application to radiative transfer and general
circulation models. Geophys. Res. Lett., 35, L13805.
Liou, K. N., Takano, Y., and Yang, P. (2010). On geometric optics and surface waves for
light scattering by spheres. J. Quant. Spectrosc. Radiat. Transfer, 111, 1980–9.
Liou, K. N., Takano, Y., and Yang, P. (2011). Light absorption and scattering by aggregates:
Application to black carbon and snow grains. J. Quant. Spectrosc. Radiat. Transfer,
112, 1581–94.
Liou, K. N., Takano, Y., Yue, Q., and Yang, P. (2013). On the radiative forcing of contrail
cirrus contaminated by black carbon. Geophys. Res. Lett., 40, 778–84.
Liou, K. N., Takano, Y., He, C., et al. (2014). Stochastic parameterization for light absorption
by internally mixed BC/dust in snow grains for application to climate models. J.
Geophys. Res. 119, 7616–32.
Liu, C., Bi, L., Panetta, R. L., Yang, P., and Yurkin, M. A. (2012a). Comparison between
the pseudo-spectral time domain method and the discrete dipole approximation for
light scattering simulations. Opt. Express, 20, 16763–76.
Liu, C., Panetta, R. L., and Yang, P. (2012b) Application of the pseudo-spectral time domain
method to compute particle single-scattering properties for size parameters up to 200.
Liu, Q. H. (1997). The PSTD algorithm: A time-domain method requiring only two cells
per wavelength. Microw. Opt. Technol. Lett., 15, 158–65.
Liu, S. and Krueger, S. K. (1998). Numerical simulations of altocumulus using a cloud
resolving model and a mixed layer model. Atmos. Res., 47–48, 461–74.
Liu, Y. and Daum, P. H. (2004). Parameterization of the autoconversion process. Part I:
Analytical formulation of the Kessler-type parameterizations. J. Atmos. Sci., 61, 1539–
48.
Liu, Y., Daum, P. H., and McGraw, R. (2004). An analytical expression for predicting
the critical radius in the autoconversion parameterization. Geophys. Res. Lett., 31,
L06121.
410 References
Livesey, N. J., Read, W. G., Lambert, A., et al. (2007). Earth Observing System (EOS),
Aura Microwave Limb Sounder (MLS). Version 2.2 Level 2 data quality and description
document, JPL D-33509. Pasadena: Jet Propulsion Laboratory, 2007-05-22.
London, J. (1957). A Study of the Atmospheric Heat Balance, New York University College
of Engineering AF19(122)-165, AFCRCTR-57-287.
Lorentz, H. A. (1880). Üeber die Beziehung zwischen der Fortpflanzungsgeschwindigkeit
des Lichtes und der Körperdichte. Ann. Phys., 245, 641–65.
Lorentz, H. A. (1906). The absorption and emission lines of gaseous bodies. In KNAW,
Proceedings, 8 II, 1905–1906. Amsterdam: Royal Netherlands Academy of Arts and
Sciences, pp. 591–611.
Lorenz, L. V. (1880). Üeber die Refractionsconstante. Ann. Phys., 247, 70–103.
Lorenz, L. V. (1890). Lysbevaegelsen i og uden for en af plane Lysbølger belyst Kugle.
Vidensk. Selk. Skr., 6, 1–62.
Lu, J. Q., Yang, P., and Hu, X.-H. (2005). Simulations of light scattering from a biconcave
red blood cell using the finite-difference time-domain method. J. Biomed. Opt., 10,
024022.
Lyot, B. (1929). Recherches sur la polarisation de la lumière des planètes et de quelques
substances terrestres. Ann. Observ. Paris (Meudon), 8, 1–161; available in English as
NASA TT F-187, Washington DC: NASA. 1964-07, pp. 1–145.
Mace, G. G., Heymsfield, A. J., and Poellot, M. R. (2002). On retrieving the microphysical
properties of cirrus clouds using the moments of the millimeter-wavelength Doppler
spectrum. J. Geophys. Res., 107, 4815.
Machta, L. and Carpenter, T. (1971). Trends in high cloudiness at Denver and Salt Lake
City. In Man’s Impact on the Climate, ed. W. H. Matthews, W. W. Kellogg, and G. D.
Robinson. Cambridge, MA: MIT Press.
Macke, A. (1993). Scattering of light by polyhedral ice crystals. Appl. Opt., 32, 2780–8.
Macke, A. (2000). Monte Carlo calculations of light scattering by large particles with
multiple internal inclusions. In Light Scattering by Non-spherical Particles: Theory,
Measurements, and Applications, ed. M. I. Mishchenko, J. W. Hovenier, and L. D.
Travis. San Diego: Academic Press, pp. 309–22.
Macke, A., Mischchenko, M. I., and Cairns, B. (1996). The influence of inclusions on light
scattering by large ice particles. J. Geophys. Res., 101, 23311–16.
Magono, C. and Lee, C. W. (1966). Meteorological classification of natural snow crystals.
J. Fac. Sci. Hokkaido Univ., 2, 321–55.
Manabe, S. (1975). Cloudiness and the radiative, convective equilibrium. In The Changing
Global Environment, ed. S. F. Singer. Dordrecht: Springer Netherlands, pp. 175–6.
Manabe, S. and Wetherald, R. T. (1967). Thermal equilibrium of the atmosphere with a
given distribution of relative humidity. J. Atmos. Sci., 24, 241–59.
Manabe, S. and Wetherald, R. T. (1980). On the distribution of climate change resulting
from an increase in CO2 content of the atmosphere. J. Atmos. Sci., 37, 99–118.
Marchuk, G. I. and Mikhailov, G. A. (1967). The solution of problems of atmospheric
optics by a Monte-Carlo method. Izv., Acad. Sci., USSR, Atmos. Oceanic Phys., 3,
147–55.
Mariotte, E. (1681). De la Nature des Couleurs. Paris: Chez Estienne Michallet.
Martens, F. (1675). Spitzbergische oder froenlandische Reise Beschreibung gethan im Jahr
1671. Hamburg: Schultzens, p. 235.
Martin, P. G. (1974). Interstellar polarization from a medium with changing grain alignment.
Astrophys. J., 187, 461–72.
References 411
Martins, J. V., Tanré, D., Remer, L., et al. (2002). MODIS cloud screening for remote
sensing of aerosols over oceans using spatial variability. Geophys. Res. Lett., 29,
1619.
Mason, B. J. (1971). The Physics of Clouds. Oxford: Clarendon Press.
Matthews, W. H., Kellogg, W. W., and Robinson, G. D. (1971). Man’s Impact on the
Climate. Cambridge, MA: MIT Press.
Mätzler, C. (2006). Microwave dielectric properties of ice. In Thermal Microwave Radia-
tion: Applications for Remote Sensing, ed. C. Mätzler, P. W. Rosenkranz, A. Battaglia,
and J. P. Wigneron. IET Electromagnetic Waves Series, 52. Stevenage, UK: Institution
of Engineering and Technology, pp. 455–62.
Maxwell, J. C. (1865). A dynamical theory of the electromagnetic field, Philos. Trans. R.
Soc. London, 155, 459–512.
Maxwell-Garnett, J. C. (1904). Colours in metal glasses and in metallic films. Philos. Trans.
R. Soc., A, 203, 385–420.
McClatchey, R. A., Fenn, R. W., Selby, J. E. A., Volz, F. E., and Garing, J. S. (1972). Optical
Properties of the Atmosphere, AFCRL Environment Research Paper ERP No. 411, 3rd
edn. Hanscom AFB, MA: Air Force Cambridge Research Laboratories, 1972-08-24.
McFarquhar, G. M. and Heymsfield, A. J. (1997). Parameterization of tropical cirrus ice
crystal size distributions and implications for radiative transfer: Results from CEPEX.
J. Atmos. Sci., 54, 2187–200.
McFarquhar, G. M., Yang, P., Macke, A., and Baran, A. J. (2002). A new parameterization
of single scattering solar radiative properties for tropical anvils using observed ice
crystal size and shape distributions. J. Atmos. Sci., 59, 2458–78.
McFarquhar, G. M., Iacobellis, S., and Sommerville, R. (2003). SCM simulations of tropical
ice clouds using observationally based parameterizations of microphysics. J. Climate,
16, 1643–64.
McMillin, L. M., Crone, L. J., Goldberg, M. D., and Kleespies, T. J. (1995). Atmospheric
transmittance of an absorbing gas, 4. OPTRAN: A computationally fast and accurate
transmittance model for absorbing gases and with variable mixing ratios at variable
viewing angles. Appl. Opt., 34, 6269–74.
Meador, W. E. and Weaver, W. R. (1980). Two-stream approximations to radiative trans-
fer in planetary atmospheres: A unified description of existing methods and a new
improvement. J. Atmos. Sci., 37, 630–43.
Mechoso, C. R., Yu, J.-Y., and Arakawa, A. (2000). A coupled GCM pilgrimage: From
climate catastrophe to ENSO simulations. In General Circulation Model Development:
Past, Present, and Future – Proceedings of a Symposium in Honor of Professor Akio
Arakawa, ed. D. A. Randall. New York: Elsevier, pp. 539–75.
Menzel, D. H. (1966). Selected Papers on the Transfer of Radiation. New York: Dover
Publications.
Menzel, W. P., Frey, R. A., Baum, B. A., and Zhang, H. (2006). Cloud Top Properties
and Cloud Phase, Algorithm Theoretical Basis Document ATBDMOD-04. 2006–10.
Washington: National Aeronautics and Space Administration.
Menzel, W. P., Frey, R. A., Zhang, H., et al. (2008). MODIS global cloud-top pressure and
amount estimation: Algorithm description and results. J. Appl. Meteorol. Clim., 47,
1175–98.
Merewether, D. E., Fisher, R., and Smith, F. W. (1980). On implementing a numeric
Huygen’s source scheme in a finite difference program to illuminate scattering bodies.
IEEE Trans. Nucl. Sci., NS–27, 1829–33.
412 References
Merwin, H. E. (1930). Refractivity of birefringent crystals. In International Critical Tables

of Numerical Data Physics, Chemistry and Technology, Vol. 7, ed. E. W. Washburn,
C. J. West, N. E. Dorsey, and M. D. Ring. New York: McGraw-Hill, pp. 16–33.
Meyers, M. P., DeMott, P. J., and Cotton, W. R. (1992). New primary ice-nucleation
parameterizations in an explicit cloud model. J. Appl. Meteorol., 31, 708–21.
Mie, G. (1908). Beiträge zur Optik trüber Medien, speziell kolloidaler Metallösungen. Ann.
Physik., 330, 377–445.
Miloshevich, L. M. and Heymsfield, A. J. (1997). A balloon-borne continuous cloud particle
replicator for measuring vertical profiles of cloud microphysical properties: Instrument
design, performance, and collection efficiency analysis. J. Atmos. Ocean. Technol., 14,
753–68.
Ming, J., Cachier, H., Xiao, C., et al. (2008). Black carbon record based on a shallow
Himalayan ice core and its climatic implications. Atmos. Chem. Phys., 8, 1343–52.
Ming, J., Xiao, C., Cachier, H., et al. (2009). Black carbon (BC) in the snow of glaciers in
west China and its potential effects on albedos. Atmos. Res., 92, 114–23.
Minnis, P., Heck, P. W., and Young, D. F. (1993). Inference of cirrus cloud properties using
satellite-observed visible and infrared radiances. Part II: Verification of theoretical
cirrus radiative properties. J. Atmos. Sci., 50, 1305–22.
Minnis, P., Smith, W. L., Jr., Garber, D. P., Ayers, J. K., and Doelling, D. R. (1995).
Cloud Properties Derived from GOES-7 for Spring 1994 ARM Intensive Observing
Period Using Version 1.0.0 of ARM Satellite Data Analysis Program. NASA Reference
Publication 1366, 58 pp. Washington: National Aeronautics and Space Administration.
Minnis, P., Young, D. F., Garber, D. P., et al. (1998a). Transformation of contrails into
cirrus during SUCCESS. Geophys. Res. Lett., 25, 1157–60.
Minnis, P., Garber, D. P., Young, D. F., Arduini, R. F., and Takano, Y. (1998b). Parame-
terizations of reflectance and effective emittance for satellite remote sensing of cloud
properties. J. Atmos. Sci., 55, 3313–39.
Minnis, P., Sun-Mack, S., Young, D. F., et al. (2011a). CERES Edition-2 cloud property
retrievals using TRMM VIRS and Terra and Aqua MODIS data. Part I: Algorithms.
IEEE Trans. Geosci. Remote Sens., 49, 4374–400.
Minnis, P., Sun-Mack, S., Chen, Y., et al. (2011b). CERES Edition-2 cloud property
retrievals using TRMM VIRS and Terra and Aqua MODIS data. Part II: Examples of
average results and comparisons with other data. IEEE Trans. Geosci. Remote Sens.,
49, 4401–30.
Minnis, P., Bedka, S. T., Duda, D. P., et al. (2013). Linear contrail and contrail cirrus
properties determined from satellite data. Geophys. Res. Lett., 40, 3220–6.
Mishchenko, M. I. (1990). Extinction of light by randomly-oriented non-spherical grains.
Astrophys. Space Sci., 164, 1–13.
Mishchenko, M. I. (1991). Extinction and polarization of transmitted light by partially
aligned non-spherical grains. Astrophys. J., 367, 561–74.
Mishchenko, M. I. and Macke, A. (1998). Incorporation of physical optics effects and
computation of the Legendre expansion for ray-tracing phase functions involving
δ-function transmission. J. Geophys. Res., 103, 1799–805.
Mishchenko, M. I. and Travis, L. D. (1998). Capabilities and limitations of a current
FORTRAN implementation of the T-matrix method for randomly oriented, rotationally
symmetric scatterers. J. Quant. Spectrosc. Radiat. Transfer, 60, 309–24.
Mishchenko, M. I., Rossow, W. B., Macke, A., and Lacis, A. A. (1996). Sensitivity of cirrus
cloud albedo, bidirectional reflectance and optical thickness retrieval accuracy to ice
particle shape. J. Geophys. Res., 101, 16973–85.
References 413
Mishchenko, M. I., Geogdzhayev, I. V., Cairns, B., Rossow, W. B., and Lacis, A. A. (1999).
Aerosol retrievals over the ocean by use of channels 1 and 2 AVHRR data: Sensitivity
analysis and preliminary results. Appl. Opt., 38, 7325–41.
Mishchenko, M. I., Hovenier, J. W., and Travis, L. D., eds. (2000a). Light Scattering by Non-
spherical Particles: Theory, Measurements, and Applications. San Diego: Academic
Press.
Mishchenko, M. I., Wiscombe, W., Hovenier, J. W., and Travis, L. D. (2000b). Overview of
scattering by non-spherical particles. In Light Scattering by Non-spherical Particles:
Theory, Measurements, and Applications, ed. M. I. Mishchenko, J. W. Hovenier, and
L. D. Travis. San Diego: Academic Press, pp. 29–60.
Mishchenko, M. I., Travis, L. D., and Lacis, A. A. (2002). Scattering, Absorption, and
Emission of Light by Small Particles. Cambridge: Cambridge University Press.
Mishchenko, M. I., Cairns, B., Kopp, G., et al. (2007). Accurate monitoring of terrestrial
aerosols and total solar irradiance: Introducing the Glory mission. Bull. Am. Meteorol.
Soc., 88, 677–91.
Mishchenko, M. I., Zakharova, N. T., Videen, G., Khlebtsov, N. G., and Wriedt, T. (2010a).
Comprehensive T-matrix reference database: A 2007–2009 update. J. Quant. Spec-
trosc. Radiat. Transfer, 111, 650–8.
Mishchenko, M. I., Rosenbush, V. K., Kiselev, N. N., et al. (2010b). Polarimetric Remote
Sensing of Solar System Objects. Kiev, Ukraine: Akademperiodyka.
Mitchell, D. L. and Arnott, W. P. (1994). A model predicting the evolution of ice particle
size spectra and radiative properties of cirrus clouds. Part II: Dependence of absorption
and extinction on ice crystal morphology. J. Atmos. Sci., 51, 817–32.
Mlawer, E. J., Taubman, S. J., Brown, P. D., Iacono, M. J., and Clough, S. A. (1997).
Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k
model for the longwave. J. Geophys. Res., 102, 16663–82.
Mlynczak, M. G., Harries, J. E. Rizzi, R., et al. (2002). Far-infrared: A frontier in remote
sensing of Earth’s climate and energy balance. Proc. SPIE, 4485, 150–8.
Moeng, C.-H. (1986). Large-eddy simulation of a stratus-topped boundary layer. Part I:
Structure and budgets. J. Atmos. Sci., 43, 2886–900.
Möller, F. (1943). Das Strahlungdiagramm. Berlin: Reichsamt für Wetterdienst (Luftwaffe).
Moore, T. G., Blaschak, J. G., Taflove, A., and Kriegsmann, G. A. (1988). Theory and
application of radiation boundary operators. IEEE Trans. Antennas Propag., 36, 1797–
812.
Moorthi, S. and Suarez, M. J. (1992). Relaxed Arakawa–Schubert: A parameterization
of moist convection for general circulation models. Mon. Wea. Rev., 120, 978–
1002.
Mugnai, A. and Wiscombe, W. J. (1986). Scattering from non-spherical Chebyshev parti-
cles. 1: Cross sections, single-scattering albedo, asymmetry factor, and backscattered
fraction. Appl. Opt., 25, 1235–44.
Mugnai, A. and Wiscombe, W. J. (1989). Scattering from non-spherical Chebyshev parti-
cles. 3: Variability in angular scattering patterns. Appl. Opt., 28, 3061–73.
Muinonen, K. (1989). Scattering of light by crystals: A modified Kirchhoff approximation.
Appl. Opt., 28, 3044–50.
Muinonen, K., Lamberg, L., Fast, P., and Lumme, K. (1997). Ray optics regime for Gaussian
random spheres. J. Quant. Spectrosc. Radiat. Transfer, 57, 197–205.
Mur, G. (1981). Absorbing boundary conditions for the finite-difference approximation of
the time-domain electromagnetic-field equations. IEEE Trans. Electromagn. Compat.,
EMC-23, 377–82.
414 References
Nakajima, T. and King, M. D. (1990). Determination of the optical thickness and effective
particle radius of clouds from reflected solar radiation measurements. Part I: Theory.
J. Atmos. Sci., 47, 1878–93.
Nakajima, T. Y. and Nakajima, T. (1995). Wide-area determination of cloud microphysical
properties from NOAA AVHRR measurements for FIRE and ASTEX regions. J.
Atmos. Sci., 52, 4043–59.
Nakajima, T., King, M. D., Spinhirne, J. D., and Radke, L. F. (1991). Determination of the
optical thickness and effective particle radius of clouds from reflected solar radiation
measurements. Part II: Marine stratocumulus observations. J. Atmos. Sci., 48, 728–50.
Nakaya, U. (1954). Snow Crystals: Natural and Artificial. Cambridge: Harvard University
Press.
NASA (National Aeronautics and Space Administration) (2012). A-Train: The Afternoon
Constellation. http://atrain.nasa.gov/ (June 12, 2012).
Needham, J. and Lu, G.-D. (1961). The earliest snow crystal observations. Weather, 16,
319–27.
Newton, I. (1704). Opticks: Or, A Treatise of the Reflexions, Refractions, Inflexions and
Colours of Light. London: Royal Society.
Noel, V., Ledanois, G., Chepfer, H., and Flamant, P. H. (2001). Computation of a single-
scattering matrix for non-spherical particles randomly or horizontally oriented in
space. Appl. Opt., 40, 4365–75.
Nousiainen, T., Muinonen, K., and Räisänen, P. (2003). Scattering of light by large Saharan
dust particles in a modified ray optics approximation. J. Geophys. Res., 108, 4025–
41.
Nussenzveig, H. M. (1992). Diffraction Effects in Semiclassical Scattering. New York:
Cambridge University Press.
Nussenzveig, H. M. (2002). Does the glory have a simple explanation? Opt. Lett., 27,
1379–81.
Nussenzveig, H. M. and Wiscombe, W. J. (1980). Forward optical glory. Opt. Lett., 5,
455–7.
Ono, A. (1969). The shape and riming properties of ice crystals in natural clouds. J. Atmos.
Sci., 26, 138–47.
Ono, A. (1970). Growth mode of ice crystals in natural clouds. J. Atmos. Sci., 27, 649–58.
Oort, A. H. (1983). Global Atmospheric Circulation Statistics, 1958–1973, NOAA
Professional Paper No. 14. Washington, DC: National Oceanic and Atmospheric
Administration.
Oshchepkov, S., Isaka, H., Gayet, J.-F., et al. (2000). Microphysical properties of mixed-
phase & ice clouds retrieved from in situ airborne “polar nephelometer” measurements.
Ou, S. C. S. and Liou, K. N. (1984). A two-dimensional radiation-turbulence climate model.
I: Sensitivity to cirrus radiative properties. J. Atmos. Sci., 41, 2289–309.
Ou, S. C. S. and Liou, K. N. (1995). Ice microphysics and climatic temperature feedback.
Atmos. Res., 35, 127–38.
Ou, S. C. S., Liou, K. N., Gooch, W. M., and Takano, Y. (1993). Remote sensing of cirrus
cloud parameters using advanced very-high-resolution radiometer 3.7- and 10.9-µm
channels. Appl. Opt., 32, 2171–80.
Ou, S. C. S., Liou, K. N., Takano, Y., et al. (1995). Remote sounding of cirrus cloud
optical depths and ice crystal sizes from AVHRR data: Verification using FIRE II IFO
measurements. J. Atmos. Sci., 52, 4143–58.
References 415
Ou, S. C., Liou, K. N., King, M. D., and Tsay, S. C. (1999). Remote sensing of cirrus cloud
parameters based on a 0.63–3.7 µm radiance correlation technique applied to AVHRR
data. Geophys. Res. Lett., 26, 2437–40.
Ou, S. C. S., Liou, K. N., Takano, Y., et al. (2002). Cloud Effective Particle Size and Cloud
Optical Thickness, Visible/Infrared Imager/Radiometer Suite, Algorithm Theoretical
Basis Document #Y2393. 2002–03. Waltham, MA: Raytheon.
Ou, S. C. S., Takano, Y., Liou, K. N., et al. (2003). Remote sensing of cirrus cloud
optical thickness and effective particle size for the national polar-orbiting operational
environmental satellite system visible/infrared imager radiometer suite: Sensitivity
to instrument noise and uncertainties in environmental parameters. Appl. Opt., 42,
7202–14.
Ou, S. C. S., Liou, K. N., Takano, Y., and Slonaker, R. L. (2005a). Remote sensing of
cirrus cloud particle size and optical depth using polarimetric sensor measurements.
J. Atmos. Sci., 62, 4371–83.
Ou, S. C. S., Liou, K. N., Takano, Y., et al. (2005b). Comparison of the University of Califor-
nia at Los Angeles line-by-line equivalent radiative transfer model and the moderate-
resolution transmission model for accuracy assessment of the National Polar-Orbiting
Operational Environmental Satellite System’s Visible–Infrared Imager-Radiometer
Suite cloud algorithms. Appl. Opt., 44, 6274–84.
Ou, S. C. S., Liou, K. N., Hsu, N. C., and Tsay, S. C. (2012). Satellite remote sensing
of dust aerosol indirect effects on cloud formation over Eastern Asia. Int. J. Remote
Sens., 33, 7257–72.
Ou, S. C. S., Kahn, B. H., Liou, K. N., et al. (2013). Retrieval of cirrus cloud properties from
the Atmospheric Infrared Sounder: The k-coefficient approach using cloud-cleared
radiances as input. IEEE Trans. Geosci. Remote Sens., 51, 1010–24.
Paltridge, G. W. and Platt, C. M. R. (1981). Aircraft measurements of solar and infrared
radiation and the microphysics of cirrus cloud. Quart. J. Roy. Meteorol. Soc., 107,
367–80.
Pan, D.-M. and Randall, D. A. (1998). A cumulus parameterization with a prognostic
closure. Quart. J. Roy. Meteorol. Soc., 124, 949–81.
Panetta, R. L., Liu, C., and Yang, P. (2013). A pseudo-spectral time domain method for
light scattering computation. In Light Scattering Reviews 8, ed. A. A. Kokhanovsky,
Berlin, Heidelberg: Springer-Verlag.
Parviainen, P., Bohren, C. F., and Mäkelä, V. (1994). Vertical elliptical coronas caused by
pollen. Appl. Opt., 33, 4548–51.
Pauling, L. (1935). The structure and entropy of ice and of other crystals with some
randomness of atomic arrangement. J. Am. Chem. Soc., 57, 2680–4.
Peltoniemi, J. I., Lumme, K., Muinonen, K., and Irvine, W. M. (1989). Scattering of light
by stochastically rough particles. Appl. Opt., 28, 4088–95.
Penner, J. E., Chen, Y., Wang, M., and Liu, X. (2009). Possible influence of anthropogenic
aerosols on cirrus clouds and anthropogenic forcing. Atmos. Chem. Phys., 9, 879–96.
Perrin, F. (1942). Polarization of light scattered by isotropic opalescent media. J. Chem.
Phys., 10, 415–27.
Petrenko, V. F. and Whitworth, R. W. (1999). The Physics of Ice. Oxford: Oxford University
Press.
Petzold, A., Ström, J., Ohlsson, S., and Schröder, F. P. (1998). Elemental composition and
morphology of ice crystal residual particles in cirrus clouds and contrails. Atmos. Res.,
49, 21–34.
416 References
Petzold, A., Döpelheuer, A., Brock, C. A., and Schröder, F. (1999). In situ observations and
model calculations of black carbon emission by aircraft at cruise altitude. J. Geophys.
Res., 104, 22171–81.
Platnick, S. (2000). Vertical photon transport in cloud remote sensing problems. J. Geophys.
Res., 105, 22919–35.
Platnick, S., King, M. D., Ackerman, S. A., et al. (2003). The MODIS cloud products:
Algorithms and examples from Terra. IEEE Trans. Geosci. Remote Sens., 41, 459–73,
doi:10.1109/TGRS.2002.808301.
Platnick, S., King, M. D., Meyer, K. G., et al. (2014). MODIS Cloud Optical Properties:
User Guide for the Collection 6 Level-2 MOD06/MYD06 Product and Associated
Level-3 Datasets, NASA GSFC MODIS. 2014-09-17. Washington: National Aero-
nautics and Space Administration.
Platt, C. M. R., Abshire, N. L., and McNice, G. T. (1978). Some microphysical properties
of an ice cloud from lidar observation of horizontally orientated crystals. J. Appl.
Meteorol., 17, 1220–4.
Platt, C. M. R., Scott, J. C., and Dilley, A. C. (1987). Remote sounding of high clouds.
Part VI: Optical properties of midlatitude and tropical cirrus. J. Atmos. Sci., 44, 729–
47.
Platt, C. M. R., Young, S. A., Manson, P. J., et al. (1998). The optical properties of equatorial
cirrus from observations in the ARM Pilot Radiation Observation Experiment. J.
Atmos. Sci., 55, 1977–96.
Poincaré, H. (1892). Théorie Mathénmatique de la Lumiere. Vol. II, Chap. XII. Paris: G.
Carré.
Prabhakara, C., Fraser, R. S., Dalu, G., et al. (1988). Thin cirrus clouds: Seasonal distribu-
tion over oceans deduced from Nimbus-4 IRIS. J. Appl. Meteorol., 27, 379–99.
Prabhakara, C., Kratz, D. P., Yoo, J.-M., Dalu, G., and Vernekar, A. (1993). Optically
thin cirrus clouds: Radiative impact on the warm pool. J. Quant. Spectrosc. Radiat.
Transfer, 49, 467–83.
Prospero, J. M., Charlson, R. J., Mohnen, V., et al. (1983). The atmospheric aerosol system:
An overview. Rev. Geophys. Space Phys., 21, 1607–29.
Protat, A., Delanoë, J., May, P. T., et al. (2011). The variability of tropical ice cloud proper-
ties as a function of the large-scale context from ground-based radar-lidar observations
over Darwin, Australia. Atmos. Chem. Phys., 11, 8363–84.
Pruppacher, H. R. and Klett, J. D. (1996). Microphysics of Clouds and Precipitation, 2nd
edn. Dordrecht: Kluwer Academic Publishers.
Purcell, E. M. and Pennypacker, C. R. (1973). Scattering and absorption of light by non-
spherical dielectric grains. Astrophys. J., 186, 705–14.
Radke, L. F., Coakley, J. A., Jr., and King, M. D. (1989). Direct and remote sensing
observations of the effects of ships on clouds. Science, 246, 1146–9.
Rajaram, B., Glandorf, D. L., Curtis, D. B., et al. (2001). Temperature-dependent optical
constants of water ice in the near infrared: New results and critical review of the
available measurements. Appl. Opt., 40, 4449–62.
Ramanathan, V. (1987). The role of earth radiation budget studies in climate and general
circulation research. J. Geophys. Res., 92, 4075–95.
Ramaswamy, V. and Detwiler, A. (1986). Interdependence of radiation and microphysics
in cirrus clouds. J. Atmos. Sci., 43, 2289–301.
Randall, D. A., ed. (2000). General Circulation Model Development: Past, Present, and
Future – Proceedings of a Symposium in Honor of Professor Akio Arakawa. New
York: Elsevier.
References 417
Randel, W. J. and Jensen, E. J. (2013). Physical processes in the tropical tropopause layer
and their roles in a changing climate. Nature Geosci., 6, 169–76.
Rap, A., Forster, P. M., Jones, A., et al. (2010). Parameterization of contrails in the UK
Met Office climate model. J. Geophys. Res., 115, D10205.
Rashed, R. (1993). Géométrie et dioptrique au Xesiècle: Ibn Sahl, al-Quhi et Ibn al-
Haytham. Paris: Les Belles Lettres.
Rayleigh, Lord (1871). On the light from the sky, its polarization and colour. Phil. Mag.,
41, 107–20, 274–9.
Rayleigh, Lord (1918). The dispersal of light by a dielectric cylinder. Phil. Mag., 36,
365–76.
Rayner, N. A., Folland, C. K., Horton, B., and Parker, D. E. (1995). A New Global Sea Ice
and Sea Surface Temperature (GISST) Data Set for 1903–1994 for Forcing Climate
Models. Internal Note 69. Bracknell, UK: Hadley Centre, Met Office, 14 pp.
Reinking, R. F. (1968). Insolation reduction by contrails. Weather, 23, 171–3.
Riédi, J., Marchant, B., Platnick, S., et al. (2010). Cloud thermodynamic phase inferred
from merged POLDER and MODIS data. Atmos. Chem. Phys., 10, 11851–65.
Riikonen, M. (1999). http://www.atoptics.co.uk/halo/spzenith.htm.
Robinson, G. D. (1980). The transport of minor atmospheric constituents between the
troposphere and stratosphere. Quart. J. Roy. Meteorol. Soc., 106, 227–53.
Rockwitz, K.-D. (1989). Scattering properties of horizontally oriented ice crystal columns
in cirrus clouds: Part 1. Appl. Opt., 28, 4103–10.
Roden, J. A. and Gedney, S. (2000). An efficient FDTD implementation of the PML with
CFS in general media. Antennas and Propagation Society International Symposium,
2000. IEEE. Salt Lake City, UT, Vol. 3, pp. 1362–5.
Rogers, R. R. and Yau, M. K. (1989). A Short Course in Cloud Physics. Oxford: Pergamon
Press.
Rolland, P. and Liou, K. N. (2001). Surface variability effects on the remote sensing of thin
cirrus optical and microphysical properties. J. Geophys. Res., 106, 22965–77.
Rolland, P., Liou, K. N., King, M. D., Tsay, S. C., and McFarquhar, G. M. (2000). Remote
sensing of optical and microphysical properties of cirrus clouds using Moderate-
Resolution Imaging Spectroradiometer channels: Methodology and sensitivity to phys-
ical assumptions. J. Geophys. Res., 105, 11721–38.
Roskovensky, J. K. and Liou, K. N. (2003a). Detection of thin cirrus from 1.38 µm/0.65 µm
reflectance ratio combined with 8.6–11 µm brightness temperature difference. Geo-
phys. Res. Lett., 30, 1985.
Roskovensky, J. K. and Liou, K. N. (2003b). Detection of thin cirrus using a combination
of 1.38-µm reflectance and window brightness temperature difference. J. Geophys.
Res., 108, 4570.
Roskovensky, J. K. and Liou, K. N. (2005). Differentiating airborne dust from cirrus clouds
using MODIS data. Geophys. Res. Lett., 32, L12809.
Roskovensky, J. K. and Liou, K. N. (2006). Simultaneous determination of aerosol and
thin cirrus optical depths over oceans from MODIS data: Some case studies. J. Atmos.
Sci., 63, 2307–23.
Roskovensky, J. K., Liou, K. N., Garrett, T. J., and Baumgardner, D. (2004). Simul-
taneous retrieval of aerosol and thin cirrus optical depths using MODIS airborne
simulator data during CRYSTAL-FACE and CLAMS. Geophys. Res. Lett., 31,
L18110.
Rossow, W. B. and Schiffer, R. A. (1999). Advances in understanding clouds from ISCCP.
418 References
Rothman, L. S., Rinsland, C. P., Goldman, A., et al. (1998). The HITRAN molecular spec-
troscopic database and HAWKS (HITRAN atmospheric workstation): 1996 edition.
Rothman, L. S., Barbe, A., Benner, D. C., et al. (2003). The HITRAN molecular spectro-
scopic database: Edition of 2000 including updates through 2001. J. Quant. Spectrosc.
Rothman, L. S., Gordon, I. E., Babikov, Y., et al. (2013). The HITRAN2012 molecular
spectroscopic database. J. Quant. Spectrosc. Radiat. Transfer, 130, 4–50.
Russell, P. B., Pfister, L., and Selkirk, H. B. (1993). The tropical experiment of the
stratosphere-troposphere exchange project (STEP): Science objectives, operations,
and summary findings. J. Geophy. Res., 98, 8563–89.
Sassen, K. (1976). An evaluation of polarization diversity lidar for cloud physics research.
Ph.D. dissertation, University of Wyoming, Laramie, WY, pp. 409.
Sassen, K. (1991). The polarization lidar technique for cloud research: A review and current
assessment. Bull. Am. Meteorol. Soc., 72, 1848–66.
Sassen, K. and Benson, S. (2001). A midlatitude cirrus cloud climatology from the Facility
for Atmospheric Remote Sensing. Part II: Microphysical properties derived from lidar
depolarization. J. Atmos. Sci., 58, 2103–12.
Sassen, K. and Hsueh, C.-Y. (1998). Contrail properties derived from high-resolution polar-
ization lidar studies during SUCCESS. Geophys. Res. Lett., 25, 1165–8.
Sassen, K. and Liou, K. N. (1979a). Scattering of polarized laser light by water droplet,
mixed-phase and ice crystal clouds. Part I: Angular scattering patterns. J. Atmos. Sci.,
36, 838–51.
Sassen, K. and Liou, K. N. (1979b). Scattering of polarized laser light by water droplet,
mixed-phase and ice crystal clouds. Part II: Angular depolarizing and multiple-
scattering behavior. J. Atmos. Sci., 36, 852–61.
Sausen, R., Isaksen, I., Grewe, V., et al. (2005). Aviation radiative forcing in 2000: An
update on IPCC (1999). Met. Zeitschr. 14, 555–61.
Schaaf, J. W. and Williams, D. (1973). Optical constants of ice in the infrared. J. Opt. Soc.
Am., 63, 726–32.
Schelkunoff, S. A. (1943). Electromagnetic Waves. New York: D. Van Nostrand.
Schiffer, R. A. and Rossow, W. B. (1983). The International Satellite Cloud Climatology
Project (ISCCP): The first project of the World Climate Research Programme. Bull.
Am. Meteorol. Soc., 64, 779–84.
Schumann, U., ed. (1990). Air Traffic and the Environment: Background, Tendencies and
Potential Global Atmospheric Effects. Lecture Notes in Engineering, 60. Berlin, Hei-
delberg: Springer-Verlag.
Schumann, U. (2002). Contrail cirrus. In Cirrus, ed. D. K. Lynch, K. Sassen, D. O’C. Starr,
and G. Stephens. New York: Oxford University Press, pp. 231–55.
Schumann, U. (2005). Formation, properties and climatic effects of contrails. Comptes
Rendus Physique, 6, 549–65.
Schumann, U. (2012). A contrail cirrus prediction model, Geosci. Model Dev., 5, 543–80.
Schumann, U. and Graf, K. (2013). Aviation-induced cirrus and radiation changes at diurnal
timescales. J. Geophys. Res., 118, 2404–21.
Schumann, U. and Wendling, P. (1990). Determination of contrails from satellite data and
observational results. In Air Traffic and the Environment: Background, Tendencies and
Potential Global Atmospheric Effects, ed. U. Schumann. Lecture Notes in Engineering,
60. Berlin, Heidelberg: Springer-Verlag, pp. 138–53.
Schuster, A. (1905). Radiation through a foggy atmosphere. Astrophys. J., 21, 1–22.
References 419
Schwarz, J. P., Gao, R. S., Perring, A. E., Spackman, J. R., and Fahey, D. W. (2013). Black
carbon aerosol size in snow. Sci. Rep., 3, 1356.
Schwarzschild, K. (1906). On the equilibrium of the sun’s atmosphere. Nach. K. Gesell,
Wiss. Göttingen. Math-Phys. Klasse, 195, 41–53.
Scoresby, W. (1820). The Arctic. Vol. I of An Account of the Arctic Regions with a History
and Description of the Northern Whale Fishery. Edinburgh: Archibald Constable and
Co., p. 551.
Seki, M., Kobayashi, K., and Nakahara, J. (1981). Optical spectra of hexagonal ice. J. Phys.
Soc. Japan, 50, 2643–8.
Selkirk, H. B. (1993). The tropopause cold trap in the Australian monsoon during
STEP/AMEX 1987. J. Geophys. Res., 98, 8591–610.
Shcherbakov, V., Gayet, J.-F., Jourdan, O., Ström, J., and Minikin, A. (2006). Light scat-
tering by single ice crystals of cirrus clouds. Geophys. Res. Lett., 33, L15809.
Sherwood, S. C. (2000). A stratospheric “drain” over the maritime continent. Geophys.
Res. Lett., 27, 677–80.
Shurcliff, W. A. (1962). Polarized Light: Production and Use. Cambridge: Harvard Uni-
versity Press.
Slingo, A. and Slingo, J. M. (1988). The response of a general circulation model to cloud
longwave radiative forcing. I: Introduction and initial experiments. Quart. J. Roy.
Meteorol. Soc., 114, 1027–62.
SMIC (1971). Inadvertent Climate Modification: Report of the Study of Man’s Impact on
Climate. Cambridge: MIT Press.
Smith, R. C. and Marsh, J. S. (1974). Diffraction patterns of simple apertures. J. Opt. Soc.
Am., 64, 798–803.
Smith, W. L., Revercomb, H. E., Knuteson, R. O., et al. (1995). Cirrus cloud properties
derived from high spectral resolution infrared spectrometry during FIRE II. Part I: The
High Resolution Interferometer Sounder (HIS) systems. J. Atmos. Sci., 52, 4238–45.
Smith, W. L., Ackerman, S., Revercomb, H., et al. (1998). Infrared spectral absorption of
nearly invisible cirrus clouds. Geophys. Res. Lett., 25, 1137–40.
Smith, W. L., Jr., Hein, P. F., and Cox, S. K. (1990). The 27–28 October 1986 FIRE IFO
cirrus case study: In situ observations of radiation and dynamic properties of a cirrus
cloud layer. Mon. Wea. Rev., 118, 2389–401.
Sobolev, V. V. (1975). Light Scattering in Planetary Atmospheres, trans. W. M. Irvine.
Oxford: Pergamon Press.
Solomon, S., Qin, D., Manning, M., et al., eds. (2007). Climate Change 2007: The Physical
Science Basis. Cambridge, UK: Cambridge University Press, 996 pp.
Solomon, S., Rosenlof, K. H., Portmann, R. W., et al. (2010). Contributions of stratospheric
water vapor to decadal changes in the rate of global warming. Science, 327, 1219–23.
Somerville, R. C. J. and Remer, L. A. (1984). Cloud optical thickness feedbacks in the CO2
climate problem. J. Geophys. Res., 89, 9668–72.
Spangenberg, D. A., Minnis, P., Bedka, S. T., et al. (2013). Contrail radiative forcing
over the northern hemisphere from 2006 Aqua MODIS data. Geophys. Res. Lett., 40,
595–600.
Spänkuch, D. and Döhler, W. (1985). Radiative properties of cirrus clouds in the middle
IR derived from Fourier spectrometer measurements from space. Z. Meteorol., 35,
314–24.
Spinhirne, J. D. and Hart, W. D. (1990). Cirrus structure and radiative parameters from
airborne lidar and spectral radiometer observations: The 28 October 1986 FIRE study.
Mon. Wea. Rev., 118, 2329–43.
420 References
Spinhirne, J. D., Hart, W. D., and Hlavka, D. L. (1996). Cirrus infrared parameters
and shortwave reflectance relations from observations. J. Atmos. Sci., 53, 1438–
58.
Stackhouse, P. W., Jr. and Stephens, G. L. (1991). A theoretical and observational study of
the radiative properties of cirrus: Results from FIRE 1986. J. Atmos. Sci., 48, 2044–
59.
Starr, D. O’C. (1987). A cirrus-cloud experiment: Intensive field observations planned for
FIRE. Bull. Am. Meteorol. Soc., 68, 119–24.
Starr, D. O’C. and Cox, S. K. (1985a). Cirrus clouds. Part I: A cirrus cloud model. J. Atmos.
Sci., 42, 2663–81.
Starr, D. O’C. and Cox, S. K. (1985b). Cirrus clouds. Part II: Numerical experiments on
the formation and maintenance of cirrus. J. Atmos. Sci., 42, 2682–94.
Stephens, G. L, Paltridge, G. W., and Platt, C. M. R. (1978). Radiation profiles in extended
water clouds. III: Observations. J. Atmos. Sci., 35, 2133–41.
Stephens, G. L., Campbell, G. G., and Vonder Haar, T. H. (1981). Earth radiation budgets.
J. Geophys. Res., 86, 9739–60.
Stephens, G. L., Tsay, S.-C., Stackhouse, P. W., Jr., and Flatau, P. J. (1990). The relevance
of the microphysical and radiative properties of cirrus clouds to climate and climate
feedback. J. Atmos. Sci., 47, 1742–53.
Stephens, G. L., Vane, D. G., Boain, R. J., et al. (2002). The CloudSat mission and the
A-Train: A new dimension of space-based observations of clouds and precipitation.
Sterle, K. M., McConnell, J. R., Dozier, J., Edwards, R., and Flanner, M. G. (2013).
Retention and radiative forcing of black carbon in eastern Sierra Nevada snow. The
Cryosphere, 7, 365–74.
Stokes, G. G. (1852). On the composition and resolution of streams of polarized light from
different sources. Trans. Cambridge Philos. Soc., 9, 399–423.
Stowe, L. L. (1977). Polarization of reflected sunlight as measured from a high-altitude
balloon. Proc. SPIE, 112, 176–83.
Stratton, J. A. (1941). Electromagnetic Theory. New York: McGraw-Hill.
Strow, L. L., Hannon, S. E., De Souza-Machado, S., Motteler, H. E., and Tobin, D. (2003a).
An overview of the AIRS radiative transfer model. IEEE Trans. Geosci. Remote Sens.,
41, 303–13.
Strow, L. L., Hannon, S. E., Weiler, M., et al. (2003b). Prelaunch spectral calibration of
the Atmospheric Infrared Sounder (AIRS). IEEE Trans. Geosci. Remote Sens., 41,
274–86.
Sun, W., Fu, Q., and Chen, Z. (1999). Finite-difference time-domain solution of light
scattering by dielectric particles with a perfectly matched layer absorbing boundary
condition. Appl. Opt., 38, 3141–51.
Suomi, V. E. (1958). The radiation balance of the Earth from a satellite. Ann. Int. Geophys.
Year, 6, 331–40.
Taflove, A. (1980). Application of the finite-difference time-domain method to sinusoidal
steady-state electromagnetic-penetration problems. IEEE Trans. Electromagn. Com-
pat., EMC–22, 191–202.
Taflove, A. (1995). Computational Electrodynamics: The Finite-Difference Time-Domain
Method. Norwood, MA: Artech House.
Taflove, A. and Brodwin, M. E. (1975). Numerical solution of steady-state electromag-
netic scattering problems using the time-dependent Maxwell’s equations. IEEE Trans.
Microw. Theory Tech., MTT–23, 623–30.
References 421
Taflove, A. and Umashankar, K. R. (1990). The finite-difference time-domain method for

numerical modeling of electromagnetic wave interactions with arbitrary structures.
Progress in Electromagnetics Research (PIER), 2, 287–373.
Takano, Y. and Asano, S. (1983). Fraunhofer diffraction by ice crystals suspended in the
atmosphere. J. Meteorol. Soc. Jpn., 61, 289–300.
Takano, Y. and Jayaweera, K. (1985). Scattering phase matrix for hexagonal ice crystals
computed from ray optics. Appl. Opt., 24, 3254–63.
Takano, Y. and Liou, K. N. (1989a). Solar radiative transfer in cirrus clouds. Part I: Single-
scattering and optical properties of hexagonal ice crystals. J. Atmos. Sci., 46, 3–19.
Takano, Y. and Liou, K. N. (1989b). Solar radiative transfer in cirrus clouds. Part II: Theory
and computation of multiple scattering in an anisotropic medium. J. Atmos. Sci., 46,
20–36.
Takano, Y. and Liou, K. N. (1993). Transfer of polarized infrared radiation in optically
anisotropic media: Application to horizontally oriented ice crystals. J. Opt. Soc. Am.
A., 10, 1243–56.
Takano, Y. and Liou, K. N. (1995). Radiative transfer in cirrus clouds. Part III: Light
scattering by irregular ice crystals. J. Atmos. Sci., 52, 818–37.
Takano, Y., Liou, K. N., and Yang, P. (2012). Diffraction by rectangular parallelepiped,
hexagonal cylinder, and three-axis ellipsoid: Some analytic solutions and numerical
results. J. Quant. Spectrosc. Radiat. Transfer, 113, 1836–43.
Takano, Y., Liou, K. N., Kahnert, M., and Yang, P. (2013). The single scattering properties of
black carbon aggregates determined from the geometric-optics surface-wave approach
and the T-matrix method. J. Quant. Spectrosc. Radiat. Transfer, 125, 51–6.
Tao, W.-K. and Moncrieff, M. W. (2009). Multiscale cloud system modeling. Rev. Geophys.,
47, 1–41.
Thevenon, F., Anselmetti, F. S., Bernasconi, S. M., and Schwikowski, M. (2009). Mineral
dust and elemental black carbon records from an Alpine ice core (Colle Gnifetti
glacier) over the last millennium. J. Geophys. Res., 114, D17102.
Toon, O. B., Tolbert, M. A., Koehler, B. G., Middlebrook, A. M., and Jordan, J. (1994).
The infrared optical constants of H2 O ice, amorphous acid solutions, and nitric acid
hydrates. J. Geophys. Res., 99, 25631–54.
Tränkle, E. and Mielke, B. (1994). Simulation and analysis of pollen coronas. Appl. Opt.,
33, 4552–62.
Travis, L. D., Coffeen, D. L., Hansen, J. E., et al. (1979). Orbiter cloud photopolarimeter
investigation. Science, 203, 781–5.
Twomey, S. and Cocks, T. C. (1982). Spectral reflectance of clouds in the near-infrared:
Comparison of measurements and calculations. J. Meteorol. Soc. Jpn., 60, 583–92.
Twomey, S. and Cocks, T. C. (1989). Remote sensing of cloud parameters from spectral
reflectance in the near-infrared. Beitr. Phys. Atmos., 62, 172–9.
Twomey, S. and Seton, K. J. (1980). Inferences of gross microphysical properties of clouds
from spectral reflectance measurements. J. Atmos. Sci., 37, 1065–9.
Twomey, S., Piepgrass, M., and Wolfe, T. (1984). An assessment of the impact of pollution
on global cloud albedo. Tellus, 36B, 356–66.
Ulanowski, Z., Hesse, E., Kaye, P. H., and Baran, A. J. (2006). Light scattering by complex
ice-analogue crystals. J. Quant. Spectrosc. Radiat. Transfer, 100, 382–92.
Ulanowski, Z., Kaye, P. H., Hirst, E., and Greenaway, R. S. (2010). Light scattering by ice
particles in the Earth’s atmosphere and related laboratory measurements. In Proceed-
ings of the 12th International Conference on Electromagnetic and Light Scattering.
Helsinki, FI, pp. 294–7.
422 References
Umashankar, K. and Taflove, A. (1982). A novel method to analyze electromagnetic scat-

tering of complex objects. IEEE Trans. Electromagn. Compat., EMC–24, 397–405.
van de Hulst, H. C. (1945). Theory of absorption lines in the atmosphere of the earth. Ann.
Astrophys., 8, 1–11.
van de Hulst, H. C. (1947). A theory of the anti-coronae. J. Opt. Soc. Am., 37, 16–22.
van de Hulst, H. C. (1957). Light Scattering by Small Particles. New York: John Wiley &
Sons.
van de Hulst, H. C. (1980). Multiple Light Scattering: Tables, Formulas, and Applications,
Vols. 1 and 2. New York: Academic Press.
van Diedenhoven, B., Cairns, B., Fridlind, A. M., Ackerman, A. S., and Garrett, T. J.
(2013). Remote sensing of ice crystal asymmetry parameter using multi-directional
polarization measurements. Part 2: Application to the Research Scanning Polarimeter.
Atmos. Chem. Phys., 13, 3185–203.
Volkovitskiy, O. A., Pavlova, L. N., and Petrushin, A. G. (1980). Scattering of light by ice
crystals. Izv., Acad. Sci., USSR, Atmos. Ocean. Phys., 16, 98–102.
Vonder Haar, T. H. and Suomi, V. E. (1971). Measurements of the earth’s radiation budget
from satellites during a five-year period. Part I: Extended time and space means. J.
Atmos. Sci., 28, 305–14.
von Koch, H. (1904). Sur une courbe continue sans tangente, obtenue par une construction
géométrique élémentaire. Arkiv foer Matematik, 1, 681–704.
Vonnegut, B. (1947). The nucleation of ice formation by silver iodide. J. Appl. Phys., 18,
593–5.
Vouk, V. (1948). Projected area of convex bodies. Nature, 162, 330–1.
Wait, J. R. (1955). Scattering of a plane wave from a circular dielectric cylinder at oblique
incidence. Can. J. Phys., 33, 189–95.
Walden, V. P., Warren, S. G., and Tuttle, E. (2003). Atmospheric ice crystals over the
Antarctic Plateau in winter. J. Appl. Meteorol., 42, 1391–405.
Waliser, D. E., Li, J.-L. F., Woods C. P., et al. (2009). Cloud ice: A climate model challenge
with signs, and expectations of progress. J. Geophys. Res., 114, D00A21.
Wallace, J. M. and Hobbs, P. V. (2006). Atmospheric Science: An Introductory Survey.
Amsterdam: Elsevier Academic Press.
Wang, C., Yang, P., Baum, B. A., et al. (2011). Retrieval of ice cloud optical thickness and
effective particle size using a fast infrared radiative transfer model. J. Appl. Meteorol.
Clim., 50, 2283–97.
Wang, M. and King, M. D. (1997). Correction of Rayleigh scattering effects in cloud optical
thickness retrievals. J. Geophys. Res., 102, 25915–26.
Wang, P.-H., McCormick, M. P., Poole, L. R., et al. (1994). Tropical high cloud character-
istics derived from SAGE II extinction measurements. Atmos. Res., 34, 53–83.
Wang, P. K. (2013). Physics and Dynamics of Clouds and Precipitation. Cambridge: Cam-
bridge University Press.
Wang, X., Liou, K. N., Ou, S. C. S., Mace, G. G., and Deng, M. (2009). Remote sensing of
cirrus cloud vertical size profile using MODIS data. J. Geophys. Res., 114, D09205.
Warner, J. and Twomey, S. (1967). The production of cloud nuclei by cane fires and the
effect on cloud droplet concentration. J. Atmos. Sci., 24, 704–6.
Warren, S. G. (1984). Optical constants of ice from the ultraviolet to the microwave. Appl.
Opt., 23, 1206–25.
Warren, S. G. and Brandt, R. E. (2008). Optical constants of ice from the ultraviolet to the
microwave: A revised compilation. J. Geophys. Res., 113, D14220.
References 423
Washington, W. M. and Meehl, G. A. (1984). Seasonal cycle experiment on the climate

sensitivity due to a doubling of CO2 with an atmospheric general circulation model
coupled to a simple mixed-layer ocean model. J. Geophys. Res., 89, 9475–503.
Waterman, P. C. (1971). Symmetry, unitarity, and geometry in electromagnetic scattering.
Phys. Rev. D, 3, 825–39.
Waters, J. W., Read, W. G., Froidevaux, L., et al. (1999). The UARS and EOS Microwave
Limb Sounder (MLS) experiments. J. Atmos. Sci., 56, 194–218.
Wegener, A. (1911). Thermodynamik der Atmosphäre. Leipzig: Verlag Von Johann Ambro-
sius Barth.
Wei, H., Yang, P., Li, J., et al. (2004). Retrieval of semitransparent ice cloud optical
thickness from Atmospheric Infrared Sounder (AIRS) measurements. IEEE Trans.
Geosci. Remote Sens., 42, 2254–67.
Weickmann, H. (1948). The Ice Phase in the Atmosphere, trans. M. G. Sutton. London:
Ministry of Supply.
Wendling, P., Wendling, R., and Weickmann, H. K. (1979). Scattering of solar radiation by
hexagonal ice crystals. Appl. Opt., 18, 2663–71.
West, R. A., Strobel, D. F., and Tomasko, M. G. (1986). Clouds, aerosols, and photochem-
istry in the Jovian atmosphere. Icarus, 65, 161–217.
Whitlock, C. H., Charlock, T. P., Staylor, W. F., et al. (1995). First global WCRP shortwave
surface radiation budget dataset. Bull. Am. Meteorol. Soc., 76, 905–22.
Wielicki, B. A., Suttles, J. T., Heymsfield, A. J., et al. (1990). The 27–28 October 1986
FIRE IFO cirrus case study: Comparison of radiative transfer theory with observations
by satellite and aircraft. Mon. Wea. Rev., 118, 2356–76.
Wielicki, B. A., Cess, R. D., King, M. D., Randall, D. A., and Harrison, E. F. (1995).
Mission to planet Earth: Role of clouds and radiation in climate. Bull. Am. Meteorol.
Soc., 76, 2125–53.
Wilcke, J. C. (1761). Rön och tankar om snö-figures skiljaktighet. K. Svenska. Vetenska-
Akad. Handl., 22, 1–7.
Wild, M., Ohmura, A., Gilgen, H., Roeckner, E. (1995). Validation of GCM simulated
radiative fluxes using surface observations. J. Climate, 8, 1309–24.
Wiscombe, W. J. (1977). The delta-M method: Rapid yet accurate radiative flux calculations
for strongly asymmetric phase functions. J. Atmos. Sci., 34, 1408–22.
Wiscombe, W. J. and Mugnai, A. (1988). Scattering from non-spherical Chebyshev parti-
cles. 2: Means of angular scattering patterns. Appl. Opt., 27, 2405–21.
Wolff, G. A. (1957). The growth of ice crystals. In Artificial Stimulation of Rain, ed. H.
Weickmann and W. Smith. London: Pergamon Press, pp. 332–9.
Wong, E., Hutchison, K. D., Ou, S. C. S., and Liou, K. N. (2007). Cirrus cloud top temper-
atures retrieved from radiances in the National Polar-Orbiting Operational Environ-
mental Satellite System: Visible Infrared Imager Radiometer Suite 8.55 and 12.0 µm
bandpasses. Appl. Opt., 46, 1316–25.
Wriedt, T. (1998). A review of elastic light scattering theories. Part. Part. Syst. Charact.,
15, 67–74.
Wu, D. L., Austin, R. T., Deng, M., et al. (2009). Comparisons of global cloud ice from
MLS, CloudSat, and correlative data sets. J. Geophys. Res., 114, D00A24.
Wylie, D. P., Menzel, W. P., Woolf, H. M., and Strabala, K. I. (1994). Four years of global
cirrus cloud statistics using HIRS. J. Climate, 7, 1972–86.
Wylie, D. P., Jackson, D. L., Menzel, W. P., and Bates, J. J. (2005). Trends in global cloud
cover in two decades of HIRS observations. J. Climate, 18, 3021–31.
424 References
Wyser, K. and Yang, P. (1998). Average ice crystal size, and bulk short-wave single-
scattering properties of cirrus clouds. Atmos. Res., 49, 315–35.
Xie, Y., Yang, P., Liou, K. N., Minnis, P., and Duda, D. P. (2012). Parameteriza-
tion of contrail radiative properties for climate studies. Geophys. Res. Lett., 39,
L00F02.
Xu, B., Cao, J., Hansen, J. E., et al. (2009). Black soot and the survival of Tibetan glaciers.
Proc. Natl. Acad. Sci. U.S.A., 106, 114–18.
Yaghjian, A. D. (1980). Electric dyadic Green’s functions in the source region. Proc. IEEE,
68, 248–63.
Yamamoto, G. (1952). On a radiation chart. Sci. Rep. Tohoku Univ. Ser. 5 Geophys., 4,
9–23.
Yang, P. and Liou, K. N. (1995). Light scattering by hexagonal ice crystals: Comparison
of finite-difference time domain and geometric optics models. J. Opt. Soc. Am. A, 12,
162–76.
Yang, P. and Liou, K. N. (1996a). Geometric-optics-integral-equation method for light
scattering by non-spherical ice crystals. Appl. Opt., 35, 6568–84.
Yang, P. and Liou, K. N. (1996b). Finite-difference time domain method for light scattering
by small ice crystals in three-dimensional space. J. Opt. Soc. Am. A, 13, 2072–85.
Yang, P. and Liou, K. N. (1997). Light scattering by hexagonal ice crystals: Solutions by a
ray-by-ray integration algorithm. J. Opt. Soc. Am. A, 14, 2278–89.
Yang, P. and Liou, K. N. (1998a). Single-scattering properties of complex ice crystals in
terrestrial atmosphere. Contr. Atmos. Phys., 71, 223–48.
Yang, P. and Liou, K. N. (1998b). An efficient algorithm for truncating spatial domain in
modeling light scattering by finite-difference technique. J. Compt. Phys., 140, 346–69.
Yang, P. and Liou, K. N. (2000). Finite difference time domain method for light scattering
by non-spherical and inhomogeneous particles. In Light Scattering by Non-spherical
Particles: Theory, Measurements, and Applications, ed. M. I. Mishchenko, J. W.
Hovenier, and L. D. Travis. San Diego: Academic Press, pp. 173–221.
Yang, P. and Liou, K. N. (2006). Light scattering and absorption by non-spherical ice
crystals. In Light Scattering Reviews: Single and Multiple Light Scattering, ed. A.
Kokhanovsky. Berlin, Heidelberg: Springer-Verlag , pp. 31–71.
Yang, P. and Liou, K. N. (2009). Effective refractive index for determining ray propagation
in an absorbing dielectric particle. J. Quant. Spectrosc. Radiat. Transfer, 110, 300–6.
Yang, P., Liou, K. N., and Arnott, W. P. (1997). Extinction efficiency and single-scattering
albedo for laboratory and natural cirrus clouds. J. Geophys. Res., 102, 21825–35.
Yang, P., Liou, K. N., Wyser, K., and Mitchell, D. (2000a). Parameterization of the scattering
and absorption properties of individual ice crystals. J. Geophys. Res., 105, 4699–718.
Yang, P., Liou, K. N., Mishchenko, M. I., and Gao, B.-C. (2000b). Efficient finite-difference
time-domain scheme for light scattering by dielectric particles: Application to aerosols.
Appl. Opt., 39, 3727–37.
Yang, P., Gao, B.-C., Baum, B. A., et al. (2001). Sensitivity of cirrus bidirectional reflectance
to vertical inhomogeneity of ice crystal habits and size distributions for two Moderate-
Resolution Imaging Spectroradiometer (MODIS) bands. J. Geophys. Res., 106, 17267–
91.
Yang, P., Baum, B. A., Heymsfield, A. J., et al. (2003a). Single-scattering properties of
droxtals. J. Quant. Spectrosc. Radiat. Transfer, 79–80, 1159–69.
Yang, P., Mlynczak, M. G., Wei, H., et al. (2003b). Spectral signature of ice clouds in
the far-infrared region: Single-scattering calculations and radiative sensitivity study.
J. Geophys. Res., 108, 4569.
References 425
Yang, P., Kattawar, G. W., Liou, K. N., and Lu, J. Q. (2004). Comparison of Cartesian
grid configurations for application of the finite-difference time-domain method to
electromagnetic scattering by dielectric particles. Appl. Opt., 43, 4611–24.
Yang, P., Wei, H., Huang, H. L., et al. (2005). Scattering and absorption property database
for non-spherical ice particles in the near- through far-infrared spectral region. Appl.
Opt., 44, 5512–23.
Yang, P., Hong, G., Kattawar, G. W., Minnis, P., and Hu, Y. (2008). Uncertainties associated
with the surface texture of ice particles in satellite-based retrieval of cirrus clouds.
Part II: Effect of particle surface roughness on retrieved cloud optical thickness and
effective particle size. IEEE Trans. Geosci. Remote Sens., 46, 1948–57.
Yang, P., Bi, L., Baum, B. A., et al. (2013). Spectrally consistent scattering, absorption, and
polarization properties of atmospheric ice crystals at wavelengths from 0.2 to 100 µm.
J. Atmos. Sci., 70, 330–47.
Yang, P., Liou, K. N., Bi, L., et al. (2015). On the radiative properties of ice clouds: Light
scattering, remote sensing, and radiation parameterization. Adv. Atmos. Sci., 32, 32–
63.
Yee, K. S. (1966). Numerical solution of initial boundary value problems involving
Maxwell’s equations in isotropic media. IEEE Trans. Antennas Propag., AP-14, 302–
7.
Yi, B., Yang, P., Liou, K. N., Minnis, P., and Penner, J. E. (2012). Simulation of the global
contrail radiative forcing: A sensitivity analysis. Geophys. Res. Lett., 39, L00F03.
Yi, B., Yang, P., Baum, B. A., et al. (2013). Influence of ice particle surface roughening on
the global cloud radiative effect. J. Atmos. Sci., 70, 2794–807.
You, Y., Kattawar, G. W., Yang, P., Hu, Y. X., and Baum, B. A. (2006). Sensitivity of
depolarized lidar signals to cloud and aerosol particle properties. J. Quant. Spectrosc.
Yuan, T. and Li, Z. (2010). General macro- and microphysical properties of deep convective
clouds as observed by MODIS. J. Climate, 23, 3457–73.
Yue, Q. and Liou, K. N. (2009). Cirrus cloud optical and microphysical properties deter-
mined from AIRS infrared spectra. Geophys. Res. Lett., 36, L05810.
Yue, Q., Liou, K. N., Ou, S. C. S., et al. (2007). Interpretation of AIRS data in thin cirrus
atmospheres based on a fast radiative transfer model. J. Atmos. Sci., 64, 3827–42.
Yurkin, M. A. and Hoekstra, A. G. (2011). The discrete-dipole-approximation code ADDA:
Capabilities and known limitations. J. Quant. Spectrosc. Radiat. Transfer, 112, 2234–
47.
Yurkin, M. A., Hoekstra, A. G., Brock, R. S., and Lu, J. Q. (2007). Systematic comparison
of the discrete dipole approximation and the finite difference time domain method for
large dielectric scatterers. Opt. Express, 15, 17902–11.
Yurkin, M. A., Min, M., and Hoekstra, A. G. (2010). Application of the discrete dipole
approximation to very large refractive indices: Filtered coupled dipoles revived. Phys.
Rev. E, 82, 036703.
Zakharova, N. T., Videen, G., and Khlebtsov, N. G. (2012). Comprehensive T-matrix
reference database: A 2009–2011 update. J. Quant. Spectrosc. Radiat. Transfer, 113,
1844–52.
Zdanowicz, C. M., Zielinski, G. A., and Wake, C. P. (1998). Characteristics of modern
atmospheric dust deposition in snow on the Penny Ice Cap, Baffin Island, Arctic
Canada. Tellus, 50B, 506–20.
Zdanowicz, C. M., Hall, G., Vaive, J., et al. (2006). Asian dustfall in the St. Elias Mountains,
Yukon, Canada. Geochim. Cosmochim. Acta., 70, 3493–507.
426 References
Zhai, P.-W., Li, C., Kattawar, G. W., and Yang, P. (2007). FDTD far-field scattering ampli-
tudes: Comparison of surface and volume integration methods. J. Quant. Spectrosc.
Zhang, C., Lee, K.-S., Zhang, X.-C., Wei, X., and Shen, Y. R. (2001). Optical constants of
ice Ih crystal at terahertz frequencies. Appl. Phys. Lett., 79, 491–3.
Zhang, F., Zeng, Q., Gu, Y., and Liou, K. N. (2005). Parameterization of the absorption
of the H2 O continuum, CO2 , O2 , and other trace gases in the Fu–Liou solar radiation
program. Adv. Atmos. Sci., 22, 545–58.
Zhang, Z., Yang, P., Kattawar, G. W., et al. (2004). Geometrical-optics solution to light
scattering by droxtal ice crystals. Appl. Opt., 43, 2490–9.
Zhang, Z., Yang, P., Kattawar, G., et al. (2009). Influence of ice particle model on retrieving
cloud optical thickness from satellite measurements: Model comparison and implica-
tion for climate study. Atmos. Chem. Phys. Discuss., 9, 1757–96.
Zhou, X.-L., Geller, M. A., and Zhang, M. (2001). Cooling trend of the tropical cold point
tropopause temperatures and its implications. J. Geophys. Res., 106, 1511–22.
Index
120° parhelion, 133, 134, see also halo A-Train satellite constellation, 259, 260
1-D climate model, 365 cloud and aerosol interactions, 366
1-D model, 51, 361, 363, 364 deposition, 192, 193
detection of thin cirrus, 286, 287
22° halo, 133, 134, see also halo direct and indirect effect of, 361, 362, 380
22° parhelion, 130, 133, see also halo historical review of polarization measurements, 302
2-D climate model, 366 radiative transfer, 332
radiative transfer results compared with satellite
46° halo, 128, 132, 134, 186, 189, 220, 233, 268, 269, observations, 346, 348
see also halo soot and sulfate aerosols, 379, 380
view from satellite instruments, 14, 16
absorbing boundary conditions, 204 Aerosol Polarimetry Sensor, 302
absorptance, 80, 255 AGCM, 366, 370, see also GCM
absorption aggregate, 20, 32, 35, 38, 39, 42, 43, 47, 49, 216, 218,
coefficient, 77, 78, 80, 325, 335, 336, 337 233, 234, 239, 271, 273, 303
cross-section, 55, 56, 156, 157, 175, 176, 330 Airborne Visible/Infrared Imaging Spectrometer, 286
efficiency, 168 AIRS, 11, 259, 277, 279, 280, 281, 300, see also
line, 250, 252, 279, 321, 323, 328, 333, 335, 336, Atmospheric Infrared Sounder
338 Airy function, 167
spectrum albedo
CH4 , 250 cloud, 344, 351, 361
CO, 250 global, 293, 390
CO2 , 250 planetary, 346
H2 O, 250 shortwave, 349
N2 , 250 single-scattering, see single-scattering albedo
N2 O, 250 solar, 341, 342, 344, 359, 360, 361, 362, 364, 373,
O2 , 250 374, 377
O3 , 250 surface, 254, 255, 257, 263, 266, 291, 293, 295,
Ac, see altocumulus 305, 321, 322, 341, 342, 344, 353, 355, 366,
accretion, 29, 33, 47, 192, 277, 360 373, 384
ACCRI, 375, 377, 380, see also Aviation Climate TOA, 347, 348
Change Research Initiative total-sky, 348
ADDA, 212, 234, 243, 246 altocumulus, 2
adding method, 112, 339 altostratus, 2
ADEOS, 302, 303, 304 AMSR2, see Advanced Microwave Scanning
Advanced Earth Observing Satellite, see ADEOS Radiometer 2
Advanced Microwave Scanning Radiometer 2, 259 AMSR-E, see Advanced Microwave Scanning
Advanced Microwave Scanning Radiometer-EOS, 259 Radiometer-EOS
Advanced Microwave Sounding Unit, 11, 259, 279 Amsterdam DDA, 212, see also ADDA
Advanced Very High Resolution Radiometer, 5, 6, 286 AMSU, 280, see also Advanced Microwave Sounding
aerosol Unit
as atmospheric composition, 249 AMSU-A, 259
427
428 Index
analyzer, 92 BDC, see Brewer–Dobson circulation

anomalous dispersion, 77 Beer’s law, 79
anthelion, 131, 134, see also halo Bessel function, 68, 71, 74, 136, 141, 144, 223
anvil, 5, 42, 50, 279, 306, 382, 383, see also tropical bidirectional reflectance
anvil climate studies, 327
APS, see Aerosol Polarimetry Sensor ice spectral, 257
Aqua, 14, 258, 259, 269, 279, 283, 285, 376, see also reflected line spectra in 1.38 and 0.76 µm bands,
MODIS, A-Train 318, 319, 324
ARM, 13, 42, 43, 49, 287, 288, 292, 295, 300, 346, reflected polarization from ice clouds, 305
348, see also Atmospheric Radiation relative position of sun–satellite system, 254
Measurement remote sensing, 248, see also remote sensing
As, see altostratus retrieval
associated Legendre function, 73 algorithm for vertical sizing, 293
asymmetry factor of cirrus cloud parameters, 325
comparison with FDTD and DDA, 173 single-scattering properties of ice spheres, 102
current MODIS operation program, 273 birefringence, 78, 79, 147
effect of ice crystal shape on cloud radiative Bishop’s ring, 117
forcing, 367 black carbon, see also BC
FDTD 1-D cloud–precipitation–climate model, 362
and DDA, 174 application to climate studies, 341
and PSTD, 220 cirrus radiative forcing, 355
for ice sphere and circular cylinder, 102 contrail cirrus, 370
invariant imbedding T-matrix method, 237 high thin clouds, 391
Koch snowflake, 184 Rayleigh–Gans–Debye approximation, 176
radiative forcing of cirrus clouds, 350 single-scattering parameterization for ice crystals,
RGD approximation, 175 333
single-scattering parameterization for ice crystals, snow grains contaminated by, 192
333 unified theory for light scattering, 180, 184
snow grain contaminated by BC and dust, 192 blackbody, 285, 342, 383, 384
surface wave approach for sphere, 167, 168, 169, blue sky, 1, 5, 6
171 BRDF (bidirectional reflectance distribution
two-channel inversion, 264 function), 254, 255, 256, 257
unified theory, 178, 179, 181, 184 Brewer–Dobson circulation, 382
asymmetry parameter, 306, 331, 348, see also brightness temperature, 270, 274, 281, 285, 307, see
asymmetry factor also temperature
atmospheric composition, 248, 357, 375 difference, 13, 15, 274, 280, 287, 288, see also
Atmospheric Infrared Sounder, 11 BTD
Atmospheric Radiation Measurement, 13, 42, 295, broadband emissivity, 336
346 BTD, 13, 15, 16, 274, 287, see also brightness
A-Train, 13, 14, 257, 275, 292, 302, 316, 386 temperature
attenuation, 59, 123, 252, 315, 339 bullet rosette
Aura, 258, 259, 386 depolarization ratio, 316
AVHRR, 5, 279, 286, 374, see also Advanced Very FDTD method, 215
High Resolution Radiometer improved geometric optics approach, 163
Aviation Climate Change Research Initiative, 375 in contrails and contrail cirrus, 370
AVIRIS, 286, 318, see also Airborne Visible/Infrared invariant imbedding T-matrix method, 234
Imaging Spectrometer phase function, 268
azimuthal angle, definition of, 107 snow grains, 193
unified theory for light scattering, 182
Babinet’s principle, 117, 134, 148, 155, 170, 171, 178
backscattering, 11, 12, 14, 45, 104, 112, 117, 147, CAGEX (CERES/ARM/GEWEX Experiment), 346,
163, 176, 247, 267, 288, 306, 308, 310, 311, 347, 348, 349, 350
312, 313, 314, 315, 316 CALIOP, 14, 16, 259, 288, 306, 316, 318, see also
backscattering depolarization ratio, see depolarization Cloud-Aerosol Lidar and Infrared Pathfinder
ratio Satellite Observations
band model, 336 CALIOP/IIR, 259
BC, 180, 192, 193, 195, 196, 355, 357, 362, see also CALIPSO, see Cloud-Aerosol Lidar and Infrared
black carbon Pathfinder Satellite Observations
Index 429
A-Train satellite constellation, 259, 260 aerosol–cloud interactions, 361

backscattering depolarization, 306, 316 anthropogenic, 277
contrails and contrail cirrus, 376 clouds and cloud feedback, 327
high thin clouds, 387, 390 contrails and contrail cirrus, 371
mechanism of stratosphere dehydration, 386 detection of thin cirrus, 285
role of cirrus in UTLS exchange, 382 greenhouse gases, 248
view from satellite instruments, 14 impacts of aviation soot on cirrus, 380
CAM, 370, 377, 378, 387, see also Community natural variability and anthropogenic perturbations,
Atmosphere Model 357
carbon dioxide, 248, 255, 359, 371 physical and microphysical properties of contrails,
carbon monoxide, 248, 259 374
CART, 346, 348, see also Cloud and Radiation RFs resulting from aviation produced contrails,
Testbed 377
Cartesian coordinate, 38, 72, 90, 94, 107, 168, 311 role of cirrus in UTLS exchange, 380
Cb, 3, 5, see also cumulonimbus role of ice clouds, 390
Cc, 2, see also cirrocumulus climate model
CCN, 180, 359, 360, 361, see also cloud condensation 1-D climate–cirrus cloud model, 363, 364
nuclei 1-D cloud–precipitation–climate model, 359, 360,
Central Equatorial Pacific Experiment, 42, see also 361
CEPEX cirrus cloud vertical profile effect, 289
CEPEX, 42, 46, 47, 266, 267, 292, see also Central cirrus radiative forcing, 351, 353
Equatorial Pacific Experiment comparison with satellite observations, 346
CERES, 11, 259, 277, 283, 346, 389, see also Clouds contrails and contrail cirrus, 373
and the Earth’s Radiant Energy System detection of thin cirrus, 286
CERES/ARM/GEWEX Experiment, 347, see also effect of interactive ice crystal size on precipitation,
CAGEX 365
CESM, see Community Earth System Model energy balance climate model, 373
CF2 Cl2 , 339 general circulation model, 45, 289, 332
CF3 Cl, 339 high thin ice clouds, 390
CFC, 249, 339 MODIS C6 products, 277
CFCl3 , 339 one-dimensional, 358, 359
CFL, 204, 216, see also Courant–Friedrichs–Lewy radiative forcing of cirrus clouds, 350
condition radiative transfer, 340
CGOM, 162, 178, see also conventional geometric single-scattering parameterization for ice crystals,
optics 332
CH4 , 101, 248, 249, 251, 252, 335, 338, 339, 343, see thin cirrus detection and comparison with GCM,
also absorption spectrum; methane 386
chemical composition cloud
of atmosphere, 380, 386 absorptance, 283, 342
of particles, 239 condensation nuclei, 180, 360, see also CCN
chlorine, 259 cover, xiv, 9, 275, 285, 348, 351, 355, 363, 371,
chlorofluorocarbon, see CFC 373, 386, 389, 390
Christiansen effect, 102, 333 overlap, 348, 377, 385, 389
Ci, 2, see also cirrus radar system, 306
circular polarization, 86, 90, 91, 92 radiative forcing, 321, 328, 350, 351, 355, 362, 370,
circumzenithal arc, 133, 134, see also halo 389, 390, see also CRF
cirrocumulus, 2, 41, see also Cc reflection, 262, 351
cirrostratus, 2, 41, 43, 57, 103, 104, 114, 319, 320, see resolving model, 51
also Cs water content, 29, 365
cirrus, 2 Cloud and Radiation Testbed, 346
CIRRUS, 323 Cloud Particle Imager, 33
cirrus radiative forcing, 351, 355 Cloud Profiling Radar, 259, 295, 387, see also CPR
CKD, 337, 338, see also correlated k-distribution Cloud-Aerosol Lidar and Infrared Pathfinder Satellite
method Observations, 14, see also CALIPSO
Clausius–Clapeyron equation, 22, 24 Cloud-Aerosol Lidar with Orthogonal Polarization,
climate change 14, 259, 316, see also CALIOP
3-D cloud structure, 365 Clouds and the Earth’s Radiant Energy System, 11,
ACCRI program, 375 259, see also CERES
430 Index
CloudSat, 13, 259, 260, 295, 386, 387, 389 continuum absorption, 338
coccolith, 237 contrails and contrail cirrus
coccolithophore, 237 CALIOP, 14
column climate issues, 370
ADDA, PSTD, and EGOH, 246 contaminated ice particles in, 192
backscattered depolarization, 312, 313 contemporary programs, 374, 375, 380
backscattering depolarization, 309 early studies, 371
classification, 32, 33 ice crystal size and shape spectra, 42, 43
comparison of GOS with FDTD and DDA, 173, 174 introduction to clouds, 5
computer-generated, 35, 38 lidar, 316
crystal size and shape distribution, 271, 272, 277, radiative forcing, 333
292 solar heating rate, 355
depolarization ratio, 316 SUCCESS, 51
effect of size distribution, 267, 268 UTLS, 380
extinction cross-section, 330 convection, 2, 8, 9, 42, 46, 50, 52, 358, 359, 366, 380,
FDTD, 214 385, 387
and PSTD, 220 convective adjustment, 359
generalized parameterization, 332 conventional geometric optics, 159, 162, 180, 325,
geometric ray-tracing, 124, 146 330, 331, 333
GOS, 173 convolution PML, 198
growth by riming, 29 coordinates
horizontal orientation, 112, 114, 129, 132, 303, body-framed, 137
310, 311 Cartesian, see Cartesian coordinate
ice crystal cylindrical, 68, 70
habits, 25 polar, 135
shape, 41 pressure, 335
IGOM, 163 rectangular, 118
II-TM right-handed, 150
and DDA, 233 spherical, 65
and IGOM, 238 spheroidal, 72
in contrails and contrail cirrus, 370 corona, 116, 117, 134, 136
measured and computed BRDF, 257 correlated k-distribution method, 257, 279, 319, see
metamorphism, 193 also CKD; k-distribution
MPACE observations, 49 Courant–Friedrichs–Lewy condition, 204
NaF, 186 CPI, 33, 42, see also Cloud Particle Imager
polarization measurements, 303 CPML, 198, see also convolution PML
pristine habit model, 269 CPR, 259, 295, see also Cloud Profiling Radar
PSTD and IGOM, 219 CRF, 321, 323, see also cloud radiative forcing
ray-by-ray and FDTD, 214 critical lapse rate, 359
relation between maximum size and width, 43 CRM, 51, see also cloud resolving model
retrieved asymmetry parameter, 306 CRS, 306, 307, see also cloud radar system
scattering and absorption properties, 344 CRYSTAL-FACE (Cirrus Regional Study of Tropical
single-scattering input to radiative transfer, Anvils and Cirrus Layers – Florida Area
264 Cirrus Experiment), 47, 292, 306
single-scattering parameterization, 328 Cs, 2, see also cirrostratus
smooth-solid crystal distributions, 283 Cu, 2, see also cumulus
unified theory, 182 cumulative probability, 257, 319, 336, 338
video cloud-scope image, 189 cumulonimbus, 3, 8, 41, 51, 249, 360, 383, see also Cb
volume and geometric cross-section, 329 cumulus, 2, 3, 9, 22, 41, 42, 46, 50, 358, 366, 382,
Community Atmosphere Model, 370, 387 387, see also Cu
Community Earth System Model, 376 Curtis–Godson approximation, see two-parameter
complex angular momentum theory, 166, 168 Curtis–Godson approximation
complex refractive index, 61, 73, 77, 115, 121, CWC, 29, see also cloud water content
165, 175, 200, 203, 255, see also refractive cylindrical coordinates, see coordinates
index
conductivity, 24, 61, 198, 200, 203, 363, see also D4S, 279, 280, 281, see also delta-four-stream
specific conductivity; thermal conductivity approximation
Index 431
DDA, 173, 178, 182, 197, 212, 233, 234, 237, 239, homogeneous and heterogeneous nucleation, 19
241, 243, 244, 246, see also discrete dipole laboratory scattering for, 186
approximation supercooled water, 22
degree of linear polarization, 93, 103, 176, 183, 225, two-channel inversion technique, 262, 264
237, 301, 302 dust
dehydration, 50, 382, 383, 386 atmospheric composition, 249
delta cloud configurations and dynamic and circulation
four-stream approximation, 341 patterns, 341
function, 110, 158, 206, 242 DDA, 239
transmission, 134, 157, 158 detection of thin cirrus, 287
dendrite, 24, 29, 35, 39, 41, 58, 125 mineral, 362
depolarization snow grains contaminated by, 192, 193, 195, 196
ratio, 104, 112, 235, 310, 311, 313, 314, 315, 316, view from satellite, 14, 16
318
technique, 11, 14, 16, 112, 315 Earth radiation budget, 345, 389
deposition, 5, 19, 20, 53, 56, 192, 193 Earth System Science Pathfinder Project, 258, 323
Descartes, 31, 117 Earth’s atmosphere, 1, 12, 46, 50, 106, 115, 116, 118,
dichroism, 113 134, 146, 175, 182, 247, 248, 250, 301, 323,
dielectric medium, 160 333, 350, 366, 390
dielectrics, 61 Earth’s orbit, 357
diffraction EBCM, 222, 223, 225, 228, 243, see also extended
Babinet, 117, see also Babinet’s principle boundary condition method
circular aperture, 134, 135, 141, 144, 171 ECMWF, see European Centre for Medium-Range
Fraunhofer, 134, 148, 155, 158, 171, 235 Weather Forecasts
hexagonal cylinder, 139, 140, 143, 144 Eddington’s approximation, 340
rectangular parallelepiped, 138, 143, 144 edge effect, 116, 117, 167, 170, 173, 214, see also
sphere, 135, 136, 144, 145 surface wave
spheroid, 140, 142, 144 effective emittance, 285
three-axis ellipsoid, 142, 144 effective particle size, 16, 264, 283, 296, 305, see also
diffusion, 22, 25, 29, 47, 53, 54, 56, 277, 360, 363, mean effective radius
366 effective size, see also mean effective size
dimethyl sulfide, 360 cirrus radiative forcing, 351
dipole moment, 75, 239, 240, 253 cloud-top and cloud-base, 289, 295
direct effect, 362 dispersion, 301
discrete dipole approximation, 173, 238, see also phase matrix, 182
DDA validation of retrieval technique, 266
discrete-ordinates method, see radiative transfer variance, 268
dispersion effective variance, 260, 263, 267
of light, 115, 333 efficiency factor, 172, 261
of size distribution, 272, 301 efficient geometric optics hybrid, 245, see also
divergence theorem, 242 EGOH
DMS, see dimethyl sulfide EGOH, 245, 246
DOE (Department of Energy), 13, 43, 49 electric
Doppler, 11, 335 charge density, 60
doubling method, 196, 264 current density, 60
droplet, see also water droplet dipole moment, 239, 240, see also dipole moment
1-D climate model, 357 displacement, 60, 75, 76
1-D cloud–precipitation–climate model, 359, field, 59, 75, 83, 86, 87, 91, 93, 96, 112, 118, 122,
361 149, 151, 152, 153, 159, 178, 201, 202, 206,
application to lidar, 315 222, 228, 239, 242, 308
backscattering depolarization, 309 vector, 60, 62, 68, 71, 77, 84, 87, 91, 93, 95, 122,
classification of ice crystals, 32 151, 207, 240, 307, 308, 310, 312, 313
contrails and contrail cirrus, 370 electromagnetic
definition of optical depth and mean effective size, equivalence theorem, 159
260, 261 field, 60, 158, 199, 201, 205
growth by accretion, 29 wave, 58, 61, 68, 76, 77, 83, 85, 86, 120, 121, 149,
growth by diffusion, 24 159, 197, 203, 205, 220
432 Index
electronic, 251, 253, 323 cross-section, 95, 97, 98, 107, 148, 157, 160, 161,
elevation angle, 130, 131 180, 189, 199, 261, 330
elliptical polarization, 87, 90, 91 efficiency, 75, 101, 148, 158, 162, 168, 173, 178,
ellipticity, 24, 87, 88, 89, 91, 93, 109, 301 184, 213, 237, 272, 333
Elsasser, 336 extraordinary wave, 78, 79
emission
absorption line, see absorption FAA, 374, 375, 380, see also Federal Aviation
absorption spectrum, 250, 251 Administration
AIRS, 280 fair-weather cumulus, 2
A-Train, 259 far field, 68, 71, 93, 96, 134, 135, 139, 146, 147, 158,
by non-black cirrus clouds, 350 159, 171, 178, 199, 207
cirrus cloud vertical profile, 293 far-infrared spectra, 210
clouds, 342 FDTD, 163, 173, 178, 180, 197, 199, 200, 201, 204,
dimethyl sulfide, 360 206, 207, 208, 209, 212, 214, 215, 218, 221,
downward thermal IR from high clouds, 244, 246, see also finite-difference time
373 domain
from underlying surface, 383 Federal Aviation Administration, 374, see also
line spectra for Earth remote sensing, 319 FAA
microwave, 386 feedback, 50, 51, 300, 327, 333, 351, 357, 359, 361,
NOx , 374 364, 365, 370, 373, 374, 377, 380, 382, 386,
radiative transfer equation, 107, 108, 254 390
water vapor from aircraft, 370 finite-difference time domain, 162, 163, 174, 198,
emissivity 292, see also FDTD
broadband, 336 FIRE, 41, 51, 264, 292, 341, 343, 344, see also First
cirrus, 373 ISCCP Regional Experiment
cirrus cloud, 259 first indirect effect, 361
cloud, 344 First ISCCP Regional Experiment, 41, 341
CO2 slicing method, 9 flux 346, see also flux density
downward, 343 broadband, 343, 346
infrared, 341 downward, 348, 357
k-coefficient, 280 eddy, 363
surface albedo at 3.75 µm, 293 forward calculation, 348
upward, 343 in the g space, 338
emittance, 283, 285 infrared, 335, 377
energy balance climate model, see climate interpretation of observed data, 344
model IR, 250, 341, 367
energy conservation principle, 107, 209 mass, 386
Environmental Protection Agency, 375 net, 355
EPA, see Environmental Protection Agency Poynting vector, 83, 120
equilibrium temperature, 250, 327, 336, 373, precipitation, 359
383 radiative, xiv, 50, 56, 102, 259, 331, 340, 341, 350,
ERB, 346, 389, see also Earth radiation budget 374, 382
ESSP, 258, 323, see also Earth System Science radiative transfer, 339
Pathfinder Project sensible and latent heat, 358
EUCREX (European Cloud and Radiation solar, 109, 110, 250, 254, 293, 332, 367
Experiment), 51, 303, 305 surface, 350
European Centre for Medium-Range Weather TOA, 328
Forecasts, 387 vertical profiles, 346
expansion coefficient, 55, 73, 222, 225, 230, 328, 331, water vapor, 383
333, 338, 340, 353 flux density, 55, 96, 339, see also flux
extended boundary condition method, 222 Forward Scattering Spectrometer Probe, 42
external reflection, 104, 114, 120, 146, 148, 151, 159, fossil fuel, 248, 249
162, 237 fountain theory, 382, 383
extinction Fourier
coefficient, 55, 113, 192, 261, 264, 281, 288, 299, spectral variable, 215
315, 318, 325, 329, 330, 331, 333, 341, 350, spectrometer, 319
353 terms, 283
Index 433
transform, 199, 206, 207, 210, 216 generalized, 146

transform spectrometer, 324 GOS approach, see GOS
four-stream approximation, see delta-four-stream ice optics, 127, 132
approximation IGOM, see improved geometric optics
Fraunhofer diffraction, see diffraction localization principle, see localization principle
freezing, 2, 19, 20, 22, 29, 53, 180, 382 modified, 292
frequency near-IR, 283
circular, 61, 76, 84 radiation pressure, see radiation pressure
CKD, 338 efficiency, 171
cloud formation, 371 radiative transfer, 266
constant, 58, 59 ray-by-ray approach, see ray-by-ray
domain, 205, 210 Snell’s and Fresnel’s laws, 118
electromagnetic wave, 203 surface wave, see surface wave
ice cloud, 281, 283 surface-integral, 147
in GHz, 252 unified theory, 178
microwave, see microwave geometric ray tracing, 124, 127, 133, 147, 313
near-field, 207 geometric shadow, 125, 136, 137, 138, 139, 140,
of all clouds and high clouds, 9 170
optical constants of H2 O, 80 geometric-optics and surface-wave, 168, 173, 195, see
resonant, 77 also GOS
response of scattering particle, 199, 206 Geophysical Fluid Dynamics Laboratory, 387
thin cirrus occurrence, 389 GEOS, see Goddard Earth Observing System Model
Fresnel Geostationary Operational Environmental Satellite, 5
coefficient, 121, 151, 165 GEWEX (Global Energy and Water cycle Exchanges),
formula, 120, 151 348
law, 118, 121 GFDL, see Geophysical Fluid Dynamics Laboratory
ray, 156, 158 Gibbs phenomenon, 216
reflection, 123, 126, 158, 159 global albedo, see albedo
Fresnelian interaction, 147 Global Change Observation Mission, 258
FSSP, see Forward Scattering Spectrometer Probe global climate model, xiv, 363, 364, 386
FTS, see Fourier transform spectrometer glory, 103, 104, 117, 127, 391
Glory satellite, 259, 302
g function, 321, see also correlated k-distribution Goddard Earth Observing System Model, 387
method GOES, 5, 347, 348, see also Geostationary
gamma size distribution, 103, 263, 273 Operational Environmental Satellite
gaseous profile, 108 GOS, see also geometric-optics and surface-wave
Gauss’ divergence theorem, 63, 161 comparison with FDTD and DDA, 173
Gaussian comparison with Lorenz–Mie, 169
probability function, 45, 127 conceptual ray-by-ray approach, 166
pulse, 206 coupled with Rayleigh–Gans–Debye, 173
system of units, 60 for ice crystals, 169
GCM, 46, 300, 332, 333, 340, 341, 350, 353, 361, for non-spherical particles, 171
365, 366, 367, 377, 380, 386, 387, 389, see for snow grains contaminated by BC and dust,
also general circulation model 195
GCOM-W1, see Global Change Observation Mission for spheres, 168
general circulation model, 45, 289, 332, see also radiation pressure, 168
GCM; climate model unified theory, 178, 184
geometric optics graupel, 22, 29, 30, 32, 315
absorption effects, 120 greenhouse effect, 350, 351, 360, 361, 362, 364, 373,
asymptotic approximation, 147, 149, 151, 153, 155, 377, 389, 390, see also IR greenhouse effect
157, 159, 161, 163, 165, 167, 169, 171, 173, greenhouse gas, xii, 248, 252, 346, 351, 357
175, 177 ground state, 253
asymptotic extinction efficiency, 237
CGOM, 162, 180, see also conventional geometric H2 O, 12, 17, 80, 248, 250, 253, 257, 264, 300, 319,
optics 323, 335, 338, 342, 350, 364, 379, 380
EGOH, see efficient geometric optics hybrid hail, 3, 29, 249
for ice crystals, 391 half-width, 77, 215, 335, 336
434 Index
halo polarization ratio and BRDF, 306

120° parhelion, 133 PSTD and FDTD, 217
22°, 117, 128 homogeneous
22° parhelion, 134 deposition, 20
46°, 117, 128 freezing, 19, 20, 53
anthelion, 133 nucleation, 19, 20, 51
circumhorizontal arc, 133 horizontal orientation, 59, 147, 303
circumzenithal arc, 133 Huygens surface, 199
diamond shape, 132 hydrogen, 17, 19
helic arc, 134 hydrological cycle, xii, 2, 248
infralateral arc, 134
lower tangent arc, 133 ICAO (International Civil Aviation Organization),
Parry arc, 133 375
Parry infralateral arc, 133 ice crystal
Parry supralateral arc, 133 habit, 25, see also ice crystal shape
subhelic arc, 133 morphology, 30, see also ice crystal shape
subsun, 132 ice crystal shape
sundog, 58, 131, 135 aggregate, see aggregate
supralateral arc, 133 as a function of temperature, 26
upper tangent arc, 134 bullet rosette, see bullet rosette
Wegener arc, 133 classification of ice crystals, 32, 33
Hankel function, 68, 74, 223 column, see column
harmonic wave, 160, 206 computer generation of 3-D ice crystals, 35, 41
He, see helium crystal size and shape distribution, 41
heat budget, 327 dendrite, see dendrite
heating rate, 50, 53, 192, 340, 355, 356, 382, 384 effect on GCM simulation, 367
helic arc, see halo growth of ice crystals, 22
helium, 313 hexagonal structure, 25, 31
Helmholtz equation, 223 historical perspective, 30
Henyey–Greenstein function, 331 hollow column, see hollow column
heterogeneous ice crystal size distribution, 41
deposition, 21, 54, see also deposition Koch snowflake, see Koch snowflake
freezing, 53, see also freezing needle, see needle
nucleation, 19, 20, 46, 366, 371 observed at South Pole, 33
processes, 363 observed from aircraft, 33
high cloud, xiv, 2, 5, 9, 247, 271, 286, 357, 362, 371, plate, see plate
372, 373, 389 ice nuclei, 2, 19, 20, 21, 22, 31, 53, 115, 175, 178,
High Resolution Infrared Spectrometer, 9, see also 180, 249
HIRS ice water content, see also IWC
High Spectral Resolution Infrared Spectrometer, 319 cirrus cloud vertical profiles, 288
HIRS, 9, 11, 389, see also High Resolution Infrared CloudSat, 13
Spectrometer correlation
HIS, see High Spectral Resolution Infrared analysis, 46
Spectrometer with ice crystal size, 45, see also IWC
hollow column with mean effective ice crystal size, 47, 48, 49,
classification of ice crystals, 32 353
computer generation of 3-D ice crystals, 35, 38, 41 definition of, 261
crystal size and shape distribution, 42, 43, 49, 266, reflected line spectra in the 0.76 µm band, 325
271, 292, 319, 344 single-scattering parameterization of ice crystals,
depolarization ratio, 316 329
geometric ray-tracing, 125 surface temperature deviation due to CO2 doubling,
ice optical property library, 277 365
IGOM and FDTD, 163 ice water path, xiv, 55, 255, 261, 332, 343, 345, 352,
II-TM and ADDA, 234 see also IWP
laboratory scattering, 189 IGOM, see improved geometric optics
phase matrix, 182 imaginary refractive index, 77, 80, 156, 330, 333, see
polarization measurement, 303 also refractive index
Index 435
improved geometric optics, xiii, 158, 173, 178, 219, vertical

229, 266, 277, 350 profile of extinction coefficient, 299
incident field, 118, 152, 199, 225, 239, 240, 242 sizing, 292
independent scattering, 58, 59, 100, 106, 112, 228 IWC–De correlation, 47, 368
indirect effect, 46, 366, 379, 380 IWP, 343, 344, 351, 353, 355, 362, 363, 364, 389, see
infralateral arc, see halo also ice water path
infrared
flux, 250, 335, 358, 363, 377, see also flux jet fuel consumption, 372
radiation, xiii, 319, 350, see also thermal infrared
radiation k-distribution, 336, see also CKD; correlated
radiative transfer, 112, 336, 340 k-distribution method
spectrum, 302, 331, 335 Kepler, 31
Infrared Radiation Interferometer Spectrometer, 319, Kirchhoff thermodynamic equilibrium, 342, see also
see also IRIS thermodynamic equilibrium
insolation, 348, 349 Koch snowflake, 35, 39, 40, 184, 316
integral equation, 138, 160, 207, 225, 228, 241 Kramers–Kronig relationship, 80
intensity function, 136, 139, 142, 145
Intergovernmental Panel on Climate Change, 371 laboratory scattering, 186, 192
internal reflection, 104, 114, 117, 124, 132, 146, 149, Lambert, 301, 341
159, 183, 237, 308, 310, 313, 315 land surface, 2, 252, 270, 271, 286, 301, 366, 386
International Satellite Cloud Climatology Program, Landsat, 262
11, see also ISCCP lapse rate, 359, 363, 382, see also critical lapse
Intertropical Convergence Zone, 8, see also ITCZ rate
invariant imbedding T-matrix method, 226 laser radar, see lidar
inversion, see retrieval; remote sensing latent heat, 23, 24, 46, 53, 358, 360, 363
IPCC, 5, 371, 375, 377, 380, 391, see also law of reciprocity, 98, 100, 101, 113, 154
Intergovernmental Panel on Climate Change Legendre polynomial, 67, 283, 331, 340
IR flux, 250, 341, 346, 357, 367, see also flux lidar
IR greenhouse effect, 360, 361, 364, 373, 377, 390, 3-D backscattering image of cirrus, 12
see also greenhouse effect active remote sensing, 247
IRIS, 319, see also Infrared Radiation Interferometer backscattering technique, 11
Spectrometer CALIOP, 259, 316
irradiance, 250, see also solar irradiance CALIPSO, 14, 376, 381, 382
ISCCP, 11, 41, 43, 51, 341, see also International cloud base height, 347
Satellite Cloud Climatology Program cloud geometric thickness, 267
ITCZ, 8, 9, 279, 281, 285, see also Intertropical depolarization technique, 14, 16
Convergence Zone detection of cirrus clouds microphysics, 184
IWC, see also ice water content differentiation of ice and water, 306, 316
1-D climate–cirrus cloud model, 362, 363, 364, equation, 315
365 from aircraft, 257
2-D cirrus cloud formation model, 55, 56 ground-based, 45, 279, 316
as a function of cirrus cloud temperature, 357, 358 II-TM, 235
cirrus radiative forcing, 353, 355 laboratory (laser) and field studies, 315
correlation with ice crystal size, 45, 46, 49 laser radar, 314
definition of, 261 MicroPulse, 348
effect of interactive ice crystal size on precipitation, orientation of ice particles, 112
365, 366, 367 PARASOL, 259
feedback, 374 retrieval of re , 283
generalized single-scattering parameterization, 332 subvisible cirrus, 383
high thin clouds, 388, 389 theoretical depolarization ratio, 316
radiative transfer versus aircraft observations, 343 vertical inhomogeneity, 288
retrieval of cirrus cloud parameters, 325, 326, see limb, 5, 114, 259, 386
also retrieval line
single-scattering parameterization of ice crystals, intensity, 335
328, 330, 331 overlap, 338, 339
thin cirrus detection and comparison with GCM, shape, 335, 336
386, 387 strength, 77, 318, 335, 336
436 Index
linear polarization, 77, 90, 110, 184, 301, 302, 303, 1-D climate–cirrus cloud model, 364
304, see also degree of linear polarization asymmetry factor, 273
linearly polarized radiance, 276 bi-spectral correlation technique, 264
line-by-line, 336 BRDF in the 1.38 µm band, 319
liquid cirrus cloud climatology, 386, 389
water content, 260, 359, 360, see also LWC construction of look-up tables, 267
water path, 260, see also LWP correlation to IWC, 353
local thermodynamic equilibrium, 109 CRF
localization principle, 146, 158, 164 at TOA, 352, 354
longwave radiation, 346, see also outgoing longwave in the 1.38 µm band, 321
radiation definition of, 42, 261, 329
Lorentz–Lorenz, 240 emissivity and solar albedo, 344
Lorenz–Mie, 216 global retrieval, 277
amplitude scattering matrix, 224 ice crystal surface roughness, 269
bidirectional reflectance, 257 in GCMs, 46, 366
comparison with linear contrails, 376
DDA, 243, 244 radiation parameterization, 55
experimental results, 186, 187 radiative heating rate, 384
FDTD, 210, 211 relation to IWC, 47
GOS, 169 retrieval, 261
PSTD, 216 single-scattering parameterization, 334
exact “solution,” 115, 118, 128, 166, 168, 171, 178, solar albedo, 344
197 spectral BRDF and absorptance, 255
invariant imbedding T-matrix method, 227 tropical ice crystal size distribution, 264
polarization of Venus, 301 UCLA GCM, 289
single-scattering properties of ice spheres, 102, mean effective radius, 188, 260, 262, 264, 277, 279,
103 289, 329, 360, see also effective size; effective
spherical coordinates, 64, 68 particle size
spheroidal coordinates, 72 mean effective size, 47, 55, 56, 101, 187, 257, 263,
water droplets, 350 271, 274, 282, 295, 305, 322, 330, 333, 343,
low cloud, 2, 271, 288, 359, 389 356, 359, 363, 365, see also effective size;
lower effective particle size; mean effective ice
atmosphere, 2, 249, 251 crystal size
tangent arc, 114, 132, see also halo meridional plane, 253
LWC, 260, 360, 361, see also liquid water content Mesoscale Atmospheric Prediction System, 346, see
LWP, 260, 261, 361, see also liquid water path also MAPS
METEOR, 319
magnetic, 118, 201 Meteosat, 377
component, 120 methane, 248, 259, see also CH4
current, 159 MFRSR, 348, see also Multi Filter Rotating
dipole, 253 Shadowband Radiometer
field, 60, 65, 68, 73, 76, 113, 134, 159, 160, 201, microwave, 11, 83, 211, 247, 250, 252, 259, 279,
205, 225 386
induction, 60, 62 Microwave Limb Sounder, 259, 386, see also MLS
permeability, 61, 62 middle cloud, 2, 5, 8, 357, 360, 371, 372, 373
poles, 60 Miller indices, 17, 19, 25
vector, 60, 62, 70, 83, 106, 115, 311 millimeter wave cloud radar, 295
MAPS, 346, 347, 348, 349, see also Mesoscale millimeter-wave radar, see radar, mm-wave
Atmospheric Prediction System Mixed-Phase Arctic Cloud Experiment, 43
MAS, 264, 266, 267, 271, 286, 306, 307, see also mixing ratio, 52, 53, 343, 360, 366, 384
MODIS Airborne Simulator MLS, 259, 386, 387, 388, 389, 390, see also
mass extinction cross-section, 107 Microwave Limb Sounder
Maxwell Garnett mixing rule, 165, 180 MMCR, 295, 297, 299, 300, see also millimeter wave
Maxwell’s equations, 58, 61, 75, 160, 199, 215, 223, cloud radar
391 mm-wave radar, see radar, mm-wave
mean effective ice crystal size, see also mean effective MODerate-resolution Imaging Spectroradiometer, 11,
size 14, see also MODIS
Index 437
MODIS, see also MODerate-resolution Imaging multiple scattering, xiii, 58, 59, 79, 106, 107, 108,
Spectroradiometer 109, 110, 112, 188, 257, 263, 281, 301, 314,
1.38 µm reflectance, 8.6–11 µm BTD, 13 319, 321, 323
1.9 µm water vapor absorption channel,
286 N2 O, 249, 251, 252, 335, 338, 339, 343, see also
Aqua, 259 nitrous oxide
A-Train, 14 Nakaya diagram, 32, see also ice crystal shape
BTD National Polar-orbiting Operational Environmental
between 11 and 12 µm channels, 14 Satellite System, 295, see also NPOESS
between 8.5 and 11 µm channels, 275 natural light, 93, 97
C5 and C6 models, 273 near field, 147, 160, 198, 205, 207
C6 model, 275 near infrared, xiii, 261, 344
CERES, 283 near-to-far-field mapping, 207
cirrus cloud vertical sizing, 289 needle, 24, 29, 32
cirrus scenes, 295 nephelometer, 187, 188, 190, 191
cloud climatology, 11 nimbostratus, 3, 4
cloud effective particle size, 17 Nimbus, 319
cloud retrieval algorithm, 294 nitrogen, 186, 193, 248
Collection 4 products, 271 nitrogen oxides, 249, 259
collocation scheme with MMCR, 299 nitrous oxide, 248
detection of thin cirrus, 287 NO, 249
difference in cloud particle size derived from NO2 , 249, 339
MMCR, 300 NOAA, 5, 9, 301, 302, 346, 374, 375, 389
differences in ice cloud optical depth derived from normal dispersion, 77
PARASOL, 302 NOx , 249, 374, 380, see also nitrogen oxide
false-color RGB image, 270 NPOESS, 295, 303, see also National Polar-orbiting
generation of look-up library, 292 Operational Environmental Satellite
global cloud climatology, 277 System
input parameters extracted from, 291 Ns, see nimbostratus
IR optical properties of cirrus, 280 nucleation, 5, 19, 20, 31, 51, 53, 56, 186, 187, 192,
linear contrails, 375 310, 363, see also heterogeneous nucleation;
MMCR observation, 299 homogeneous nucleation
optical depth and ice crystal size, 260 nuclei, 5, 20, 21, 51, 178, see also cloud condensation
product evolution (C6 versus C5), 388 nuclei; ice nuclei
retrieval of contrail properties, 376
seasonal ice cloud optical depth and ice crystal O2 , 12, 247, 248, 249, 251, 252, 258, 264, 323, 324,
mean effective radius, 279 335, 338, 339
solar channels, 263 O3 , 249, 250, 251, 257, 335, 338, 339, 346, 374, 379,
surface albedo at 3.75 µm, 293 380
Terra, 269 OCO, 251, 258, 319, 323, see also Orbiting Carbon
tropical ice crystal size distribution, 264 Observatory
two-channel inversion, 262 OLR, 346, 348, 350, 367, 368, 377, see also outgoing
validation of cloud product, 296 longwave radiation
water clouds, 264 OMI, 259, see also Ozone Monitoring Instrument
weighting function, 290 one-parameter scaling approximation, 336
MODIS Airborne Simulator, 393 optical depth, definition of, 260
modulator, 92 OPTRAN, 279
monochromatic intensity, 107 Orbiting Carbon Observatory, 258, 323, see also OCO
monochromatic radiative transfer, 321, 338 ordinary wave, 78, 79
Monte Carlo, xiii, 124, 126, 127, 146, 147, 155, 170, outgoing longwave radiation, 346, 349, see also OLR
195 outgoing wave, 199, 204
MPACE, 43, 49, see also Mixed-Phase Arctic Cloud overtone, 318
Experiment oxygen, xiv, 17, 19, 248, 249, 319, 323, 325, see also
Mueller matrix, 92 O2
Multi Filter Rotating Shadowband Radiometer, ozone, 249, 255, 259, 336, 343, 355, 366, 384, 390,
348 see also O3
multiple reflection, 293 Ozone Monitoring Instrument, 259
438 Index
PARASOL, 259, 275, 302, see also Polarization & plane wave, 68, 71, 84, 87, 118, 119, 135, 136, 139,
Anisotropy of Reflectances for Atmospheric 198
Sciences coupled with Observations from a plane-parallel atmosphere, xiv, 106, 108, 339
Lidar planetary albedo, see albedo
parhelic circle, see halo planetary boundary layer, 250, 363, 366, see also PBL
parhelion, see halo plate
Parry arc, 131, 133, 134, see also halo backscattering depolarization, 312
particle size distribution, 271, 272, 329, 359, see also classification, 32, 33
size distribution comparison of GOS with FDTD and DDA, 173
PARTNER, 374 comparison with
Partnership for AiR Transportation Noise and experiment, 189
Emissions Reduction, see PARTNER Koch snowflake, 184
PBL, 250, 366, see also planetary boundary layer computer generation of 3-D ice crystals, 35, 39
perfectly matched layer, 198, 204 contrails and contrail cirrus, 371
permeability, 61, 75, 198, see also magnetic correlation between BRDF and polarization ratio,
permeability 305
permittivity, 61, 62, 75, 77, 165, 198, 200, 203, 204, crystal size and shape distribution, 41, 42, 43, 49,
241 266, 268, 271, 272, 292, 319, 344
phase function, see also scattering phase function depolarization ratio, 316
3-D distribution of scattered intensity, 108 detailed drawing, 31
ADDA, 243 extinction cross-section, 330
backscattering, 315 FDTD, 214
bullet rosette, 214 and PSTD, 220
columns and plates, 189 for snow grains, 195
delta-function adjustment, 340 Fraunhofer diffraction, 148
enhancement of 22° halo, 235 generalized parameterization, 332
FDTD, 215 geometric ray-tracing, 124
and Lorenz–Mie, 210 geometrical similarity with column, 311
for horizontally oriented ice particles, 111 GOS, 171
for polydisperse ice spheres, 103 coupled with RGD, 173
for randomly oriented spheroids, 104 growth by riming, 29
four size distributions, 267 halos and arcs, 132
Henyey–Greenstein function, 331 horizontal orientation, 45, 112, 114, 129, 131, 303,
IGOM and FDTD, 163 310
II-TM and DDA, 233 ice crystal habit, 25, see also ice crystal shape
importance of ice crystal, 267 ice optical property library, 277
laboratory measurements, 186 ice particle surface roughness, 269
Legendre polynomial expansion, 331 IGOM, 163
MODIS C5 and C6 models, 273 II-TM, see invariant imbedding T-matrix method
normalization, 188 II-TM and ADDA, 234
optically thin cirrus, 254 measured and computed BRDF, 257
particle-ensemble-averaged, 271 phase matrix, 182
PSTD, 216, 220 polarization measurements, 303
Rayleigh scattering, 338 polarization ratio and BRDF, 305
ray-tracing technique, 146 relation between maximum size and length, 45
RGD approximation, 176 scattering and absorption properties, 344
rough surfaces, 183, 269 single-scattering input to radiative transfer, 264
smooth surfaces, 182 spectral single-scattering, 328, 333, 334
three effective variances, 268 sundog, 130
three habit models, 268 unified theory, 182
T-matrix, 225 volume and geometric cross-section, 329
phase matrix, see scattering phase matrix PML, 198, 199, 204, 205, see also perfectly matched
photon, 125, 126, 170, 195, 196, 289, 291, layer
323 Poincaré sphere, 90
Planck, 107, 108, 110, 250 polar nephelometer, see nephelometer
function, 55, 109, 280, 335, 336 polar stratospheric cloud, 5
plane of stratification, 108 polarizability, 75, 175, 238, 240, 243
Index 439
Polarization & Anisotropy of Reflectances for budget, 14, 327, 345, 346, 373, 390
Atmospheric Sciences coupled with pressure, 165, 167, 168, 171
Observations from a Lidar, 259 radiative equilibrium, 383
POLarization and Directionality of the Earth’s radiative forcing, 192, 251, 286, 328, 333, 344, 351,
Reflectances, 259 353, 355, 358, 359, 360, 362, 365, 375, 376,
polarized light, 93, 236, 312 377, 386, 389, 390, see also cloud radiative
polarizer, 92, 313 forcing
POLDER, 259, 275, 276, 301, 302, 303, 304, 305, see radiative heating, 50, 53, 57, 289, 355, 359, 382, 384,
also POLarization and Directionality of the 385
Earth’s Reflectances radiative transfer
polydisperse, 103, 146 1.38 µm H2 O band, 319
potential temperature, 52, 366, see also temperature AIRS data, 279
Poynting vector, 84, 120, 156, 160, 209, 310, 311 bi-spectral correlation technique, 264
precipitation, xiv, 2, 5, 8, 56, 192, 248, 249, 259, 327, CAGEX, 346
355, 359, 360, 361, 362, 365, 367, 380, 387, CAGEX biases for LW flux, 349
390 CERES-2, 283
principle of reciprocity, 229, see also law of cirrus cloud formation, 55
reciprocity cirrus radiative forcing, 351
PSC, see polar stratospheric cloud correlated k-distribution method, 257
pseudo-spectral time domain method, 215, see also delta-four-stream approach, 332
PSTD discrete-ordinates method, 339
PSTD, 215, 216, 217, 220, 221, 244, 245 Eddington’s approximation, 340
equation, 108
quarter-wave plate, 92, 301 flux gain at the surface, 374
for oriented and complex ice crystals, 339
radar for tropical anvil, 382
94-GHz, 259, see also CloudSat Fu–Liou code, 348
Cloud Profiling Radar, see Cloud Profiling Radar in horizontally oriented ice particles, 112
cloud radar system, see cloud radar system in plane-parallel atmosphere, 108
cloud reflectivity, 13, 307, 387 in the g-domain, 321
CloudSat k-coefficient approach, 280
vertical inhomogeneity, 295 Legendre polynomial, 331
lidar (or radar) equation, 315 model results compared with satellite observations,
mm-wave, 11, 12, 13, 287, 295, 297, 299, 389, see 345
also millimeter-wave radar model results versus aircraft observations, 341
retrieval of re , 283 MODIS channels, 265
validation, 280 multiple scattering, 106
vertical inhomogeneity, 288 offline simulation, 370
windows, 12 radiative forcing analysis, 389
radiance radiative heating, 53
3.75 µm, 291, 294 reflected
a look-up library, 293 intensity, 254
clear, 281 polarization, 303
cloud-cleared, 279, 280 retrieval of cirrus cloud parameters, 326
cloudy, 279, 280 solar reflection techniques, 254
incident, 280 vector, 109
Level-1B, 275 vertical sizing algorithm, 296
measured, 294 radiative–convective model, 362, 363, 365, see also
polarized, 276, 301 climate model
simulated, 293, 295 radiometer, 247, 257, 259, 262, 275, 302, 323, 346,
solar, 285, 289, 293 386, 389
spectral, 280, 346 rainbow, 31, 103, 104, 116, 117, 127, 188, 257, 303
total, 291, 293 random orientation, 129, 154, 177, 214, 224, 310,
transmitted, 280 311, 313
upwelling, 280, 319 ray tracing, see geometric ray tracing
radiation ray-by-ray, 160, 163, 168, 170, 213, 245
balance, 285 Rayleigh scattering, 175, 177, 252, 289, 293, 301, 337
440 Index
Rayleigh–Gans–Debye approximation, 173, see also rotational

RGD band, 251, 336
Rayleigh–Wait theory, 72 matrix, 151, 153, 308
RBRI, see ray-by-ray RSP, 302, 306, see also Research Scanning
real refractive index, 77, 80, 115, 169, 175, 180, see Polarimeter
also refractive index
reciprocity, see law of reciprocity; principle of SARB, 346, see also surface and atmospheric
reciprocity radiation budget
rectangular parallelepiped, 136, 138, 139, 144, 171 SARTA, 280, see also Stand-Alone clear-radiance
reflected intensity, 254 Radiative Transfer Algorithm
reflection Sc, see stratocumulus
coefficient, 126 scaling approximation, see one-parameter scaling
function, 253, 254, 325 approximation
matrix, 151, 309 scattering amplitude matrix, 208, 209, 224
reflectivity, 13, 262, 276, 280, 287, 306 scattering angle, definition of, 92
refraction matrix, 152 scattering matrix, 152, 155, 163, 207, 208, 209, 224
refractive index, 79, 80, 81, 117, 118, 119, 120, 128, scattering phase function, 214, 267, 283, 325, see also
149, 152, 156, 168, 180, 186, 210, 211, 212, phase function
216, 223, 225, 227, 228, 236, 237, 240, 243, scattering phase matrix
244, 261, 301, see also complex refractive 10 independent parameters, 229, 233
index; imaginary refractive index; real ADDA, PSTD, and EGOH, 246
refractive index circular-polarization representation, 229
remote sensing, see also retrieval contribution of diffraction, 156
active, 247 definition, 94
AIRS, 279 FDTD, 212
atmospheric absorption spectrum, 250 for an ensemble of ice crystals, 97, 99, 100, 101,
atmospheric composition and structure, 250 146, 157, 182
A-Train satellite constellation, 257 for horizontally oriented ice particles, 113
CALIPSO, see Cloud-Aerosol Lidar and Infrared for randomly oriented soot aggregates, 176
Pathfinder Satellite Observations IGOM, 229
CloudSat, see CloudSat II-TM method, 229, 230
delta-four-stream, 279 near-to-far-field, 207, 209
ice clouds, 248, 254 RGD approach, 177
ice crystal phase function, 268 vector radiative transfer, 109, 110
lidar, 11 scattering plane, 92, 107, 152, 153, 154, 207, 241,
MODIS, 275 276, 308
optical depth and mean effective size, 260 Schelkunoff, see electromagnetic equivalence
passive, 247 theorem
POLDER, 302 Schuster, 106, 340
radiative transfer, 254 Schwarzschild, 106, 340
reflected line spectra, 324, 326 sea surface temperature, 259, 300, 366, see also SST
reflected polarization, 248, 301, 303 second indirect effect, 361, 362
satellite, 259 semiconductors, 61
validation of the vertical sizing, 300 semi-direct effect, 362
vertical sizing, 292 sensible heat, 363
representation of a light beam, 301 separation of variables, 228
Research Scanning Polarimeter, 302, see also shortwave radiation, 370
RSP SIM, 207, 208, see also surface-integral method
resonance, 180, 253 similarity, 32, 103, 308, 311, 332, 340
retrieval, 253, 260, 262, 263, 264, 266, 267, 268, 271, single-scattering albedo
275, 277, 279, 280, 281, 283, 286, 289, 291, aerosol, 348
292, 293, 295, 296, 297, 300, 302, 303, 306, combined ice crystals and absorbing gases, 338
326, 348, 386, 387, 388, 390, see also remote comparison with FDTD and DDA, 173
sensing conventional geometric optics, 331
RGB (red, green, blue), 14, 16, 269, 270, 271 definition of, 101, 330
RGD, 176, 177, 178, 180, 184, see also effect of water vapor, 325
Rayleigh–Gans–Debye approximation FDTD, 199
Index 441
and ray-by-ray, 213 spectrum, 250, 285, 333, 338, 339

and PSTD, 220 zenith angle, 114, 131, 255, 257, 263, 276, 283,
functions of only De , 353 293, 304, 305, 321, 323, 324, 325, 341, 344,
generalized single-scattering parameterization, 333 352, 353, 355, 384
geometric ray tracing, 127 solid angle, 95, 106, 108, 109, 127, 159, 316, 346
Glory satellite, 259 soot, 147, 176, 180, 193, 362, 370, 379, 380, see also
GOS, 169, see also geometric-optics and BC; black carbon
surface-wave source function, 107, 108, 109, 110
horizontal orientation, 113 specific conductivity, 61
ice sphere, 102 specific humidity, 52, 360, 363, 373
II-TM and IGOM, 237 spectral single-scattering parameterization, 328
Koch snowflake, 184 spectral transmittance, 335, 336, 337
N homogeneous layers, 341 spectrometer, 247, 258, 319, 323
normalized phase function, 158 spherical
parameterization using mean effective size, albedo, 264, 293
55 coordinates, see coordinates
particle-ensemble-averaged, 272 harmonic, 67
radiative particle, 31, 94, 95, 100, 101, 110, 115, 165, 166,
forcing of cirrus clouds, 350 167, 175, 184, 261, 301, 344, 362
transfer equation, 108, 254 wave function, 223, 225
surface wave, 172 spheroidal coordinates, see coordinates; Lorenz–Mie
two-channel inversion, 264 split-window, 283
unified theory, 178, 179, 181, 184 SST, 366, see also sea surface temperature
size distribution, 42, 56 St, 2, see also stratus
aerosol, 348 Stand-Alone clear-radiance Radiative Transfer
analytic, 268 Algorithm, 280, see also SARTA
counted crystals, 189 standard atmosphere, 2, 251, 252, 255, 355, 359, see
droplet, 187, 260, 360 also U.S. Standard Atmosphere
dust aerosol, 193 STEP, 382, see also Stratosphere-Troposphere
for contrail, 43 Exchange Project
for spherical particles, 145 Stefan–Boltzmann constant, 383
gamma, see gamma size distribution Stokes parameters, 86, 87, 89, 90, 91, 93, 94, 95, 99,
ice crystal, 41, 46, 47, 51, 52, 53, 55, 114, 261, 264, 100, 109, 155, 157, 301, 303
267, 292, 319, 329, 330, 332, 344, 351, 362, stratocumulus, 2, 262, 341
363 Stratosphere-Troposphere Exchange Project, 382
in situ, 49 stratus, 2, 52, 289, 359, see also St
log-normal, 188 subhelic arc, see halo
particle, see particle size distribution sublimation, 19, 53, 56, 289, 367
vertical, 296 Subsonic Aircraft: Contrail and Cloud Effect Special
size parameter, definition of, 102 Study, 370, see also SUCCESS
Snell’s law, 118, 120, 123, 126, 128, 129, 148, 149, subsun, 114, 131, 303, see also halo
310 SUCCESS, 42, 51, 266, 370, 374, see also Subsonic
snow grain, 192, 193, 196 Aircraft: Contrail and Cloud Effect Special
snowflake, 21, 29, 30, 31, 32, 39, 145, 184, 249, 315, Study
see also Koch snowflake sulfur dioxide, 249
S-NPP (Suomi-National Polar-orbiting Operational sundog, see halo
Environmental Satellite System (NPOESS) supralateral arc, see halo
Preparatory Project), 280, 295, 296, 305 surface
SO2 , 249, 339, see also sulfur dioxide albedo, see albedo
solar and atmospheric radiation budget, 346
albedo effect, 360, 361, 364, 365 flux, see flux
flux, 339, 341, 342, 346, 358 integral method, 134, 207
heating rate, 355, see also heating rate reflection, 262, 319
insolation, 357, 371 temperature, see temperature
irradiance, 276, 323, 346, see also irradiance wave, 116, 117, 118, 166, 168, 170, 171, 172, 180,
radiation, 111, 113, 180, 255, 257, 280, 293, 339, 308, 350
342, 352, 353, 356, 359, 362, 367 susceptibility, 240
442 Index
temperature radiative balance equation, 383

1-D climate–cirrus cloud model, 364 rate coefficients, 360
2-D cirrus cloud model, 53 Rayleigh optical depth, 338
at crystal surface, 24 reference, 336
atmospheric, 259, 343, 359 sea water, 259
brightness, 270, 271 secondary ice particle production, 22
classification of ice crystals, 32, 33, 41 slope of linear fit of, 373
cloud, 285 snowflake shape, 31
cloud top, 11, 271, 277 sounding, 346
cloud–climate model, 373 standard atmosphere, 364, see also standard
correlation with De , 49 atmosphere
crystal size versus temperature range, 42 stratosphere dehydration, 386
decreasing with height, 5 surface, 8, 24, 250, 291, 293, 350, 359, 360, 364,
dependence of 365, 371, 373, 390
conductivity on, 61 to determine De , 46
depolarization ratio on, 316 triple point, 23
ice crystal size on, 358, 367 tropopause, 383
ice refractive index on, 80, 83 tropopause layer, 380
IWC on, 357 tropopause region, 383
line profile on, 337 upper atmosphere, 250
line strength on, 335 vapor pressure over ice at, 25
dependent contrail parameterization, 376 Terra, 11, 13, 14, 269, 277, 283, 284, 285, 296, 297,
depolarization as a function of, 316 376, see also MODIS
depolarization in relation to, 315 TES, see Tropospheric Emission Spectrometer
differential between cloud and surface, thermal
357 conductivity, 24, 363
distributions, 353 equilibrium, 107
field, 355 infrared radiation, 57, 106, 112, 350, 371, see also
freezing, 2 infrared radiation
greenhouse radiative perturbation, 358 thermodynamic equilibrium, 336, 342, see also
heterogeneous nucleation, 20 Kirchhoff thermodynamic equilibrium
homogeneous thin cirrus clouds, 386
freezing, 22 Tibetan Plateau, 193
nucleation, 19 time harmonic, 84, 87
horizontal averages, 363 T-matrix, 176, 178, 197, 216, 222, 223, 224, 225, 226,
horizontal distribution, 381 227, 228, 237, 243
ice crystal, 54 TOA, 250, 254, 276, 280, 281, 283, 285, 328, 339,
ice crystal habit, 25 345, 348, 350, 351, 352, 354, 358, 370, 373,
ice crystal shape, 55 389, see also albedo; flux
ice crystal size, 42 total reflection, 124, 149
ice crystal size and shape, 33 transformation matrix, 92, 94, 95, 98, 99, 109, 139,
ice nuclei concentration, 54 153, 154, 156, 307, 308, 311
infrared flux as a function of, 363 transmission, 79, 80, 113, 123, 126, 155, 226, 251,
land surface, 366 252, 293, 308, 341, 357
linear contrails, 376 function, 292, 325
moderated, 371 matrix, 308
negative correlation with depolarization, 317 transmissivity, 281
nucleation, 19, 186 transmittance, 279, 280, 325, 335, 336
nucleation threshold, 20 transverse vector, 149
of planetary atmosphere, 336 triple point, 23
photospheric, 250 TRMM, 11, 42, 283, see also Tropical Rainfall
plate habit, 189 Measuring Mission
positive correlation with IWC, 389 tropical anvil, 42, 50, 382, see also anvil
potential, 52, see also potential temperature Tropical Rainfall Measuring Mission, 11, 42, see also
pristine and irregular habits, 49 TRMM
probability of adhesion, 29 tropopause, 249, 250, 357, 380, 382, 383, 385, 386,
profile, 250 387
Index 443
troposphere, 3, 5, 249, 250, 252, 300, 323, 328, 335, VIM, 207, 208, see also volume-integral method
346, 355, 357, 359, 371, 373, 380, 382, 383, VIPS, 42, 47, see also video ice particle sampler
385, 386, 387, 389 VIRS, 283, 285, see also Visible and Infrared
Tropospheric Emission Spectrometer, 259 Scanner
two-parameter Curtis–Godson approximation, 336 visibility, 3, 250
two-stream approximation, 340 Visible and Infrared Scanner, 11, 283, see also
VIRS
U.S. Standard Atmosphere, 293, 343, see also visible spectrum, 59, 262
standard atmosphere Visible-Infrared Imaging-Radiometric Suites, 295
UCLA AGCM, 341, 366 VISST (Visible Infrared Shortwave-infrared
ultraviolet radiation, 249, see also UV Split-window Technique), 283
uncinus, 43, 264 volcanic dust, 249, see also dust
uniaxial crystal, 78, 79 volume-integral method, 207
unified theory for light scattering by ice crystals, 178,
179, 181, 184, 186, 192, 316 water droplet, xiv, 2, 59, 134, 184, 187, 188, 248, 260,
unit vector, 36, 63, 65, 73, 118, 121, 148, 149, 150, 261, 264, 279, 283, 308, 314, 315, 329, 350,
152, 153, 159, 160, 162, 207, 208 359, 362, see also droplet
unpolarized light, 93, 313 wave equation, 59, 115, 122, 205
upper scalar, 65, 66, 68, 70, 71, 73
atmosphere, 5, 50, 383 vector, 58, 62, 64, 73
tangent arc, see halo wave vector, 121, 122
troposphere and lower-stratosphere, 380 wavelength, 61
upwelling radiance, 280, 281, 319 wavenumber, definition of, 61
UTLS, 371, 380, 383, 386, see also upper-troposphere Weather Research Forecast, 341
and lower-stratosphere Wegener arc, see halo
UV, 59, 252, 257, see also ultraviolet radiation weighting function, 290, 291
wet deposition, 192
velocity of light, 60, 75, 76, 77 window, 9, 12, 253, 275, 279, 280, 281, 289, 319, 327,
vibrational 339
mode, 17 WRF, see Weather Research Forecast
rotational transition, 17, 323 Wu-Shing, 30
transition, 251, 318
video ice particle sampler, 42 Yin-Yang, 30
VIIRS, 295, 296, see also Visible-Infrared
Imaging-Radiometric Suites zenith angle, definition of, 107

Light Scattering by Ice Crystals - Fundamentals and Applications PDF

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Light Scattering by Ice Crystals - Fundamentals and Applications PDF

Hochgeladen von

Copyright:

Verfügbare Formate

LIGHT SCATTERING B Y ICE CRYSTA LS

Fundamentals and Applications

kuo-nan liou is a Distinguished Professor of Atmospheric Sciences and founding

Cambridge University Press is part of the University of Cambridge.

This publication is in copyright. Subject to statutory exception

To all the happy ice crystals in planetary atmospheres

Let there be light.

Preface page xii

1.8 Numerical modeling of cirrus cloud formation 50

3.1.4.3 Horizontally oriented plate and column crystals 131

4.3 Discrete dipole approximation 238

6 Application of light scattering by ice crystals to climate studies 327

In Chapter 2, “Fundamentals of light scattering by ice crystals,” a number of fundamental

“Let there be light,” and there was light.∗

1.1 Introduction to clouds

The pavilion of Heaven is bare,

1.2 Geographical and temporal distributions of ice in the Earth’s atmosphere

1.2.2 An example of global cloud climatology

pixels by measured radiances is a complex decision-making process based on computer

1.2.3 View from ground-based instruments

is a three-dimensional backscattering image of cirrus clouds by a scanning lidar system

(a) mm-wave radar reflectivity dBZe

18:00 Time (UTC) 17:00

1.2.4 View from specific satellite instruments

(b) RGB (c) τ3 (Ocean)

Longitude (oE) Longitude (oE)

1.3 Formation of ice crystals

1.3.1 The structure of ice crystals

1.3.2 Homogeneous and heterogeneous nucleation

deposition appears to be impossible because liquid droplets will nucleate homogeneously

1.3.3 Atmospheric ice nuclei

(c) Contact Freezing

(d) Immersion Freezing

1.3.4 Secondary ice particle production

1.4 Growth of ice crystals

fK = L2I /(KRv T 2 ), fD = Rv T /[Dei (T )], (1.4.5)

1.4.2 Growth by accretion

1.5 Illustration of ice crystal morphology

1.5.2 Classification of ice crystals

1.5.3 Computer generation of three-dimensional ice crystals

number of computer-generated 3-D ice crystal shapes introduced in previous sections,

where n = 1, . . . , 6. The parameter α is an angle between a unit vector to an upper

cos[(n − 7)π]z − c = 0, n = 7, 8. (1.5.6)

n=1 n=2 n=3 n=4

Side View Top View Side View Top View

Side View Top View Side View Top View

An = An−1 + (3 × 4n−1 /32n )A0 . (1.5.9)

1.6 Ice crystal size and shape distributions in cirrus clouds

1.6.2 Ice crystal size and shape spectra

1.6.3 Correlation between ice crystal maximum dimension and width

1.7.2 Correlation analysis

MPACE (10/17-10/18, 2004)

Ice Water Content (IWC, g m–3)

ln(De ) = a + b ln(IWC) + c[ln (IWC)]2 , (1.7.3)

1.8 Numerical modeling of cirrus cloud formation

1.8.2 A two-dimensional cirrus cloud model

1.8.2.1 Ice microphysics

1.8.2.2 Radiative transfer

In this chapter we present a number of fundamental subjects related to light scattering by

2.1 Introductory remarks

2.2 Maxwell’s equations, wave equations, and boundary conditions

where σ is referred to as specific (or electrical) conductivity, ε is the permittivity, and

Since ∇ × ∇ × E = ∇(∇ · E) − ∇ 2 E and ∇ · E = 0, we can substitute ∇ × H from

= a cos φ + b sin φ, (2.2.40)

= pn (μ) = pn ( cos θ ). (2.2.42)

ψ(r, φ, z) = R(r)(φ)Z(z). (2.2.58)

x = (1 − η2 )1/2 (ξ 2 − 1)1/2 cos φ,

where is the semifocal distance, −1 ≤ η ≤ 1, 1 ≤ ξ ≤ ∞, and 0 ≤ φ ≤ 2π. For the

E(t; ; ε) = El cos + Er e−iε sin

I (, ε) = E(t; ; ε)E ∗ (t; ; ε)

I (, ε) = 12 [I + Q cos 2 + (U cos ε − V sin ε) sin 2]. (2.4.26b)