1shimoda H Gholz H L Nakane K Eds The Use of Remote Sensing I

THE USE OF REMOTE SENSING IN THE MODELING
OF FOREST PRODUCTIVITY
FORESTRY SCIENCES
Volume 50
The titles published in this series are listed at the end of this volume.
The Use of Remote Sensing
in the Modeling
of Forest Productivity
Edited by
HENRY L. GHOLZ
University of Florida, Gainesville, Florida, U.S.A.
KANEYUKINAKANE
Hiroshima University, Hiroshima, Japan
and
HARUHISA SfflMODA
Tokai University, Tokyo, Japan
k4
SPRINGER SCIENCE+BUSINESS MEDIA, B.V.

A C L P . Catalogue record for this book is available from the Library of Congress
ISBN 978-94-010-6290-9 ISBN 978-94-011-5446-8 (eBook)

DOI 10.1007/978-94-011-5446-8
Printed on acid-free paper
All Rights Reserved

1997 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1997
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
Contents
Preface VII
Section One: Stand-Level Analyses

CHAPTER ONE
Assessing Leaf Area and Canopy Biochemistry of Florida Pine
Plantations Using Remote Sensing 3
H.L. Gholz, PJ. Curran, J.A. Kupiec and G.M. Smith
CHAPTER TWO
Modeling Radiative Transfer through Forest Canopies: Implications
for Canopy Photosynthesis and Remote Sensing 23
T. Nilson and J. Ross
CHAPTER THREE
Estimating Forest Canopy Characteristics as Inputs for Models of
Forest Carbon Exchange by High Spectral Resolution Remote Sensing 61
M.E. Martin and J.D. Aber
Section Two: Landscape/Regional-Level Analyses

CHAPTER FOUR
Detecting Structural and Growth Changes in Woodlands and Forests:
The Challenge for Remote Sensing and the Role of Geometric-Optical
Modeling 75
DLB. Jupp and J. Walker
CHAPTER FIVE
Integrating Remotely Sensed Spatial Heterogeneity with a
Three-dimensional Forest Succession Model 109
J.F. Weishampel, R.G. Knox, K.J. Ranson. DL Williams and J.A. Smith
CHAPTER SIX
Combining Remote Sensing and Forest Ecosystem Modeling: An Example
Using the Regional HydroEcological Simulation System (RHESSys) 135
J.e. Coughlan and J.L. Dungan
CHAPTER SEVEN
Forest Vegetation Classification and Biomass Estimation Based on
Landsat TM Data in a Mountainous Region of West Japan 159
N.J. Lee and K. Nakane
V
CHAPTER EIGHT
Forest Structure and Productivity along the Oregon Transect 173
DL Peterson
CHAPTER NINE
Use of Remote Sensing to Model Land Use Effects on Carbon
Flux in Forests of the Pacific Northwest, USA 219
D.O. Wallin, M.E. Harmon, WE. Cohen, M. Fiorella
and WK. Ferrell
Section Three: Global-Level Analyses

CHAPTER TEN
Global Biospheric Monitoring with Remote Sensing 241
s.N. Goward and D. G. Dye
CHAPTER ELEVEN
Energy Conversion and Use in Forests: An Analysis of Forest Production
in Terms of Radiation Utilisation Efficiency (E) 273
J.J. Landsberg, S.D. Prince, PG. Jarvis, R.E. McMurtrie,
R. Luxmoore and B.E. Medlyn
Color Plates 299
Index 311
VI
Preface
This project developed in the wake of a half-day, five-paper symposium entitled "The
Remote Sensing of Structure and Chemistry of Forest Vegetation," which was
organized for the XV International Botanical Congress held in Yokahama, Japan, on
September 2, 1993. The symposium participants expressed their commitment to a
book over dinner at a Chinatown restaurant, and eventually contributed half of the
chapters. However, the scope of the symposium was obviously too limiting: structure
and chemistry were seen as intermediate steps in the linking of remote sensing with
the modeling of forest carbon dynamics ("productivity," as used loosely in this book).
Hence, subsequent to the congress the remaining authors were recruited and the final
organization of the book emerged.
Although the theme of the book is not "scaling," per se, changes in spatial scale
obviously are used as the primary organizational tool: the first section of the book
focuses on leaf- to canopy-level radiative transfer, leaf biochemistry and their links to
photosynthesis and canopy carbon gain; the second section addresses additional
environmental and ecosystem changes at regional scales; and the final section takes a
stab at issues related more to the global scale. However, readers would likely be well
advised to look as much for the overlap across the sections as for the contrasts, for it
seems this is where many of the scientific challenges remain. In this context, some
allowance should be made for redundancies which, in our opinion, highlight the
commonalities among what at fust may seem to be very different perspectives.
Forest ecosystems are an obvious choice for applying both remote sensing and
simulation modeling: their physical structure varies much more than that of any other
type of ecosystem over both space and time, creating tremendous logistical difficulties
for in-field analysis. Many of the world's remaining forests are largely inaccessible for
routine field study in any case. But the approaches discussed here certainly are not
limited to forests. Our hope is that this volume catalyzes a discussion among the wide
range of scientists, engineers, students, policymakers and land managers whose
separate expertise, but also full collaboration, is needed to confront the challenges in
sustainable management of the world's natural resources.
Acknowledgments - Financial support for the Congress symposium and subsequent

book preparation was provided by EBS San Kou, Ltd., Hiroshima, and Hiroshima
University, Japan. Partial support for printing was also provided by the Florida
Agricultural Experiment Station. The layout artist for the book was Candace Hollinger
of Gainesville, Florida. Kristin King, the technical editor, also resides in Gainesville.
Henry L. Gholz, Gainesville, Florida, USA

Kaneyuki Nakane, Hiroshima, Japan
Haruhisa Shimoda, Tokyo, Japan
June 1996
VII
SECTION ONE
Stand-Level Analyses
ONE
Assessing Leaf Area and Canopy Biochemistry of Florida Pine

Plantations Using Remote Sensing
Henry L. Gholz, Paul J. Curran, John A. Kupiec and Geoff M. Smith
Gholz, H.L., Curran, PJ., Kupiec, J.A. and Smith, G.M. 1996. Assessing leaf area
and canopy biochemistry of Florida pine plantations using remote sensing. - In:
Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the
Modeling of Forest Productivity. Kluwer Acad. Pub!., Dordrecht, The Netherlands,
pp.3-22.
Leaf area index (LAI) and the ~iochemical makeup of forest canopies are critical
determinants of carbon (C) gain by forests. However, both variables may be highly
unstable in space and time and high spatial resolution measurements made on the
ground are difficult and expensive to obtain. We estimated LAI and canopy
biochemistry using remotely sensed imagery and data from a field site in north central
Florida. Ground measurements indicated a high seasonal fluctuation in LAI, and
annual variation of approximately 10%, correlated with the Normalized Difference
Vegetation Index (NDV!) derived from Thematic Mapper (TM) imagery. Repeated
fertilization affected concentrations of nitrogen (N) and chlorophyll in the pine
foliage, but had little effect on concentrations of water, lignin or cellulose.
Biochemical differences among samples of whole fresh pine needles were related to
reflectance properties determined in the laboratory using a portable
spectroradiometer. We then explored coupling laboratory spectral measurements of
foliar biochemistry with field spectra by analyzing the signal-to-noise ratio (SNR) of
the Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Results indicate that
the AVIRIS provides data with an SNR barely sufficient to estimate foliar
biochemistry; maximum SNRs for slash pine are suggested. Finally, using actual
AVIRIS data and simultaneously obtained field samples, stepwise regression of
corrected imagery indicated that three wavebands accounted for 94% of the spectral
variation, all in the spectral region of the reflectance red edge. These collective results
indicate the feasibility of parameterizing process-level models of primary productivity
of P elliottii stands using remotely sensed data.
H.L. Gholz, School of Forest Resources and Conservation, University of Florida,

Gainesville, FL 32611, USA. Pl. Curran, Department of Geography, University of
Southampton, Highfield, Southampton, S09 5NH, UK. I.A. Kupiec, Scottish Natural
Heritage, Research and Advisory Services Directorate, 2 Anderson Place, Edinburgh,
EH6 5Np, UK. G.M. Smith, Department of Geography, University College of
Swansea, Singleton Park, Swansea, SA2 8PP, UK.
3
4 The Use ofRemote Sensing in the Modeling of Forest Productivity
Introduction
A primary determinant of the productive capacity of forests is the leaf area that can be
supported per unit area of ground (leaf area index, or LA!). This has been shown to be
especially true for pine forests, where physiological adjustments to changing
resources (e.g., nutrient availability) are relatively small (Teskey et al. 1994).
However, although physiological adjustments may be minor, nutrient availability is
often the main factor limiting LA!, so that the feedback from tree nutrition to soil
nutrient availability (i.e., nutrient cycling) is also critical. Furthermore, water
availability interacts with nutrient availability to limit overall forest productivity on
many sites. Therefore, it is essential to understand the interactive effects of water and
nutrient availability on carbon (C) cycling in order to predict the net primary
production (NPP), or net C gain, of forests.
Plantation stands of slash pine (Pinus elliottii var elliottii Engelm.) in northern
Florida, USA, have been the subject of intensive research since 1981 with the aim of
understanding the factors that control C cycling processes (e.g., LA! dynamics, C
fixation and C allocation). Recent work has focused on synthesizing this information
into computer simulation models of ecosystem C dynamics (e.g., Ewel and Gholz
1991, Cropper and Gholz 1993a, 1994).
Measurements of the LA!, foliar biochemistry and plant water status of forests,
made on the ground and replicated in space and time, have been essential to this effort
but are very difficult and time consuming to carry out. Since slash pine has only two
age classes of needles, it exhibits large seasonal and annual changes in LA! (Fig. 1,
Gholz et al. 1991). Therefore, we also obtained various remotely sensed images of the
research site several times during the study. We hypothesized that key structural
1986 - 1987 1987 - 1988 1988 - 1989

6
4
X
W
0 2
Z

w 0
~ 10

u-

8
y ;'-++ +
W
-l
6
. rYH "';
9l/i ~?
0
.-.0 - 0 CON1'WOL
J 0 0 , A J A o 0 F A J A o 0 F
DATE
Figure 1. Seasonal LAI over three years for control (C) and fertilized (F) P.
elliottii stands in northern Florida as assessed through ground measurements, for
each age class (1-4, top panel) and for the whole canopy (bottom panel) (from
Gholz et aI. 1991).
5
(specifically LAI) and biochemical attributes of slash pine canopies could be

estimated indirectly using remote sensing, thereby enabling more efficient
extrapolation of the estimates made on the ground in both space and time. Precedent
for this approach existed as a result of studies that had estimated LA! and canopy
biochemistry remotely over broad transects of diverse vegetation (e.g., Peterson et al.
1988, Spanner et al. 1990a). However, this was the first attempt to apply remote
sensing to the detection of seasonal and treatment-induced changes in the same forest
stands over time. We also utilized spectroradiometers in both the laboratory and the
field to measure the spectral properties of slash pine needles and canopies as an
intermediate step between measurements made on the ground and airborne estimates.
Study site
The focus of the remote sensing research was a 60-ha slash pine plantation in north
central Florida (29 44'N, 82 9.30'W). The soil is sandy and characterized by low
organic matter and nutrient status. The predominant soil type is an UItic Haplaquod,
although the relative development of the subsurface spodic and argillic horizons is
highly variable (Gaston et al. 1990). A water table fluctuates between the surface and
a depth of ca. 2 m over a typical year, depending on precipitation patterns. The
elevation of the site is 39.5 1.8 m. The mean annual (1955-1987) temperature is
21.7C and the mean annual precipitation is 1320 mm (NOAA 1989). Annual rainfall
totals at the site for May 1987 through April 1989 were 1051 and 1125 mm,
respectively, concluding three years of extended drought.
The vegetation was dominated by even-aged planted (second-rotation) slash pines,
21 years old in 1986. Slash pine is native to these sites; plantations of this species
currently amount to nearly 4 x 106 ha in Florida. Site preparation consisted of
chopping the residues, broadcast burning, bedding and planting after the stem-only
harvest of the previous stand. There were no further treatments applied after planting.
The mean plot basal area in August 1986 was 25.9 3.3 m2 ha'[, the mean tree density
was 1190 118 ha't, the mean tree diameter (at 1.3 m height) was 17.3 0.5 cm and
the mean tree height was 15.4 0.8 m (Gholz et al. 1991). Understory vegetation
consisted of native species reestablished naturally after site preparation, dominated by
saw palmetto (Serenoa repens) and gallberry (flex glabra).
Sixteen 50 x 50-m plots were established in the study area in 1986, with ground
measurements confined to internal 25 x 25-m subplots. Fertilizers were added
quarterly to eight plots for one year beginning in February 1987 and then
semiannually through June 1992; the other eight plots served as controls. The
fertilizer was added in dry form and consisted of a complete mix of macronutrients;
micronutrients were also added during 1987. Greater detail on stand structure and C
dynamics in relation to the fertilization can be found in Gholz et al. (1991) and
Cropper and Gholz (1993a,b, 1994).
Remotely sensed data

Three electro-optical sensors with a range of spectral bandwidths were used in this
research. The broad-band Thematic Mapper (TM) was carried by the Landsat satellite,
0\
~
~
~
~
Table 1. The three sensors used in this research. Note that the thermal infrared waveband on the Landsat TM was omitted. .s;,
::tl
~
Sensor Platform Swath Instantaneous Spectral Bandwidth Number Vegetation ::l

o
width field of view coverage (nm) of variables ~
(km) (m) (nm) wavebands under ~
;:
investigation '"
~.
S
Thematic Mapper Landsat 180 30 450-2350 60-270 6 Leaf
(TM) satellite Area Index ~
(canopy)
Airborne Visible ER-2 11 20 410-2450 9.4-9.7 224 Chlorophyll,

~
Infrared Imaging aircraft water, nitrogen,
~
.s;,
Spectrometer (AVIRIS) lignin, 61
cellulose (canopy) ~
....
'""'tl
Infrared Intelligent Laboratory - =0.1 350-2500 2-5 665 Chlorophyll, C:l
Spectroradiometer (IRIS) water (needle) ~
<")
....
~.
~.
7
the narrow-band Airborne Visible Infrared Imaging Spectrometer was carried by a

high-altitude aircraft and the very narrow-band Infrared Intelligent Spectroradiometer
(IRIS) was used in laboratory studies (Table 1). These sensors were used over a period
of five years with varying frequency (Table 2). In addition, a Spectron SE-590
spectroradiometer was used for the measurement of reflectance spectra in the field
(Curran et al. 1992, Curran 1994).
Table 2. Foliar samples and remotely sensed data collected over a five-yr
period. Y = data were recorded, N = data were not recorded. Full names and
descriptions of the sensors are given in Table 1.
Sensor
Date Number of samples TM AVIRIS IRIS
February 1988 124, Y N N

September 1988 162 Y N N
March 1989 Y N N
July 1989 166 N N N
March 1990 160 N y y
September 1990 160 N Y y
August 1991 320 N N Y
July 1992 360 N Y N
Relating LAI to forest canopy reflectance

In near-infrared (NIR.: 760-900 nm) wavelengths, within-leaf scattering is dominant;
therefore, reflection from the canopy is high. In red (R: 630-690 nm) wavelengths,
however, pigment absorption is dominant, resulting in low reflectance (Jensen 1983).
Consequently, LAI is usually related positively to an increase in the difference
between NIR and R, at least up to the reflectance asymptote of the canopy (Peterson
and Running 1989). A number of environmental factors (e.g., solar angle, understory
vegetation, atmospheric conditions and phenology) cause variability in this
relationship (Curran 1983).
The remote sensing of forest LA! has been hampered by the need to measure LA!
(or leaf mass) with an accuracy adequate for deriving a predictive relationship.
Despite the logistical problems involved, several groups have managed to produce
such relationships (e.g., Jensen and Hodgson 1985, Running et al. 1986, Danson
1987, Peterson et al. 1987, Spanner et al. 1990a,b, Danson and Curran 1993).
Unfortunately, in collecting data on the ground upon which these relationships were
based, LA! measurement techniques inherently unresponsive to seasonal change were
used. For example, most were derived using regression relationships between leaf area
(or leaf mass) per tree and a seasonally stable tree dimension (e.g., diameter), with
both variables measured at various times throughout the growing season and often
over different years (e.g., Gholz et al. 1979). Therefore, these relationships, although
potentially very useful for estimating LAI over large areas for one point in time,
cannot be used to study the seasonal changes in LAI of a forest stand. The only way
to obtain accurate, seasonal, remotely sensed estimates of LAI is to derive
relationships between remotely sensed measurements and LAI that are based on
actual values of LAI recorded near the time of image acquisition.
We related TM data for three dates over a 13-month period to LAI data collected on
the ground within two weeks of image acquisition in February 1988, September 1988
and March 1989. These data were corrected radiometrically and atmospherically and
standardized in terms of irradiance (Curran et al. 1992). Two wavebands of the TM
data were used to calculate the Normalized Difference Vegetation Index (NDVI =
(NIR-R)I(NIR + R)), which was then related to LAI measurements for the 16 plots
made on the ground.
LAI estimates for each plot were obtained from measurements made on the ground
(Fig. 1) through destructive sampling in August 1986, 1987 and 1988. These estimates
were then combined with monthly estimates of accretion of new leaf area, obtained
through measurements of needle elongation on a subset of plots, and decreases in old
leaf area, obtained through monthly measurements of needle litterfall on all plots, to
construct monthly estimates of LAI over the three years (Gho1z et al. 1991). The effect
of the fertilizer on LAI was first statistically manifest in the spring (new foliage)
production of 1988.
Measurements of specific leaf area (cm 2 g'l) indicated that there were no vertical or
azimuthal patterns within a needle age class for this variable within tree crowns. The
specific leaf area of newly expanding needles began at a high value and decreased
linearly until it reached the value of the old foliage by early fall, after which time it
remained constant until needles were shed the following year (Gholz et al. 1991). LAI
of old needles was calculated as the product of monthly plot estimates of old-needle
biomass and the mean annual specific needle area for old needles. New-needle LAI
was calculated as the product of monthly plot estimates of biomass and the mean
monthly specific area for new needles.
There was a positive linear relationship between NDVI and LAI on all three dates
with R 2 values of 0.35,0.75 and 0.86 for February 1988, September 1988 and March
1989, respectively. The constants and NDVI coefficients varied as a result of a number
of factors, the main one hypothesized to be seasonal changes in understory LAI (Fig.
2).
Three predictive relationships were used to estimate LAI on the other plots using
data from half of the plots on each date (Fig. 3), with a root-mean-square error of
0.74 LAI. These estimates, accurate to 15.6% of the mean LAI, tracked the LAI
increase from spring to autumn and the LAI decrease from autumn to spring (Curran
et al. 1992). This initial study demonstrated the potential of using Landsat TM
imagery for estimating the stand-level LAI required for input to the labile C model
(Cropper and Gholz 1993a).
Cle~ed foresl ::I ' Road (hI p1'olJ!" i
t I.. "
."I .
-- .
~
.1. 111-
.. ~
'r COntrolled slash Pine torest I I
itt
I' !
l,
~
"..---- ~
- - Pasture [
.. i
. ,
~
... -...".......i.. I ...

- ,
Figure 2. Radiance spectra and first derivative radiance spectra for 3 x 3-pixel areas (unless otherwise stated) of six land covers.
The AVIRIS image was recorded on 6 September 1990 and covers approximately 25 km 2 to the north of Gainesville, Florida,
USA. Image data courtesy of NASA/JPL.
\0
10 10
February 1988 a March 1989 C
8 8
3'06 3'0 8
A
i!!
::J
i!! A A
::J
&0
~~A~A
la4
CD
0
fO 000
~4
CD
:::E :::E AA
0 2
2
LAI. -5.41 + 15.95 NOVI LAI. -9.BS + 24.75 NOVI
0 0
0.50 0.54 0.58 0.62 0.66 0.70 0.50 0.54 0.58 0.62 0.66 0.7e
NOVI NOVI
10 10
September 1988 c b o February 1988 d
a
8 8 o September 1988 III 0
~o e
3'0 6 oeO
a c
~
'0 6
A March 1989
a
i!! c CD 0 0 00
::J C OJ %
~ 4
CD
0 ,E4
.n et>~
:::E I>
2 2
LAI. '4.21 + 33.99 NOVI
0 0
0.50 0.54 0.58
NOVI
0.62 0.66 0.70 0 2
"
Measured LAI
6 8 10
Figure 3. The relationship between NDVI recorded by the Landsat Thematic

Mapper and LAI for 16 plots on three dates (a-c). Panel d illustrates the
relationship between measured and estimated LAI for four control and four
fertilized plots for three dates. The estimated LAI in d was derived using Landsat
TM data and three predictive regression relationships (a-c), each based on data
from four control and four fertilized plots.
Patterns in the canopy biochemistry of slash pine stands

Foliar samples were obtained from the upper, middle and lower thirds of the crowns
of 36 randomly selected control plot trees in August 1986 and 1987. Approximately
six branches per crown section of each tree were sampled. All of the foliage from each
sampled branch was removed separately by age class and thoroughly mixed. Three
replicate subsamples were selected randomly. These subsamples were subjected to
routine laboratory biochemical analyses for N, P, K, Mg and Ca, as described in Gholz
et al. (1985a). Analyses of variance of the crown level means for each tree were then
undertaken to determine whether nutrient concentrations were affected by vertical
caJ10py position. Foliage-bearing twigs (new and old) and unfoliated branches were
analyzed similarly.
Foliage samples were also collected from the upper third of the canopies of all 16
plots in August 1991. The aim of this sampling was to describe the variation in
biochemical (N, chlorophyll, lignin, cellulose and water) concentrations of slash pine
needles after the fertilization effects were fully manifest and then to relate
11
biochemical concentrations to reflectance spectra measured in the laboratory (IRIS)

and remotely (AVIRIS). Ten samples of each age class were collected using a rifle to
shoot down branches. All samples were immediately bagged, labeled and placed in
cold (OC) boxes in the field. The samples were then taken to a nearby laboratory and
stored in freezers at -10C. A total of 320 samples were taken to the UK for analysis.
All subsequent analyses were performed on the defrosted needles.
Water concentrations for these samples were calculated by oven drying needles for
24 hours at 70C. Chlorophyll concentration was determined using standard
spectrophotometry of tissue extracts in a 90% acetone solution (Mackinney 1941).
Nitrogen concentration was measured using an elemental analyzer (Roboprep-CN).
Lignin and cellulose concentrations were measured from acid detergent fiber using
the permanganate procedure of Van Soest and Wine (1968).
An assumption basic to the application of remote sensing to detect biochemical
differences among vegetation plots is that there are no concentration gradients within
the canopies. The significance of this is that although the sensor records radiation
reflected primarily from the upper portion of the canopy (especially in stands with
high leaf areas), the objective generally is to characterize the canopy as a whole.
There were some clear concentration gradients within the crowns of slash pine,
depending on tissue type and age class. Nitrogen, P, K and Mg all decreased with
canopy depth for unfoliated branches, while Ca concentrations increased (Table 3).
The only significant gradient for old twigs was an increase of Ca with depth; no
significant gradients were observed for new twigs. Gradients were also observed for
new and old needles, although not all were significant. Nitrogen in new needles
tended to decrease with depth while old needle N increased, although neither trend
was significant at the p =0.05 level. Phosphorus and K concentrations for new needles
both increased slightly (not significant). In old needles, however, P, K, Ca and Mg all
increased significantly with depth. Where significant effects of depth were observed,
the use of upper-canopy concentrations or means unweighted by vertical needle
biomass distribution could introduce significant bias into whole-canopy estimates of
nutrient concentrations and contents. We did not investigate how the fertilization
treatment may have affected vertical distributions of nutrients; this remains an
additional concern.
Examination of the August 1991 analyses (Table 4) indicated that the strongest
correlations among biochemicals were positive and occurred between concentrations
of lignin, cellulose and water (Table 5). There was a weaker but statistically significant
correlation between N and chlorophyll concentrations. This may be due to increased
demand for N-rich proteins in the photosynthetic pathway caused by higher
chlorophyll concentrations. T-tests showed that the fertilization treatment only
significantly affected (p =0.05) chlorophyll and N concentrations (Table 6). However,
concentrations of water and of the structural elements lignin and cellulose did not vary
with fertilization treatment.
There was a greater difference in the biochemical concentrations, except for N,
between needle ages than between treatments (Table 7). Old needles were drier and
contained higher concentrations of chlorophyll. There was no correlation between
chlorophyll and N concentrations, perhaps due to the translocation of N out of leaf
12 The Use of Remote Sensing in the Modeling of Forest Productivity
tissue prior to senescence. Significantly lower concentrations of lignin and cellulose

in older needles were counterintuitive, but were very slight in any case.
Table 3. Nutrient concentrations (% mean 1 std. dev.) for each of three

canopy layers (U =upper third, M =middle third, L =lower third) from control
(unfertilized) trees sampled in August 1986 and August 1987. Where significant
differences (p = 0.05) did not occur, simple arithmetic means for the whole
canopy are provided.
Tissue Level N P K Ca Mg
New needles U 0.1130.036

M 0.77IO.086 0.085O.015 0.3450.095 0.1 250.032 0.1350.029
L 0.1 550.045
Old needles U 0.056O.OO8 O.l350.019 0.2160.079 0.1030.028

M 0.77IO.066 0.0580.008 0.1380.021 0.25 \0.079 O.l120.028
L 0.0640.012 0.1470.032 0.2530.092 0.1 280.036
New twigs U
M 0.3950.054 0.0590.018 0.2170.070 0.1780.053 0.1 080.025
L
Old twigs U 0.1900.059

M 0.3230.041 0.0320.008 0.1120.032 0.2250.079 0.0970.027
L 0.2360.087
Unfoliated U 0.2440.040 0.022O.OO7 O.O85O.030 O.I920.079 O.084O.O17

branches M 0.209O.O30 O.018O.O04 0.078O.052 O.2490.O98 O.O64O.015
L O.199O.040 O.014O.O04 O.055O.012 0.266O.On 0.051O.OI2
Table 4. Biochemical concentrations for 80 slash pine foliage samples (August

1991).
Biochemical Mean Standard Minimum Maximum

deviation
Water (%) 58.3 4.4 37.6 78.0

Chlorophyll (mg g.l) 1.90 0.78 0.70 4.37
Nitrogen (%) 0.59 0.14 0.23 1.00
Lignin (%) 22.1 1.7 18.8 25.5
Cellulose (%) 36.1 2.4 31.9 42.2
13
Table 5. Simple correlations (r) among biochemical concentrations from

needles sampled in August 1991.
Water Chlorophyll N Lignin Cellulose
Water 1.0
Chlorophyl -0.48* 1.0
N -0.09 0.52* 1.0
Lignin 0.74* -0.21 0.12 1.0
Cellulose 0.73* -0.20 -0.02 0.51 * 1.0
* Indicates significance at the p = 0.01 level.
Table 6. Differences in mean biochemical concentrations between control and

fertilized treatments for needles sampled in August 1991 (figures in parentheses
represent standard deviations). ,
Treatment
Biochemical Control Fertilized
Water (%) 58.6 (2.7) 58.0 (3.1)

Chlorophyll (mg gO') 1.43 (0.32) 2.37 (0.70)*
N (%) 0.51 (0.10) 0.68 (0.11)*
Lignin (%) 21.8 (1.6) 22.4 (1.6)
Cellulose (%) 35.8 (2.1) 36.3 (2.6)
* Significantly different from the control plot mean at the p = 0.05 level.
Table 7. Differences in mean concentrations between needle ages for needles

sampled in August 1991 (figures in parentheses represent standard deviations).
Since there was an interaction between age class, treatment and N, separate
means for each age class and treatment are listed.
Needle age class
Biochemical New Old
Water (%) 60.8 (1.7) 55.9 (1.9)

Chlorophyll (mg g"l) 1.49 (0.59) 2.32 (0.80)
N(%) 0.60 (0.12) 0.59 (0.15)
Control 0.52 (0.11 ) 0.49 (0.10)
Fertilized 0.67 (0.09) 0.69 (0.12)
Lignin (%) 23.3 (1.3) 21.0 (1.1)
Cellulose (%) 37.5 (2.1) 34.6 (1.7)
Laboratory reflectance spectra of slash pine foliage

Laboratory reflectance measurements were made using a Geophysical Environmental
Research Corporation (GER) IRIS MK IV (Milton and Rollin 1989). The instrument
was set up in a constant-temperature (20C) laboratory. A voltage-stabilized full
spectrum artificial lighting source was used to illuminate the target and reference
panel combination. The light source was positioned at a 45 angle approximately 1 m
from the target/reference panel. All light except that from the source was excluded.
The optical head unit was positioned directly over the target/reference combination at
a height of 30 cm. At this height, reflectance spectra were measured from
2
approximately 3 cm of the foliage target. The needles were aligned side by side on
adhesive tape placed on a matte black background and were oriented parallel to the
artificial light in order to suppress shadowing effects. This arrangement was adopted
after a large number of trials that assessed the effects of background, influence of the
tape, needle orientation, number of levels of stacked needles needed, heating and
drying effects of the lamp and area of sample included in the sensor's field of view.
A total of320 reflectance spectra were recorded from the samples that were later to
undergo analysis for biochemical composition.
Freezing and thawing are known to influence the spectral properties of leaves.
Woolley (1971) described two effects observed on defrosted soybean leaf tissue: (i)
the overall reflectance was lowered slightly, an effect attributed to the formation of ice
crystals in the tissue, which disrupted cell membranes and allowed some water to flow
into the intercellular spaces; and (ii) water absorption bands were shifted to longer
wavelengths. Therefore, we also conducted a test to compare the effect of freezing and
thawing on slash pine needles using the IRIS. We could determine no measurable
effect, possibly because conifer needles possess greater mechanical strength than do
soybean leaves.
The reflectance spectra for needles from control and fertilized plots (Fig. 4) showed
the main difference occurring around the 600 nm (green) reflectance region. Needles
from the fertilized plots reflected less radiation than those from the control plots. The
larger concentrations of chlorophyll a and b did not result in large increases in the
absorption of blue and red light. Rather, these absorption features were broadened and
resulted in decreased reflectance of green radiation. A similar effect could be seen in
the reflectance spectra for new and old needles: the reflectance of green radiation is
reduced for old needles, which have a larger concentration of chlorophyll.
Old needles had a higher reflectance than new needles for wavelengths of more
than 800 nm (Fig. 5), due to the greater water concentration offirst-yr needles. Water
absorbs radiation strongly at four specific wavelengths (approximately 975, 1175,
1400 and 1900 nm) and also over the entire near-infrared region. These results for
water are similar to those obtained by Carter (1991), who also used an IRIS to
measure the reflectance from slash pine needles.
Other biochemicals such as N/protein, lignin and cellulose are known to absorb
radiation at specific wavelengths and can be readily detected in dried and fresh foliar
material using near-infrared reflectance spectroscopy (NIRS) (Wessman et al. 1988).
15
60
g
l
E 4G
u
u
c
.2
u
u
'i
a:: 20
0
'GO 1m 1200
V\ 1600 2000 2'GO
Wavelength (nm)
Figure 4. Mean laboratory reflectance spectra of 40 control plot samples and 40

fertilized samples of slash pine needles collected in August 1991 (needle age
differences are ignored).
60

-Old
g 40
u
u
c:
0
-0
.!!
t; 20
a::
~
0
400 800 1200 1600 2000 2400
Wavelength (nm)
Figure 5. Mean laboratory reflectance spectra of 40 new and 40 old slash pine
needle samples collected in August 1991 (treatment differences are ignored).
Research is still in progress to investigate the potential of extending the approach used
in NIRS to laboratory spectral measurements using fresh whole samples of slash pine.
Coupling the AVIRIS measurements and foliar biochemistry: The

problem of sensor noise
The coupling of laboratory and field spectra with foliar biochemistry has met with
considerable success in other studies (Curran 1989). In the laboratory, there is
typically a constant, controllable and strong radiation source illuminating a
homogeneous leaf sample located a few centimeters from a detector. In the field, there
is typically a variable, uncontrollable and relatively weak radiation source

illuminating a heterogeneous canopy located a few meters from its detectors.
Fortunately, the measurement (or dwell) time in both the laboratory and the field can
be adjusted to ensure that a large supply of photons reaches the detector(s) and
therefore that the signal is large. In contrast, an imaging spectrometer, such as the
AVIRIS, relies on a variable, uncontrollable and relatively weak radiation source
typically illuminating a heterogeneous canopy located several tens of kilometers from
a detector. In addition, the spectrometer has a very short dwell time, further reducing
the strength of the recorded signal. Together, these factors suggest a number of
potential problems in the coupling of the AVIRIS data with foliar biochemistry. To
minimize these problems, there is a need to (i) correct for atmospheric effects, (ii)
adequately sample the spatial variability of foliar biochemistry in a three-dimensional
canopy for an area representative of an AVIRIS pixel, (iii) accurately co-locate these
ground data with the AVIRIS data and (iv) select an optimum data processing
procedure (e.g., modeling, decomposition or stepwise regression). However, all of
these steps will not enable the AVIRIS data to be coupled with foliar biochemistry if
the SNR ratio of the AViRIS is too small. Therefore, we first investigated the SNR at
which key foliar biochemical absorption features of the slash pine needles could be
detected.
The reflectance spectra of slash pine needles from the 1991 sampling were used to
estimate the SNR needed before the AVIRIS could be used to accurately estimate
foliar biochemical concentrations. Reflectance spectra were recorded in the
laboratory under conditions in which instrument noise could be accounted for. The
spectra were then degraded to the radiometric resolution of the AVIRIS and first and
second derivatives of reflectance spectra were obtained. The magnitudes of both the
first and the second derivative spectra were then correlated, one waveband at a time,
with the concentrations of each biochemical. The largest correlation coefficients
identified the most useful wavebands for the remote sensing of foliar biochemicals
(Table 8).
Table 8. The wavebands of maximum correlation between foliar biochemical

concentration and derivative reflectance. The biochemicals associated with
absorptions at these particular wavebands are derived from Curran (1989) and
Peterson (1991). The r values are Pearsons correlation coefficients and all are
significant at p =0.01.
First derivative Second derivative
Foliar Waveband Waveband

biochemical (nm) Feature r (nm) Feature r
Nitrogen 1491 Protein 0.5 1195 Protein 0.4

Lignin 1689 Lignin 0.4 1709 Lignin -0.4
Cellulose 1287 Cellulose -0.4 1719 Cellulose -0.3
17
To determine the degree of noise at which these features were masked, random
noise was added to the "pure" spectra and the correlation procedure was repeated a
number of times, each time with an increasing amount of noise. The wavebands were
ranked by their correlation coefficients after each addition of noise. The threshold
mean degree of noise (MSNR) at which a waveband from Table 8 was said to be
masked was the degree at which its position in the ranking first changed by more than
one place.
The MSNR at which the absorption features of N, lignin and cellulose were
masked varied from 37: 1 to 65: 1 (Table 9). NASA's Jet Propulsion Laboratory (JPL)
reported the SNR of the AVIRIS for a range in wavebands and a 50% reflectance
target (Table 10, column 2, from Green et al. 1992). However, these values are larger
than for any part of the observed laboratory vegetation spectra (Table 10, column 3).
Therefore, we converted JPL SNRs at 50%, albeit approximately, to values that would
be obtained when recording vegetation spectra using the laboratory reflectances we
observed (Table 10, column 4).
Table 9. The SNR threshold (MSNR) at which the strongly correlating

absorption features were masked by noise.
First derivative Second derivative
Foliar Waveband Waveband

biochemical (nm) MSNR (nm) MSNR
Nitrogen 1491 37:1 1195 61:1

Lignin 1689 50:1 1709 46:1
Cellulose 1281 65:1 1719 48:1
When the MSNRs from the laboratory (Table 10, column 5) are compared with the
approximate JPL SNRs at vegetation reflectance, it appears that the AVIRIS provides
data with an SNR sufficient to estimate foliar biochemical concentrations. The only
waveband that did not have a sufficiently large SNR was at 1491 nm. This waveband
probably would not be selected when using the AVIRIS data, however, since it is on
the flanks of a major water vapor absorption feature. Four- to tenfold increases in the
SNR should provide data with more than sufficient SNR to estimate foliar
biochemical concentrations.
The MSNR values for slash pine (Table 10) should be used only as a guide to
formulating minimum requirements for airborne systems. The MSNR values would
certainly be increased if atmospheric, canopy and sampling effects were taken into
account. But, in general, the results of this analysis support the viability of using
remotely sensed data to estimate foliar biochemical concentrations.
Table 10. A comparison of the SNR reported by the JPL (Green pers. comm.
1993) for the AVIRIS at the start of the 1993 flight season and the SNR threshold
(MSNR) at which the absorption features could no longer be detected (Table 9).
JPLSNR@ Vegetation Approx. JPL

Waveband 50% reflectance SNR@ MSNRin
(nm) reflectance in laboratory vegetation laboratory
(%) reflectance
1491 175:1 7.1 25:1 37:1

1689 300:1 17.1 100:1 50:1
1281 325:1 36.8 240:1 65:1
1195 300:1 35.3 210:1 61:1
1709 290:1 16.2 95:1 46:1
1719 290:1 15.8 90:1 48:1
Coupling the AVIRIS measurements and foliar biochemistry: An

example of chlorophyll concentration
The latest results for the remote sensing of slash pine needle biochemistry using the
AVIRIS data are based on the overflight of the Florida site in July 1992. At this time,
a total of 360 lIeedle samples were collected from seven control and seven fertilized
plots. The average chlorophyll concentrations were 1.33 and 2.10 mg gol for the
control and fertilized needles, respectively (ignoring age class effects).
The AVIRIS imagery was preprocessed by JPL and corrected for atmospheric
effects at the Center for the Study of Earth from Space (CSES) at the University of
Colorado using the Atmospheric Removal Program (ATREM, Gao et al. 1992). The
pixels corresponding to the study plots were extracted from the imagery at the
University College of Swansea using ground control points and an Affine
transformation. Examination of the error in locating a series of ground checkpoints
showed that the extracted pixels were within plot boundaries. Water absorption bands
(1322-1482 and 1789-1963 nm) were removed from the AViRIS spectra. Stepwise
regression (BMDP) was then used to select the wavelengths that accounted for the
variation in the average chlorophyll concentration of each plot (n = 14).
The first three waveband selections (790.4, 800.0 and 819.3 nm) accounted for
94% of the spectral variation. These wavebands are located in the spectral region of
the red edge, the point of maximum slope in a reflectance spectrum of vegetation.
Although it has been well documented that this region contains information on
chlorophyll concentration (Gates 1980, Baret et al. 1987, Curran et al. 1990, 1995),
we believe this is the first study in which the AVIRIS data were used to establish a
quantitative relationship. Research continues to investigate the relationships between
the AVIRIS spectra and leaf water, N, lignin and cellulose concentrations (Curran and
Kupiec in press).
19
Conclusions
Field data obtained during these studies indicated that the physiology of slash pine
remains relatively unaffected by nutrient availability. The main response to
fertilization is an increase in stand LA!, which then leads to increased C gain and stem
growth by the stands. Cropper and Gholz (1994) investigated the possible sources of
C necessary for this initial LAI increase and concluded that as little as a 5%
reallocation of C from fine root biomass could account for it - a percentage
impractical to measure directly in the field. A 15% increase in net photosynthesis rates
of fertilized trees could also lead to the requisite "extra c." However, there is no
evidence of such a change for slash pine that is of sufficient magnitude to be detected
if it did occur (Teskey et al. 1994). In any case, the effect of fertilization seems to be
mainly an increase in the rate of tree growth along the normal developmental
trajectory, preserving the allometry and allocation patterns normally observed
(Colbert et al. 1990). In this case, the importance of accurately monitoring LA! is
paramount for determining changes in seasonal and annual stand C gain, while the
additional assessment of associated changes in canopy biochemistry may not be.
In nature, large changes in site nutrient availability over short time intervals are not
usually expected. However, nutrient availability does in fact decrease markedly over
one (25-yr) rotation of slash pine, as needle litter accumulates because of low
decomposition rates (Gholz and Fisher 1984, Gholz et al. 1985a,b); this process may
explain the lower LA!s of older stands (Gholz and Fisher 1982). On the other hand,
marked changes in water availability in forest soils occur seasonally and from year to
year. Decreased water availability was postulated as the reason for LA! declines on
control plots of approximately 10% per year from 1986 through 1989, during a
prolonged drought period in the southeastern United States (Gholz et al. 1991); this
pattern was totally masked on the fertilized plots (Fig. 1).
Although no relationship between water availability and LA! could be obtained
with the available data, it is conceivable that the relationships between LA! and
TMlNDVI obtained in the earlier studies can be used to explore much longer-term
fluctuations in LAI and perhaps to develop a model relating LAI to climatic
conditions as well as to nutrition. In addition, our results suggest that chlorophyll
concentrations (and likely water and N concentrations) of slash pine needles may be
determined through analysis of high spectral resolution data. Routine determinations
of such variables are well beyond normal capabilities of research conducted on the
ground and highlight the key role remote sensing may play in long-term ecological
analyses of forests.
Acknowledgments - Gholz acknowledges support from USDA Cooperative

Agreement Asfs-9, 961-69 and NSF Grant 8919647. Field sites and support were
provided in cooperation with Container Corp. of America. Curran acknowledges
NERC Grant GRSI7647, NERC Studentship ET4/90fILS/59 to Smith and NERC
EPFS for the loan of the IRIS Spectroradiometer; an NRC Senior Research
Associateship; and NASA support through Grant 93120 for the AVIRIS overflights, for
partial funding (from the Earth Science Application Division) and for help with image
processing by Jennifer Dungan.
References
Baret, E, Champion, L., Guyot, G. and Podaire, L. 1987. Monitoring wheat canopies with a
high spectral resolution radiometer. - Rem. Sens. Environ. 22: 367-378.
Carter, G.A. 1991. Primary and secondary effects of water content on the spectral reflectance
ofleaves. -Am. J. Bot. 78: 916-924.
Colbert, S.R, Jokela, E.I. and Neary, D.G. 1990. Effects of annual fertilization and sustained
weed control on dry matter partitioning, leaf area and growth efficiency ofjuvenile loblolly
and slash pine. - For. Sci. 36: 995-1014.
Cropper, w.P, Jr., and Gholz, H.L. 1993a. Simulation of the carbon dynamics of a Florida
slash pine plantation. - Ecol. Model. 66: 231-249.
Cropper, W.P., Jr., and Gholz, H.L. 1993b. Constructing a seasonal carbon balance for a
forested ecosystem. - Clim. Res. 3: 7-12.
Cropper, W.P., Jr., and Gholz, HL 1994. Evaluating potential response mechanisms of a forest
stand to fertilization and night temperature: A case study using Pinus elliottii. - Ecol. Bull.
(Copenhagen) 43: 154-160.
Curran, P.J. 1983. Multispectral remote sensing for the estimation of green leaf area index. -
Phil. Trans. Roy. Soc. ,London (Ser. A) 309: 257-270.
Curran, P.I. 1989. The remote sensing of foliar chemistry. - Rem. Sens. Environ. 29: 271-278.
Curran, P.J. 1994. Attempts to drive ecosystem simulation models at local to regional scales. -
In: Foody, G.M. and Curran, PJ. (eds). Environmental Remote Sensing from Regional to
Global Scales. 1. Wiley and Sons, Chichester, UK, pp. 149-166.
Curran, Pl. and Kupiec, J.A. 1995. Imaging spectrometry: A new tool for ecology. - In:
Danson, EM. and Plummer, S.E. (eds). Advances in Environmental Remote Sensing. J.
Wiley and Sons, Chichester, UK (in press).
Curran, P.I., Dungan, J.L. and Gholz, H.L. 1990. Exploring the relationship between
reflectance red edge and chlorophyll content in slash pine. - Tree Physiol. 7: 33-48.
Curran, P.J., Dungan, JL and Gholz, H.L. 1992. Seasonal LA! in slash pine estimated with
Landsat TM. - Rem. Sens. Environ. 39: 3-13.
Curran, P.J., Windham, W.R. and Gholz, H.L. 1995. Exploring the relationship between
reflectance red edge and chlorophyll concentration in slash pine leaves. 2. - Tree Physiol.
15: 203-206.
Danson, EM. 1987. Preliminary evaluation of the relationships between SPaf-1 HRV data
and forest stand parameters. - Int. J. Rem. Sens. 8: 1571-1575.
Danson, EM. and Curran, P.J. 1993. Factors affecting the remotely sensed response of
coniferous forest plantations. - Rem. Sens. Environ. 43: 55-65.
Ewel, K.C. and Gho1z, HL 1991. A simulation model of the role of belowground dynamics in
a Florida pine plantation. - For. Sci. 37: 397-438.
Gao, B-C., Heidebrect, K.B. and Goetz, A.EH. 1992. Atmospheric Removal Program
(ATREM) Users Guide. - Center for the Study of Earth from Space, Cooperative Institute
for Research in Environmental Sciences, University of Colorado, Boulder, CO.
Gaston, L., Nkedi-Kizza, P., Sawka, G. and Rao, P.S.C. 1990. Spatial variability of
morphological properties at a Florida flatwoods site. - Soil Sci. Soc. Am. J. 54: 527-533.
Gates, D.M. 1980. Biophysical Ecology. - Springer-Verlag, New York. 611 pp.
Gho1z, H.L. and Fisher, RE 1982. Organic matter production and distribution in slash pine
(Pinus elliottii) plantations. - Ecology 63: 1827-1839.
Gho1z, H.L. and Fisher, RE 1984. The limits to productivity: Fertilization and nutrient cycling
in coastal plain slash pine forests. - In: Stone, E.L. (ed). Forest Soils and Treatment
Impacts, Proceedings of the Sixth North American Forest Soils Conference, June 1983,
University of Tennessee, Knoxville, TN, pp. 105-120.
21
Gholz, H.L., Grier, C.c., Campbell, AG. and Brown, AT. 1979. Equations for Estimating
Biomass and Leaf Area of Plants in the Pacific Northwest. - Research Paper 41, Forest
Research Lab, Oregon State University, Corvallis, OR
Gholz, HL, Fisher, RF. and Pritchett, WL. 1985a. Nutrient dynamics in slash pine plantation
ecosystems. - Ecology 66: 647-659.
Gholz, H.L., Perry, C.S., Cropper, WP, Jr. and Hendry, L.C. 1985b. Litterfall, decomposition
and nitrogen and phosphorus dynamics in a chronosequence of slash pine (Pinus elliottii)
plantations. - For. Sci. 31: 463-478.
Gholz, H.L., Vogel, SA, Cropper, WP., Jr., McKelvey, K., Ewel, KC., Teskey, R.O. and
Curran, PJ. 1991. Dynamics of canopy structure and light interception in Pinus elliottii
stands, North Florida. - Ecol. Monogr. 61: 33-51.
Green, R.O., ConeI, J.E., Bruegge, CJ., Margolis, J.S., Carrere, V., Vane, G. and Hoover, G.
1992. In-flight calibration of the spectral and radiometric characteristics of AVIRIS in
1991. - In: Summaries of the Third Annual JPL Airborne Geoscience Workshop, NASA,
Jet Propulsion Lab, Publication 92-14, Pasadena, CA, pp. 1-4.
Jensen, J.R 1983. Biophysical remote sensing. -Ann. Assoc. Am. Geog. 73: 111-132.
Jensen, J.R and Hodgson, M.E. 1985. Remote sensing of forest biomass: An evaluation using
high resolution remote sensor datil and loblolly pine plots. - Prof. Geog. 37: 46-56.
Mackinney, G. 1941. Absorption of light by chlorophyll solutions. - J. BioI. Chern. 140:
315-322.
Milton, EJ. and Rollin, E.M. 1989. The Geophysical Environmental Research, Inc. IRIS Mark
IV Spectroradiometer. A guide for UK users. - Department of Geography, University of
Southampton, UK
NOAA. 1989. Climate data - Florida. - National Climate Data Center, National Oceanic and
Atmospheric Administration, Asheville, NC.
Peterson, DL 1991. Report on the workshop Remote Sensing of Plant Biochemical Content:
Theoretical and Empirical Studies. - NASA White Paper, NASA, Ames Research Center,
Moffett Field, CA. 25 pp.
Peterson, D.L. and Running, S.W. 1989. Applications in forest science and management. - In:
Asrar, G. (ed). Theory and Applications of Optical Remote Sensing. J. Wiley and Sons,
New York, pp. 429-473.
Peterson, D.L., Spanner, M.A., Running, S.W, and Teuber, K.B. 1987. Relationship of
Thematic Mapper Simulator data to leaf area index of temperate coniferous forests. - Rem.
Sens. Environ. 22: 323-341.
Peterson, D.L., Aber, J.D., Matson, P.A., Card, D.H., Swanberg, N., Wessman, C.A and
Spanner, M.A. 1988. Remote sensing of forest canopy and leaf biochemical contents. -
Rem. Sens. Environ. 24: 85-108.
Running, S.W., Peterson, D.L., Spanner, M.A. and Teuber, KB. 1986. Remote sensing of
coniferous forest leaf area. - Ecology 67: 273-276.
Spanner, M.A, Pierce, L.L., Peterson, D.L. and Running, S.W. 1990a. Remote sensing of
temperate coniferous forest leaf area index: The influence of canopy closure, understory
vegetation and background reflectance. - Int. J. Rem. Sens. 11: 95-111.
Spanner, M.A., Pierce, L.L., Running, S.W and Peterson, D.L. 1990b. The seasonality of
AVHRR data of temperate coniferous forests: Relationship with leaf area index. - Rem.
Teskey, RO., Gholz, H.L. and Cropper, W.P., Jr. 1994. The influence of climate and nutrition
on net photosynthesis of mature slash pine. - Tree Physiol. 14: 1215-1227.
Van Soest, PJ. and Wine, R.H. 1968. Determination of lignin and cellulose in acid detergent
fiber with permanganate. - J. Assoc. Off. Anal. Chern. 51: 780-785.
Wessman, C.A., Aber, J.D., Peterson, D.L. and Melillo, J.M. 1988. Foliar analysis using near
infrared reflectance spectroscopy. - Can. J. For. Res. 18: 6-11.
Woolley, J.T. 1971. Reflectance and transmittance of light by leaves. - Plant Physiol. 47:
656-662.
TWO
Modeling Radiative Transfer through Forest Canopies:

Implications for Canopy Photosynthesis and Remote Sensing
Tiit Nilson and Juhan Ross
Nilson, T. and Ross, J. 1996. Modeling radiative transfer through forest canopies:
implications for canopy photosynthesis and remote sensing. - In: Gholz, H.L.,
Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the Modeling of
Forest Productivity. Kluwer Acad. Pub!., Dordrecht, The Netherlands, pp. 23-60.
In order to theoretically describe the radiative transfer inside a forest canopy,

information must be obtained pn the following basic geometrical and optical
characteristics: the geometrical cross section of foliage elements, three-dimensional
distribution of their area volume density and the phase function. In coniferous trees
and stands, it is reasonable to consider one-yr-old shoots main foliage elements.
Theoretical problems related to the determination of optical parameters in the
hierarchical levels of needle, shoot, crown and canopy are discussed and a few
examples demonstrating the structural and optical complexities of forest communities
are presented. A brief description of the basic components of a new forest ecosystem
radiation model and some examples of the results obtained are given.
T Nilson and J. Ross, Institute of Astrophysics and Atmospheric Physics, Estonian

Academy ofSciences, EE2444 Observatory Toravere, Estonia.
23
Introduction
Radiation models are needed to quantitatively describe the absorption and partitioning
of solar energy by ecosystems and their components in assessing biosphere-
atmosphere interactions. Models are also needed to interpret remotely sensed data on
forest ecosystems and to simulate the rates of plant physiological processes in the
ecosystem. Several other applications of radiation models exist, including the analysis
of problems arising in the determination of leaf area indices (LAI) of forest canopies
by optical methods (e.g., Goel and Norman 1990, Stenberg et al. 1994).
A theoretical treatment of solar radiation penetration, absorption and reflection in
plant (particularly forest) communities involves the solution of a typical physical
problem. However, the medium in which the physical process takes place is extremely
complex because of its biological nature. Therefore, to handle the physical problem,
many simplifying assumptions concerning the community structure must be made. As
a result, simulated community structures often differ totally from those existing in
nature. Most of the simplifications are made intuitively, without estimating their
quantitative effect on the solar radiation field. Clearly, these community structures will
never be simulated in full detail, but it is appropriate to quantitatively estimate the
contributions of various simplifications to errors in the radiation field, especially in
quantities derived from the radiation field. This paper is an attempt to underline the
many problems and difficulties arising in the theoretical treatment of radiative transfer
in forest communities, rather than to provide final results ready for application.
Basic concepts
Most of the existing theoretical models of solar radiation in plant canopies are based
on the radiative transfer concept (Ross 1981). Accordingly, the central questions are
how to describe the plant (forest) community as an optical medium, to determine the
necessary optical parameters, to formulate the equation of transfer and to solve the
equation. Alternative treatments are based on geometrical optics (e.g., Li and Strahler
1985, Jupp and Walker 1996), Monte-Carlo simulation techniques (e.g., Ross and
Marshak 1991) or the radiosity equation (Gerst! and Borel-Donohue 1992). Many
extensive review papers on this subject have appeared in recent years (e.g., Goe11988,
Myneni et ai. 1989, Asrar 1989, Myneni and Ross 1991, Nilson 1992).
Essential structural properties of different tree species and tree distribution patterns
can be derived effectively by computer simulation, for example, by using the fractal
geometry approach (Borel-Donohue 1988). But while trees and the whole forest
canopy are easily generated by computer technology, the solution of the related three-
dimensional radiative transfer problem in this extremely complex absorbing and
scattering medium remains problematic, or at least impractical, for routine
calculations.
The basic definitions needed in the radiative theory of forest canopies are the same
ones introduced for large atmospheric particles (Van de Hulst 1981). The interaction
cross section and phase function as well as the volume density of the particles must be
described and also, for nonspherical particles, the distribution of their orientation in
space. Leaves, needles, shoots or branches may appear in the role of particles. After
25
these fundamental characteristics have been defined, the equation of transfer must be
solved.
Specific features of the forest radiation regime

Radiation penetration, absorption and scattering are treated in essentially the same
way in forest communities as in other types of vegetation. Structurally, however,
typical forests are considerably more complex than agroecosystems, a fact relatively
well covered by research. Although a homogeneous leaf canopy model can be
appropriate in some limited cases (e.g., a closed canopy of broadleaf trees), spatial
inhomogeneity caused by such factors as the presence of many storeys, variability of
species, size and form and distribution patterns of individual trees may significantly
affect forest radiation characteristics. The main feature of inhomogeneity is many
clustering (clumping) levels in the structure of forest communities. In conifer trees, for
example, clustering can be seen at the shoot, branch, whorl and crown levels. Because
of the considerable effects of needle clustering in shoots, conifers have a smaller
interaction cross section per unit leaf area and need to be treated differently from
broadleaf trees.
Generally, radiation models are somewhat specific: for instance, one model is
applied to radiation interception or transmission, another to photosynthetic
calculations and a third to canopy reflectance calculations. Although specific models
are obviously most useful, different models should have common parameters and
submodels, since, in any specific model, we have to calculate exactly the same
quantities. For instance, any good stand reflectance model should estimate the
radiation transmitted to the forest floor. Knowing the latter, it becomes relatively easy
to calculate the radiation intercepted by the tree storey and absorbed by the ground
layer of vegetation.
The main geometrical-optical characteristics needed for applying the radiative
transfer approach are as follows:
1. Geometrical cross section (GCS) of an element. This is defined as its
projection area (m2) on the plane perpendicular to the view direction. The
GCS determines the shadow area cast by the element and, consequently, the
range of screening of parallel rays of light in that direction. The GCS of a
nonspherical element depends on its orientation relative to the view
direction. The GCS may be defined for a single leaf or needle as well as for
a shoot, branch or trunk.
Often GCS is replaced by the relative geometrical cross section (RGCS) by
normalizing GCS with the total surface area of the particle. RGCS is
relatively insensitive to changes in element size, especially if the form of
"big" and "small" elements remains the same. By definition, RGCS
coincides with the silhouette-to-total-area ratio (STAR) introduced by
Carter and Smith (1985). However, it appears reasonable to normalize
RGCS on the basis of half the total surface area (Chen and Black 1992,
Stenberg et al. 1994). For planar leaves, RGCS defined by the half-area
basis is equivalent to Ross' (1981) G-function (the projection of unit foliage

area).
2. The volume density for the same structural elements: leaf, needle, shoot and
branch. It is also common to define the amount of leaf area (needle, woody
2
parts) per unit volume (m m- 3 ) on the half-area basis. An alternative
treatment uses the shoot number volume density (m- 3). It is important to
note that the volume density is a random function of location inside the tree
crown.
3. Phase function of elements. This describes the angular distribution of
radiation scattered by an element. This function may also be defined per
unit surface area or per element.
To demonstrate this approach, we next consider problems of radiative transfer in
coniferous forests. In conifers, the specific structure of one-yr-old shoots makes it
possible, more clearly than in broadleaf trees, to define a shoot as a clearly
distinguishable hierarc/1ical level. From the theoretical point of view, existing
theoretical models seem inadequate to describe conifers.
The structure of the tree layer of a coniferous forest may be simulated using one of
several models (e.g., Oker-Blom 1986, Nilson 1992):
1. Homogeneous layer(s) of needles, analyzing needle geometry and optical
properties
2. Homogeneous layer(s) of shoots, utilizing the shoot optical model and the
average volume density of shoots or its vertical distribution
3. An array of individual tree crowns acting as subcanopies with a random
(uniform) distribution of shoots inside
4. The same as in model 3 but also considering three-dimensional nonuniform
distribution of shoots, branches and trunks inside the tree crowns
5. The same as in model 4 but also considering three-dimensional distribution
of shoot dimensions
In the future, this list will probably be further refined to include structural models of
such elements as branches and whorls.
In cases 1 and 2, the one-dimensional radiative transfer equation approach may be
applied, while in cases 3, 4 and 5, the three-dimensional transfer problem must be
solved to the extent possible. For case 1, the main structural and optical unit is a needle
and for cases 2-5, it is a shoot.
A radiative transfer equation is formulated as a balance equation for incident and
outgoing photons for a so-called elementary volume. In case 1, the interaction of solar
radiation inside the elementary volume is considered with needles, twigs and trunks;
in the rest of the cases, with shoots, twigs and trunks.
In principle, radiation models may be derived for each hierarchical level, from the
single needle to the shoot, branch, whorl, crown and canopy. Models of lower
hierarchical levels can be nested within the higher-level models.
In the following section we consider the main optical parameters for structural units
of coniferous forests at different hierarchical levels.
27
Optical model of a needle

There seem to be no principal differences either in the internal cellular and cuticular
structures of broad leaves and needles or in their physiology and function. Optically,
however, some essential discrepancies can be observed and summarized as follows:
1. The thickness:width ratio of leaves and needles is different, which affects
needle GCS calculations. The leaf has mostly been considered a plane-
parallel scattering and absorbing medium. When leaf GCS is calculated,
leaf thickness is ignored; usually this assumption will not introduce
considerable error. For needles, the thickness:width ratio is typically from
0.2 to I; consequently, neglecting needle thickness would result in a
considerable underestimation of average needle GCS.
2. For the same reason, radiation scattering to the side cannot be neglected for
needles. It is essential to consider this fact when measuring needle optical
characteristics. Often reflectance and transmittance are measured for a
matrix of needles, where needles are laid tightly side by side. Then some
part of the radiation is scattered from one needle to another. At least for
needles with large thickness:width ratios, especially in the near-infrared
part of the spectrum, the resulting reflection and transmission
characteristics are different from those of a single needle.
3. Although not confirmed by measurements (Moldau 1965, Brakke et al.
1989, Walter-Shea and Norman 1991), the bi-Lambertian assumption
concerning reflection (excluding the specular component) and transmission
of radiation in leaves is widely applied. The physical justification for using
the bi-Lambertian assumption for needles is still more doubtful. When
illuminated by parallel rays, different points on a needle's surface have
different levels of irradiance, owing to the needle's curved surface. If we
disregard these problems and apply the bi-Lambertian assumption to
needles, the angular regions of "reflection" and "transmission" must be
defined in another way than for leaves.
4. Because of surface curvature, the specular reflection from a typical needle
surface is directed into a considerably larger region of view angles than
from a piece of planar leaf. Consequently, in order to describe the angular
distribution of specular reflection from a single needle, it is necessary to
examine the distribution function of needle (local) surface normals. The
shapes of needle cross sections vary for different coniferous species,
ranging from shapes resembling semicircles or sickles to those
approximating triangles and parallelepipeds.
5. For a needle, the penumbral effect is important at relatively short distances
from the needle. From a distance of 10 cm, the angular diameter of a needle
with a I-mm width is nearly the same as the angular diameter of the sun. As
a result, the shadow cast by a coniferous shoot can never be treated as full
shade.
Geometrical model of a needle

For a proper description of radiative transfer in coniferous trees and forests, we must
consider several geometrical characteristics of needles. Although they are
fundamentally simple, it is very inconvenient to introduce them into a theoretical
treatment. The most essential geometrical characteristics of a single needle have been
listed below and formulas for an ideal cylindrical needle of length, In' and diameter, dn'
have been given:
total two-sided area,
one-sided area,
0.5 Sn =1t In d/2 + 1t (d/2) 2;
silhouette area (measured by an optical planimeter)
Ssilh = In dn;
geometrical cross section, GCSn(a) (where a is the angle between the view
direction and needle axis),
GCSn(a) =In d n sin a + 1t (d/2)2 cos a;
inclination, en' and azimuth, n' of needle axis;
distribution function, g!Sn(e sn )' of needle surface local normals' inclination,
esn'
1
g/sn(e sn ) = . 2esn - cos 2e n ;
1t'JSIn esnt [1t/2 - en' 1t/2 + en];
distribution function, g2sn(sn-n)' of needle surface local normals'

azimuth, sn'
 sn - n) + cos e,rn sin (sn - n)] ,
needle inclination, an' relative to shoot axis and azimuth, 1\;

distribution function, glsn(asn )' of needle surface local normals' inclination,
asn' relative to needle axis,
glsn<a sn ) = 8(a sn - 7tl2),
where 8 is the Dirac 8-function;

distribution function, g2sn(Psn)' of needle surface local normals' azimuth,
Psn' in coordinates associated with the needle,
g2sn(Psn) = l/21t.
29
In the expressions for the distribution functions, the top and bottom parts of the needle
cylinder have been ignored. Here, e sn and <l>sn cannot be treated as independent
variables. On the contrary, a functional relationship exists between them. Hence, the
joint probability distribution, gs.(esn,sn) '# glsn(e sn ) g2sn(sn - <l>n).
Here and in the following expressions, all angular distribution functions are
normalized to unity (or one). For the inclination angle, we have
f f
1t12 1tI2
glsn(e) sin e de = 1, or glsn(a) sin a da = 1,
o 0
and for the azimuth angle
f
21t
g2s.( = 1.
o
Similar geometrical characteristics of an elementary volume (or a shoot) containing
a set of needles of different axis angles but (usually) ofthe same size are
2 3
volume density of needle area, un(x,y,z) (m m- ), defined on the half-area
basis or 011 the basis of needle number volume density, nn(x,y,z) (m- 3);
distribution function, gln.cen,)' of needle axis (na) inclination, en.;
distribution function, g2n.(n.)' of needle axis azimuth, <l>n.;
distribution function, gln.(an.), of needle axis inclination, an.' relative to
shoot axis;
distribution function, g2n.(~n.)' of needle axis azimuth, ~n.' in the

coordinates associated with shoot axis.
For a shoot in which all needles have the same inclination angle, vn' and no
preferred azimuth around the axis,
gln.(an.) =D(an - vn) and g2n.(~n,) = 1I(21t).

Then the GCS ev ' averaged over nonoverlapping cylindrical needles (of the elementary
volume, eJ, is
f f
21t 21t
/
GCse/e,v n) = In d n sin r \ d<l>n + 1t(d/2/ Icos l\ld<l>n' (1)
o 0
where r = (e, - <l>n).
In a more general case, when the GCS is averaged over an arbitrary needle axis
inclination angle distribution, glna(an.), with no preferred azimuth, Eg. (1) can be used
as an auxiliary function to calculate the GCS ev ' namely,
1tI2
GCS ev (8) = f GCS n(8,a n.) glna(ana ) sin ana dana'
o
From Eg. (1), the relative geometrical cross section (RGCS) of needles having the
same inclination angle but distributed uniformly with respect to azimuth and
calculated on the half-area basis is
RGCS eJ8,v n) = GCS eJ8,v n)/(0.5 Sn) = GCS eJ8,v n)/[1t d/2 (In + d/2)].
An example of RGCS ev as a function of view angle, 8, calculated using the last
expression, is given in Figure 1. Two versions of needle angle relative to shoot axis,
35 and 40, respectively, have been used. The needle cylinder length was assumed to
be 3.2 cm and the diameter 0.1 cm.
It can be shown that for a spherical needle axis distribution (glna(an.) = 1)
RGCS eJ8) = 0.5,
i.e., it is the same as for spherically distributed leaves.
f
The integral in Eg. (1), Icos /\ldn' can be analytically evaluated. It coincides
with the expression of the RGCS (G-function) of leaves with fixed inclination angle
and uniform distribution with respect to azimuth (Ross 1981).
0.6
---
4(J0 __ . . -::
(/)
o
,-,0.4
a:::: 1....---35
w
~w
~0.2
0.0 +---r-,....----,---,r---r--,--.---.---,
30 60 90
VIEW ZENITH ANGLE, degrees
Figure 1. Average relative geometrical cross section of cylindrical needles as a

function of view zenith angle. All needles have the same inclination angle of 35
or 40, respectively; the azimuthal distribution of the needles is uniform.
31
Phase function of a needle

Unfortunately, too little information about the scattering properties of single needles is
available. Here, we consider a small surface element of a needle that may be treated as
a piece of plane. For this surface element, the phase function could be defined as it is
for a leaf (Ross 1981). The whole needle might be viewed as a sum of individual facets
and the needle phase function can be calculated as a sum of contributions from all
illuminated and visible facets. However, the photons entering the needle from one
facet and emerging from another also must be dealt with in some way. To properly
describe the needle phase function, the entire three-dimensional radiative transfer
problem for the needle must be solved.
We can define the phase function (SPF) for the needle, Yn(rna,rn~r)dru1t, as the
probability that a photon will be scattered by the needle of unit (one-sided) area from
the incident radiation direction, ro' into a solid angle, dQ, about direction r. If the
intensity of illuminating radiation is 10, then the energy scattered by the whole needle
into the solid angle, dQ, is
10 RGCSn(ro,r n.) Yn(rna,ro~r) dQ/1t 0.5 Sn RGCSn(r,rna).
The integral over all view angles yields the scattering coefficient of the needle
(O(ro,rn.) =f Yn(rna,ro~r) RGCSn(r,rn.) dru21t.
41t
For an elementary volume containing a set of nonoverlapping needles with the axis
distribution function g(rn.) = g(e na ,<1>n.)' the SPF, r(ro,r), is given as
f
r(ro,r) = g(rn.) Yn(rna,ro~r) RGCSn(ro,rn.) RGCSn(r,r n.) dQ n/21t.
21t
As in planar leaves, the scattering act may be divided into "reflection" and
"transmission" depending on whether the point of interest is on the illuminated or the
reverse side of the needle.
Reflectance and transmittance

For flat leaves, Yis often given by a bi-Lambertian approximation assuming Y== R for
reflection and y== T for transmission: one flat side of the leaf represents reflection and
the other transmission. In more sophisticated models, the specular reflection
component from the leaf surface is considered separately. For a needle, the transition
region from "reflection" (illuminated side) to "transmission" (reverse side) is
determined by the perpendicularity of the incident ray and the local surface normal. At
the transition point from "reflection" to "transmission," the local surface normal is
perpendicular to the incident radiation direction, that is,
cos r~rsn = cos eo cos e sn + sin eo sin e sn cos (<1>0 - <1>sn) = O.
Similarly, the visible and invisible transition point from the view direction parts of the
needle are separated by the condition cos rl\rsn = O. The bi-Lambertian assumption for
a needle surface point yields
_{R' if cos rs~rn > 0and cos r~r > 0;

Yn,(rsn,ra~r) - 1\ 1\ (2)
T, if cos rsn ra < 0 and cos rsn r > O.
The irradiance at the illuminated point on the needle surface is proportional to cos
rs~ra; hence, the radiances emerging from these points must also be proportional to R
cos rs~ra' To obtain the SPF for the whole needle, we must total the radiances of all
surface points. We may assume that
Yn(rna,ra~r) = Rff cos rs~ra gsn(esn'cPsn ) sin e sn desn dcPsn

Q
+ ff
1
T gsn(esn'cPsn ) sin e sn desn dcPsn ' (3)
~
where the regions of integration Q, and Q 2 correspond to upper and lower conditions
in Eq. (2), respectively.
Although it may be inconvenient to calculate integrals in Eq. (3) for needles of a
complex shape, it allows the relative portions of reflection and transmission for any
combination of incident and view angles to be determined. Thus, the needle phase
function will be represented as a certain linear combination of needle reflectance and
transmittance.
Except for the near-infrared part of the spectrum, the transmission of radiation
through typical needles is considerably smaller than the reflection. As a result, the
larger the illuminated part of the needle exposed in the view direction, the larger the
value of the SPF. Consequently, the maximum needle SPF must be in the direction of
backward scattering and the minimum in the forward direction. This is somewhat
analogous to the "hot-spot phenomenon" (Jupp and Walker 1996) and forms an
essential part of radiation scattering in needles and coniferous shoots.
As with leaf optics (e.g., Vanderbilt et al. 1991), at least when remote sensing
applications are considered, it is reasonable to distinguish between the specular and
the internal (diffuse) reflection components. The specular component, Ys p ' is caused by
reflection on the needle's cuticular wax surface, while the internal component, Yd' is
due to interactions ((multiple) refractions and reflections) with internal mesophyll
structures. Hence, we have
Yn = Ysp 8(r-rp) + Yd'

where r p defines the direction of the needle surface normal of specular reflection. Note
that here, with a fixed incident radiation direction, we can find local surface normal
directions satisfying the conditions of specular reflection for more than one view
direction, because the needle surface may be curved. For a planar leaf, if small-scale
surface undulations are neglected, only one specular reflection direction will be noted.
33
At this point, the importance of local normal distribution functions of the needle
surface becomes evident.
Specular reflectance
In calculating the SPF of specular reflection from a needle surface, formulas derived
for the elementary volume of a leaf canopy may be applied (Nilson 1991, Vanderbilt
et al. 1991), namely,
(4)
where
2 2
Ylsp(aO,n) = [sin\ao-i)/sin\ao+i) + tan (ao-i)/tan (ao+i)]I2,
and sin i = (sin ayn
is calculated by the Fresnel formula. Here, n is the refractive index of the cuticular
wax and a o is the angle between the local surface normal and the view direction.
Sometimes a factor K is introduced into Eq. (4) to account for possible deviations
from the Fresnellaw (Brakke 1994). The direction, rp' of the surface normal for any
pair of incident and view directions is calculated on the condition that incidence angle
and specularly reflected radiation angles are equal and all directions, ro' rand r p' are
situated in one plane. However, here the distribution function, g(rp)' must be
interpreted as the distribution of surface normals (in the part of the needle that is
illuminated and visible at the same time).
If the needle is modeled as a cylinder, the distribution function of its surface
normals inclination is a O-function and, as a consequence, specularly reflected
radiation intensity for a given pair of ro and r is either zero or infinity (i.e., also a 0-
function). However, if we consider angular regions of incident or view angles to be
finite, as is the case in real situations, or to be averaged over a set of needles with some
nonsingular distribution, the intensity of specular reflection becomes finite.
In one example, we calculated the average SPF of the specular reflection
component of cylindrical needles having the same inclination angle, v = 35, and a
uniform azimuth distribution, thus simulating a Scots pine (Pinus sylvestris) shoot
with a vertical axis (Fig. 2). In this case, Eq. (1) for the surface normal distribution and
Eq. (4) for the specular reflectance were applied. The refractive index for cuticular
wax was assumed to be lA. In order to obtain finite values of the specular reflectance
phase function, calculations were made with the angular dimensions of the radiation
source approximately equal to that of the sun (0.5 in diameter). Three different
directions of incident radiation were considered: 8 0 = 0 (full lines, with illumination
parallel to shoot axis), 8 0 = 45 (dashed line) and 8 0 = 90 (dotted line, with
illumination perpendicular to shoot axis). The view directions are in the principal
plane (the vertical plane through the illuminating source), as shown in Figure 2a, and
in the perpendicular to the principal plane, as shown in Figure 2b. Note that in many
view directions, specular reflection from the surfaces of cylindrical needles of the
given configuration is not possible. For instance, at 8 =90 and forward reflection (
0
= 180), there is only one peak intensity at 8 = 90, but in all other view angles it is
I
I
I
I
I
I
0.20 a I
0.20 b
z I Z
0 I
0
i= i=
u I U
5 015
l.L
I
I
I
5 015
l.L
W I W
(j) I (j)
/45
I 0.10 /
I 0.10
0- 0-
I
tl\J tAJ
0::: / 0:::

.-J ;90
.-J
3 0 .05 I
I 3 005
w I W
90
~
0- 0- ,90 ~5: ,
(j)
,~O ,45
, : I
(j)
. , '
I
0.00 I ' ii, i I

:'- /
i i ~ I I 0.00
~ __ .. __ .. _\.. -J - - - - :.... .......__.'. . ..
0 45 90 135 180 225 270 315 360 o 45 90 135 180 225 270 315 360
VIEW ANGLE, degrees VIEW ANGLE, degrees
Figure 2. Calculated specular phase function of cylindrical nonoverlapping

needles having the same inclination angle (35) and distributed uniformly with
respect to azimuth: (a) view directions in the principal plane; (b) in the
perpendicular to the principal plane. View angle 8 = 0 corresponds to the z-axis
("shoot axis"). D1umination angles, 80' are given in the figure at the respective
curves.
zero. In the perpendicular plane, the specular phase function is symmetrical. The
analysis shows us that both factors in Eq. (4) - the angular distribution of surface
normal inclination and the angular distribution of the specular component according
to the Fresnel law - appear to be important in forming the averaged phase function.
These results might be interpreted as representing the specular SPF of the model
shoot, in case there were no mutual shading of needles in the shoot. We would expect
a relatively large forward reflection when the illumination direction and shoot axis
angle coincide. However, this is the angular region of minimum transmission through
the shoot; a considerable number of small "mirrors" are screened by other needles in
the shoot. At the same time, the large forward reflection in the case of 8 0 = 90 is
screened to a lesser extent and may have an important effect on the shoot phase
function. In measurements, however, it is difficult to distinguish forward scattering
from the transmitted (nonintercepted) radiation. The effect of mutual shading will be
considered subsequently.
For remote sensing applications, it is important to remember that, for spherically
oriented needles, the distribution of surface normals is also spherical. If the shoot
structure effects are abandoned, the specular phase function derived for spherically
oriented leaves (Nilson 1991) becomes applicable.
35
Optical model of a coniferous shoot

Optical properties of shoots can be derived from those of the needles and shoot
structure. The geometrical characteristics necessary for deriving the GCS of a shoot
and its envelope size and form are
total amount (area) of needles;
average dimensions of needles or their distribution function;
distribution of needle axis inclination and azimuth with respect to shoot
axis (in some cases, joint probability distribution of needle size and
azimuth, e.g., the shoot of Norway spruce, Picea abies); and
distribution of needle attachment points along the shoot axis.
GCS and RGCS

Radiation intercepted by a shoot is proportional to its GCS or projected area. Usually,
a considerable mutual overlapping of needles takes place, so that the actual GCS of a
shoot is considerably less than the total GCS of all needles belonging to the shoot and
having the same inclination and azimuth angles as those within the shoot. As a result,
the extinction coefficient of radiation is considerably less in conifers than in broadleaf
trees for the same amount of leaf area, even if one-sided needle area is considered.
This problem and its implications for radiation extinction in coniferous forests,
evaluation of shoot and canopy photosynthesis and determination of needle area index
from optical measurements have been adequately covered in several papers (Norman
and Jarvis 1975, Carter and Smith 1985, Oker-Blom et al. 1986, Oker-Blom and
Smolander 1988, Chen et al. 1991, Gower and Norman 1991, 1991, Stenberg et al.
1994). Usually, a certain shading or overlapping index or STAR (Carter and Smith
1985) is introduced to account for this phenomenon.
The extent of overlapping depends on illumination angle relative to shoot axis, on
needle angles relative to shoot axis and on volume density and distribution pattern of
needles along the shoot axis.
As a first approximation, the envelope of a typical Scots pine shoot might be
modeled as a cylinder. Thus, its cross-sectional areas, GCS env or RGCS env ' can be
calculated easily. The shoot is always semitransparent, so the GCS of a shoot, GCS sh '
may be expressed as
GCS Sh = GCS env (1 - a 1),
where a 1 is the transmission coefficient through the shoot or the probability of gap in
the shoot envelope projection region. Both GCS env and a 1 depend on the view
direction. In addition, an elementary volume consisting of identical shoots with a
fixed inclination angle and a uniform azimuth distribution of shoot axes might be
considered. Average RGCS env ' a j and RGCS sh were calculated for a model shoot with
the following parameters: length of cylindrical needles = 3.2 cm, width = 0.1 cm,
needle angle relative to shoot axis = 35, shoot axis length = 10 cm, number of needles
2
= 200 and total one-sided needle area = 102 cm . The transmission coefficient was
calculated as the probability of no overlap according to the Poisson distribution, while

the Poisson parameter - the mean value of overlaps - was calculated from
geometrical considerations as the ratio of total GCS of needles to GCS env . The average
coefficient of transmission as a function of shoot axis inclination angle at different
view zenith angles and the average RGCS of the shoot are illustrated in Figure 3. The
average transmission through the shoots is surprisingly constant, except when the
vertical shoot is viewed from the tip. The RGCS of spherically distributed shoots was
found to be approximately 0.32. This value is relatively close to the measured
estimates for Scots pine shoots (average 0.284 and standard deviation 0.044) obtained
by Oker-Blom and Smolander (1988). A possible explanation for the difference in
these values may be that the shoot transmission coefficient was calculated using the
Poisson formula, which may not be an adequate descriptor for the distribution of
needle overlapping.
~
U
SO.4 ~~~90~~3~~_~
8
__~~~- 75
45 60
0.4
z
o
~
~02
l/l
~
I-
tl
o
+-........----.-,--.-~---,r_-.-....,....__,
60
SHOOT INCUNATION ANGLE, degrees
90
~ 0.0 +----r-~..___::r_---r-~.--.-....,....----.,........, 0.0
60
SHOOT INCUNATION ANGLE, degrees
90
Figure 3. Calculated average transmission coefficient (left) and relative

geometrical cross section (right) of cylindrical shoots of Scots pine as functions of
shoot axis inclination angle. The shoots are uniformly distributed with respect to
azimuth. Numbers at the curves show the values of the view zenith angle.
Shoot phase function

The importance of the shoot phase function (SPF) has been clearly underestimated in
the interpretation of remotely sensed data on coniferous forests. Systematic
measurements of shoot SPF are practically lacking (Ross et al. 1994). Moreover, in
spite of their apparent simplicity, deriving a good optical model for coniferous shoots
poses an inconvenient theoretical problem.
We can define the shoot SPF, ys(rsa,rn,r)/rt dQ, where rsa is the direction of the shoot
axis, as the probability that a photon travelling in the direction, ro ' will be scattered by
the shoot into the solid angle, dQ, about r, rsa being the direction of the shoot axis. The
phase function, Ys' may be defined on the basis of individual shoots or as an area phase
function. In the latter case, it is reasonable to normalize Ys with the GCS of the shoot
37
in the view direction. Then the energy scattered by a shoot from the incident radiation
direction, ro' into the view direction, r, is given by
10 RGCS(ro,rs.) y.<r5O,ro~r) dQht 0.5 Sn RGCS(r,r5O).
The integral SPF over all view angles gives us the scattering coefficient of the shoot
f
(O(ro,r5O) = ys(rsa,ro~r) RGCS(r,rsa ) dQ/21t.
41t
The SPF for an elementary volume that contains a set of nonoverlapping shoots
having an axis distribution function, g(rs.) = g(asa'Psa)' is expressed as
r(ro,r) = f g(rs.) y.<rsa,ro~r) RGCS(ro,rsa ) RGCS(r,rsa ) dQ./21t.
21t
Following are some preliminary results of the Scots pine shoot SPF measurements
made at the Toravere Observatory in Estonia in 1993 (see also Ross et al. 1994).
Shoots were illuminated with a horizontal monochromatic beam with a diameter of
120 mm. Scattered radiation was measured from the illuminated part of the shoot with
a view field diameter of 100 rom in the horizontal plane. Three values of shoot axis
inclination were used: a = 0 (vertical shoot), a sa = 90 (horizontal shoot) and a sa =
50
45 (inclined shoot). The shoot azimuth, P5o (measured clockwise from the incident
radiation beam), had the following values: P5o = 0, with the shoot parallel to the
incident beam and the shoot top directed backwards to the beam; Psa = 90, with the
shoot twig perpendicular to the incident beam and the left side of the shoot
illuminated; and P 5o = 270, with the shoot twig perpendicular to the incident beam
and the right side of the shoot illuminated. While the shoot cross section in the plane
perpendicular to the twig was not fully circular (owing to a smaller needle density on
the lower side than on the upper part), the shoot position was characterized by an
additional angle qt, with the shoot rotation angle around the shoot axis. The angular
distribution of radiance was measured in the incident radiation (horizontal) plane at
the angle intervals 5-170 and 190-355 relative to the incident beam.
Measurements were made in the blue (485 nm), green (555 nm), red (650 nm) and
near-infrared (790 nm) spectral regions. Since these measurements were conducted
only on a single plane, normalization of the SPF was accomplished by dividing the
individual output signal at a certain scattering angle by the sum of the signals from all
measured directions.
The shoots of Scots pine were collected in December from the top part of a 13-yr-
old tree growing in the south side of a young forest near the Observatory (shoot IT)
and from the top part of a IO-yr-old tree (shoot 4.1) growing by the roadside of the
Tallinn-Tartu motorway. The structural characteristics of current-year shoots used in
the measurements are given in Table 1.
Table 1. Values of structural characteristics for the shoots used in the

measurements. Is: shoot length (cm); d s: shoot cylinder diameter (cm); db: twig
diameter (cm); nn: number of needles per 1 cm twig; an: the angle between twig
and needle; In: needle length (mm); d n: needle width (mm); d n2 : needle thickness
2
(mm); ~: needle dry mass (mg); Sn: needle area (mm ); and Sos: needle total
2
area per 1 cm twig (cm cm,I).
Shoot
IT 26 8 0.69 10.4 20-40 68 1.75 0.79 27.9 298 31

4.1 28 5 0.7 17.2 30 47 1.7 0.8 17.1 191 32
Following are some results of the SPF measurements:

Experiment 1. Vertical shoot perpendicular to the incident light beam (8 s = 0). The
upper part of the shoot is illuminated ('I' = 0). Figure 4 shows that in all
wavelengths the maximum scattering is in the backscattering direction. Moving
away from the backscattering direction, the phase function decreases. In the blue
and red region, the scattering is very low between scattering angles 80 0 and 170 0
In the green and near-infrared regions, the decrease in SPF is slower. The angular
dependence can be explained by changes in the ratio of illuminated ("reflection")
and shaded ("transmission") parts of needles. In the green and near-infrared
regions, the brightness contrast between illuminated and shaded parts of needles is
relatively lower than in the blue and red parts of the spectrum and, consequently,
the angular dependence of SPF is less significant.
O.OJ
Z
Q
0
z
::>
.....
00;1
.,

VJ ,

T
0-
: I
0 , J
~OOI ,
..
" I
~
0' "
I
1-
,
/

.~
0
V1 ... --
... - __
000
~-----........ ------~~-v- 1 ,
0 60 1;10 80 '40 300 .3 0
"r/l'1f t; A ',1 r ",'oren,
Figure 4. Phase function of a Scots pine shoot (shoot 4.1) in four different
spectral regions: blue (B), green (G), red (R) and near-infrared (NIR). The shoot
is in the vertical position.
39
Experiment 2. The influence of the angle of incidence on SPF. Measurements were

carried out in four shoot positions: (i) vertical shoot, perpendicular to the incident
beam (a sa = 0); (ii) horizontal shoot, with the twig parallel to the incident beam
and the shoot top directed toward the incident beam (as. = 90, = 0); (iii)
inclined shoot, with the shoot top inclined toward the incident beam (as. = 45,
 = 0); and (iv) inclined shoot, with the shoot top inclined to the opposite
direction of the incident beam (a sa = 45, = 180). In all experiments the upper
side of the shoot was directed toward the incident beam ('P = 0). In the blue region
(Fig. 5) in cases (i), (iii) and (iv), the shape of shoot SPF is similar. With an increase
of up to 90 in the scattering angle, the SPF will monotonously decrease; a local
SPF maximum exists between scattering angles 110 and 250 in case (iii). In case
(ii) with top illumination, the shape of the SPF is essentially different: from the
backscattering maximum, the SPF rapidly decreases practically to zero at 20 and
remains zero between scattering angles 20 and 120 and between angles 240 and
340. The rapid vanishing of the maximum near the backscattering direction (hot
spot) compared with other ill~mination configurations may be explained by the
smaller angular correlation radius in the bidirectional gap probability. (The view
direction nearly coincides with the axes directions of many needles.) In forward
scattering directions between scattering angles 150 and 240, a large SPF
maximum exists. In the red region, the situation is similar. In the green region (Fig.
6), the difference between (ii) and other positions is not so significant, though great
forward scattering in case (ii) still remains. With an increase in the scattering angle,
the SPF will decrease up to 150. The shoot SPF is least variable in the infrared
region (Fig. 7). Note that between certain scattering angles (70 and 290), the
influence of the shoot position on the SPF is particularly small.
I
1
O. I
,/ I
...o
u
I
,
I
~OOJ 1\
"
I
Vl 12 I
<l: I
l002 I I~ I
111 I j
(.)
Z
IE
~\:\"I. , ~
t:?O
<l:
U
~ .... : .' '4 .:.....:
(/) ~.-Y~,
-::...,,:;:: ,..
,
.L
0.00
o 120 180 240 .~ 360
SeA' RI GAG E. degreHs
Figure 5. Phase function of a Scots pine shoot (shoot 1T) at different directions of
the shoot axis with respect to the incident beam. 1: a sa = 0; 2: asa = 90, <l>sa = 0; 3:
as. = 45, <l>sa = 0, asa = 45, <l>s. = 180. Blue spectral region.
om5
Z
o
i=
U
z
::;)
l<-
0010
w
Vl
<l:
Q..
"
I
o
a::
w
~OOO5
"i
l-
I-
<l:
U
Vl
0.000
o 60 120 180 240 300 360
SeA ERI G A GLE. degrees
Figure 6. The same as in Figure 5, but for the green spectral region.
0.010
Z
o
i=
~0008
::;)
l<-
w 0006
~
Q..
0 0 .004
Z
CE
w
:= 0002
<l:
U
VJ
0.000
a 60 120 180 240 JOO 3&0
SCATTERI G A GLE. degrees
Figure 7. The same as in Figure 5, but for the near-infrared spectral region.
The curves with 8sh = 90 and = 0 in Figures 5-7 are comparable to those with
80 = 0 in theoretical calculations (Fig. 2). Smaller maximums at scattering angle
120 (and 220 on the other side of the shoot) are close to those predicted by the
Fresnel formula, using the distribution of surface normals of cylindrical needles.
The maximum around the forward scattering angle (180) can be explained by the
specular reflection on needle surfaces, since it is hardly detectable in the near-
infrared. The angular width of the latter maximum is about the same as predicted
by theory.
Experiment 3. The purpose of this experiment was to clarify whether illuminating
the top or the middle part of the shoot had any effect on the SPF. It appeared that
illumination affected the angular scope of the hot-spot reflection: the region of
greatest backscattering is larger when the middle part of the shoot is illuminated.
41
Between scattering angles 20 0 and 1700 as well as between angles 200 0 and 3500 ,
scattering is very small in the blue and red regions. In the near-infrared, between
scattering angles 40 0 and 1600 as well as between angles 250 0 and 3200 , the SPF is
greater when the top is illuminated than when the middle part of the shoot is
illuminated. The results of these measurements have demonstrated that the shoot
SPF is a variable quantity that depends on the internal structure of the shoot, the
geometry of illumination and the wavelength. The shape of the SPF is greatly
influenced by the proportion of illuminated and shaded needles visible in the view
direction (i.e., the hot-spot phenomenon), by the brightness contrast in needle
reflectance and transmittance and by the specular reflectance from individual
needles. The latter effect can be observed as small, bright flecks in certain view
directions. In spectral regions where needles transmit practically no radiation (blue
and red), the angular course of the SPF is very sensitive to hot-spot and specular
reflection. These effects are less notable in the green region and much less notable
in the near-infrared. In the near-infrared, because of considerable radiation
transmission through needles and multiple radiation scattering inside the shoot, the
angular dependence of SPF is,'markedly lower.
Theoretical modeling of shoot phase function

To elaborate a theoretical model of shoot SPF, the radiative transfer problem for a
single shoot must be solved, including the calculation of multiple scattering between
needles. Assuming that we have a shoot geometrical model and all necessary
parameters to construct a shoot (i.e., needle number and size, distribution of needle
angles relative to shoot axis and distribution of needle attachment points on the shoot
axis), we may calculate the single phase function of the shoot as follows.
By fixing its attachment point and axis direction, we choose a needle from the
shoot. This may be done systematically for all needles in a sequence or by random
rule. We divide the needle into smaller portions representing inner, medium and outer
parts of the needle in the shoot. For each part of the needle (at a point (x,y,z, we
calculate the bidirectional gap probability, p(x,y,z,ro,r), in the incident radiation
direction, ro' and the view direction, r, and multiply it by the volume density of one-
sided needle area, u(x,y,z), and the value of needle SPF, Yn(x,y,z;rn,ro,r). The phase
function will include the diffuse as well as the specular reflectance components
discussed previously. The global first-order-scattering radiance of the shoot is
calculated by combining all contributions from different portions of the same needle
and from different needles. In other words, we must calculate the integral
JJJYsh(rs,ro,r) =n(x,y,z;rn,ro,r)p(x,y,z,ro,r)u(x,Y,z)dxdydz,
V(rs)
where the integration is performed over the whole shoot envelope volume, V(r.).
Calculation of the bidirectional gap probability

Here, it is reasonable to refer to a direct analogy between the radiative transfer
problem for a single shoot, a single crown and the forest canopy as a whole (Nilson
1992). We must consider the bidirectional gap probabilities to correctly solve the
radiation scattering problem and to integrate the hot-spot phenomenon.
The bidirectional gap probability has an extremely important effect on the angular
distribution of scattered radiation from the shoot. It magnifies the contribution of
needles well exposed to the view and illumination directions. In spite of small needle
width and thickness, the mutual dependence problems of the gap probabilities in the
illumination and view directions cannot be avoided. The geometrical significance of
mutual dependence may lie mainly in the fact that a single needle can screen both the
view and the radiation source directions at the same time. Principally, its theoretical
treatment is the same as for leaf canopies (Kuusk 1991). However, additional
complications arise from a large difference in needle length and width. As a result, the
angular correlations in the gap probabilities decrease rapidly in directions
perpendicular to the shoot axis (perpendicularly to the needles) and much more slowly
in the plane coinciding with the shoot axis. Numerically, however, it is possible to
calculate the bidirectional gap probability for any point within the shoot and in any
two directions by identifying a certain "vulnerability" area on the plane of two
coordinates - the distance between the shoot base and the needle azimuth at the point
of needle attachment on the shoot axis. If a randomly chosen needle falls inside this
vulnerability region, both directions will be screened at the same time.
The major factors assumed to determine the shape of a shoot phase function are the
projected shoot area in the view direction and the incident radiation direction and the
probability of seeing illuminated needles, especially those parts of needles whose
specular reflectance components occur in the view direction.
Because of mutual shading, the outermost needles will contribute most to the fust-
order phase function. While multiple radiation scattering inside a shoot is often
ignored in the visible (PAR) region (Ross 1981), it is important in the near-infrared.
For theoretical treatment of multiple scattering, the associated three-dimensional
transfer equation must be solved. Since calculation of both single and multiple
scattering inside a shoot causes computational problems, we hope that some kind of
parameterization in a shoot phase function can be developed for routine use.
In forest reflectance models, shoot phase functions must be averaged over the
respective shoot axes distribution function (i.e., for an elementary volume). Initially,
the chief question is how to calculate the phase function for spherically oriented
shoots. To date (e.g., Nilson and Peterson 1991), the phase function of spherically
oriented leaves has been used for conifers. No estimates of the error that may be
caused by disregarding shoot structure are available.
Next, we consider and compare two hypothetical experiments: (i) measurement of
the radiance of a real coniferous shoot, and (ii) measurement of the radiance from the
same amount of needles (and the twig), but with needles randomly dispersed in a
larger volume with average volume density of needles in a typical crown, preserving
needle inclination and orientation angles. The resulting radiances will be different in
these two cases because of the following reasons:
1. The amount of needles visible in the view direction is always larger for
dispersed needles because of extensive mutual shading of needles within
the shoot. Invisible needles do not contribute to the measured (single
43
scattering) radiance. Therefore, in order to compensate for the effect of

needle clustering in shoots, needle amount must be diminished by means of
a specific correction or shading factor. The value of this correction factor is
a function of shoot angle with respect to view direction.
2. For the same reason, the amount of illuminated needles is larger for
dispersed needles. As a result, the average brightness of needles is larger for
dispersed needles than for a shoot. To compensate for this effect, the
average reflectance factor of a single needle must be diminished. However,
no reliable quantitative estimates for this correction have been provided
thus far. Similar statements have been made by Norman and Jarvis (1975).
More precisely, the bidirectional gap probability in the incident radiation
and view directions must be considered for all needles. The two random
events (the occurrence of gaps in the view and incident radiation directions)
can be mutually dependent, especially in the backward scattering direction.
This dependence gives rise to the shoot hot-spot effect. The angular scope
of this dependence is .iarger for a shoot, since the angular dimensions of
shadowing neighbouring needles is larger than in dispersed needles. Thus,
the hot-spot effect for dispersed needles is considerably sharper than for a
shoot.
3. The radiance caused by multiple scattering between needles is different in
these two cases. Since the number of photons in a single scattering is larger
for dispersed needles than for a shoot, there are more radiation "sources"
available for multiple scattering. But the probability of a photon's escape
from the volume of dispersed needles is larger than from a shoot at any
scattering angle, again due to the smaller angular dimensions of adjacent
needles. In other words, an intercepted photon is "trapped" by the shoot.
The mean scattering order inside a shoot is clearly higher than for dispersed
needles. However, the probability of a photon's escape from a tree crown
consisting of uniformly distributed needles is smaller than from a crown
containing an equivalent number of shoots. At the same time, when the
order of local multiple scattering increases because of needle clustering in
shoots, the order of global multiple scattering decreases. These two
mechanisms compensate for each other to some extent. It is not trivial to
ascertain which of the effects will be dominant in reflectance from the
whole canopy. Some preliminary estimates have shown (Nilson 1991) that
clustering will decrease the reflectance of multiple scattering in the near-
infrared region, where multiple scattering is most significant.
Optical model of a tree crown
Form of crown envelope

Usually, simple geometrical surfaces are used as models for crown envelopes (cones
for conifers and ellipsoids for broadleaf trees). As a rule, however, these models are
unrealistic. Because crown form adapts to that of neighbouring trees, most tree crowns
are not rotationally symmetrical; consequently, asymmetrical models (e.g., Koop
1989) are preferred. Unfortunately, applying asymmetrical and thus more realistic
models introduces additional parameters that need to be determined in one way or
another. To date, no good estimates of the magnitude of error in radiation
characteristics caused by approximations of crown form are available.
Three-dimensional distribution of needle area

In most theoretical models, volume densities of needle area or shoot number are
assumed to be uniform over the crown envelope, though measured needle area
distributions are quite irregular. To illustrate this point, we consider here some
unpublished data on Scots pine crown structure.
Measurements were carried out in June 1982 in Haapajarvi, 50 km north of
Petrozavodsk in the former USSR, near the Forest Field Station of the Forest Institute
in the Karelian Branch of the Academy of Sciences. The site was of low fertility
(Calluna type). A detailed description of the study area has been published (Hari et al.
1985). Data are presented for a model tree with parameters corresponding to those of
the average tree of the stand: tree age 35 years, height 5.6 m, crown length 4.0 m,
2
crown radius 0.85 m, total (one-sided) needle area 7.58 m and tree needle area index
3.37.
For the structure of the whorls see Figure 8. The shape of the branches represents
the mean of all branches of the whorl considered. The crown envelope of this tree can
be modeled approximately as a cone in the upper half and as a cylinder in the lower
half. The needle area distribution inside the envelope is far from being uniform either
NEEDLE AREA PER 1 OM LAYER. dm2

o 0+-:::::-r'~~_~-:l20c.....~_~~
tOO

:i 2oo
e;o
300
400 50 4000
DISTANCE FROM TRUNK, em
Figure 8. Left: crown structure of the model tree. Right: vertical distribution of
needle area in the model tree. The vertical axis, depth, is measured downwards
from the top of the tree. T =total area, needles of different age: 1 =current year;
2 = previous year; 3 = third year; 4 = fourth year; 5 = fifth year.
45
vertically (Fig. 8) or radially (Fig. 9). Isolines of average needle area density are given
in Figure 10. We can see large differences in needle area density over the crown. The
absolute maximum needle area volume density is located in the upper part of the
crown. As well as increasing the depth, the maximum needle area density at fixed
heights is shifted away from the tree trunk. In the lower half of the crown, there is a
nearly conical space devoid of needles around the tree trunk.
'25 a 1.25 b
E
1)
"-1.00 ~
.00
>-"
I-
Ui Ui
675 675
a a
<{
~0.50
Q' l}t050
<! <{
W w 7,
5025 5025
w w
w w
Z Z
0.00 000
0 '00
Figure 9. Radial distribution of needle area in the model tree crown in different
layers from the top of the tree. a: 1 = 0-40; 2 = 40-80; 3 = 80-120; 4 = 120-160;
5: 160-200. b: 6 = 200-240; 7 = 240-280; 8 = 280-320; 9 = 320-360; 10 = 360-400
em.
DISTANCE FROM TRUNK, em

20 40 60 80 100
o~~~--'---'--'----'----'---'
100
E
o
200
l-
n..
w
o
300
400
o
Figure 10. Smoothed isolines of the needle area volume density (dm2 dm-3) inside
the model tree crown. Bold line (max) shows the location of the density maximum
at each height.
In addition to the nonunifonn distribution of the mean value of the needle area, a
mixed type (clustered-regular) random distribution of needle area volume density
must be taken into account. If we divide the crown into smaller volume cells, many
empty cells appear (i.e., there is clustering). At the same time, clusters are often
distributed regularly compared with entirely random distributions. All of these
structural peculiarities affect the distribution of gaps inside the canopy as well as
radiation penetration, interception and scattering. Unfortunately, only limited
evaluations of the function of these structural effects in transmitting radiation to the
forest floor are available (e.g., Chen and Black 1991, 1992). Smith et al. (1993)
applied the LI-COR LAI-2000 instrument in estimating LAI of Douglas-fir
(Pseudotsuga menziesii) and, as is typical in conifers, found that this optical method
underestimated LA! (by 62%). Smith et al. explained this underestimation both by
needle clustering in branches (74% contribution) and by nonrandom branch
distribution (26% contribution).
If we consider shoots as minimum structural units, then the variability of shoot
dimensions and total needle area per shoot must be described quantitatively. Shoots
and needles in well-illuminated parts of the crown are larger than those in deep shade.
Additional problems arise from structural changes shoots undergo with age. As a rule,
current-year shoots have smaller diameters because needle angle relative to shoot axis
increases with shoot age. Other effects of aging include needle loss and increased
shoot transparency.
Angular distribution of shoot axes is an additional important distribution function
that may affect the radiation field. Figure 11 represents the distribution function of
shoot axis inclination in different layers of the same Scots pine tree crown considered
previously. In the uppennost layer, most of the shoots were vertical. Moving toward
2.5 a 25 b
t;20
i= 2 6 70
I::
U U
Z Z
::J .5 ::J1.5
t.. t..
Z
g::J .0
Z
g::J .0
aJ m
i= i=
~0.5 ~05
0.0 00
60 90 0 .:so 60 90
ITH ANGLE. de rees S'~OOT AXIS ZE ITH A 'CLE. degrees
Figure 11. Weighted by the needle area distribution function of shoot axis
inclination at different layers of the crown of a Scots pine tree. Numbers of layers
are the same as in Figure 9; curve T corresponds to the distribution averaged
over all layers.
47
lower layers, the average shoot inclination angle increased. Averaged over all layers
and weighted by needle area, maximum shoot inclination distribution occurred in the
interval between 10 and 40. The shoots were distributed more or less uniformly
relative to azimuth angle. This type of shoot axis distribution results in more radiation
transmission from the uppermost layers to deeper layers of the crown in vertical and
near-vertical directions compared with a spherical shoot axis distribution. This also
means that deeper layers have better exposure to remote sensing instruments
measuring reflectance from the nadir direction than would be expected from the
spherical distribution. This kind of shoot axis distribution clearly refers to some light
adaptation phenomenon in the crown structure: vertical upper shoots effectively
capture direct sunlight when the sun is low and radiation cannot penetrate into deeper
layers because of screening by neighbouring trees; at medium solar elevations, shoot
inclination angle does not affect radiation interception (uniform azimuth distribution
is assumed); direct solar radiation at high sun positions and diffuse sky radiation can
penetrate to deeper layers and the vertical distribution of absorbed radiation is more
uniform than with any other distribution.
Relatively little attention has been focused on the role of the woody parts in
forming the radiation field in forests. To date, only a few speculations on their
essential function in the formation of remotely sensed signals have been presented.
However, the same types of quantities should be described for the woody parts as for
leaves or needles. An additional specific feature is the joint distribution of leaves or
needles and branches. The fine branches (shoot axes) are preferably located inside the
needle clusters (shoots), but larger branches are located outside the smallest clusters
and inside the next level of clusters - branches. All of these features should be
analytically described unless computer-drawn trees are considered.
Optical model of the forest community

The next level in the hierarchy is that of the whole stand. In addition to describing the
size distribution of trees, it is important to characterize the tree distribution pattern. In
sample plot areas or transects where all tree locations and dimensions are mapped, it
is possible to derive detailed estimates of radiation penetration to any point of the
canopy (Hari et al. 1985, Koop 1989). Although it is convenient to assume a random
tree spatial distribution pattern, such simulations are inadequate. More sophisticated
distribution patterns are either computer-simulated or analytic extensions of the
random Poisson distribution. Usually, the tree distribution pattern is characterized by
the distribution of tree trunk locations in the horizontal plane. However, in all kinds of
radiation simulations (penetration, interception and reflection) it is important to
consider the crown distribution pattern as well. Because crowns are asymmetrical and
their forms become adapted to those of neighbouring trees, the distribution pattern of
crowns is usually more regular than that of tree trunks.
A forest ecosystem radiation model (FERM)

In the following example, we consider some details of a particular forest ecosystem
radiation model. The purpose of this model is to derive the most important radiation
characteristics of a forest ecosystem: the albedo, the reflectance factor and its angular
and spectral distributions, as well as photosynthetically active radiation (PAR) and
integral radiation fractions absorbed by different overstorey and understorey
components and by the soil.
The general concept of FERM is based on our earlier forest reflectance model
(Nilson and Peterson 1991, Nilson 1992). The study object is either a homogeneous
forest compartment or a stand. Different tree species and tree size classes as well as
the ground-layer vegetation are introduced as forest ecosystem components (see Fig.
12). For each component, the set of input parameters is given, including tree size (tree
height, crown length and radius and dbh) and horizontal distribution pattern
parameters, leaf and branch area indices for each overstorey component, single-leaf
reflectance and transmittance coefficients and branch and trunk bark reflectance. The
angular distribution of all canopy elements (leaves, shoots and supporting tissues) is
assumed to be spherical. For conifers, an additional parameter describing the mutual
shading of needles in shoots is introduced. The ground-layer vegetation (herb layer) is
treated as (locally) homogeneous vegetation with the given LA! and leaf inclination
distribution. Optical properties of leaves and underlying soil (i.e., the optical
characteristics of the herb layer) are calculated by means of a homogeneous canopy
model (Nilson and Kuusk 1989, Kuusk 1994).
m Plno
dom1neolmc
25
Pme Sprut'e. E
domln.led Pine oYtlralort!)'
0
20 .uppr~ ed 5 20
~
15 15
~
10 ~1O
Spruce. ~ 4
r~ltlnerehon
8
I
5
Figure 12. A schematic representation of the structure of a 65-yr-old Scots pine-

dominated forest community (left). Vertical distribution of the fraction of
incident radiation intercepted per 1 m canopy layer by different ecosystem
components (right).
The basic submodels of FERM were originally derived for remote sensing
applications. However, calculating the reflectance of the forest canopy requires
solving the entire radiative transfer problem within the ecosystem. In other words, in
the reflectance model many important canopy radiation characteristics are calculated
as byproducts. For instance, to estimate radiance from the ground layer we must also
estimate radiation transmitted to the forest floor, among other characteristics.
49
In this model, the total scene is divided into sunlit and shaded parts of the canopy
and the ground layer. A probabilistic-analytical radiative transfer approach is applied
to calculations of the penetration of nonintercepted direct and diffuse radiation and of
the first-order phase function in the tree layer (Nilson 1991). Multiple scattering of
radiation inside the ecosystem is calculated by an approximate solution of the
radiative transfer equation in an equivalent homogeneous plant canopy. Spectral
albedos of the forest community are calculated by numerical integration of reflectance
factors over the whole hemisphere, integral albedos by integrating the spectral albedos
over the spectrum.
The number of model input parameters, particularly structural community
parameters, is considerable. Because of interrelations and feedbacks existing in the
ecosystem, many of the input parameters are related to each other. Therefore,
application of the model causes numerous problems. A possible solution lies in
integrating the FERM into a growth model of the ecosystem components. Existing
empirical forest growth models such as yield tables are insufficient, since they are
unable to provide values for the most important forest ecosystem structural
parameters from the radiative tF~nsfer point of view (i.e., LA! and canopy cover
contributed by all ecosystem components). Therefore, two stages in this modeling
effort, the model validation stage and the routine use stage, can be outlined. At the
validation stage of model quality testing, we should estimate as many parameters as
possible by in situ measurements. We hope to create for routine use species-based,
site-specific standardized data sets for many of the input parameters, using the
regression relationships between structural forest parameters and those of the growth
models.
Although first tests of the reflectance submodel of FERM have yielded acceptable
results (Nilson and Peterson 1991), more extensive tests are needed. The development
ofFERM is still underway. Some parts of the model, such as calculation ofthe angular
distribution of reflectance in the principal plane and of radiation interception by
ecosystem components, are in place. To illustrate the capability of the FERM, some
examples of preliminary results are given in Figures 12-14 and Tables 2 and 3.
The forest ecosystem considered in these examples is relatively simple: a Scots
pine stand dominated by 65-yr-old trees growing on a Vaccinium site type. The
ecosystem components in the tree layer (Table 2, Fig. 12) include three size classes of
Scots pine, subsequently referred to as the dominating, the dominated and the
suppressed, and a few Norway spruces in the overstorey. Ground-layer vegetation is
dominated by a dwarf shrub, Vaccinium vitis-idaea. Norway spruce regeneration is
also considered as a special ecosystem component in the understorey. Most of the
PAR is intercepted by dominating and dominated pines (Fig. 12). These two
components, especially their sunlit fractions, are also the largest contributors to the
nadir reflectance factor (Table 3). However, the role of sunlit ground-layer vegetation
cannot be ignored.
Table 2. Some structural data for a 65-yr-old Scots pine-dominated forest

ecosystem and the percentages of radiation intercepted by different ecosystem
components (solar zenith angle 40).
Ecosystem Number of Crown closure Leaf (needle) Radiation

component trees ha'i (soil cover) area index intercepted
Tree layer
Pine dominating 246 24.1% .279 37.2% 1.68 38.7% 47.4%
Pine dominated 364 35.7% .275 36.7% 1.79 41.2% 31.8%
Pine suppressed 138 13.5% .081 10.8% 0.44 10.1% 4.7%
Overstorey spruce 13 1.3% .021 2.8% 0.26 6.0% 1.7%
Spruce regeneration 258 25.3% .093 12.4% 0.17 3.9% 0.6%
Tree layer total 1019 100% .749 100% 4.34 100% 86.3%
Ground layer .74 2.7 11.3%

Soil 2.4%
Table 3. Contributions of different single- and multiple-scattering components

to the nadir reflectance factor in the red and near-infrared spectral regions at the
solar zenith angle 40. This is the same ecosystem shown in Table 2.
Reflectance or Contributions to reflectance

scene component
Red Near-infrared
First-order scattering
Tree crowns
Pine dominating .0092 47.9% .0302 23.9%
Pine dominated .0049 25.5% .0160 12.7%
Pine suppressed .0006 3.1% .0019 1.5%
Overstorey spruce .0001 0.5% .0003 0.2%
Spruce regeneration .0002 1.0% .0007 0.6%
(Sunlit) ground .0023 12.0% .0147 11.6%
Multiple scattering
Tree crowns .0018 9.4% .0563 44.5%
Ground .0001 0.5% .0063 5.0%
Total reflectance .0192 100% .1264 100%

51
Figure 13 represents the angular distribution of reflectance in the principal plane

predicted by the FERM for the same forest community described in Figure 12. In
general, these angular distributions resemble those of homogeneous plant canopies.
Since the view directions used in the calculations are sparsely distributed, the detailed
angular structure in Figure 13 (especially near the hot spot) cannot be represented very
accurately.
0.15 0.8
0:::
o
I-
0:::
o
I-
u
~ ;::06
wO.10
U w
Z u
Z
~ ~04
U U
W
W
-' -'
~0.05 l.L
w
0:::
0:::0.2
o 0:::
w
0::: Z
0.00 +o~..,..,~..,..,~..,..,~-.--.~-.--.---,--, 0.0 +O~-.--.~-.--.~-,--,~-,--,~..,..,---,---,
-90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90
VIEW NADIR ANGLE. degrees VIEW NADIR ANGLE. degrees
Figure 13. The angular distribution of the red (a) and near-infrared (b)
reflectance factors on the principal plane. The ecosystem is the same as in Figure
12. Solar zenith angles follow. 1: 25.3; 2: 39.7; 3: 54.0; 4: 68.4; 5: 82.8.
Negative view angles correspond to the sun's side.
The FERM makes it possible to carry out very different kinds of numerical
experiments. The effects of successional and management changes in the reflectance
of a Scots pine forest growing on a fertile Vaccinium site (Nilson and Peterson 1994)
and in the proportions of absorbed radiation by ecosystem components have
comprised a large class of simulations. Figure 14 illustrates both simulated
successional changes in nadir reflectance and management effects. The latter includes
a removal of deciduous trees at the age of 5 years and two thinnings, one at 15 years
and another at 40 years of age. In both thinnings, 30% of the basal area was removed.
Successional changes in the structural forest parameters were modeled using yield
tables for Estonian forests. For details on determining input parameters, see Nilson
and Peterson (1994). The time step used in the simulations was 5 years up to a stand
age of 20 years and 10 years thereafter. When the experiment incorporates the effects
of thinnings, a model of the thinning response of stand structural parameters is
needed. In this particular case, we assumed that after thinning the stand parameters
recovered during the next time step.
0.04 a 0.4 b
n:: n::
a a
I-
() I- DR

u..
()
0.3
DR u..
w w
()
()
Z
z
ti 0.02 ~02 -
()
w W
....J
u.. ....J
u..
W
n:: w
n:: 01
0
w n::
n:: z
0.00 0.0
0 20 40 60 80 100 0 20 40 60 80 100
STAND AGE, years STAND AGE, years
Figure 14. Changes in nadir reflectance of a Scots pine-dominated forest

community from ciearcutting through 100 years for the red (a) and the NIR (b)
region. Stand developPtent was assumed to proceed along the values predicted by
the yield table (smooth curve and dashed parts of the line). Rapid reflectance
changes (full line) represent the simulated effects of management treatments:
deciduous removal (DR) and two thinnings (THI and TH2).
Implications for canopy photosynthesis: Influence of shoot

structure on photosynthesis
The structure of coniferous shoots and its effect on radiation conditions is essential in
the theoretical treatment of shoot and whole-canopy photosynthesis. The problem has
been well formulated by Smolander et al. (1987), Oker-Blom et al. (1991) and
Gutschick (1991). In addition, we offer the following summary of experimental
results obtained by Ross et al. (in press).
The measurement of shoot radiation conditions and net photosynthesis is a
complicated methodological and experimental problem for which no adequate
solution has yet been found. Typically, in measurements of the photosynthetic
radiation response, photosynthesis is expressed by the absorbed CO2 per unit leaf area
per unit time. Radiation is characterized by PAR irradiance or, more precisely, by the
PAR absorbed on the same leaf area unit. What is the leaf area in the case of a
coniferous shoot? Theoretically, to correctly determine the shoot photosynthesis light
response curve, we need the absorbed PAR and net photosynthesis for each needle
surface element. By totaling the absorbed PAR and net photosynthesis of needles, we
can determine the total PAR absorbed by the shoot and the shoot net photosynthesis.
Dividing these values by the shoot's needle area, we can obtain the mean absorbed
PAR and net photosynthesis per unit needle area per unit time.
Experimentally this cannot be done. Usually, the shoot photosynthesis light
response curve is determined in the following way: the shoot or some part of it is
placed into a leaf chamber where CO 2 exchange and net photosynthesis are recorded.
The leaf chamber is oriented so that the shoot axis is perpendicular to direct solar
radiation. PAR is measured with a flat sensor, also perpendicular to direct solar
53
radiation, and placed either inside or outside the chamber. There are several ways to
express the results and construct the shoot photosynthesis light response curves. The
simplest is to calculate the net photosynthesis rate per unit of projected shoot area. A
disadvantage of this procedure is that different needle areas may correspond to a unit
of projected shoot area. Therefore, the photosynthesis light response curves of shoots
with different architectures are not comparable. An alternative is to calculate the net
photosynthesis rate per unit needle area (or dry weight) of a shoot. This method of
presenting shoot net photosynthesis is probably the most widely used in the literature.
However, it has some disadvantages. For example, we consider two shoots with the
same needle area (Fig. 15), shoot A having a smaller needle angle, a smaller shoot
diameter and a greater needle density than shoot B. We assume that the illumination
conditions and the photosynthesis light response curve of needles are the same for
both shoots. Needles on shoot B have better illumination, absorb more PAR and have
a higher net photosynthesis rate than needles on shoot A. Consequently, at the same
incident PAR the net photosynthesis rate per unit needle area should be higher for
shoot B than for shoot A. Even ~f individual needles have invariant photosynthetic
responses, differences in the degree of mutual shading of needles result in different
shoot photosynthesis light response curves.
A B
Figure 15. Illustration of two shoots, A and B, with different geometrical

characteristics but the same needle area.
The measurement of irradiance variations on needle surfaces in a shoot and their

dependence on shoot architecture is an extremely challenging theoretical problem. In
an actual forest community, it would require the use of fibre optics techniques. The
influence of shoot architecture on the irradiance distribution and photosynthesis of a
shoot in a direct radiation field has been analyzed theoretically by Smolander et al.
(1987). They found that in a direct radiation field, the mean irradiance on the needle
surface, determined by the ratio between shoot silhouette area and total needle area
(STAR), is the major source of variability in shoot photosynthetic response. Although
the STAR can be determined using photography, it cannot be determined directly from
measurements of shoot characteristics.
To obtain an approximate characterization of the variation of illumination

conditions inside the shoot cylinder depending on shoot architecture parameters, a
new parameter has been proposed (Ross et al. in press): the needle living space (NLS)
inside the shoot. The shoot cylinder volume may be expressed as
2
Vs = n(ds /2)2 Is = n(ln sin vn + d,I2)2 Is "" n Is I~ sin vn , (5)
where Is and ds are the length and diameter of the shoot cylinder, d, is the twig
diameter, In is the needle length and vn is the needle angle (d t is small compared to dJ
The total needle surface area of a shoot is equal to the needle surface area of a needle,
SnOn) (here expressed as a function of its length), multiplied by needle density, nN , and
shoot length, Is'
(6)
The NLS is expressed as the volume inside the shoot cylinder per unit needle area:
I2 . 2
NLS =n n sm vn (7)
Sn(ln)n N
Thus, the larger the NLS, the better the illumination conditions on the needle surface.
As can be seen in Eq. (7), the NLS is a function of three parameters of shoot
architecture: needle length, needle density and needle angle. All these parameters can
be obtained from simple phytometrical measurements.
Measurements of the diurnal course of net photosynthesis in a shoot trap cuvette
and of shoot structure were carried out in August 1989 at the Hyytiiilii Experimental
Forest Station in Central Finland using the measuring system described by Hari et al.
(1990). Diurnal curves of net photosynthesis for 17 two-yr-old shoots were recorded.
From these measurements, shoot photosynthetic light response curves were
constructed by visual fitting. Examples of these curves are presented in Figure 16.
From the light response curves, the values of shoot net photosynthesis, Po' at
irradiation values of 100, 300, 500 and 1000 /lIDol m-2s- 1 were estimated, where Po is
expressed in relative units cm- 2 needle area. The relationship between Po and NLS was
found to be linear and is expressed by the following regression lines:
2 l
1= 100 /lIDol m- s- , Po = 4.30 + lAO NLS, r=0.71,
2 1
1= 300 /lIDol m- s- , Po = 6.66 + 3.26 NLS, r = 0.85,
2 l
1= 500 /lIDol m- s , Po = 8.57 + 3.98 NLS, r = 0.90,
I = 1000 /lIDol m-2s- 1, Po = 8.26 + 5.31 NLS, r = 0.89.
These results demonstrate an important relationship between shoot net photosynthesis
and shoot structure, characterized by NLS. At all values of irradiance, shoot net
photosynthesis increases linearly with an increase in NLS by improving the
illumination conditions of individual needles inside the shoot cylinder.
55
15
o(J)
. 0
!!! (J)
~ u
i!:z ~()
c
)0"
V)I
o 0E
.....
~ $!
Q. ~
W ~
z '5
~
-2 -1
PAR IRRADIANCE. J.Lmol em sec
Figure J6. Photosynthetic light response curves of four different Scots pine
shoots with different needle living space (NLS). Experimental points are shown
with fitted curves. NLS is indicated at the respective curves.
The photosynthetic light curves were also approximated by means of the

Michaelis-Menten equation, expressed as
a J Pm.,
PG = - RII'
aJ+Pmax
where a is the initial slope, PmolX is the light-saturated value of PG and Rg is dark
respiration. The correlation between a and NLS as well as that between Pmax and NLS
was studied. The analysis revealed a relatively weak correlation between a and NLS
compared with the correlation between PmolX and NLS. The correlation is linear and is
expressed by the following regression lines:
a = 0.063 + 0.020 NLS, r =0.660,

PmolX = 11.86 + 6.41 NLS, r = 0.822.
Here, the question arises whether a correlation between shoot photosynthesis and
NLS exists or whether other factors are involved. Needle age affects the
photosynthetic rate, which is at its maximum for one-yr-old shoots and decreases with
age. On the other hand, NLS is generally larger for two-yr-old shoots. Consequently,
the correlation between PG and NLS would have been still higher if shoot age were
also considered.
Conclusions
Forest ecosystems are probably the most complex subjects in studies of plant canopy
radiation regime. The main challenges associated with these studies are
floristic composition of the overstorey and understorey;
age and size differences among trees of the same species;
nonhomogeneous and semirandom distribution patterns of trees;
adjustment of crown shapes to those of neighbouring trees;
clumped or semiregular distribution of leaves or coniferous shoots
inside tree crowns;
specific structure of coniferous shoots and variability in vertical and
horizontal directions as well as with shoot age;
small needle dimension, which causes the penumbral effect, and
difficulties in experimental determination of needle optical properties;
difficulties in obtaining adequately averaged radiation measurements in
mature forests.
Because of these complications, mathematical models of forest radiation regime
tend to be oversimplified and their application value in forestry relatively limited.
Nevertheless, such models are useful conceptual tools and provide a better
understanding of the forces driving the formation of radiation fields within and
reflected from forests. In addition, they promote the study of certain aspects of
relationships between forest structure and radiation, radiation and photosynthesis and
forest structure and ecosystem function. Recent development of remote sensing
applications in vegetation studies has greatly increased the need to elaborate different
mathematical models of forest radiation regime and spectral reflectance.
The main objectives of applying radiative transfer models in remote sensing studies
of forest ecosystems may be outlined as follows:
recognition of how a reflected signal is formed;
identification of the primary factors that determine a reflected signal and
its temporal and spatial variability;
simulation of the effects on reflectance of various scenarios of
ecosystem development, including successional changes and
management effects;
determination of various ecosystem parameters from remotely sensed
data by means of model inversion;
interpretation and normalization of remotely sensed data, (e.g.,
extending the measured data to another solar elevation, phenological
stage or the like).
Until recently, very little attention has been focused on the inversion of forest
reflectance models. The existing reflectance models seldom contain main structural
forest parameters. Moreover, stand structure descriptions used in traditional forest
57
mensuration (production-oriented forestry) are different from those used in

reflectance models (e.g., stem volume and basal area vs canopy closure and LAI).
From this point of view, reflectance models of the FERM type could be preferred,
since there is considerable overlap between the model input data set on stand structure
and traditional forest mensuration data.
The reflectance submodel of the FERM can be used for inversion of any model
input parameters. In many practical cases, this cannot be done effectively because of
the many parameters that remain undetermined. The number of undetermined input
parameters may be considerably reduced if available a priori information about the
forest under study is used, such as GIS data and relationships among the parameters.
Specular reflectance is not of the greatest importance in photosynthesis and
boundary layer energetics. Even in remote sensing applications, to the extent that
nadir measurements are used, it is relatively insignificant. However, in the
interpretation of results from multiangle measurement systems or from polarization
and depolarization measurements, it cannot be ignored.
As a matter of fact, penumbral issues have not been considered in detail in the
preceding analyses. When penumbral effects are ignored, however, errors in the
theoretical estimation of photosynthesis can be marked. We agree that
parameterization models, as presented by Pfreundt (1988), together with further
modeling efforts may provide acceptable solutions to the problem.
There is a great need for optical models of needles. An example of a good model
for broadleaf trees is that of Jacquemoud and Baret (1990). This model permits
calculation of the whole leaf reflectance and transmittance spectrum using only three
input parameters (chlorophyll, water content and a structural parameter). We hope that
in future forest applications it will be possible to relate the input parameter values to
species and site conditions.
In this paper we discussed only the problems associated with the determination of
important geometrical and optical characteristics of forests. Details on solving the
radiative transfer problem are well elaborated elsewhere (e.g., Myneni and Ross
1991). Our main objective was to demonstrate the structural complexity of forest
communities compared with naive structural models that generally form the basis of
current radiative transfer models. Much work remains to be done in testing and
modifying existing radiative transfer models, despite the fact that acceptable models
for several applications are already available.
References
Asrar, G. (ed). 1989. Theory and Applications of Optical Remote Sensing. - 1. Wiley and Sons,
New York. 734 pp.
Borel-Donohue, C.c. 1988. "Models for backscattering of millimeter waves from vegetation
canopies." - Ph.D. thesis, University of Massachusetts, Boston.
Brakke, T. W. 1994. Specular and diffuse components of radiation scattered by leaves. - Agric.
For. Meteorol. 71: 283-295.
Brakke, T.w., Smith, 1.A. and Harnden, 1.M. 1989. Bidirectional scattering of light from tree
leaves. - Rem. Sens. Environ. 29: 175-183.
Carter, G.A. and Smith, W.K. 1985. Influence of shoot structure on light interception and
photosynthesis in conifers. - Plant Physiol. 79: 1038-1043.
Chen, J.M. and Black, T.A. 1991. Measuring leaf area index of plant canopies with branch
architecture. -Agric. For. Meteorol. 57: 1-12.
Chen, J.M. and Black, T.A. 1992. Defining leaf area index for non-flat leaves. - Plant Cell
Environ. 15: 421-429.
Chen, J.M., Black, T.A. and Adams, R.S. 1991. Evaluation of hemispherical photography in
determining plant area index and geometry of a forest stand. - Agric. For. Meteorol. 56:
129-143.
Gerst!, S.W. and Borel-Donohue, c.c. 1992. Principles of the radiosity method versus
radiative transfer for canopy reflectance modeling. - Trans. Geosci. Rem. Sens. 30:
271-275.
Goel, N.S. 1988. Models of vegetation canopy reflectance and their use in estimation of
biophysical parameters from reflectance data. - Rem. Sens. Rev. 4: 1-212.
Goel, N.S. and Norman, J.M. (eds). 1990. Instrumentation for studying vegetation canopies for
remote sensing in optical and thermal infrared regions. - Rem. Sens. Rev. 5(1): 1-360.
Gower, S.T. and Norman, J.M. 1991. Rapid estimation ofleaf area index in conifer and broad-
leaf stands using the LI-COR LAI-2000. - Ecology 72: 1896-1900.
Gutschick, v.P. 1991. Joining leaf photosynthesis models and canopy photon-transport models.
- In: Myneni, R.B. and Ross, J. (eds). Photon-Vegetation Interactions. Applications in
Optical Remote Sensing and Plant Ecology. Springer-Verlag, Berlin, pp. 504-535.
Hari, P., Kaipiainen, L., Korpilahti, E., Makela, A., Nilson, T., Oker-Blom, P., Ross, J. and
Salminen, R. 1985. Structure, radiation and photosynthetic productivity in coniferous
stands. - Research Notes 54, Department of Silviculture, University of Helsinki. 233 pp.
Hari, P., Korpilahti, E., Pohja, T. and Rasanen, P. 1990. A field system for measuring the gas
exchange of forest trees. - Silva Fenn. 24: 21-27.
Jacquemoud, S. and Baret, F. 1990. A model of leaf optical properties spectra. - Rem. Sens.
Environ. 34: 75-91.
Jupp, D.L.B. and Walker, J. 1996. Detecting structural and growth changes in woodlands and
forests: The challenge for remote sensing and the role of geometric optical modelling. - In:
Gholz, HL, Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the
Modeling of Forest Productivity. Kluwer Academic Publishers, Dordrecht, The
Netherlands, pp. 75-108.
Koop, H. 1989. Forest dynamics. SILVI-STAR: A Comprehensive Monitoring System. -
Springer-Verlag, Berlin. 229 pp.
Kuusk, A. 1991. The hot spot effect in plant canopy reflectance. - In: Myneni, R.B. and Ross,
J. (eds). Photon-Vegetation Interactions. Applications in Optical Remote Sensing and Plant
Ecology. Springer-Verlag, Berlin, pp. 139-159.
Kuusk, A. 1994. A multispectral canopy reflectance model. - Rem. Sens. Environ. 50: 75-82.
Li, X. and Strahler, A.H. 1985. Geometric-optical modeling of a conifer forest canopy. - Trans.
Geosci. Rem. Sens. 23: 705-721.
Moldau, H. 1965. On the use of polarized radiation to analyse the reflection indicatrixes of
leaves. - In: Investigations on Atmospheric Physics 7: Questions of Radiation Regime of
Plant Stand (in Russian). Academy of Sciences ESSR, Institute of Physics and Astronomy,
Tartu, Estonia, pp. 96-101.
Myneni, R.B. and Ross, 1. (eds). 1991. Photon-Vegetation Interactions. Applications in Optical
Remote Sensing and Plant Ecology. - Springer-Verlag, Berlin, 565 pp.
Myneni, R.B., Ross, J. and Asrar, G. 1989. A review on the theory of photon transport in leaf
canopies. - Agric. For. Meteorol. 45: 1-153.
59
Nilson, T. 1991. Approximate analytical methods for calculating the reflection functions of
leaf canopies in remote sensing applications. - In: Myneni, RB. and Ross, J. (eds). Photon-
Vegetation Interactions. Applications in Optical Remote Sensing and Plant Ecology.
Springer-Verlag, Berlin, pp. 161-190.
Nilson, T. 1992. Radiative transfer in nonhomogeneous plant canopies. - In: Stanhill, G. (ed).
Advances in Bioclimatology I. Springer-Verlag, Berlin, pp. 60-88.
Nilson, T. and Kuusk, A. 1989. A reflectance model for the homogeneous plant canopy and its
inversion. - Rem. Sens. Environ. 27: 157-167.
Nilson, T. and Peterson, U. 1991. A forest canopy reflectance model and a test case. - Rem.
Nilson, T. and Peterson, U. 1994. Age dependence of forest reflectance: Analysis of main
driving factors. - Rem. Sens. Environ. 48: 319-33 I.
Norman, J.M. and Jarvis, P. 1975. Photosynthesis in sitka spruce (Picea sitchensis (Bong.)
Carr.). Part 5. Radiation penetration and a test case. - J. Appl. Ecol. 12: 839-878.
Oker-Blom, P. 1986. Irradiance distribution and photosynthesis of a Scots pine shoot as
influenced by shoot structure and solar radiation field geometry. - In: Fujimori, T. and
Whitehead, D. (eds). Crown and Canopy Structure in Relation to Productivity. Forestry and
Forest Products Research Institut~, Ibaraki, Japan, pp. 382-395.
Oker-Blom, P. and Smolander, H. 1988. The ratio of shoot silhouette area to total needle area
in Scots pine. - For. Sci. 34: 894-906.
Oker-Blom, P., Kotisaari, A., Kellomaki, S., Ross, J. and Smolander, H. 1986. Crown
projection area of young Pinus sylvestris: A model and its test. - Scand. J. For. Res. 1:
67-74.
Oker-Blom, P., Lappi, J. and Smolander, H. 1991. Radiation regime and photosynthesis of
coniferous stands. - In: Myneni, RB. and Ross, J. (eds). Photon-Vegetation Interactions.
Applications in Optical Remote Sensing and Plant Ecology. Springer-Verlag, Berlin, pp.
469-499.
Pfreundt, J. 1988. Modellierung der raumlichen Verteilung von Strahlung,
Photosynthesekapazitat und Produktion in einem Fichtebestand und ihrer Bezeihung zur
Bestandsstruktur. - Univ. Gottingen, Berichte des Forhungszentrums Waldokosystemel
Waldsterben, Reiche A, Bd. 39. 163 pp.
Ross, J. 1981. The Radiation Regime and Architecture of Plant Stands. - DR Junk Publishers,
The Hague, The Netherlands. 391 pp.
Ross, J. and Marshak, A. 1991. Monte Carlo methods. - In: Myneni, RB. and Ross, J. (eds).
Photon-Vegetation Interactions. Applications in Optical Remote Sensing and Plant
Ecology. Springer-Verlag, Berlin, pp. 442-467.
Ross, J., Meinander, O. and Sulev, M. 1994. Spectral scattering properties of Scots pine shoots.
- In: Proceedings of the IGARSS '94 Symposium, August 8-12, 1994, California Institute
of Technology, Pasadena, CA, Vol. 2, pp. 1451-1454.
Ross, J., Stenberg, P., Berninger, F. and Hari, P. 1995. The influence of shoot architecture on
net photosynthesis. - In: Hari, P., Ross, J. and Mecke, M. (eds). Production process of Scots
pine: Geographical variation and models. Acta For. Fenn. (in press).
Smith, N.J., Chen, J.M. and Black, T.A. 1993. Effects of clumping on estimates of stand leaf
area index using the LI-COR LAI-2000. - Can. J. For. Res. 23: 1940-1943.
Smolander, H., Oker-Blom, P., Ross, J., Kellomaki, S. and Lahti, T. 1987. Photosynthesis of a
Scots pine shoot: Test of a shoot photosynthesis model in a direct radiation field. - Agric.
For. Meteorol. 39: 67-80.
Stenberg, P., Linder, S., Smolander, H. and Flower-Ellis, J. 1994. Performance of the LAI-
2000 plant canopy analyzer in estimating leaf area index of some Scots pine stands. - Tree
Physiol. 14: 981-985.
Van de Hulst, H.e. 1981. Light Scattering by Small Particles. - Dover Publishers, New York.
470 pp.
Vanderbilt, v.e., Grant, L. and Ustin, S.L. 1991. Polarization of light by vegetation. - In:
Myneni, R.B. and Ross, J. (eds). Photon-Vegetation Interactions. Applications in Optical
Remote Sensing and Plant Ecology. Springer-Verlag, Berlin, pp. 191-228.
Walter-Shea, E.A. and Norman, J.M. 1991. Leaf optical properties. - In: Myneni, R.B. and
Ross, J. (eds). Photon-Vegetation Interactions. Applications in Optical Remote Sensing and
Plant Ecology. Springer-Verlag, Berlin, pp. 229-251.
TH R E E
Estimating Forest Canopy Characteristics as Inputs for Models

of Forest Carbon Exchange by High Spectral Resolution
Remote Sensing
Mary E. Martin and John D. Aber
Martin, M.E. and Aber, J.D. 1996. Estimating forest canopy characteristics as inputs
for models of forest carbon exchange by high spectral resolution remote sensing. -
In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in
the Modeling of Forest Productivity. Kluwer Acad. Pub!., Dordrecht, The
NASA's Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) was used to derive

input parameters for a model of forest ecosystem carbon (C) balance. These
parameters include canopy nitrogen (N) concentration and foliar biomass predicted
with multiple linear regression equations using selected spectral bands, and species
composition determined by means of a supervised image classification. The model
predicted total net photosynthesis for the study area, the Harvard Forest in Petersham,
Massachusetts, with a spatial resolution of 20 m. The model was simulated five times
using the input variables of species, foliar N concentration and foliar biomass derived
from either field sampling or spectral data. Although the mean value for net
photosynthesis over the 400-ha study site was similar when derived from both existing
field data and remotely sensed data (656 g C m'2 y{1 and 630 g C m'2 y{', respectively),
the latter provided information on the spatial variability of photosynthesis throughout
the study area that was not evident when coarse-scale field data were used.
Mary E. Martin and John D. Aber; Complex Systems Research Center; Institute for the
Study of Earth, Oceans and Space, University of New Hampshire, Durham, NH
03824, USA.
61
Introduction
The relative perceived costs and benefits of high spectral resolution remote sensing for
the estimation of regional to global carbon (C) balances have affected both the
direction of research and the selection of instruments for major environmental
programs. Several existing models applied at the global scale (e.g., Prince 1991,
Sellers et al. 1992, Running and Hunt 1993, Goward and Dye 1996) use some
combination of reflected red and infrared radiation (e.g., NDVI or simple ratios) as the
principal descriptor of canopy condition, since this index can be derived over large
areas from Advanced Very High Resolution Radiometer (AVHRR) data. However, it is
still somewhat uncertain which canopy characteristic is most strongly related to
NDVI. Good correlations between NDVI and total leaf area index (LAI) have been
reported within biomes (Peterson et al. 1987, Gholz et al. 1996), but generalized
algorithms for converting broad-band reflectance to LA! across several growth forms
(e.g., broadleaf and needleleaf forests) have not been reported. Physically, intercepted
photosynthetically active radiation (PAR) has the most direct meaning (Sellers et al.
1992) and may be used without conversion to a canopy structural value, but LAI
remains a central component of many models. Whether vegetation indices relate to
total foliar biomass, total leaf area or some descriptor of total chlorophyll or its
concentration (e.g., Myneni et al. 1995) is not clear.
Research carried out under NASA's Accelerated Canopy Chemistry Program has
suggested that accurate predictions of foliar nitrogen (N) concentrations can be
obtained from high spectral resolution remote sensing data (1O-J1Il1 resolution in the
0.4-2.4 nm range) (Martin and Aber in review). To date, however, no field-based
calculations have been made to estimate increases in the accuracy of model
predictions of ecosystem C gain that might be expected with the availability of spatial
N concentration data. Nor has the potential to improve the accuracy of canopy biomass
estimation been assessed.
The purpose of this paper is to combine information on species composition, foliar
biomass and foliar N concentration generated from high spectral resolution remote
sensing data with a simple, validated model of forest canopy photosynthesis to
estimate total net C gain for the Prospect Hill Tract of the Harvard Forest in Petersham,
Massachusetts. We will compare both total C gain for the 400-ha area modeled and the
degree of spatial variation captured by the progressive addition of spatial information
on species, canopy mass and foliar N concentration.
Methods
Site description
The data described in this paper were collected at the Harvard Forest in Petersham,
Massachusetts (latitude 4232'N, longitude nOll'W). Field sampling was done on
twenty 50 x 50-m plots, and remote sensing data were acquired at a resolution of 20 m
for a 10 x 30-km area including the main region of interest - the Prospect Hill tract
of the Harvard Forest. The species composition at this site is mixed hardwood
(primarily Quercus rubra [red oak] and Acer rubrum [red maple]), with conifer stands
63
of Pinus strobus and P. resinosa (white and red pine), Picea abies (Norway spruce)
and Tsuga canadensis (hemlock). The red pine and Norway spruce stands were
planted in the early 19OOs, while the hardwoods, white pine and hemlock were mature,
naturally occurring stands. There is also a large Picea mariana (black
spruce)-dominated bog in the center of the tract (Plate la).
Field-measured canopy characteristics

Field measurements of canopy biomass, foliar chemistry and LA! were collected for
20 sites at the Harvard Forest in 1992 at the time of remote sensing data acquisition.
During the same year, data from NASA's Airborne Visible/Infrared Imaging
Spectrometer (AVIRIS) and field data were collected for an additional 20 plots at
Blackhawk Island, Wisconsin. Foliage samples were collected within 10 days of the
remote sensing acquisition and analyzed for N and lignin concentration (Newman et
al. 1995). Litterfall collections were used to determine canopy biomass, with litter
from a single year determining the biomass of deciduous species. Conifer species'
litterfall was multiplied by foliar r~tention time (in years) to determine foliar biomass.
Canopy-level foliar N concentration was derived from the measured N concentration
for each species weighted by the fraction of that species in the canopy. A detailed
description of the methods used for field sampling can be found in Martin and Aber (in
review).
Leaf area index was determined for each plot using the litterfall collection and
specific leaf weights (SLW, g cm-2). SLW for each hardwood species was determined
by measuring projected leaf area with a LiCor LI 3100, and oven-dried (70C) leaf
weight. Conifer species' SLWs were calculated from data reported by Johnson and
Lindberg (1992) and Reich et al. (in review). All LAI values are expressed on a
projected area basis.
Data planes derived by remote sensing (AVIRISj

Remote sensing data were acquired with AVIRIS during the 1992 field data collection
period (15 and 21 June 1992 for the Harvard Forest and Blackhawk Island,
respectively). This instrument measures radiance in 224 contiguous bands between 0.4
and 2.5 flIIl, each with a width of 10 nm. Spatial resolution of this instrument is 20 m
and data are collected in a swath with a width of 10 kID. Data were corrected to ground
reflectance for both scenes using a radiative transfer model combined with an
empirical line correction (Gao et al. 1991, 1992, Clark et al. 1994).
Atmospherically corrected reflectance data were transformed to first difference
spectra and correlated with field-measured canopy N concentration using multiple
linear regression techniques (Martin 1994). A calibration equation relating remotely
sensed AVIRIS fust difference spectral data to field-measured N concentration was
developed with data from both the Harvard Forest and Blackhawk Island. These
equations are of the form
n
L
Nitrogen (%) = bo + b),,: (1)
i= 1
where bo and bj are fitting coefficients and A./ represents the first difference AVIRIS
reflectance data for n bands. This equation has been used to estimate foliar N
concentration for all of the Prospect Hill pixels in the AVIRIS scene, with bands
centered at 783 and 1681 run.
We have used the same calibration technique in this paper to estimate foliar
biomass for Prospect Hill using first difference AVIRIS absorbance (log R 1) data.
Separate calibration equations were developed for needle- and broadleaf stands.
Multiple linear regression equations relate two to three selected bands of first
difference AVIRIS absorbance data to field-measured foliar biomass at the Harvard
Forest and Blackhawk Island.
In previous work (Martin 1994), AVIRIS data were used in conjunction with
existing GIS field data on species distribution to develop an improved map of species
composition for the Prospect Hill study area. A supervised classification using four
AVIRIS bands (a subset of bands previously identified for the estimation of canopy N
and lignin concentration) was used to classify the study area into 10 stand types (Table
I). This classification identifies conifer stands by species and allows more accurate
parameterization of the'~odel for these pixels. All hardwood stands in the study area
are composed of several species and are identified only as mixed hardwood stands.
Table 1. Description of species categories classified by AVIRIS data.
Stand class Stand description
I open, not forested

2 hemlocklhardwood
3 mixed softwood
4 Norway spruce
5 white pine
6 red pine
7 black spruce bog
8 hardwood bog
9 mixed hardwoods
10 75% hardwood, 25% softwood
11 25% hardwood, 75% softwood
Normalized Difference Vegetation Index (NDVI) was calculated for each pixel in
the AVIRIS image as in Eq. (2),
[IR-RED]
NDVI = [IR+RED] (2)
where RED and IR are AVIRIS reflectance values equivalent to Thematic Mapper
(TM) Bands 3 (RED, 630-690 nm) and 4 (IR, 760-900 nm).
65
Application of the PnET-Day model

The PnET-Day model (Aber et al. 1995) was developed to utilize two recently
produced data sets to increase the accuracy and simplicity of predicting whole-forest
canopy C gain. The first is a greatly expanded set of data on the relationship between
foliar N concentration and maximum rates of net photosynthesis (Am.) for temperate
and tropical woody species (Reich et al. 1991b, 1992). The second is a set of daily C
balance measurements collected over three years using the eddy correlation technique
(Baldocchi et al. 1988, Verma 1990), as part of the Harvard Forest Long-Term
Ecological Research (LTER) and National Institute for Global Environmental Change
(NIGEC) programs (Wofsy et al. 1993).
The PnET-Day model uses a linear relationship from Reich et al. (in review) to
calculate instantaneous absolute rates of A max from foliar N concentrations. Empirical
data suggest that dark respiration rates average 10% of A max ; thus, gross
photosynthesis is set to 1.1 times A max . Maximum gross photosynthesis is reduced for
suboptimal temperature and vapor pressure deficit (VPD) using equations in the
original PnET model (Aber and Federer 1992). A layered canopy is simulated, with
both radiation intensity and specific leaf weight (SLW) declining with canopy depth.
Realized rates of photosynthesis decline nonlinearly with decreasing light intensity.
Leaf respiration increases with temperature (QlO =2.0) and is calculated separately for
daytime and nighttime temperatures. Maximum (summer) and minimum (winter) leaf
mass are input parameters. The onset of canopy development in spring is driven by an
accumulated growing-degree-day algorithm, and canopy senescence results from
negative C balances in autumn. Both of these are responsive to weather patterns
unique to a specific year (see Aber et al. 1995 for a complete description of the model).
PnET-Day is a canopy-only model and does not simulate soil or nonleaf respiration.
The gross photosynthesis term in PnET-Day has been validated against 295 daily
estimates of gross C exchange (GCE) obtained by eddy correlation measurements at
the Harvard Forest (Aber et al. 1995). Eddy correlation methods actually measure net
C exchange (NCE) of an entire ecosystem. Adding measured soil and stem respiration
terms to this value yields GCE, equivalent, in this case, to our gross photosynthesis
(Wofsy et al. 1993, Fan et al. 1995). Agreement between model and eddy correlation
estimates of GCE (Aber et al. 1995) suggests that this kind of simple, daily time-step
model based on physiological measurements at the leaf level can accurately predict
seasonal changes in gross C exchange by a forest canopy. Aggregation analyses
suggest that using monthly mean climatic data to drive the canopy model produces
results similar to those achieved by averaging daily eddy correlation measurements of
GCE (Aber et al. 1995).
We used the PnET-Day model to estimate net photosynthesis for each pixel within
the Harvard Forest AVIRIS scene for which foliar N concentration, foliar biomass and
forest type were estimated. Of the 10 forest types identified in the scene (Table 1),
categories 8 and 9 were simulated as pure hardwood stands, and 4, 5, 6 and 7 were
simulated as either pure spruce (4,7) or pure pine (5,6) stands. For these, only foliar N
concentration and maximum foliar biomass changed between pixels, using
AVIRIS-derived estimates for each cell. The remaining input parameters remained
constant (Aber et al. 1995).
For pixels identified as mixed (2, 3, 10 and 11), AVIRIS-derived estimates of N

concentration were used to determine the ratio of each species group present. For
example, if N concentration for an AVIRIS-classified mixed hardwood/conifer pixel
was greater than 2.0%, the pixel was defined as all hardwood and the model was
simulated with hardwood parameters and N set equal to the AVIRIS-derived value. If
the N concentration was less than 1.4%, the pixel was defined as all conifer and
simulated with red pine parameters. For pixels with N concentrations between 1.4%
and 2.0%, PnET-Day was simulated for each of the species groups represented in the
pixel, and the weighted-mean average of the two became the value for that pixel. This
weighting was a function of AVIRIS-derived N concentration and an average conifer
and hardwood N concentration (Martin and Aber in review).
All pixels were simulated for three years (1991-1993) using measured average
monthly climate conditions at the Harvard Forest for those years. Values reported are
the average annual totals for this three-yr period.
In order to test the increase in specificity of GeE predictions with the addition of
spatial data, the PnET-Oay model was simulated using data input variables of species,
foliar N concentration and foliar biomass from a combination of sources (Table 2).
The first model run used the best ground measurement data available without the use
of remote sensing. The source of the field-measured species distribution is a GIS
database of a Harvard Forest stand survey completed in 1986 (Harvard Forest
Archive). Foliar biomass and N concentrations used were the mean values derived
from field sampling of plots with similar species composition. The second run used
AVIRIS-derived species and field-measured N concentration and foliar biomass. The
third and fourth model runs used AVIRIS-derived species and AVIRIS-derived N
concentration or foliar biomass values, respectively. The final model run was done
using data derived from AVIRIS for all three input variables: species, foliar biomass
and N concentration.
Table 2. Sources of model iuput parameters used in the five model runs, mean
value of modeled net photosynthesis and mean difference in a pixel-by-pixel
comparison of runs 2-5 with run 1.
Model Species FoliarN Foliar Mean Mean

run biomass g C m-2 yr-I difference
1 Field data Field data Field data 656

2 AVIRIS-derived Field data Field data 615 141
3 AVIRIS-derived Field data AVIRIS-derived 621 224
4 AVIRIS-derived AVIRIS-derived Field data 629 235
5 AVIRIS-derived AVIRIS-derived AVIRIS-derived 630 242
67
Results and discussion
Interactions of canopy characteristics

Unlike other studies that have predicted LAI from NDVI, our study found that NDVI
and LAI were not highly correlated among the species in this region (Fig. 1). Although
NDVI differed between needleleaf and broadleaf stands, there was little variability in
NDVI within these two stand types. LA! for all hardwood stands varied from 3 to 6
and conifer species ranged in LA! from 2 to 8, with pine stands having low values and
Norway spruce stands having high values. The relationship between NDVI and LA!
was not significant within plots of similar leaf type or over all plots. An inverse
(although not significant) relationship existed between NDVI and LAI across plots of
both leaf types (e.g., broad- and needleleat). Lack of a significant relationship within
the needleleaf stands may have resulted from our use of published SLW values instead
of measurements taken at the Harvard Forest. However, field data were available for
all dominant broadleaf species on each plot. Within the diversity of species
composition in this study, NDVI was more strongly correlated with foliar N
concentration (R2 = 0.92) than a~y other measured parameter (LAI, foliar biomass,
total N, see Fig. 1). Foliar N concentration was significantly related to NDVI across all
plots and within the broadleaf plots. This is consistent with the results of Myneni et al.
(1995), who showed that vegetation indices such as NDVI should be indicative of the
concentration of foliage constituents in optically thick canopies.
AVIRIS-derived data planes

Canopy N concentration and species distribution for the study area have been derived
from AVIRIS data in earlier studies (Plates la, Ib, Martin 1994, Martin and Aber in
review). Two separate multiple linear regression equations were developed to predict
foliar biomass from AVIRIS first difference absorbance data - one for conifer and
2
one for hardwood stands (Table 3), with R = 0.90 and 0.86, respectively. TheAVIRIS
estimate of foliar biomass for the entire Prospect Hill region was calculated by
identifying each pixel as hardwood or conifer, and applying the appropriate foliar
biomass calibration equation to that pixel (Plate lc).
AVIRIS-derived N concentration and foliar biomass for each stand type indicate
that the majority of values falls within the expected range (i.e., conifers have high
foliar biomass and low N concentration), as illustrated in Figure 2. However, due to
errors in either species classification or N concentration prediction, it is possible that
some pixels will have a combination of N concentrationlbiomass/species values that
are not reasonable. For instance, a mixed pixel misclassified as a needleleaf pixel
could have a higher N value than expected and therefore fall out of the range of the
N-A max prediction equation in the model. To illustrate, a number of pixels are identified
as needleleaf species and have an N value> 1.8%. These N concentration values may
be higher than expected for needleleaf species, since a pixel could be classified as
needleleaf if it included up to 20% broadleaf species (Martin 1994) - causing an
increase in overall foliar N concentration for the pixel. Additionally, a number of
pixels are identified as broadleaf, with N values < 1.6% (Fig. 2). These are classified
0\
00
N
E 2000
$ eB R20 6~" ;;i
:; 1~ R2 tt1 J110.01 ~
E 0 R2 ..0 10 c:::
9 1000 0 0 t.o::l
D)!5QO 0 C't)
Q 0 ~ <:l
~ 0 ~
072 0.76 060 0.6- ::>0

~
o R2.008 R'.0.67.. 0 ~
'l> R2 ... 0.oo 1 R' ..0 98 'il 
2 Rl ..0.l0 V:l
~
'r 0 0Rl ..061 
5 5 0 ;::,:
oQ
- 80 o ~o
~~2 0 0 .76 060 0.- 20 500 '000 1500 2000 S

2_ R2.0.n.. 2 2
S-

22 R2 ... 01~ ~ 2'; ~ R2059 2; 8 i 0 R'011 ~
'" 20 R2 .. 0.~5 ~ 20 ~ Rl .. O.oo 20,Q 0 R' .,0.01 <:l
~ 1.8 0 '_IS?J R2 M-O.08 I e ~ R2 bI-O.06 ~
g '6 16 1,6 ::::
; 1,4 CO 14. 0 0 ,. 0 ~
1.2 0 cSb 0 1,2 0 0 0 ~ 1.2 00 0 00
'8 72 0.7. 0 .0 0." 100 ~ 100;) '500 2000 '.02 J _ 5 1
.Q,
~
<:l
,.. 2SOO 0 Rl.0 47.. 2~ 0 2~ 0 2~ 0 R'.O,41. ~
E 2000 CO R~ ,..0.00 2000 0 2000 00 2000 cP R2,..-0.04 .....
'" R2 .. =021 R' ... 0.~2.. '"tl
'"'; 1500 1$00 0 ISOC 0 R2.0.79.. 1500 0 (3
&l 1000 00 1000 OOR2.0.9S.. 1000 &0 R'tllO.98.. 1000 0 ~
e 0 AfS:i .D. 0 fl2 0 . 96.. 0 ~ 0 R' ..0.51.. 0 <0 ;:;
~ 500 0 (J/3"'Q ~ .-u R2 0 83.. !OO 0: ~ 0 000 ~ 0 .....
0 0 0 ;::: .
8.12 0.76 0.60 0.. 0 500 .'000 1500 2000 2 J _ 5 6 7. 1012 .._ 16 1. 2.0 2.2 2._ ~.
NOVI fol,or 8oomoss (gm- 2) 1AI Nitrogen (7.)
Figure 1. Relationships between canopy characteristics measured at 20 study plots in the Harvard Forest. R2 stands for all plots,
R201 for needleleaf species and R2bl for broadleaf species. Circles represent needleleaf plots, diamonds represent broadleaf plots.
00
:::-
'600 ~
'600
-;.
..
-i '60:[
~ , 200 e
() '200
,, !!
e
OO~ 800 ~
r."OJ ."
r.
sao
C C
Cl :-
0 400 0 ~oo
Q 'tuvr" . "
0 ......~W1.
0. r.
0.
0000
"
~
.. 0, a 4
r'
.0 12 26 30
"
;;
;z
0
clIe a IE Ob6
CO('llie
:::- :::- :::-
1500 '600 1600
...' ~
. ~
~ , 2CO E E
u '200[ u
~ ~ _-r;~~.\. "
. ~
00 ....
"c ,__.
---r~~*-~~ . . . __ a c",.
'" ..
<:
0
Q: - r.
,,[ -" . . .::: .!~~.~.:.~ "-
OJ
. 0000 ~OOOO ;Z
(~ ".-2)
Figure 2. Plots of PnET-predicted net photosynthesis for needle- and broadleaf pixels vs AVIRIS-predicted foliar biomass, N
concentration and NDVI. 0\
\0
as hardwood bog pixels, typical of an area with an extremely sparse canopy. The
broadleaf plots used to develop the N calibration equation had LAI values of 3 or
more; for this reason, the N equation does not predict well on sparse-canopy pixels.
Table 3. Calibration equation predicting foliar biomass for needleleaf and

broadleaf species using first difference AVIRIS absorbance data. SEC =
Standard Error of Calibration.
2
Leaf type Coefficient A(nm) R SEC
Needleleaf bo 778.94 0.90 1653

bl 5576.72 636
b2 2991.47 469
b3 16201.23 420
Broadleaf bo ' 21575.45 0.86 169

b2 -488486.35 1280
b3 -132796.17 1501
Model estimates of net photosynthesis

Effects of increased information. The mean values for net photosynthesis calculated
for each of the five model runs were similar - 656,615,621,629 and 630 g C m- 2 y{1
for runs 1-5, respectively. The model run using all field data yielded the highest
estimate, 656, compared with a lower estimate of 615 when all three input parameters
were derived from AVIRIS data. However, model runs varied widely in pixel-to-pixel
comparisons, indicating that the model runs using averaged field data missed a great
deal of fine-scale variability (Plate 2). In a pixel-to-pixel comparison of run 1 (using
all field data) and runs 2-5 (using combinations of AVIRIS-derived data), the runs
using foliar biomass and/or N concentration data derived from AVIRIS (runs 3-5)
differed most from the model run driven only by field data. These differences were
almost twice as high as the difference between the model run using AVIRIS-derived
species and mean field-measured N concentration and foliar biomass (run 2). The
model run using only field-measured parameters at coarse resolution provides an
adequate estimate of net photosynthesis for the entire region. (The mean value for net
photosynthesis was the same as for the other, more detailed, model runs.) However, to
truly understand the spatial variability in net photosynthesis, information is needed for
all three input variables for each pixel within the scene (Plate 2b).
Sensitivity ofPnET-Day predictions to foliar biomass and N concentration. Estimated
annual net photosynthesis varied widely between pixels and between forest types
within the Harvard Forest in response to estimated foliar biomass and N concentration.
In general, the deciduous forest types were apparently more sensitive to foliar N
concentration than to total foliar biomass (Fig. 2). This is due in part to a relatively
71
narrow range of foliar biomass values, and to the relatively steep slope of the
relationship between A max and foliar N concentration (Aber et al. 1995). The opposite
is true for evergreen stands, where a large range in biomass values and a shallow
relationship between foliar N concentration and maximum net photosynthesis results
in great sensitivity to biomass and relatively little to foliar N concentration (Fig. 2). In
both cases, estimated net photosynthesis would be predicted more accurately from
foliar biomass and/or foliar N concentration than from NDVI.
Conclusions
Several results reported here may alter the way we approach the regional estimation of
forest C balances. First, the traditional use of NDVI as an indicator of LAI may not
apply to optically dense forests, where the best correlation appears to be with foliar N
concentration. This could alter the use of NDVI in models of both C and water
exchange. Second, the use of relatively coarse-scale data (e.g., Plate 2a) derived
entirely from field collections provides a similar estimate of total net photosynthesis
over the region, as does the finer-scale data derivable from AVIRIS imagery. However,
this may be due to the relatively homogeneous nature of the landscape used for this
example. Third, coarse-scale data do not capture the complexity of fine-scale
differences actually occurring over the landscape, which can be defined by
AVIRIS-type imagery. The power or limitation of each combination of data planes
used here (Table 2) for any particular application will depend on the scale or degree of
resolution required.
References
Aber, J.D. and Federer, C.A. 1992. A generalized, lumped-parameter model of photosynthesis,
evapotranspiration and net primary production in temperate and boreal forest
ecosystems. - Oecologia 92: 463-474.
Aber, J.D., Reich, P.B. and Goulden, M.L. (in review). Extrapolating leaf CO 2 exchange to the
canopy: A generalized model of forest photosynthesis validated by eddy correlation.
Baldocchi, D.D., Hicks, B.B. and Meyers, T.P. 1988. Measuring biosphere-atmosphere
exchanges of biologically related gases with micrometeorological methods. - Ecology
69: 1311-1340.
Clark, RN., Swayze, G., Heidebrecht, K., Goetz, A.F.H. and Green, RO. 1993. Comparison
of methods for calibrating AVIRIS data to ground reflectance. - In: Green, RO. (ed).
Summaries of the Fourth Annual JPL Airborne Geoscience (AVIRIS) Workshop,
October 25-29, 1993, NASA, Jet Propulsion Lab, Pasadena, CA, 1: 35-36.
Fan, S.M., Goulden, M.L., Munger, J.w., Daube, B.C., Bakwins, P.S., Wofsy, S.C., Amthor,
J.S., Fitzjarrald, D.R., Moore, K.E. and Moore, T.R. (in review). Environmental controls
on the photosynthesis and respiration of a boreal lichen woodland: A growing season of
whole-ecosystem exchange measurements by eddy correlation.
Gao, B., Goetz, A.F.H. and Zamudio, J.A. 1991. Removing atmospheric effects from AVIRIS
data for surface reflectance retrievals. - In: Green, RO. (ed). Proceedings of the Third
Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop, May 20-21,1991,
NASA, Jet Propulsion Lab, Pasadena, CA, pp. 80-86.
Gao, B., Heidebrecht, K.B. and Goetz, A.EH. 1992. Atmosphere removal program (ATREM)
user's guide. - Center for the Study of Earth from Space/CIRES, University of Colorado,
Boulder, CO, 24 pp.
Gholz, H.L., Curran, PJ., Kupiec, J.A. and Smith, G.M. 1996. Assessing leaf area and canopy
biochemistry of Florida pine plantations using remote sensing. - In: Gholz, H.L.,
Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the Modeling of
Forest Productivity. Kluwer Academic Publishers, Dordrecht, The Netherlands, pp.
3-22.
Goward, S.N. and Dye, D.G. 1996. Global biospheric monitoring with remote sensing. - In:
Johnson, D.W. and Lindberg, S.E. 1992. Atmospheric Deposition and Forest Nutrient Cycling.
- Springer-Verlag, New York. 707 pp.
Martin, M.E. 1994. "Measurements of laboratory and airborne high spectral resolution visible
and infrared data." - Ph.D. dissertation, University of New Hampshire, Durham, NH. 97
pp.
Martin, M.E. and Aber, J:D. (in review). Estimation of forest canopy lignin and nitrogen
concentration and ecosystem processes by high spectral resolution remote sensing.
Myneni, R.B., Hall, EG., Sellers, PJ. and Marshak, A.L. 1995. The interpretation of spectral
vegetation indices. - IEEE Trans. Geosci. Rem. Sens. (in press).
Newman, S.D., Soulia, M.E., Aber, J.D., Dewey, B. and Ricca, A. 1995. Near infrared analyses
of forest foliage. 1. Proximate carbon fraction and nitrogen analyses for the Accelerated
Canopy Chemistry Program: Methods and quality control. - J. Near Infra. Spectr. (in
press).
Peterson, D.L., Spanner, M.A., Running, S.w. and Teuber, K.B. 1987. Relationship of
thematic mapper simulator data to leaf area index of temperate coniferous forests. -
Prince, S.D. 1991. A model of regional primary production for use with coarse-resolution
satellite data. - Int. J. Rem. Sens. 12: 1313-1330.
Reich, P.B. and Walters, M.B. 1992. Leaf life-span in relation to leaf, plant, and stand
characteristics among diverse ecosystems. - Ecol. Monogr. 63: 365-392.
Reich, P.B., Kloeppel, B., Ellsworth, D.S. and Walters, M.B. (in review). Different
photosynthesis-nitrogen relations in deciduous hardwood and evergreen coniferous tree
species.
Reich, P.B., Walters, M.B. and Ellsworth, D.S. 1991. Leaf age and season influence the
relationship between leaf nitrogen, leaf mass per area, and photosynthesis in maple and
oak trees. - Plant Cell Environ. 14: 251-259.
Running, S.W. and Hunt, E.R. 1993. Generalization of a forest ecosystem process model for
other biomes, BIOME-BGC, and an application for global-scale models. - In:
Ehleringer, J.R. and Field, C.B. (eds). Scaling Physiological Processes: Leaf to Globe.
Academic Press, San Diego, CA, pp. 141-158.
Sellers, P.J., Berry, J.A., Collatz, G.J., Field, C.B. and Hall, EG. 1992. Canopy reflectance,
photosynthesis, and transpiration. 3. A reanalysis using improved leaf models and a new
canopy integration scheme. - Rem. Sens. Environ. 42: 187-216.
Verma, S.B. 1990. Micrometeorological methods for measuring surface fluxes of mass and
energy. - Rem. Sens. Rev. 5: 99-115.
Wofsy, S.C., Goulden, M.L., Munger, J.w., Fan, S.-M., Bakwin, PS., Daube, B.C., Bassow,
S.L. and Bazzaz, EA. 1993. Net exchange of CO2 in a mid-latitude forest. - Science 260:
1314-1317.
SECTION TWO
Landscape/Regional-Level Analyses
FOUR
Detecting Structural and Growth Changes in Woodlands and

Forests: The Challenge for Remote Sensing and the Role of
Geometric-Optical Modelling
David L.B. Jupp and Joe Walker
Jupp, D.L.B. and Walker, J. 1996. Detecting structural and growth changes in
woodlands and forests: The challenge for remote sensing and the role of geometric-
optical modelling. - In: Gholz, HL, Nakane, K. and Shimoda, H. (eds). The Use of
Remote Sensing in the Modeling of Forest Productivity. Kluwer Acad. Publ.,
Dordrecht, The Netherlands, pp. 75-108.
Vegetation structure can be defined as the vertical and horizontal distribution of plant
material and individual plants within a plant community. The structure of forests and
woodlands is sometimes regarded merely as a factor for determining the function of
total biomass, leaf area or leaf area index (LAI). However, in vegetation dynamics as
well as in resource and habitat assessments, structure is a key factor. In the same way,
biomass, vegetation floristic association and LAI are routinely derived from remotely
sensed data, while structural effects are largely ignored or regarded as a nuisance. In
semiarid woodlands in particular, high-image variance due to crown spacing and
shadowing is a mirror of the high spatial variance and gappy structure that characterise
this type of land surface. Since the amount of shadowing depends on sun and view
angles as well as on the proportion and size of the plants, a range of reflectance values
can be obtained for exactly the same land surface viewed at different times and using
different view perspectives. This effect is generally called the Bidirectional
Reflectance Distribution Function (BRDF). Interpreting significant temporal trends in
the structure and composition of vegetation requires the application of physical
models - both to interpret remotely sensed data and to understand the elements
creating the vegetation dynamics. Detecting these trends assumes particular
importance in assessing the impacts of global climate change. The remote sensing
measurement models discussed here belong to the class of geometric-optical, or GO,
models for vegetation types in which the crowns of the upper stratum (trees or shrubs)
are not touching, as occurs in open forests, woodlands and shrublands. These
vegetation types occupy a large proportion of the earth's surface. For example, in the
Murray-Darling Basin in Australia, these systems make up nearly 75% of an area
comprising 1 million sq. km. To illustrate structural change, the dynamics that
occurred in an area of woodland vegetation over a lO-yr period were simulated and
expressed by changes in the height and cover of trees, shrubs and grasses and the cover
of bare soil. A vegetation dynamics model (RESCOMP) based on availability of water,
light and nutrients was used to simulate the structural changes. Over this period, the
75
shrubs increased dramatically (from 2% to 20%), tree cover fluctuated slightly and
grass or bare soil fluctuated dramatically depending on seasonal rainfall conditions.
These data form the vegetation input into the GO model. We focused on shrub
invasion, and asked whether reflectance data (even assuming effective atmospheric
correction) could detect the increase in shrub cover in the woodland. While total cover
changes were clearly detectable, trends in shrub cover as opposed to tree cover could
not be identified using only spectral data. We propose that separation could be
achieved by quantifying and monitoring BRDF effects for pure woodlands, pure
shrublands and mixtures of trees and shrubs; modelling changes in image texture for a
range of pixel sizes; or both. Ultimately, the underlying relationships among spatial
variance, ecological condition and future dynamics in visible/near-infrared, thermal
and radar images seem to indicate that image texture and variance are the tools that can
best monitor vegetation dynamics and structure for spatially variable vegetation types.
At this time, however, BRDF seems to provide the most accessible operational tool for
monitoring structural changes at multiple scales.
D.L.B. Jupp and J. Walker, CSIRO Division of Water Resources, P.O. Box 1666,
Canberra, ACT 2601 Australia.
77
Introduction
Remotely sensed images from satellites and aircraft have been used successfully both
to characterise spatial variation in land cover and to monitor temporal changes in
cover, or effective biomass. Successful applications at regional scales include land
clearing, agricultural use and delineation of large wildfIres. At generalised levels of
parameter resolution, issues regarding seasonal differences in greenness, flooding and
siltation, forest disturbance and grassland productivity can be addressed using orbiting
satellite data. These examples represent signifIcant advances in monitoring capability
and utilise the strengths of current remote sensing systems to provide a synoptic view
with complete spatial coverage. However, the potential to use temporal scenes to
monitor subtle changes in the structure (the vertical and horizontal distribution of
vegetation components) of specifIc land surfaces may be a far more signifIcant task for
regional monitoring. Indeed, structure rather than integrated measures of cover and
biomass must be considered if the monitoring is to be matched to the scale of the
activities generating many of the changes. One example is forestry, where the
consideration of structure is a, primary economic factor and structural changes
resulting from forest operations constitute an issue that must be addressed to fully
assess sustainability. At a broader scale, the same considerations apply to monitoring
the impacts of global climate change.
Given the current worldwide focus on impacts of climatic or land use changes on
the sustainability of a great number of landscape systems, there is an especially urgent
need to establish the means to monitor surface changes at a scale suitable for land
management and in terms of parameters directly related to the ecological condition of
the surface. In this chapter we propose that structural information, in addition to
biomass or leaf area index (LAI), is needed if this is to be done effectively. This is
because interpretation of change needs to be made in terms of the degree of
degradation or amelioration in order to detect early warnings or indicators of
environmental change. Structural properties are often the measures that are needed. A
case in point is the Murray-Darling Basin of southeastern Australia (Fig. 1), where
considerable effort is being directed to develop methods to combat dry land
salinization and unwanted shrub invasions. The Basin is large (1 million km2 ) and has
been subjected to massive tree clearing for cropping and livestock grazing purposes
over the past 200 years (Walker et al. 1993). Such changes in land cover are relatively
easy to detect and quantify over much of the Basin. However, subtle changes are
occurring in the remaining native vegetation and in some agricultural systems
requiring a more sophisticated approach. For example, in the Murray-Darling Basin,
tree removal has affected the water balance, increasing recharge into groundwater
systems. The subsequent increases in saline water table levels has resulted in
salinization and waterlogging of the lower parts of the landscape. Detecting changes
in the vegetation response before salt patches occur on the surface represents a
powerful management tool as well as an early warning system. Increased grazing
pressure can also result in land degradation, two marked effects being soil erosion and
invasions of woody weeds into semi-intact woodland systems. The latter problem is
expressed as a structural change rather than one of total biomass change.
Figure 1. The location of the Murray-Darling Basin and the Wycanna field site in
Australia.
Because of the size of the Basin, remotely sensed data are ideally suited to monitor
land surface changes and identify areas requiring land rehabilitation or, in
combination with other information, predicting areas with a high risk of salinisation
and degradation. However, semiarid ecosystems (woodlands and shrublands) occupy
some 75% of the Basin. In these systems, the upper canopy is discontinuous (i.e.,
crown separation is 0.5-20.0 times the crown width, Walker and Hopkins 1990), and
rapid fluctuations in the composition and cover of the ground layer can occur as a
result of sporadic rainfall events. The high spatial and temporal variance is
characteristic of woodlands and is present in remotely sensed images as variance
created by contrasts introduced by the shadowing of objects (trees, shrubs and grass
tussocks) and background variation, in addition to variation in the spectral signatures
of the components themselves. Since the magnitude of these shadowing and
composition effects is determined by the proportion and size of the vegetation
components as well as by sun and view angles, a wide range of values can be obtained
for an area of a scene that remains physical1y unaltered but is viewed at different times.
Because this variation contains the very information pertinent to ecological change in
these systems, methods that utilise the variance, rather than suffer from it, are urgently
needed.
In this chapter we discuss the proposition that physical models based on scene
components at the scale of individual tree crowns are the appropriate technology to
develop for assessing and monitoring such discontinuous environments using
remotely sensed data. Physical vegetation dynamics models that allow for competition
between individuals allow us to define the expected outcomes of disturbance,
rehabilitation or "business as usual" in terms of vegetation growth and succession.
Using compatible remote sensing measurement models where the primary geometric
79
objects are trees, we can also interpret images directly in terms of the essential
indicators of ecological change and economic yield and at the scale where disturbance
and regeneration occur. As an outcome of this approach, when we compare physical
scene component models with vegetation dynamics models that have similar scale and
component structure, we can ask such relevant questions as: (i) is the vegetation
composition and biomass actually diverging from the expected path, given its previous
structure, composition and the actual recent climatic regime? and (ii) how much
change must occur in vegetation structure before the change can be detected by
remotely sensed reflectance images?
Many existing models can simulate scene reflectance and optical properties in
terms of variations in land surface composition and structure. There are also many that
simulate land surface changes in response to driving environmental forces such as
sunlight and rainfall, and changes in the near-surface properties of the soil. One
approach used for optical models is to treat the canopy as a turbid medium consisting
of layers of scattering elements. The canopy elements are parameterized by
reflectance and transmittance of individual leaves, LAI and leaf angle distribution.
This approach has been comprehensively reviewed by Myneni and Ross ( 1991). Such
models tend to be one dimensional, with leaf density varying with height above the
ground. The turbid medium approach is most useful in closed canopy situations but
has limited applicability for our region of interest where, although closed canopies can
occur at certain times in croplands, discontinuous canopies prevail. Matching these
models is a class of ecological ones often called "green sponge" models, where the
vegetation layer is a diffuse blend of phytomass (Running et al. 1989, Pierce et al.
1993). These models can simulate future trajectories of green leaf and biomass, but
competitive effects among individual trees are largely absent.
Relatively few remote sensing and ecological models can be consistently combined
to simulate the level and scale of competition, anthropogenic disturbance and
ecological interaction that exist among individual trees. An important class of models
that can be comprises the geometric-optical (GO) models for scene radiance
modelling, such as those described in Jupp et al. (1986), Strahler and Jupp (l99Ia,b)
and Nilson and Peterson (1991), and individual tree-scale models that can simulate
land cover and the development of vegetation structure, such as the RESCOMP model
of Penridge et al. (1987) and the gap models described in Shugart (1984). Here we
discuss the use of such models in the development of a methodology to monitor land
cover and structural changes.
The ecological significance of cover and structure

Community structure is determined by the competitive growth of species present in
the vegetation in response to environmental and anthropogenic constraints.
Environmental constraints include bioclimatic and biophysical regimes as well as
disturbance regimes such as fire and insect attack. Anthropogenic constraints include
certain types of land use imposed on the vegetation, such as agriculture, grazing,
forestry and, to some degree, protection of wilderness areas.
Vegetation structure is defined here as the horizontal and vertical distribution of
components within a plant community. The vertical aspect of the vegetation structure
can be described by the stratification of vegetation into layers and the horizontal aspect
of vegetation structure by the arrangement (spacing) and density of foliage within a
given vegetation layer. The components that comprise a plant community, known as its
"growth forms," refer to the general character of the vegetation (e.g., trees, shrubs,
maBee, grasses), as weB as to its sizes and shapes. An area of vegetation within a
similar environmental/anthropogenic disturbance regime and exhibiting a repeatable
pattern of vegetation components is often referred to as a community. Individual
communities can be characterised as a mixture of growth forms of varying height and
spacing, which is referred to coBectively as "architecture." The architecture of a
community accounts for most of the characteristics used to classify vegetation
according to its structure.
Where distinct vegetation layers exist, vegetation samples can be divided into
different vertical strata. AB vegetation samples have an upper stratum, but lower layers
may or may not occur. In some cases, distinct layering may not occur even though
vegetation is present below the upper strata. Vegetation structure has been recognised
as a key component of Australian vegetation through its use in various vegetation
classifications. The current widely used map of Australian vegetation by Carnahan
(1976) and the field guide of Walker and Hopkins (first edition 1984, revised 1990)
can be traced back through Specht (1970), to WiIliams (1955) and finaIly to Wood
(1937). These classifications recognise a triplet nomenclature for vegetation
classification comprising the upper, middle and lower layers to develop structural
classes and structural formations. Such classifications can be directly related to
function. For example, plant size is generaBy related to the volume of soil that can be
exploited for water and nutrients. Likewise, crown size can be used to modify
radiation penetration onto the ground as weB as to provide an indication of the area a
plant will probably influence through shading, litter deposition and other effects.
A common way to quantify the amount of vegetation present is to specify foliage
area (crown area, leaf area or both) in the identified strata by species. The vertical and
horizontal distribution of this foliage area is a structural factor and determines the
relationship between foliage area and light penetration. In open forests and woodlands
this relationship is often referred to as "gappiness," in that foliage clumping creates
gaps through which light may penetrate to the lower layers (Nilson and Ross 1996).
Ecological factors related to the gappiness of vegetation include spatial interactions
among individual trees, shrubs and grass tussocks. The spatial distribution of
individuals of the species present is also a function of tolerance as weIl as competition.
Tolerance is the ability of a species to withstand the conditions of its local
environment. If environmental conditions move beyond the tolerance level for a
species, it wiB die. Competition refers to a plant's contest for space and nutrients both
within and outside its own species. Thus plant survival depends upon the suitability of
a given site as weIl as on the outcome of competition for nutrients and space both
within and between species. The biotic and abiotic features of a site, especiaBy
tolerance and competition, combine to exert a controIling and selective influence on
the geographical distribution of taxa (Kuchler and Zonneveld 1988) and are revealed
in the structure of the plant community.
81
A theoretical approach to the study of plant competition, especially between plants

of different sizes, is "ecological field theory" (Walker et al. 1989). Ecological field
theory explicitly utilises structure and hence potential resource acquisition for
individuals in the community as well as in different layers that can be described in the
field. This approach follows Milne (1961) and Grime (1973) in seeking to define
competition in terms of its mechanisms rather than its effects. Interference is the
influence of a plant upon its neighbours' environment through resource competition.
Interference potential is the interference a newly germinated seedling must overcome
to establish itself and subsequently grow at a site within the influence of all
neighbouring plants. This potential is calculated in terms of the effects of the
neighbours' crowns, stems and roots upon the spatial distribution of water, nutrients
and light. Other work, notably by Shugart (1984), Running et al. (1989) and Pierce et
al. (1993), has followed the same general theme in emphasising structure as a key
component in vegetation functioning, but to varying mechanistic degrees.
Measurement models for forests and woodlands
Cover and structure measurement models

The different perspectives of the many applications being considered, such as forest
resource assessment, ecological assessment of degradation and amelioration, as well
as monitoring sustainability and factors that affect remotely sensed data, must be
brought into a common framework if we are to successfully match the needs of the
assessment with the capabilities of the tool. To do this, we define a forest or woodland
as a collection of trees occupying a number of height strata and varying in shape and
size.
Within the strata, trees are labelled by species (e.g., species t); for a given species in
a given stratum (e.g., stratum i), the density of trees (i.e., of their stems, or trunks) will
be denoted \i (Fig. 2). The overall height is h (to the top of the tree), the height from
the ground to the start of the crown is I, the crown thickness is T (= h-l) and the crown
diameter is D. The diameter of the trunk (at a nominated height - normally about 1.5
m or, more systematically, 0.1 I) - is denoted d. In this idealised tree, the crown is
regarded as a volume that the tree commands but does not necessarily densely fill with
leaves and stems. Crown openness is expressed by the crown factor, F, which is 1.0
when the crown is densely filled by leaves and 0.0 when it is vacuous. The crown
factor is measured as one minus the fraction of sky vertically visible through the crown
(Walker and Hopkins 1990).
At the right of Figure 2 is a summary of the foliage area or sum of fractional foliage
area by height in an example canopy. This is a quantitative representation for the
foliage profile for the canopy and provides a structural description that can be used to
define canopy layering (Penridge 1984, Walker and Penridge 1987, Penridge 1987).
This measurement model allows both remote sensing and ecological vegetation
dynamics to be related to the interests of foresters. Forest assessment for logging
depends on the style of logging (e.g., clearfelling for woodchips or selective logging)
and on the end use of the timbers. However, general measurements used in forest
resource assessment and monitoring are basal area (BA) and timber volume (V). In
folil" Profll.
" D.ftllty
Figure 2. Notation used for forest measurement model, RESCOMP.
selective operations, tree size distributions for the different species classes are needed
(or at least means and coefficients of variations of d) to assess commercial content as
well. In terms of the above notation, the BA and V values associated with a given
species in a given stratum are
BA,; = A,; a,i (I)

1 1r2
= A,i 4 d,;
and
V,; =fBA,;l (2)
= f(l-TIh) BA,; h
where a,i is the mean cross sectional area of trees in stratum i, species t, f is the form
factor for the trunks (a measure of taper for the species) and the ratio of T to h is a
canopy structural parameter, which is usually relatively stable for species in similar
age (i.e., height) classes and will be referred to later as a scale-free parameter in
canopy BRDF or "hotspot" models.
The importance of the size distribution is that BA can be the same whether there are
a large number of small trees or a few large trees. For a complete forest resource
assessment the distribution of d is needed, or at least the mean value of d in a given
species/height class and a measure of the variance such as coefficient of variation.
From a remote sensing platform, most forested areas (apart from very sparse,
woody canopies) are dominated by the crown cover. As the parallel to BA, there is a
crown area index (C), which is defined as
83
(3)
where D is the crown diameter. C multiplied by F is the effective crown area index,
taking into account within-crown gaps and openness. In many areas (Heller and
Ulliman 1992), it has been observed that C and BA are closely correlated within
species and structural forms (e.g., emergents vs mid-canopy trees). For composite,
discrete canopies, C and BA are often related to about 60-70% of the variation. Even
in the diverse, open-forest areas of Australia dominated by Eucalyptus, McDonald
(1993) has recently shown the statistical persistence and usefulness of this
relationship. From an ecological point of view, these characteristics of the canopy are
significant as measures of the way photosynthetically available light is distributed in
the canopy (Nilson and Ross 1996) and of the spatial distribution of pathways and
storages in nutrient cycling.
Many of the fundamental issues of remote sensing of forest and woodland cover
and structure can be illustrated by considering the simplest situation, where the trees
have no vertical structure but lilre just "disks" of one colour (R d ) scattered on a
background of another colour (R b ) (Fig. 3). In remote sensing of the disk scene, the
colours of the objects and background in an area called the instantaneous field of view,
or pixel, are integrated to provide the image data for the single pixel (r,,) according to
(4)
where kd is the fraction of area of the pixel that is disk and kb is the fraction that is
background, or "gap," between disks. Since there are only two components, kb + kd = 1.
Figure 3. Disk model with 50% cover, constant size and Poisson distribntion of
disk centres.
The cover, or disk area, index C, can be defined as before. However, the relationship
between the gappiness, kb , and C depends on the spatial distribution of trees or disks.
There have been many studies of this distribution, which has been variously modelled
as random (by which is normally meant Poisson-distributed) or as more-or-Iess
clustered. Models for distributions ranging from extreme clustering to uniform and
regular have been developed, such as the parametric models described in Penridge
(1986). Generally speaking, if the trees or disks are clustered, there will be a higher
gap fraction, kb , for the same C than if they are Poisson-distributed. In the case of a
Poisson distribution, the relationship is simple and
(5)
where K indicates the mean of k over all pixels. Using a similar notation, we will
define a pixel based on crown area index, c, as:
1 n 1t 2
c=-E-D (6)
A p i=1 4 I
where the sum is over ,crowns with centres in the pixel and Ap is the pixel area. C is
clearly the mean of c over the pixels in the simple scene. The relationship between c
and kb also depends on the distribution of the trees or disks and needs to be evaluated
statistically over groups of pixels rather than for individual pixels. If crown size and
stem size are related, then c can be considered equivalent to the BA measured for a
single field site during timber cruising and C equivalent to the whole-stand BA.
If the colours are grey levels, the histogram of rs will show how the colour of the
scene varies. It has been suggested (and supported by experiment, Key 1994) that
when the distribution of centres is Poisson, a Beta distribution can be used to model
the grey-scale distribution. In common with the Beta distribution, when pixel sizes are
very small, the distribution of rs will be bimodal, concentrating at Rd and R b in
proportion to the disk cover and gap fraction with variance:
(7)
Eventually, as the pixel becomes larger, the histogram becomes unimodal,

concentrating at the mean (R), where
(8)
and with variance

(9)
where Var(kb ) depends on disk size and pixel size (Jupp et ai. 1988, 1989). This
change is illustrated in the sequence (i)-(iv) of Figure 4. In this case, the trees have
14% colour and the background 29%, chosen to represent reflectances in the red
region for trees plus shadows and sunlit background, as described later in the chapter,
and the ratio of pixel length to disk diameter is 1:8, 1:4, 1:2 and 1:1, respectively.
The main message in this simple model is that information about structure (disk
size and distribution of disk centres) is contained in second-order statistics such as
variance and in the way these statistics change with pixel size (for alternative second-
85
order statistics based on run lengths, see Chen et al. 1993). The spatial variance in
images and the ecological significance of variance are closely related and should be
addressed in terms of the same underlying measurement models.
The opportunity of using image processing to determine structural parameters
through variance has been investigated theoretically during research prompted by
successful development of the Li-Strahler inversion model. The Li-Strahler inversion
model (Strahler and Li 1981, Li and Strahler 1985) was developed by establishing a
formula for the variance of e, although their work was not restricted to the simple disk
model. If the variance is denoted V( e) and the trees are approximately circular in shape
with lognormal size distribution, characterised by the coefficient of variation C;, r
it
follows that the mean area of tree crowns may be found from the formula
A = Vee) (10)
(1 + c9
r-
C
Figure 4. Effect of pixel size on the disk model: (i) full resolution, (ii) 2 x 2
blocking, (iii) 4 x 4 blocking and (iv) 8 x 8 blocking.
The use of this formula depends on how well the measure e can be derived from field
or image data. In practice, this involves resolving the image data into their component
fractions by unrnixing and relating e to the variation in the background component
fraction. For the simple disk model, this means estimating kb from knowledge of the
"end-member" colours R d and Rb , then converting it to an equivalent c value (see
details in Li and Strahler 1985). Even here, however, numerical studies have shown
that good results are very dependent on knowing or accurately estimating R d and Rb
However, despite the clear opportunity for estimating structural information from
image data, this pathway has not been very successful when taken from the research
level of single, well-measured stands to regional levels (Woodcock et al. 1994, 1995).
Perhaps the fact that the corresponding varianc"e formula is not commonly used by
foresters using plots with fixed sizes to estimate mean tree size from basal area surveys
conducted on the ground is an indication of the difficulty of handling variance in
practice. We have discussed the variance because second-order statistics still represent
a significant and incompletely exploited opportunity for inferring structure from
remotely sensed visiblelNIR, thermal or radar data.
GO models and modifications

The disk model is not, however, very realistic. A more realistic, but still simple model
for the remote sensing of a canopy made up of crowns like those in Figure 2 is the
geometric-optical (GO) model. The simple GO (or hotspot) model for scenes that
describe open-forest or woodland areas is based on the one described in Jupp et al.
(1986), Strahler and Jupp (1991a,b) and Li and Strahler (1992). In this model, four
kinds of ground cover are visible from a given direction. These are referred to as scene
components and consist of sunlit canopy (symbol se), shaded canopy (she), sunlit
background (sb) and shaded background (shb). Each component is assumed to have a
characteristic radiance and the radiance of a pixel (r,) is modelled as the area-weighted
combination (or linear mixture) of the characteristic component radiances:
(11)
where the subscripts se, she, sb and shb indicate the radiances of the four components
named above, Rj represents the (mean) radiance of component j and k indicates the
sensed proportion of each component within the pixel from the given view direction.
The mean radiance over the scene (R,,), assuming the view and sun directions are
constant, can be written as
(12)
where Kj represents the mean or expected value of the varying proportions kj over the
scene for j as the components se, she, sb or shb. The mean value (R,), as a function of
sun and observer position, defines the BRDF of the scene.
To compute the scene BRDF model, a description of the objects including their
sizes and shapes, their densities and their distribution patterns over the background is
needed, and the geometrical relationships between the objects and the expected values
of the four components must be established. Jupp et al. (1986), Strahler and Jupp
(1991a,b) and Li and Strahler (1992) describe such a model for spheroidal crown (not
87
necessarily opaque) volumes that is valid for any view or illumination angles using the
Boolean model of Serra (1982). In the Boolean model, the object centres are assumed
to be randomly distributed in a Poisson distribution. By defining the geometry and the
distributions, expressions for ~ may be derived. Strahler and Jupp (1991a,b) use a
simple model for spheroids that is adequate for moderate sun and view zenith angles
and Li and Strahler (1992) provide some more general alternative models for resolving
the Kj" These basic scene BRDF models are easily implemented in various forms such
as mathematical packages or spreadsheets.
In the woodlands and open-forest areas typical of the area of Australia where the
model studies are being made, the pixel-to-pixel behaviour of the image is
conveniently (if not as accurately) described by a simpler form of the model in which
the shaded background, sunlit (but still relatively dark) tree and shaded tree
components are combined so that
(13)
where X is a composite component combining sunlit and shaded tree and shaded
background and R x is computed a$
R _ KscRfC + K,"cR'hc + KshbR'hb

(14)
X - I - K'b
For this discussion, this simpler model has the advantage of clearly demonstrating
how, in many woodlands, the image pixel-to-pixel variation is driven primarily by the
variation in the proportion of sunlit background visible in the pixels and the contrast
between this sunlit background and the other components. It also provides a simple
estimate for k'b from images where R'b and R x are known for an appropriate image
channel, or channel combination, as
r, - R x
k"b = R.,b - RX (15)
For such a model, the mean radiance (i.e., BRDF) over all pixels in a patch with the
same basic underlying cover type and structure is therefore
E(r,) = R, = R x + (R,b - Rx)K'b (16)
where K,b is the mean value of k'b' or the expected proportion of visible sunlit
background for the particular sun and view positions. Moreover, the variance of the
pixel radiance is
(17)
From this it is clear that the variance of a scene, for which this simplified form of
the model is appropriate, is defined by that of ksb and the spectral contrast between the
sunlit background and the composite of tree and shadow, and behaves very much like
the simple disk model introduced before if "disk" is replaced by the composite object
of a tree plus its shadow (Fig. 4). Image simulations have shown that the variance of
k'b is also well approximated by a Beta distribution, with the same mean and variance.
Hence, image variance as a function of pixel size seems to be a sufficient texture

statistic for this simple model.
In the simplified model, the viewed scene consists of objects made up of a
composite of projected tree-plus-shadow silhouettes scattered over a sunlit
background. The sensor integrates the radiance over a pixel so the model for the
resulting image is one of objects on a contrasting background regularised by the
integration into pixels. Such models, and their properties of scale variance, may be
handled by the tools described in Jupp et al. (1988, 1989), where expressions for the
way var(ksb ) changes with pixel size can be found. In general, the expressions are quite
complex and must be computed numerically. However, they can be used effectively to
implement the Li-Strahler inversion in situations where k'b can be estimated from the
image data by end-member analysis.
The end-member analysis technique is similar to but more general than the
estimation of components, as described above. It has been the subject of research and
application in Australia (Pech et al. 1986, Pickup and Foran 1987) and at regional
scales where pixels are mixtures of land covers (Cross et al. 1991). End-member
analysis assumes each pixel to be a composition, or mixing, of a few base components,
or "end members." The result of such mixing is illustrated in Figure 5 for a pixel of
Landsat MSS scale (30 x 30 m), in which an imaginary trajectory starts out with a low
cover of trees and moves with increasing grass and shrub cover to a final point with
low grass cover but more shrubs. The pixel signature is assumed to be a linear sum of
reflectances (in two bands, 5 and 7) from each of n end members weighted in
proportion to its cover (kj ) in the pixel
n
rs=J:,kjR j (18)
j =1
kj~Oforj= l,n
k
j =1
j = 1.
End-member analysis seeks to invert this mixing by deriving the proportions (k) of
each component in the pixel signature. This can feasibly be derived from the remotely
sensed data provided that if there are n components (trees, shrubs, grass, etc.) there are
at least (n-I) channels of data that separate the end members spectrally. The key
assumptions of the end-member method are that (i) the end members (pure examples
of total cover by trees, shrubs, grass and background) are spectrally consistent
between sites and (ii) reflectance values for end members (Rj ) are available from
remotely sensed data or can be derived by other means (such as field spectral
measurements).
There has been considerable work aimed at deriving end members from spectral
data (a form of principal components analysis, Boardman 1990) and employing high-
resolution spectral data to effect separation of more than a few components (Adams et
al. 1989). However, with a lack of available high-resolution spectral data, the approach
suffers from several significant limitations to its applicability in the Basin: (i) available
broad-band signatures of the tree and shrub crowns over much of the Basin, whilst
different, are not markedly spectrally distinct; (ii) even if spectrally distinct crowns did
89
green Jeaf
f- .....
/ ""'--...-." dry bright soil
/
/ ------.---
dry grass + Iitler
I
I
, tree
I
I
\
\
\ dark soil
\
\
\
water
Bend 5
Figure 5. An example of the trajectories used in end-member analysis.
exist for the available bands, their distinction is confounded by the effects of
shadowing within crowns and shadows cast on the background (with bigger plants
shading smaller plants). This makes the signature of the end members difficult to
estimate, because the signature depends on the proportions of crowns and shadows
present and variations in sun and view angles; and (iii) relatively low covers of trees
and shrubs, together with shadowing, introduce such high spectral variance into the
data relative to the spectral contrasts between end members that the numerical
methods used in the end-member analysis become highly unstable.
Shadow effects obviously depend primarily on sun angle (Fig. 6). Although the
crown cover may be the same, lower sun angles clearly decrease image brightness.
Differences due to shadowing can be taken into account in end-member analysis,
provided the end-member values are recalculated for each temporal image and one or
more components labelled "shade" are added to the list. However, successful
application of end-member analysis still depends on an assumption of linear scaling
along cover gradients due to changes in sun position and sensor view angle. These
assumptions are erroneous in structured vegetation (e.g., vegetation with
discontinuous tree or shrub cover), limiting the application of such methods to general
synoptic estimates of changes in vegetation cover.
The alternative is to model vegetation cover directly as an assemblage of various
sizes and shapes of three-dimensional objects (trees, shrubs, grass tussocks, herbs,
etc.) scattered on a background that may be uniform or heterogeneous (Li and Strahler
1985, Jupp et al. 1986). The GO model may then be used to model the bidirectional
reflectance of the canopies. In this approach, effects due to shadowing on the overall
reflectance (or infrared temperature) from a scene become important, useful features
and the correlated interactions between shaded and sunlit components are built into the
analysis, although it now becomes nonlinear. The directional radiance of the
(0) (b) (e)
Figure 6. The same woodland plot comprising randomly located trees, showing
the extent of shadow at sun angles of (a) 15, (b) 30 and (c) 60 offzenith.
vegetation is then a mixture of four components (sunlit and shaded tree crowns, and
sunlit and shaded backgrounds) seen from a given viewing angle (Fig. 7). The areal
proportions of these four components, for given illumination and viewing directions
(which can be off-nadir), will be a function of the sizes, shapes, orientations and
placements of the objects (i.e., individual plants) within the scenes. The various plant
attributes needed to quantify a scene model are given in Figure 7 and clearly match the
field and vegetation model parameters shown in Figure 2.
Figure 7. The three-dimensional structure of the woodland is defined by S (mean

gap between trees), D (mean tree diameter), T (mean crown depth) and H (mean
height to midcrown). The reflectance components, as proportions used in the
simplest geometric-optical (GO) model, are C (sunlit crowns), T (shaded crowns),
G (sunlit background) and Z (shaded background).
91
A GO model is most appropriate to woodlands or vegetation in which the cover is

discontinuous, that is, where tree and shadowed background interactions account for a
large proportion of the variance in the image. The further advantage of these models is
that they are potentially invertible, capable of providing structural as well as cover
information. The invertibility of GO models was demonstrated in studies by Strahler
et al. (1988), Franklin and Strahler (1988) and Wu and Strahler (1993), in which tree
size and density were estimated from reflectance data. When size, shape and
orientation are fixed or characterised by distributions of known parameters, and the
object centres are randomly distributed, the proportions of the four components can be
estimated using the Boolean model of Serra (1982). This GO model is termed the
Boolean version (Strahler and Jupp 1991a,b, Li and Strahler 1992) and accounts for
the changes in proportions that occur with random overlapping objects as the density
of objects increases and can easily model scale effects and changing sun and view
directions. The GO aspect of the model implies that multiple scattering of radiation in
the vegetation layer is neglected. While visual evidence supports this, there are
wavelengths (particularly the near-infrared) where multiple scattering is very
significant (this has been recently addressed by Li et al. 1995).
In simulations in which size, shape and orientation are not fixed and the spatial
distribution of the objects ranges from regular, through random, to clumped, a more
empirical approach can be used. Values from simulated rasterized images can be used
to calculate the proportions of the components. The simulations can be highly realistic,
with computer-simulated trees and leaves, or simply extensions of the simpler
analytical models. One such simulation was used to extend the type of analytical
model used here to include wide ranges of spacing and crown size distributions. This
model, called IMSIM (Penridge 1991), was based on ray tracing (Foley et al. 1990).
The analytical and simulation versions of the simple GO model have been compared
for many random distributions of plants and were found to give similar results when
the assumptions of the analytical model were satisfied and for moderate departures
from them. The analytical model has provided very powerful results, especially in
terms of mechanisms and spatial variance, and of scaling effects of the models.
However, simulation and computer graphics play a very important role when the
mathematics becomes intractable (see, for example, Goel et al. 1991).
A spatially explicit vegetation dynamics model (RESCOMPj

A process-based vegetation model aims to describe the role of vegetation in energy
and water balances and in biogeochemical cycles occurring in the land surface layer.
In order to model vegetation dynamics, we must also consider the growth of
vegetation components and the environmental constraints on plant productivity. An
effective vegetation dynamics model can start with the current cover and generate a
future series of changes in land cover given inputs of light and water and a regime of
possible environmental and anthropogenic disturbances. Well-established examples
are the FORET group of models (Shugart et al. 1980, Shugart 1984).
The main requirements for linking a vegetation dynamics model with GO models
are that (i) the vegetation dynamics model contain ecological elements that provide for
mechanisms of structural change over time; and (ii) the output from the generated time
series of changes include the sizes, shapes and spatial arrangements of the plants as
well as foliage biomass.
Several vegetation dynamics models that include structure as a factor and satisfy (i)
and (ii) above are available. We found it convenient to use a spatially explicit
vegetation dynamics model available at CSIRO called RESCOMP (RESource
COMPetition, Penridge et al. 1987). Substantial field data collected over a 30-yr
period suggest that this model provides accurate trends in woody weed invasions and
woodland dynamics for the research site studied and that its incorporation of structure
at the individual tree level is a key to its success. The details of the model and its
ecological validation are not, however, specifically an issue for this chapter because it
was used simply to generate realistic input for the GO model.
In brief, RESCOMP is driven by water, light, nutrients, temperature and salinity,
and by competition between individuals. In particular, the spatial interaction
mechanism built into RESCOMP allows the occurrence of a range of interactions,
from competitive exclusion to facilitation (Walker et al. 1989, Walker and Dowling
1990). A variety of spatial and compositional scenarios can be simulated using various
combinations of the driving variables. Importantly, the vegetation structural output is
in the form of plant size, shape and spacing (as shown in Fig. 7). The output from
RESCOMP is therefore suitable to interface directly with the GO model. The plot
sizes normally used range from 400 m2 to 10,000 m2 , allowing estimation of spatial
variation, although this was not done for this chapter.
The RESCOMP model provides outputs in terms of structural and compositional
trends over a specific period. The GO model can be used to convert these data into
mean reflectance values, provided the colours (i.e., reflectances) of the primary
components are known for a realistic sequence of times of day and dates (as well as
sensor view angles, if appropriate). It is quite possible for the component reflectances
to respond, for example, to moisture availability in the root zones of different layers,
which is also an output of the vegetation dynamics model (as is done here by
estimating the varying proportions of green and dry grass). Examination of reflectance
trends can therefore indicate whether underlying environmental vegetation trends can
be detected.
The changes of main concern in the Murray-Darling Basin relate to plant responses
to salinization, grazing and fire. In the example given here, the focus is on woody weed
increases associated with overgrazing.
A simulated vegetation change scenario

A 40 x 40-m plot was selected from within a Eucalyptus populnea (poplar box)
woodland at Wycanna, Talwood, in the northern part of the Basin (Fig. 1) for the
simulation. The spatial location, as well as the leaf areas and mean leaf angles of trees,
shrubs, grass, litter and bare soil, were defined by field measurement. RESCOMP was
used to predict variation in leaf areas, litter and bare soil at 13-week intervals for a 10-
yr period (Fig. 8). The climatic data used were actual data for the area during the
period 1970-1980 and grass response to wet periods was measured by fraction of
green grass. The modelled trends shown in Figure 8 conform to vegetation responses
previously observed in the woodland studied. The essential features of the simulation
93
40....------------------------,
30
10
.. '
0L---'-52---'10-4--'5'-6--20..1-- ---'--0---'31-2--36.L..4--4..1-16---'46-6--'520
6 26
Time (Weeks)
Figure 8. Variation in the projected percentage of cover for trees, shrubs and
grasses over a lO-yr period as predicted using RESCOMP.
are relatively minor fluctuations in leaf area of trees, large variations in leaf areas and
greening of grass and a gradual increase in shrub cover. In this example, the variations
are mainly controlled by rainfall events.
Remaining questions include the following: will the trend in shrub cover be
detected by reflectance imagery and, if not, how can the trend be identified?
The Boolean version of the GO model was used to produce a trajectory at yearly
intervals for the woodland cover series shown in Figure 8. In addition, a trajectory was
produced for the same tree and shrub composition but with complete removal of grass,
as would be the case with heavy grazing. These trajectories are referred to as
"ungrazed" and "overgrazed" in the following discussion. The components and
signatures used are based on those reported in Jupp et aL 1986 (Table 1, Fig. 9).
The signatures in Table 1 were based on radiances obtained by regression methods
using field data from Wycanna in southern Queensland (Australia) and a Landsat MSS
image (Jupp et aL 1986). Although there are apparently 12 components in the mixing
model, the (nonlinear) GO model and redundancy between litter and soil reduces it to
five effective variables when soil and litter are combined. Such a model may, in
principle, be resolved by four channels of data. However, because of the complexity of
the mixing model even for five components and the lack of significant spectral
separation between end members, inversion of this model is quite difficult. The values
listed probably underestimate actual contrasts between crowns, between sunlit and
shaded components and between litter and soil and therefore provide a conservative
model for evaluating separation of effects. Figure 9 shows this redundancy as well as
illustrating how a simple two-component (sunlit background and tree plus shadow)
model is too limited to describe this situation, as is a model with only one "shade"
component. In the two-layer GO model used, within-crown volume foliage and its
contribution to the off-nadir BRDF were also modelled using crown openness, crown
foliage area index (FAI) and mean foliage angle, with stems and leaves not
differentiated.
Table 1. Component signatures used in the GO model to convert component

proportions to reflectance values.
Component Red NIR
Sunlit tree crown (sc) 16.1 24.0

Shaded tree crown (she) 6.1 14.5
Sunlit shrub crown (ss) 17.5 26.8
Shaded shrub crown (shs) 6.8 16.2
Sunlit green grass (sgr) 15.0 42.0
Shaded green grass (shgr) 9.0 36.0
Sunlit dry grass (sdr) 27.2 32.6
Shaded dry grass (shdr) 19.4 31.2
Sunlit litter (sl) . 30.7 30.3
Shaded litter (shl) 14.9 25.4
Sunlit soil (sg) 27.8 30.8
Shaded soil (shg) 14.9 25.4
GO Model Components
45,---------------------------,
.~
40 ..... / .........
.
35 ...........
~------
"--"-~._-------
20
15 . . . . . . - . - - . - - - - - - - - - - - -
10+-----,-----.-------,----,-----..,------1
5 10 15 20 25 30 35,
Red Reflectance (%)
Figure 9. Spectral components of the geometric-optical model. Sunlit (s) and

shaded (sh) trees (t), shrubs (s), green (gr) and dry (dr) grass, litter (I) and soil (g).
95
Despite the extra complexity of shadowing in Table 1 and Figure 9, the main
underlying trend in the composite reflectance trajectory (Fig. 10) is still due to the
decrease in the amount of visible sunlit soil (with signature RSg )' which decreases as
the composite of woody vegetation cover and associated shade (with signature R x)
increases. The ungrazed woodland departs from the trend because of periodic
increases in green vegetation in the background, which is visible in gaps through the
trees and woody vegetation. Figure 11 shows the time series of NDVI for the same
data. Although Figures 10 and 11 were generated by a model, many practical examples
also support the proposition that the use of current image processing methods can
successfully monitor these major effects in a woodland system, provided there is
effective atmospheric correction, the local variance is reduced and the actual cover
values are calibrated by field observations for the types of communities present. Green
grass and tree covers are also well separated spectrally so that (as illustrated in Fig. 10)
the responses of these factors can be separated over time.
G
(sunlit grass)
40
35
a;
c:
c:
P-
C)
a: /ungrazed woodland
Z 30 3 J/ (sunlit background) B
.~~~,
10 Il-a--' lI-
25 10...... "overgrazed woodland
X (composite tree/shrub)
15 20 25
Red Channel
Figure 10. Changes in percentage of reflectance values modelled from the

RESCOMP output using the Boolean version of the GO model.
What is unclear from Figure 10, even more unclear in Figure 11 and equally unclear
in real imagery is the structural nature of the cover changes. The increase in shrub
cover is almost indiscernible. Thus an almost identical cover trajectory can be
produced by trees or shrubs, mixtures of trees and shrubs and single- and multiple-
layered canopies, provided they all have the same total cover. This is so despite the fact
that these changes in structure provide important indicators of system change as well
as creating much of the variance found in remotely sensed images. Figures 12(a) and
12(b) illustrate this effect, for which the same trajectories were computed, assuming
all trees or all shrubs and the ungrazed green grass signal. Figure 12(a) shows the
NDVI plots and Figure 12(b) the albedo plot (in this case just the average of the red
and near-infrared reflectances). NDVI is not sensitive to structural changes, although
there is an albedo change because the lower shrubs show less cast shadow and have
lO-yr Cover Gradient

NDVl over the 10 years
0.42
032
:>
~
0.22
0.12
0.02
0 100 200 300 400 '00
TIme (Weeks)
...... ungrazed ..... overgrazed
Figure 11. NDVI of the'10-yr growth trajectory.
more sunlit background for the same overall crown cover. Over the 10 years, the
albedo shows a significant fall because the dark signature of the vegetation is being
modelled as well as the increasing shadowing. It is possible that in situations where
there is more spectral contrast between components the separation will be greater.
However, for the woodland being modelled, the combination of low spectral contrast
between vegetation components and significant variations in structure was the basis
for our interest. Figures 10, 11 and 12 also suggest that, despite the powerful addition
of phenology into the information base provided by NDVI composite data (Holben
1986), the single-value time series is not enough to take advantage of the well-resolved
vegetation and background effects observed here. A parallel albedo series seems to be
needed if some of the (simpler) effects are to be effectively monitored.
Extracting structural information from remotely sensed data

There are two main approaches to unraveling structural and compositional changes
using remotely sensed data. The first is to use spatial variance as information rather
than removing it as "noise." Figure 13 shows a theoretical calculation of the spatial
standard deviation in red reflectance for the cover at t lO (ungrazed) of Figure 10 (44%
cover) as a function of image pixel size. That is, it represents the expected standard
deviation in small patches of pixels (of the given size) in areas of the scene with this
type of cover, and depends (as described in GO models and modifications) on the
spectral contrast between components as well as on the spatial variation in pixel
composition. If the varying composition of the cover were due to trees only (6-m
crowns), the local spatial variation as a function of pixel size created by the discrete
nature of the data would be shown by the upper curve. If the composition were due to
shrubs only (1.05-m crowns), the standard deviation would be similar to the lower
curve (i.e., the variance relative to cover and pixel size is a potential indicator of crown
97
IO-yr Cover Gradient

NOVI fOf tlttS. shrubs and mixed
0.42
032
~
Z
02'
0.12
0.02 +----,------,------,----,------j
o 100 '00 300
nme(Weeks)
........ composilc ..... Irc(s -6- shrubs
Figure 12(a). NDVI trajectory f~r all trees, aU shrubs and the modelled mixture.
IO-yr Cover Gradient

Albedo over the 10 years
",-----------------------,
26
"
24
23
22+----,.----,------,.-----,-------j
o '00 '00 300 400 sao
nme(Weeks)
...... composile-e- trees -&- shrubs
Figure 12(b). Albedo trajectory for aU trees, aU shrubs and the modelled mixture.
size). If the variance observed in practice were due to the same effects as that for the
theoretical model used here, in the case of the Wycanna woodland data we would
conclude that changes in cover of the smaller crowns (shrubs) had occurred and that a
shrub invasion had thereby been successfully detected.
The effect of object size is the basis for the inversion of structure developed by
Strahler and Li (1981), Li and Strahler (1985, 1986), Strahler et al. (1988), Franklin
and Strahler (1988) and Wu and Strahler (1993). However, in recent attempts to
operationalise these methods from tests at the stand level to regional mapping, success
cover: 44%
crown SIlO 6 m
c:
.2
Oi 2
.~
o
'E
'"
---_
'g
in'" 1
..........
....
..... . ... .
crown SI'8 1 05 m;Jrr- - - - - - - _
ol-------,30:':------5-'-o---------!70--~
Pixel SIze (m)
Figure 13. Variance estimated between pixels at increasing pixel size for trees
(crown size 6 m) and shrubs (crown size 1.05 m) with a vertical view. Plots are for
trees and shrubs alone, and the woodland comprising mixtures of trees and
shrubs at tx (13 weeks) and txo (520 weeks). .
has been mixed. Woodcock et aL (1994) reported an extensive trial of regional

mapping in which basal area and changes in basal area were very well mapped through
their relationship with crown area index, but tree size estimates based on the Li-
Strahler inversion model were poor. In an examination of the reasons for the mixed
success, Woodcock et aL (1995) concluded that the problem arises from poor
relationships between image variance and field measurements of tree size even where
the local mean of the variable c defined above correlates well with local cover. It seems
that success may depend on isolating the component of the image variance due to the
varying mixing of crowns in pixels. In highly variable areas this may not be an easy
task. Indeed, Jasinski and Eagleson (1990) have successfully used the image variance
to help establish cover. In their work, the structural aspect of the variance Var(k sb ) is
ignored relative to its relationship with spectral contrast (Rsb-Rxl In a study of the
ability of remote sensing to map old-growth forests, Nel et aL (1994) showed how gap
size variance is much higher in old-growth forests and is mirrored in the image texture.
They also pointed out that visual interpretation makes use of this to a much greater
degree than current image processing methods to define forest type. Perhaps the issue
is that the gappiness directly creates the variance rather than object size. For the
Boolean model, mean-gap size (dg ) is closely related to object size and spacing such
that larger objects for the same cover create a much more diverse gap size distribution
as well as a larger mean-gap size. Using calculations from Chen et aL (1993), it may
be shown for the Boolean model that
1tD
dg ="4c (19)
99
so that, for a given cover (i.e., given C), the mean size of the gaps is directly
proportional to D.
In practice, however, the distribution of gap sizes between crowns may not relate so
closely to tree size. Non-Poisson distributions and gaps created by old trees dying or
disturbances such as fire (Shugart 1984) may not create the same effect as gaps left by
default because trees grow randomly. In Chen et at. (1993) it was shown how texture
statistics based on pixels were limited in their information about the underlying
structure (in that case a simple disk model) by both the object-to-pixel-size ratio and
the ratio of mean-gap size to pixel size, with the second often being the more
important. The main problem may well be that we have not yet learned to model or use
this kind of texture in image processing.
In contrast to the difficulty of using texture at the subpixel-to-pixel scale, use of
texture variations as a function of pixel size has provided a means to assess scaling and
natural scales of interaction for many years (Woodcock and Strahler 1987, Woodcock
et al. 1996, Collins et al. 1995). The patch structure of images between about 10 pixels
in linear measure and less than about one-tenth of the linear extent of the image is
itself an aspect of the land surface that is open to monitoring and ecological
interpretation. Texture measures such as the variogram (Woodcock et at. 1988a,b,
Collins et al. 1995) seem adequate for these measurements. Currently, therefore,
image spatial structure at the intermediate scale and spectral information on resource
use and biomass provide effective division of operational areas of remote sensing.
A second approach to obtaining essentially structural information is available using
airborne data (Barnsley et at. 1990) and may become fully accessible from space with
future satellite sensors. It involves the relationships between structure and changes in
image intensity relative to sun and sensor view positions. That is, it makes use of the
image BRDF. When the sun and view angles are the same, the image is at maximum
brightness. This geometry, which is well known as a problem for high sun
photography, creates the image hotspot. As the angle between the sun and view
positions changes, different proportions of sunlit and shaded vegetation and
background are sensed leading to a highly characteristic BRDF that depends on
structure. The shadowing effects created by changing sun position can be used to
separate covers to some degree, even for nadir viewing (Walker et at. 1986, Jupp et at.
1986). However, BRDF effects due to sun position changes alone are difficult to
separate from other temporal effects. Of particular interest, therefore, are methods that
allow observations of the same area from different view angles to be made at the same
or very similar times. Such methods can use the hotspot effect as a tool to monitor
changes in the structure of cover.
The hotspot effect takes its name from the strong brightening that occurs when the
observer and sun are aligned so that no shadow is visible - except possibly the
shadow of the observer. However, the way this brightening decreases away from the
hotspot point and the degree of darkening observed when the sun and observer are
opposite are all significant indicators of canopy structure. In the example used here,
greater angular change is required to make the shadow beneath a shrub visible than is
necessary for the shadow beneath a tree. The underlying geometry of the hotspot effect
and some models that describe it for the type of scene model used here have been
described in Strahler and Jupp (1991a,b) and Jupp and Strahler (1991). However, the
hotspot effect and BRDF models that incorporate it have been studied widely in many
areas of science. BRDF and hotspot effects for soils were established by Hapke (1968,
1986) and those for atmospheric hotspot phenomena by Greenler (1980). For
vegetation, various approximate and empirical hotspot functions have been developed
(Suits 1972, Kuusk 1985, Gerstl et al. 1986, Nilson and Kuusk 1989) and incorporated
into inversion models (Liang and Strahler 1993). Fully explicit geometrically based
models for the hotspot effect for tree crowns were presented by Nilson (1977), Jupp et
al. (1986) and Strahler and Jupp (1991a,b). The use of explicit geometric optics for
leaf canopies was outlined in Jupp and Strahler (1991) and extended to include leaf
shape and stem effects by Qin (1993). The benefit of the geometric approach is
illustrated in the development of relationships between hotspot size and shape, and of
scale-free ratios such as crown width to height and diameter to depth for trees and leaf
diameter to height and leaf length to width for canopies. These essentially structural
ratios affect BRDF even when pixel size is very large relative to crown and leaf size.
The theoretical utility of the hotspot effect for assessing structure is illustrated in
Figure 14. Here the cover for each case shown is the same (44-%) and the sun position
is at 52 zenith angle. Away from the hotspot point, reflectance for the shrub case
0
shows much slower fall away and its broad hotspot width is due to the higher ratio of
shrub crown diameter to shrub height. The trajectories for t l and t lO in Figure 14 are
between the shrub and tree case. The t lO trajectory's greater proximity to the pure shrub
line indicates an increase in shrub cover. A characteristic difference between the trees
and shrubs is that the ratio of crown diameter to height (D/h) is about 1.0 for the shrubs
and about 0.3 for the trees. Together, crown size and the crown-diameter-to-height
ratio can separate the tree and shrub combinations we have considered. In other
situations where crown shape varies, the BRDF also contains information on the shape
ratio of crown diameter to crown depth (DfT). It may be shown mathematically that for
the simple three-dimensional Boolean model we have used, the combination of local
spatial variance (to obtain mean crown size) and the hotspot effect (diameter-to-height
ratio, Jupp and Strahler 1991) can provide information of special value for forest
resource assessment. The challenge now is to develop the methods within routine land
cover image processing considering a higher level of complexity and variability of
covers and structures than that used in simple models.
An example of how aircraft data may be used to monitor structure through the
hotspot effect is given in Figure 15. Here, by flying with the scan of a Daedalus DATM
1268 along the principal plane of the sun or at 90 to the principal plane, it is possible
0
to find the average response for a large patch of open forest (Jupp et a1. 1994). The
resolution of the crown-
width-to-height ratio was very high, with the (model) response to variations in the
ratio plotted on the graph. Flying scanners, videos and cameras mounted specifically
to include the hotspot effect can therefore provide useful structural information if
atmospheric and land surface effects are correctly separated. In some cases, end users
are more interested in compensating for the hotspot effect than in using it. Fitting the
model shown in Figure 15 provides an effective means for doing this as well.
101
22
shrubs
20
--~--
50 30 10 0 10 30 50 70
V,ew Angle
Figure J4. Simulated reflectance values in the red band at different view angles
and for a sun zenith angle of 52. Plots are for trees and shrubs alone, and the
woodland comprising mixtures of trees, shrubs at t] (13 weeks) and ~o (520
weeks).
cu rye for the

principal plane
curv for 90 10
principal plane
-50 -30 -10 10 30 50

5canAngle
Figure J5. BRDF along principal plane and 90 to principal plane. Modelled data
(with varying structure ratios) compared with DATM scanner data.
Discussion
The work presented here suggests a pathway to monitor structural changes in
woodland and open-forest systems. In many parts of the world, these systems are
being degraded by fuewood harvesting and domestic animal grazing and are facing
increasing salinization and erosion. The sequence of steps in the proposed analysis,
using complementary vegetation and reflectance models with a time series of
atmospherically and geometrically corrected images, is as follows: (i) predict
temporal cover changes using a process-driven vegetation dynamics model and
available field and meteorological data to provide expected changes; (ii) use an
effective remote sensing measurement model to convert predicted cover changes to
reflectance values, image variance and image BRDF; and (iii) compare the modelled
information (the expected outcome) with satellite-derived values (the observed
outcome).
A similar system for well-managed forest areas has been explored recently by
Nilson and Peterson (1994). However, in that case the comparison could be made with
growth information collected by forest managers or derived from a cross section of
stands in different stages of growth. Interpretation of remotely sensed data from more
extensive natural open-forest and woodland systems in terms relevant to monitoring
changes in vegetation structure can be accomplished only by combining vegetation
dynamics and measurement models for remotely sensed data that use the same land
cover components. Complementary modelling approaches of this type have been
presented and used in combination to describe how remote sensing can be used to
monitor changes in composition and structure in a specific Australian woodland
system.
Although current image processing methods and images available from satellites
and aircraft provide a useful and essential baseline for operational monitoring,
improved model-based methods are needed to effectively monitor open-forest,
woodland, shrubland and grassland systems like those in the Murray-Darling Basin,
where changes in structure are dominant indicators of land degradation or
rehabilitation. We have described how image and vegetation dynamics models based
on the same description of the land cover can be combined and how image spatial
variance and changes in reflectance with sun and observer positional effects may be
used to extract structural information characteristic of the changing woodland system.
A number of algorithms and effective, atmospherically corrected standard products
are being developed for the EOS/MODIS series that either take advantage of this
opportunity or provide data with which it could be developed (Running et al. 1994).
However, it appears that the use of textural analysis in determining structural change
could be improved. Because a primary problem is the accurate unmixing of pixel
components to estimate the spatial covariances between them, the opportunity of using
high spectral resolution for this step (such as that available in the EOS/MODIS data)
should be explored. Another current approach involves combining BRDF and angular
variance (Jupp and Woodcock 1992, Jupp et al. 1994). However, advances in handling
image texture and variance probably will be necessary to maximize their potential as
tools for mapping and monitoring image structure. A particularly interesting area
103
where the use of texture is yet to be fully exploited and where considerable potential
for structural information exists is radar data. GO models have recently been
developed for radar scattering (UIaby et al. 1990, Wang et al. 1993) and studies linking
structural parameters to radar backscatter are increasing (Pierce et al. 1994). Radar
data do not have BRDF unless operated in bistatic mode, so texture may be even more
important in the successful use of modern multipolarisation, multilook C-, L- and P-
band systems than it is for visible region data (see also Peterson 1996).
Changes derived from a monitoring system can be either expected or unexpected. If
exploratory methods such as classification are used, all changes are unexpected.
Change maps derived in this way are complex records of environmental and other
changes. One way to incorporate a degree of expectation into monitoring is through a
reference trajectory based on a vegetation dynamics model combined with effective
remote sensing measurement models. The time since the last correction can provide
rainfall and sunlight data. The evolution may be tracked to the present and into the
future using conditional simulations of rain and sunlight. The remote sensing
measurement model applied to this series provides the normal base against which to
measure change with remote sen'sing - provided atmospheric correction and BRDF
effects are well handled. Such model-based change analysis (rather than exploratory
methods) would be the most effective use of the wealth of data soon to be derived from
the MODIS series (Running et al. 1994). Even if acceptance of the models used is not
unanimous, as long as they serve as explicit benchmarks and can identify
environmental change outside the envelope of normal expectation they will provide an
effective standard product.
Conclusions
Many studies support the proposition that satellite or aircraft remote sensing can
effectively and operationally monitor the time series of albedo and greenness in terms
related to total biomass, LAI, general resource use and health at the regional scale.
Structure, while a highly significant component of the ecology of the world's forests
- as well as the primary source of information for economic and sustainability
monitoring - is less easily obtained from integrated time series based on large pixels
such as those used in NDVI compositing. For the current types of remote sensing
instrumentation, texture and BRDF provide two basic opportunities for improved
structural discrimination.
In this chapter, we have described how a vegetation dynamics model may be used
to simulate the comparative development of the structure of a land surface and how
some current BRDF models may be used to estimate the remotely sensed baseline
implied. To use this approach effectively for sensing changes in structure, further
development of analytical methods is necessary. At the general scale, BDRF seems the
most promising measure of land surface structure. Mathematically derived scaling
effects are clearly fundamental to the behaviour of the data and make it a clear
candidate for regional monitoring. At the high-resolution end (pixels preferably less
than 10 m and definitely no more than 20 m, for example), texture has not been
successful as a tool with which to map tree size at a regional scale (Woodcock et al.
1995), although it seems to help provide useful cover and basal area estimates.
Applying the "empty bottle criterion" (according to which, despite what has been said
about the wines, the empty bottle is probably the best indicator) to texture and BRDF,
we should consider (i) that texture methods have not been used with field survey data
or quantitative air photo interpretation in the Li and Strahler style, and (ii) that, despite
30 years of opportunity, the hotspot effect has not been used by air photo interpreters
to assess structure - but texture has.
The conclusion could be that texture may be ideal for visual structural assessment
from high-resolution data but that we have not yet found a way to handle it
mathematically, computationally or both. This conclusion is strengthened by
observing the shadow of an aircraft crossing forested areas with a range of structures.
Hotspot width and shape certainly change, but textural changes are even more obvious
to the eye. Clearly, there is both opportunity and justification for more research.
Finally, the use of models or standard sites as measures of normal community
development, and the development of measures for changes in cover, condition and
structure away from what is determined as expected or desired, provide a model for
monitoring that promises to be more effective than exploratory or inductive methods.
The work by Nilson and Peterson (1994) represents a decisive step towards
implementation of such a model for managed forests. Use of vegetation dynamics
models (where such a database is not available) in the open-forest and woodland areas
of the world, matched by a remote sensing data stream like that planned for
EOSIMODIS, may provide the means to implement it globally.
Acknowledgments - The CSIRO Division of Water Resources and the CSIRO Office
of Space Science Applications (COSSA) have supported this strategic research and
enabled airborne applications to Australian woodlands to contribute to the
development of BRDF studies. Elizabeth McDonald contributed unpublished data,
advice and valuable assistance to this research and this chapter. Wally Wu
(Biosystems, Texas A & M University) and Les Penridge (Pen Info) provided
technical and theoretical help in the RESCOMP simulations.
References
Adams, J.B., Smith, M.O. and Gillespie, A.R. 1989. Simple models for complex natural
surfaces: A strategy for the hyperspectral era of remote sensing. - In: Proceedings of the
IGARSS'89 Symposium: Remote Sensing, an Economic Tool for the Nineties, July 10-14,
1989, Vancouver, B.C., Canada. IEEE, New York, pp. 16-21.
Barnsley, M., Morris, K. and Reid, A. 1990. Preliminary analysis of a multiple view angle
image data set. - In: Proceedings of the NERC 1989 Airborne Remote Sensing Campaign
Symposium, December 18-19, 1990, Keyworth, UK, pp. 49-68.
Boardman, J. 1990. Inversion of high spectral resolution data. - In: Proceedings of the SPIE,
Imaging Spectroscopy and Terrestrial Environment, vol. 1298, April 16-20, 1990, Orlando,
FL. International Society of Optical Engineers, Bellingham, WA, pp. 222-223.
Carnahan, J.A. 1976. Natural vegetation (map with accompanying booklet and commentary).
- In: Atlas of Australian Resources, 2d ser., Department of Natural Resources, Canberra,
Australia.
105
Chen, R., Jupp, D.L.B., Woodcock, C.E. and Strahler, A.H. 1993. Nonlinear estimation of
scene parameters from digital images using zero-hit run-length statistics. - IEEE Trans.
Collins, J.B., Woodcock, C.E. and Jupp, D.L.B. 1995. Spatial dependence and nested
hierarchical scene models. - In: Proceedings of the ASPRS/ACSM, vol. 3, February 1995,
Charlotte, NC, pp. 535-544.
Cross, A.M., Settle, J.J., Drake, N.A. and Paivinen, R.T.W. 1991. Subpixel measurement of
tropical forest cover using AVHRR data. - Int. J. Rem. Sens. 12: 1119-1129.
Foley, J.D., Van Dam, A., Feiner, S. and Hughes, J. 1990. Computer Graphics. Principles and
Practice. - Addison-Wesley Publishing Co., Reading, MA. 1174 pp.
Franklin, J. and Strahler, A.H. 1988. Invertible canopy reflectance modeling of vegetation
structure in semi-arid woodland. - IEEE Trans. Geosci. Rem. Sens. 26: 809-825.
Gerstl, S.A.W., Simmer, C. and Powers, B.J. 1986. The canopy hot-spot as crop identifier. - In:
Proceedings of Symposium on Remote Sensing for Resources Development and
Environmental Management, August 1986, Enschede, The Netherlands, pp. 261-263.
Goel, N.S., Rozehnal, I. and Thompson, R.L. 1991. A computer graphics based model for
scattering from objects of arbitrary shapes in the optical region. - Rem. Sens. Environ. 36:
133-144.
Greenler, R. 1980. Rainbows, Halos, and Glories. - Cambridge University Press, Cambridge,
UK. 195 pp.
Grime, J.P. 1973. Competition and diversity in herbaceous vegetation - A reply. - Nature 242:
344-347.
Hapke, B. 1968. On the particle size distribution of lunar soil. - Planet. Space Sci. 16:
101-110.
Hapke, B. 1986. Bidirectional reflectance spectroscopy IV: The extinction coefficient and the
opposition effect. - Icarus 67: 264-280.
Heller, R.c. and Ulliman, J.J. 1992. Forest resource assessments. - In: Colwell, R.N. (ed).
Manual of Remote Sensing. American Society of Photogrammetry, Falls Church, VA, pp.
2229-2324.
Holben, B.N. 1986. Characteristics of maximum-value composite images from temporal
AVHRR data. - Int. 1. Rem. Sens. 7: 1417-1434.
Jasinski, M.P. and Eagleson, P.S. 1990. Estimation of subpixel vegetation using red-infrared
scattergrams. - IEEE Trans. Geosci. Rem. Sens. GE28: 253-267.
Jupp, D.L.B. and Strahler, A.H. 1991. A hotspot model for leaf canopies. - Rem. Sens.
Environ. 38: 193-210.
Jupp, D.L.B. and Woodcock, C. 1992. Variance in directional radiance of open canopies. - In:
Proceedings of the IGARSS'92 Symposium: International Space Year: Space Remote
Sensing, May 26-29,1992, Clear Lake City, TX. IEEE, New York, pp. 1490-1492.
Jupp, D.L.B., Walker, J. and Penridge, L.K. 1986. Interpretation of vegetation structure in
Landsat MSS imagery: A case study in disturbed semi-arid eucalypt woodland. Part 2.
Model based analysis. - J. Environ. Manage. 23: 35-57.
Jupp, D.L.B., Strahler, A.H. and Woodcock, c.E. 1988. Autocorrelation and regularization in
digital images. 1. Basic theory. - IEEE Trans. Geosci. Rem. Sens. 26: 463-73.
Jupp, D.L.B., Strahler, A.H. and Woodcock, C.E. 1989. Autocorrelation and regularization in
digital images. 2. Simple image models. - IEEE Trans. Geosci. Rem. Sens. 27: 247-258.
Jupp, D.L.B., McDonald, E.R., Harrison, B.A., Li, X., Strahler, A.H. and Woodcock, C.E.
1994. Prospects for mapping canopy structure using Geometric-Optical models. - In:
Proceedings of the Seventh Australian Remote Sensing Conference, Melbourne, Australia,
pp. 263-270.
Key, J.R. 1994. The area coverage of geophysical fields as a function of field of view. - Rem.
Kuchler, A.w. and Zonneveld, I.S. (eds). 1988. Vegetation mapping. - In: Lieth, H. (series ed).
Handbook of Vegetation Science. Kluwer Academic Publishers, Dordrecht, The
Netherlands, 635 pp.
Kuusk, A. 1985. The hot spot effect of a uniform vegetative cover. - Sov. J. Rem. Sens. 3(4):
645-658.
Li, X. and StraWer, A.H. 1985. Geometric-optical modeling of a conifer forest canopy. - IEEE
Trans. Geosci. Rem. Sens. GE23(5): 705-721.
Li, X. and Strahler, A. H. 1986. Geometric-optical bidirectional reflectance modeling of a
coniferous forest canopy. - IEEE Trans. Geosci. Rem. Sens. GE24(6): 906-919.
Li, X. and Strahler, A.H. 1992. Geometric-optical bidirectional reflectance modeling of the
discrete-crown vegetation canopy: Effect of crown shape and mutual shadowing. - IEEE
Trans. Geosci. Rem. Sens. 30: 276-292.
Li, X., Strahler, A.H. and Woodcock, C.E. 1995. A hybrid geometric optical-radiative transfer
approach for modeling albedo and directional reflectance of discontinuous canopies. -
IEEE Trans. Geosci. Rem. Sens. 33: 466-480.
Liang, S. and Strahler, A.H. 1993. An analytic BRDF model of canopy radiative transfer and
its inversion. - IEEE Trans. Geosci. Rem. Sens. 31: 1081-1092.
McDonald, E.R. 1993. "Estimating vegetation structure from remotely sensed images." -
Masters thesis, University of Canberra, Australia, November 1993. 112 pp.
Milne, A. 1961. Definition of competition among annuals. - In: Multhorpe, F.L. (ed).
Mechanisms in Biological Competition. Cambridge University Press, London, pp. 40-161.
Myneni, R.B. and Ross, J. (eds). 1991. Photon-Vegetation Interactions: Applications in Optical
Remote Sensing and Plant Ecology. - Springer-Verlag, Heidelberg. 565 pp.
Nel, E.M., Wessman, C.A. and Veblen, T.T. 1994. Digital and visual analysis of thematic
mapper imagery for differentiating old growth from younger spruce-fir stands. - Rem.
Nilson, T. 1977. A theory of radiation penetration into non-homogeneous plant canopies. - In:
The Penetration of Solar Radiation into Plant Canopies (in Russian). Estonian Academy of
Sciences, Institute of Physics and Astronomy, Tartu, Estonia, pp. 5-70.
Nilson, T. and Kuusk, A. 1989. A reflectance model for the homogeneous plant canopy and its
inversion. - Rem. Sens. Environ. 27: 157-167.
Nilson, T. and Peterson, U. 1991. A forest canopy reflectance model and a test case. - Rem.
driving factors. - Rem. Sens. Environ. 48: 319-331.
Nilson, T. and Ross, J. 1996. Modeling radiative transfer through forest canopies: Implications
for canopy photosynthesis and remote sensing. - In: Gholz, H.L., Nakane, K. and Shimoda,
H. (eds). The Use of Remote Sensing in the Modeling of Forest Productivity. Kluwer
Academic Publishers, Dordrecht, The Netherlands, pp. 23-60.
Pech, R.P., Davis, A.w. and Graetz, R.D. 1986. Reflectance modelling and the derivation of
vegetation indices for an Australian semi-arid shrubland. - Int. J. Rem. Sens. 7: 389-412.
Penridge, L.K. 1984. FOLIG: A computer program to generate foliage profiles. - CSIRO,
Division of Water and Land Resources, Technical Memorandum TM 84/29, Canberra,
Australia. 20 pp.
Penridge, L.K. 1986. LPOINT: A computer program to simulate a continuum of two-
dimensional point patterns from regular, through Poisson to clumped. - CSIRO, Division of
Water and Land Resources, Technical Memorandum TM86/12, Canberra, Australia. 15 pp.
107
Penridge, L.K. 1987. FOL-PROF: A fortran-77 package for the generation of foliage profiles.
Part 2. Programmer Manual. - CSIRO, Division of Water Resources, Technical
Memorandum TM 87/10, Canberra, Australia. 75 pp.
Penridge, L.K. 1991. IMSIM: Shaded three-dimensional images for woodland scenes using a
two-dimensional buffer approach. - CSIRO, Division of Water Resources, Technical
Penridge, L.K., Walker, J., Sharpe, PJ.H., Spence, D., Wu, H. and Zou, G. 1987. RESCOMP:
A resource competition model to simulate the dynamics of vegetation cover. - CSIRO,
Division of Water and Land Resources, Technical Memorandum TM 87/5, Canberra,
Australia. 178 pp.
Peterson, D.L. 1996. Forest structure and productivity along the Oregon transect. - In: Gholz,
H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the Modeling of
173-218.
Pierce, L.L., Walker, J., Dowling, T.!., McVicar, T.R., Hatton, T.T., Running, S.W and
Coughlin, J.L. 1993. Ecohydrological changes in the Murray-Darling Basin. 3. A
simulation of regional hydrological changes. - J. Appl. Ecol. 30: 283-294.
Pierce, L.E., Sarabandi, K. and Ulaby, F.T. 1994. Application of an artificial neural net work in
canopy scattering inversion. - lilt. J. Rem. Sens. 15: 3263-3270.
Pickup, G. and Foran, B.D. 1987. The use of spectral and spatial variability to monitor cover
change on inert landscapes. - Rem. Sens. Environ. 23: 351-63.
Qin, W 1993. Modelling bidirectional reflectance of multicomponent vegetation canopies. -
Running, S.W., Nemani, R.R., Peterson, D.L., Band, L.E., Potts, D.F., Pierce, L.L. and
Spanner, M.A. 1989. Mapping regional forest evapotranspiration and photosynthesis by
coupling satellite data with ecosystem simulation. - Ecology 70: 1090-1101.
Running, S.W, Justice, S., Salomonson, v., Hall, D., Barker, J., Kaufman, Y., Strahler, A.H.,
Huerte, A., Muller, J-P., Vanderbilt, v., Wan, Z., Teillet, P. and Carneggie, D. 1994.
Terrestrial remote sensing science and algorithms planned for EOS/MODIS. - Int. J. Rem.
Sens. 15: 3587-3620.
Serra, J. 1982. Image Analysis and Mathematical Morphology. - Academic Press, London. 610
pp.
Shugart, H.H. 1984. A Theory of Forest Dynamics: The Ecological Implications of Forest
Succession Models. - Springer-Verlag, New York. 278 pp.
Shugart, H.H., West, D.C. and Emanual, WR. 1980. Patterns and dynamics of forests: An
application of simulation models. - In: West, D.C., Shugart, RH. and Botkin, D.B. (eds).
Forest Succession. Springer-Verlag, New York, pp. 74-94.
Specht, R.L. 1970. Vegetation. - In: Cheeper, G.W (ed). The Australian Environment. 4th ed.
CSIRO and Melbourne University Press, Melbourne, Australia, pp. 44-67.
Strahler, A.H. and Li, X. 1981. An invertible coniferous forest canopy reflectance model. - In:
Proceedings of the 15th International Symposium on Remote Sensing of the Environment,
Ann Arbor, MI, pp. 1237-1244.
Strahler, A.H. and Jupp, D.L.B. 1991a. Geometric-optical modeling of forests as scenes
composed of three-dimensional discrete objects. - In: Myneni, R.B. and Ross, J. (eds).
Photon-Vegetation Interactions: Applications in Optical Remote Sensing and Plant
Ecology. Springer-Verlag, Heidelberg, pp. 415-440.
Strahler, A.H. and Jupp, D.L.B. 1991 b. Modeling bidirectional reflectance of forests and
woodlands using Boolean models and geometric optics. - Rem. Sens. Environ. 34:
153-166.
Strahler, A.H., Wu, Y. and Franklin, J. 1988. Remote estimation of tree size and density from
satellite imagery by inversion of a geometric-optical canopy model. - In: Proceedings of
the 22nd International Symposium on Remote Sensing of the Environment, October 20-26,
1988, Abidjan, Ivory Coast.
Suits, G.H. 1972. The cause of azimuthal variations in directional reflectance. - Rem. Sens.
Environ. 2: 175-182.
Ulaby, FT., Sarabandi, K., McDonald, K., Whitt, M. and Dobson, M.e. 1990. Michigan
microwave canopy scattering model. - Int. J. Rem. Sens. 11: 1223-1253.
Walker, J. and Penridge, L.K. 1987. FOL-PROF: A fortran-77 package for the generation of
foliage profiles. Part 1. User Manual. - CSIRO, Division of Water Resources, Technical
Walker, J. and Dowling, T.!. 1990. A non-stochastic, physiologically based model of plant
invasion using Ecological Field Theory. - Plant Protec. Quart. 5: 85-93.
Walker, J. and Hopkins, M.S. 1990. Vegetation. - In: McDonald, R.e., Isbell, R.F., Speight,
J.G., Walker, J. and Hopkins, M.S. (eds). Australian Soil and Land Survey Field Handbook.
2d ed. Inkata Press, Melbourne, Australia, pp. 67-89.
Walker, J., Jupp, D.L.B., Penridge, L.K. and Tian, G. 1986. Interpretation of vegetation
structure in Landsat MSS imagery: A case study in disturbed semi-arid eucalypt woodland.
Part 1. Field data analysis. - J. Environ. Manage. 23: 19-33.
Walker, J., Sharpe, P.I.H., Penridge, L.K. and Wu, H. 1989. Ecological field theory: The
concept and field tests. - Vegetatio 83: 81-95.
Walker, J., Bullen, F. and Williams, B.G. 1993. Hydro-ecological changes in the Murray-
Darling Basin. Part 1. The number of trees cleared over two centuries. - J. Appl. Ecol. 30:
265-273.
Wang, Y, Day, J. and Sun, G. 1993. Santa Barbara microwave backscattering model for
woodlands. - Int. J. Rem. Sens. 14: 1477-1493.
Williams, R.I. 1955. Vegetation regions of Australia. - In: Atlas of Australian Resources, 1st
ser., Department of National Development, Canberra, Australia.
Wood, J.G. 1937. The Vegetation of South Australia. - Government Printer, Adelaide,
Australia.
Woodcock, e.E. and Strahler, A.H. 1987. The factor of scale in remote sensing. - Rem. Sens.
Environ. 21: 311-322.
Woodcock, e.E., Strahler, A.H. and Jupp, DLB. 1988a. The use of variograms in remote
sensing. 1. Scene models and simulated images. - Rem. Sens. Environ. 25: 323-48.
Woodcock, e.E., Strahler, A.H. and Jupp, D.L.B. 1988b. The use of variograms in remote
sensing. 2. Real digital images. - Rem. Sens. Environ. 25: 349-80.
Woodcock, e.E., Collins, J.B., Gopal, S., Jakabhazy, VD., Li, X., Macomber, S., Ryherd, S.,
Harward, V.I., Levitan, J., Wu, Y and Warbington, R. 1994. Mapping forest vegetation
using Landsat TM imagery and a canopy reflectance model. - Rem. Sens. Environ. 50:
240-254.
Woodcock, e.E., Collins, J.B., Jakabhazy, VD., Li, X. and Macomber, S. 1995. Inversion of
the Li-Strahler model for mapping forest vegetation: What works and what doesn't. - In:
Proceedings of the ASPRSIACSM, vol. 2, February 1995, Charlotte, NC, pp. 299-308.
Woodcock, e.E., Collins, J.B. and Jupp, DLB. 1996. Scaling remote sensing models. - In:
Van Gardingen, P., Foody, G. and Curran, P. (eds). Scaling Up. Society of Experimental
Biology, Cambridge University Press, Cambridge, UK (in press).
Wu., Y and Strahler, A.H. 1993. Remote estimation of crown size, stand density and foliage
biomass on the Oregon transect. - Ecol. Appl. 4: 299-312.
FIVE
Integrating Remotely Sensed Spatial Heterogeneity with a

Three-dimensional Forest Succession Model
John F. Weishampel, Robert G. Knox, K. Jon Ranson, Darrel L. Williams and James
A. Smith
Weishampel, J.F., Knox, R.G., Ranson, KJ., Williams, D.L. and Smith, J.A. 1996.
Integrating remotely sensed spatial heterogeneity with a three-dimensional forest
succession model. - In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of
Remote Sensing in the Modeling of Forest Productivity. Kluwer Acad. Publ.,
To address questions concerning the effects of naturally occurring or

anthropogenically induced spatial heterogeneity on forested stands and landscapes, it
is necessary to couple spatially explicit simulators with georeferenced data. In this
study, we quantified heterogeneity within a forested scene from central Maine by
using edge detection algorithms designed for synthetic aperture radar (SAR)
imagery. A spatially extended forest gap model that incorporates a three-dimensional
solar radiation routine was used to simulate forest stand productivity as a function of
edge orientation. The model results were then scaled up to the landscape based on the
extent of edge and edge orientations detected in the SAR scene. The estimated
increase in productivity, as a function of light-mediated edge effects, averaged :0:26%
to :0:4% at the pixel (12.1 x 12.l-m) and scene (:o:3600-ha) scales, respectively, over a
120-yr period after the edge was created.
l.F Weishampel, Department of Biology, P.O. Box 2368, University of Central

Florida, Orlando, FL 32816-2368, USA. R.G. Knox, K.l. Ranson and D.L. Williams,
Biospheric Sciences Branch, Code 923, NASA Goddard Space Flight Center,
Greenbelt, MD 20771, USA. J.A. Smith, Laboratory for Terrestrial Physics, Code 920,
NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA.
109
Introduction
In ecological systems, local heterogeneity can playa comparably central role in
regulating stability and diversity. For instance, fine-scale variation in resource
availability may contribute to global changes such as desertification (Schlesinger et
al. 1990). Heterogeneity among patches is believed to generally increase species
coexistence and, at larger scales, have a stabilizing effect on regional dynamics
(Czaran and Bartha 1992). Occurrences of high resource availability that support
fugitive species (and potentially allow nutrient losses) are more likely when the
competitively dominant species are unable to track the spatiotemporal pattern in
variation of resources. In spatial models of plant populations, species can partition
heterogeneity in resource levels caused by local demographic stochasticity (Pacala
and Tilman 1994). A similar mechanism explains why shade-intolerant tree species
persist in the absence of large-scale disturbance in the quasi-spatial context of
traditional gap model simulations (Shugart 1984, Botkin 1993, Clark 1993). It has
even been proposed that the scales of heterogeneity present regulate the distribution
of body sizes or organisms found in that environment (Holling 1992). In these and
many other ecological' contexts, heterogeneity is itself a "signal" with important
consequences, rather than "noise" potentially interfering with accurate estimates of
coarse-scale properties.
Relevance of spatial heterogeneity

Ecosystem and community ecologists classically viewed natural and managed
landscapes as supporting patchworks of vegetation. Differences among climates,
soils, hydrological conditions and disturbance histories were thought to produce
repeatable differences among vegetation patches (McIntosh 1985). This view
provides a rationale for vegetation classification schemes and for the generation of
thematic maps showing the spatial disposition of areas of different vegetative cover
types. Conceptually, such map units show areas expected to have similar mean values
for properties of interest, such as rates of uptake or exchange with the atmo-, hydro-
and lithospheres. Classical remote sensing followed this lead, emphasizing the
correct classification of patches and the generation of thematic views.
The study of heterogeneity is not novel or even particularly new, although Fisher
(1925) noted that "... until comparatively recent times, the vast majority of workers
[in statistics] appear to have had no other aim than to ascertain aggregate, or average,
values. The variation itself was not the object of study, but viewed instead as a
troublesome circumstance which detracted from the value of the average." Although
quantitative remote sensing estimates patch properties as continuous functions of
measurements at a sensor rather than through correlation with a classification, it
tends to stress the recovery of mean values for relatively homogeneous areas. To a
first approximation, vegetation gradients and successional sequences have also been
seen as context-free realizations of nonspatial processes that can be understood at the
patch level (e.g., Major 1951, Whittaker 1953). Thus, most attention has focused on
the effects of average values (i.e., first moments) for patches rather than on the effects
of heterogeneity within patches (i.e., higher moments), the structure of interpatch
111
boundaries or the spatial arrangement of landscape elements. However, ignoring

spatial heterogeneity produces vast uncertainties when we attempt to scale our
understanding of dynamics of relatively homogeneous patches (i.e., where most
ecological or remote sensing studies have been concentrated) to dynamics of
complex landscapes (Mooney and Chapin 1994). Given the increasing awareness of
the ecological importance of spatial heterogeneity (Kolasa and Pickett 1991,
Caldwell and Pearcy 1994) and transitional zones or ecotones (Holland et al. 1991,
Hansen and di Castri 1992, Risser 1993), it has become critical to include the effects
of particular spatial processes when modeling the mechanisms that define forest
productivity.
Spatial heterogeneity in forested systems

A fundamental structural unit in a forested ecosystem is the forest edge, which
functions as a border between heterogeneous areas. While canopy gaps have received
much attention in the forestry literature (Platt and Strong 1989, Denslow and Spies
1990) as the spatial unit of forest dynamics, each gap is fuzzily defined by its
surrounding edge. As such, the presence of within-stand edges may reflect natural
fine-scale patch cycling (i.e., gap phase dynamics; 0.01-0.20 ha) forming the
boundaries separating gaps and non-gaps as well as all the in-between gradations of
patch types (e.g., seedling, sapling, pole and mature tree or stand developmental
stages). Remotely sensed textural properties of a forest canopy that detect the
presence of these fine-scale patterns may therefore correspond to the age or
successional state of the stand (Cohen et al. 1990, Weishampel et al. 1992, Wallin et
al. 1996). Changes in textural properties due to changes in gap or edge frequency
reflect forest turnover; for instance, a recently measured increase in canopy
heterogeneity in a tropical rainforest has been attributed to fertilization effects from
elevated CO 2 levels (Phillips and Gentry 1994).
At typically coarser resolutions, forest edges may represent borders between
naturally occurring, relatively slowly changing landscape elements (e.g., those
resulting from edaphic or topographic differences) as well as more rapidly occurring
natural (e.g., fire, windthrow) and typically more rectilinear anthropogenic (e.g.,
harvesting, controlled burning) disturbances. Boundaries between vegetation patches
from both natural and managed landscapes represent distinct regions that differ
abiotically and biotically from the interiors of the two intersecting patches. Such
small-scale ecotones have been compared to cell membranes protecting the interior
from the outside environment and regulating the flow of energy and matter into and
out of the interior (Wiens et al. 1985). Because forest productivity is strongly
influenced by microclimatic forcing functions (e.g., soil temperature, relative
humidity, solar radiation) that have been found to be more variable at a forest-clearcut
edge than within either the forest or the clearcut (Wales 1967, Williams-Linera 1990,
Chen et al. 1993, Matlack 1993, Young and Mitchell 1994), edge productivity tends
to differ from that of the patch interior (Ghiselin 1977). Although increases in solar
radiation at an edge may promote increases in net primary production (NPP), this
may be countered by corresponding increases in dessication. Hence, edge formation
may act to the benefit or the detriment of certain species (e.g., shade-intolerant
xerophiles or mesophiles, respectively), possibly yielding higher or lower species

diversity as well as more differential successional patterns in productivity than are
found in intact or managed areas (Soule and Kohm 1989).
Empirical studies have found that forest-clearcut edges exhibit differences in
stem densities, basal areas and mortality rates and permit the persistence of shade-
intolerant species (Wales 1972, Williams-Linera 1990, Chen et al. 1993, Esseen
1994). The extent of edge effects adjacent to a clearcut is site dependent and has been
estimated from 20 m in an eastern North American deciduous forest (Wales 1972) to
over 240 m for an old-growth North American Pacific Northwest forest (Chen et aI.
1992). This variation in the extent of the zone of edge influence is a product of
multiple factors, such as the latitude, slope and aspect influencing the solar regime,
the surrounding climate, the method by which the forest boundary is maintained and
the forest's species composition, physiognomy and successional stage at the time of
edge formation, as well as the presence or absence of herbivores (Ranney 1977). As
such, edges, like gaps (Liebermann et al. 1989), cannot be simply defined, since they
portray a diverse array of forest structural units that represent a variety of processes
covering a range of spatiotemporal scales.
Remote sensing of spatial heterogeneity
Modeling spatial processes of forests

Remote sensing is the primary tool for the synoptic analysis of forests at the stand-to-
landscape level. It allows researchers to address such general questions as (i) what
elements are present, (ii) what spatial arrangements these elements have and (iii)
what their temporal dynamics are (Quattrochi and Pelletier 1991). Because remote
sensing affords the ability to classify forests based on species composition (e.g.,
evergreen vs deciduous, Pastor and Broschart 1990, Lee and Nakane 1996), canopy
chemistry (Wessman et al. 1988, Martin and Aber 1996), structural attributes (e.g.,
LA!, density, biomass, Running et aI. 1986, Wu and Strahler 1994, Ranson and Sun
1994) and phenological differences (e.g., seasonal greenup and leaf drop, and freeze
and thaw cycles, Spanner et al. 1994, Way et al. 1994) and to detect and monitor
natural (Hall et al. 1991) and human-induced (Kachi et al. 1986) forest dynamics and
disturbances (Miller and Williams 1978), it readily permits the detection and
monitoring of spatial heterogeneity. Thus, in a limited sense, remote sensing has been
applied to characterize spatial patterns and processes of vegetation such as the
dynamics of biome boundaries (Tucker et al. 1985), the distribution of elements
within a forested landscape (e.g., Syrjanen et al. 1994) or the canopy properties
within a forest stand (Cohen et al. 1990).
At the finest scale, remotely sensed heterogeneity is simply pixel-to-pixel
variance or tonal contrast (i.e., texture). Textural measurements from a window of
pixels (Musick and Grover 1991) have been used as a method to enhance forest
classification (Ulaby et al. 1986, Cohen and Spies 1992, Ranson and Sun 1993).
These responses typically vary with the resolution of either the sensor or the analysis
(Cohen et al. 1990, Weishampel et al. 1994). The pixel-to-pixel variance may be
113
spatially organized, forming clumped, random or regular patterns. Clumped patterns

(i.e., patches) may exhibit either different scales of organization or anisotropy (Kux
and Henebry 1994). With clumped patterns, there are edges. If edges are considered
distinct entities, they are typically examples of anisotropic clumped distributions.
Remotely sensed edges are boundaries between areas of different tonal contrast.
Thus, in the presence of scene texture, edges can be defined. Because edges, like
coastlines and textural patterns, exhibit fractal qualities, the detection and delineation
of edges is highly scale dependent (Johnston et al. 1992).
In the coming decades, the primary global environmental changes will involve
land use patterns (Mooney and Chapin 1994). Of the two basic applications of remote
sensing to forest ecology, (i) to monitor inaccessible or spatially extensive regions,
and (ii) to detect spatial patterns and processes, less effort has concentrated on the
latter (Iverson et al. 1989). While certain changes in heterogeneity may not be
obvious via conventional remote sensing techniques and may require the
development of new sensors (e.g., laser altimeters, Harding et al. 1994) or
information extraction technique,s (e.g., wavelet analysis, Bradshaw and Spies 1992,
Deren and Juliang 1994; wombling, Fortin 1994), many of these changes are
presently detectable via established remote sensing methodology (Miller and
Williams 1978), from the Pacific Northwest patchwork of blockcuts (Vitousek 1994)
to the fishbone-shaped cuts along the Brazilian Amazon (Moran et al. 1994).
Although numerous studies have quantified the extent of regional deforestation (e.g.,
Woodwell et al. 1983, Nelson et al. 1987, Skole and Tucker 1993), not until recently
have studies related remotely sensed deforestation to other ecologically relevant,
quantitatively measured spatial patterns (Ghiselin 1977, O'Neill et al. 1988, Ranson
and Williams 1992), such as patch size and amount of edge environment (Johnston
and Bonde 1989, Ranson and Sun 1993, Syrjanen et al. 1994, Ranson and Smith in
press) or to the dynamics of such landscape indices (Krummel et al. 1987, LaGro
1991, Spies et al. 1994).
Inherent in the use of remote sensing to detect spatial patterns for extrapolation
of ecological processes is the need to develop parameters that relate environmental
patterns to biophysical processes (Krummel 1986, Wessman 1992). Thus,
quantifying landscape patterns is only part of the effort, and process-based models
that explicitly incorporate spatial dynamics are necessary (Steffen et al. 1992, Wallin
et al. 1996).
One type of model, termed the forest "gap model" (Shugart et al. 1992, Botkin
1993), can be adapted to such a use (Levine et al. 1993, Walker 1994). In this case,
each tree growing on a plot whose size relates to the zone of influence of a canopy
dominant (or the "gap" formed when it dies) is modeled as a unique entity. The
models generally run at annual time steps to simulate vegetation dynamics at decadal
to millennial temporal and 0.01 to 0.2-ha spatial scales. The dynamic basis of a gap
model is the ongoing feedback between the prevailing environmental conditions
(e.g., light, temperature, soil moisture and nutrient availability), which modify the
biotic responses (e.g., establishment, growth, regeneration and mortality), and the
subsequent stand structure which, in turn, modifies the environmental conditions.
Gap models have been used primarily to simulate forest succession in a limited
area (i.e., ::; 0.2 hal and have been spatially explicit in the vertical dimension (i.e.,
taller plants shade shorter plants below them). However, more recent models embed
patch-scale processes into a larger landscape (Smith and Urban 1988, Urban et al.
1991, Busing 1991, Pacala et al. 1993). By using a spatially explicit method of direct
extrapolation (King 1991), which juxtaposes a series of common crown-sized (0.01-
hal plots into a grid, a forested region can be viewed as a hierarchical system (Urban
et al. 1987) composed of interacting, grid-based components (i.e., gap-scale units).
Although within-plot horizontal homogeneity is maintained, such a model permits
lateral, plot-to-plot interactions (i.e., a large tree can shade trees on adjacent or
nearby plots as well as trees beneath it on the same plot). Thus, gap-scale patterns can
be propagated across a gridded landscape that may not be inferred from the average
behavior of individual units (Smith and Urban 1988).
Increasing the spatial extent of forest models increases the logistical difficulties
of collecting data for model calibration, parameterization and confirmation. Remote
sensing may provide a means to obtain this spatial data (Running et al. 1989, Hall et
al. 1991, Levine et al. 1993). An initial comparison of "scaled-up" gap model output
with remotely sensed data used a transect version of a gap model parameterized for
the H.J. Andrews Experimental Forest in Oregon to generate structural properties of
young, mature and old-growth Douglas-fir forests (Weishampel et al. 1992).
Operating at a scale of kilometers, the model was able to capture temporal aspects of
landscape dynamics associated with canopy texture of Pacific Northwest forests
found in remotely sensed imagery. However, this was successful, in part, because
environmental heterogeneity such as that resulting from different climatic, soil-
moisture, topographic or disturbance regimes was unusually lacking at the scale of
analysis. In this chapter, we emphasize environmental heterogeneity, utilizing an
adaptation of the ZELIG model (Weishampel and Urban in press) to simulate local
edge effects induced by harvesting practices. These spatial effects are scaled to a
remotely sensed landscape that represents a patchwork of successional stages and
species assemblages.
Quantifying heterogeneity with synthetic aperture radar
The Forest Ecosystems Dynamics study site

The Forest Ecosystems Dynamics (FED) project at NASA's Goddard Space Flight
Center (GSFC) is primarily concerned with the application of remote sensing to the
modeling of patterns and processes of forested systems (Levine et al. 1993, Prihodko
et al. 1994). This project emphasizes the detection of features at local to regional
spatial scales (10 1 to 105 m) and the exploration of mechanisms involving temporal
4
scales ranging from physiological processes to long-term ecological processes (10.
3
to 10 years). Thus far, the FED project has concentrated on studying forests at the
southern boreal/northern hardwood transition zone (e.g., Voyageurs National Park,
Minnesota, and Howland, Maine). Data for this chapter are from International
Paper's Northern Experimental Forest (NEF) near Howland, Maine (45 15' N, 68
45' W). The area comprises ==7000 ha consisting of hemlock-spruce-fir (i.e., primarily
115
Tsuga canadensis, Picea rubens, P. mariana and Abies balsamea), aspen-birch (i.e.,
primarily Populus tremuloides and P. grandidentata and Betula papyrifera and B.
allegeniensis) and hemlock-hardwood mixtures (Levine et al. 1994). The flat to
gently rolling terrain of the NEF with a maximum elevation change of 135 mover 10
km minimizes the need to topographically correct imagery or adjust for hillslope
effects. As an experimental forest, the NEF has a variety of forest spatial structures
(e.g., block cuts and strip cuts of various ages and orientations on a range of soil
types). Additionally, due to the region's glacial history, soil drainage classes within a
small area may vary greatly, from excessively drained to poorly drained (Levine et al.
1994), leading to a mosaic of stands of varying species composition and density. An
area occupying approximately half of the NEF (i.e., ",3600 ha) was used for this study
(Fig. 1a).
Edge detedion in radar images at the NEF

Synthetic Aperture Radar (SAR) imagery over the NEF was acquired from the
airborne SAR (AIRSAR; Held et,al. 1988) flown after leafout in late June 1993. The
SAR records radar return as a measure of backscatter from the surface. The Jet
Propulsion Laboratory (JPL) AIRSAR has three wavelength bands (-6 cm = C-band,
-24 cm = L-band and -68 em = P-band) and transmits and receives radar signals with
four combinations of horizontal (H) and vertical (V) polarizations (HH, VV, HV,
VH). For the following analyses, we used P-band, HH polarization (PHH) data. The
longer wavelength does not saturate with higher levels of biomass as quickly as the
shorter wavelengths (Ranson and Sun 1994) and the PHH data have been found to
possess the highest single channel separability among dense regenerating patches and
other cover type categories (i.e., forest openings, thinned and older cuts; Ranson and
Sun 1993).
Because microwave sensors are more sensitive to variation in total above-ground
biomass and moisture content than optical sensors, which are more sensitive to green
leaf biomass (Ustin et al. 1991), their ability to detect and characterize canopy gaps
and forest patches of different ages has been deemed superior to that of optical
sensors (Waring et al. in press). Additionally, the double-bounce backscatter (i.e.,
trunk to ground to sensor or ground to trunk to sensor) present at a forest edge should
yield a distinct signal. Hence, at comparable spatial resolutions, SAR should provide
better detection of structural edges of forests than optical sensors. Edge detection in
SAR images, however, is not straightforward due to the multiplicative nature of
"speckle," an artifact of the coherent addition of backscatter (Adair and Guidon
1990). Therefore, conventional edge detection techniques that involve determining
the difference of pixel values in a window may yield false edges (Touzi et al. 1988,
Lopes et al. 1993). Other techniques that filter out speckle and delineate boundaries
have proven more successful (Adair and Guidon 1990).
To remove low-contrast edges, a problem in delineating patches in an L-band
(23-cm) Seasat SAR agricultural scene (Adair and Guidon 1990), low-backscatter
responses, (i.e., < 40th percentile) were grouped together. This group corresponded to
low 10 kg m ) above-ground biomass patches in the NEF scene (Ranson and Sun
2
1994). We employed the ratio-of-averages operator (Touzi et al. 1988) to initially

b)
c) d)
Figure 1. (a) AIRSAR PHH image over the NEF study area, with lighter
responses indicating higher backscatter; (b) binary image depicting the edges
detected by the 5 x 5 ratio operator; (c) remaining edges following application of
the morphological thinning, pruning and filtering operations; (d) overlay of the
detected edges with the SAR image.
define the forest edges. Given a square window comprising an odd number of pixels
on a side (i.e., 3 x 3, 5 x 5, etc.), this operator averages the pixels at the two edges on
opposite sides of the window (i.e., Xl and x2) for the four edge orientations (i.e.,
vertical, horizontal and the two obliques). The value of the difference, 1- xl /x 2 ,
where xl < x2 ' of the orientation with the lowest Xl: X2 ratio is calculated and then
replaces the value of the center pixel. This operator was applied to each center pixel
within the window across the image. Figure Ib represents a binary version of the new
117
image with the edge threshold set at the 80th percentile after a 5 x 5 ratio operator was
applied. The choice of the chosen percentile is arbitrary and subsequently is adjusted
with the application of the following morphological filters.
First, a morphological thinning operation (Dougherty 1992) was applied to
reduce the thickness of the edge to a single pixel where feasible. We used the "hit-or-
1 2 3 4
miss" or "Serra" transform to accomplish this. The four matrices E , E , E , E :
[!ill[!ill[TIQ][Q]]J
[Q]]J DDITD [!ill DDJD
were each paired with the matrix F:
[!ill
DDJD
2 3 4
to form the four structuring pairs B " B , B , B for the sequential thinning of image S
(i.e., Fig. I b). The image S was thinned as follows:
S' = S B', S2 = SI B2, S3 = S2 B3, S4 = S3 B 4, S5 = S4 B' ...
The expression S B I equals:
S - [ n {S + e' : e I E 1}] n [ n {S" + f : f F} ]
1 1
where SC is the complement of S, and e and f are the 180 0 rotations of E and F,
1
respectively, around their origins. Thus, E probed the outside of S, while F probed
the inside; if there was a match, the extent of S was reduced.
Next, a sequential pruning operation (Dougherty 1992) was used to excise
extruding one-pixel-wide segments oriented in different directions, one direction at a
time, by cycling through the matrices:
000 000 o1 0 00 0
o1 0 110 o1 0 o1 1
o1 0 000 00 0 o0 0
00 0 1 0 0 00 1 00 0
o1 0 o1 0 o1 0 o1 0
1 0 0 o0 0 00 0 o0 1
where the center pixel indicates the origin. For this pruning routine, if a 3 x 3-pixel
window of the image matched one of the above matrices, the center pixel was
eliminated. Although the thinning operation stops when the detected edge is one pixel
wide, the pruning operation was used sparingly because nonspurious, disconnected
linear features can be systematically eliminated by this process. Finally, a cleaning
filter was used to delete any "edge" that fit within a 3 x 3 matrix. The results of the
morphological thinning, pruning and cleaning operations on the binary edge image
are depicted in Figure lc.
Quantification of edge extent and orientation

Given the product of the edge detection procedures, the question becomes: What is
the extent of the edge structure in this forested area? Also, because edge orientation
can determine the magnitude and the spatial extent of the environmental forcing
functions (Wales 1967, 1972, Ranney 1977, Chen et al. 1992, 1993, Matlack 1993,
1994), how are the edges distributed in space? Using the techniques presented above,
we found the extent of edges to be primarily dependent on the arbitrarily chosen
threshold values. Changing the initial 40th percentile segmentation based on the
backscatter values, or the 80th percentile based on the strength of the edge signal,
could produce vast differences in the amount of edge and magnitude of tonal contrast
detected. Thus, the choice of threshold values depends on the objectives of the study.
This study was concerned with high-strength edges that represent sharp boundaries
between regions of high (> 10 kg m,2) and low biomass 10 kg m'2). Of the total
pixels in the 500 x 500 scene, nearly a tenth represented edge pixels (i.e., 23,813),
which means that half of the initially detected pixels (i.e., > 80th percentile) using the
5 x 5 ratio operator Were eliminated by the morphological filters. Because the
resolution of the AIRSAR pixel was 12.1 m, these edge pixels, if the edge can be
considered to be one pixel deep, comprised ",350 ha. Or, in linear terms, the edges
extended ",290 km throughout the scene.
To determine the orientation of each edge pixel, the following routine was used.
Each edge pixel was given a value (i.e., x) of 0; the value of the non-edge pixels was
the PHH backscatter (Fig. Id). A 7 x 7-pixel matrix composed of the relative -->
Cartesian coordinates was centered on each edge pixel (i.e., xu,u)' The vector E,
where:
--> 3 3
E=I. I.
j=,3 ;=,3
(i e x.,j e x.)l49
I.J I.J
was used to describe the direction and magnitude of each edge pixel. The angle (8)
-->
associated with each E value was rectified to true north. Because certain edges may
reflect regions where the morphological thinning did not reduce--> the thickness of the
edge to a single pixel, magnitudes of the standard deviations of E < 0.5 were ignored.
This further reduced the -->
number of edge pixels by about -->
one tenth. Figure 2 shows a
plot of one thousand I E I values> 0.5. By grouping the E values by orientation into
S-facing, E-facing, N-facing and W-facing aspects (i.e., 45 :::; 8 < 135, 135 :::; 8 <
225, 225 :::; 8 < 315, 315 :::; 8 < 45, respectively), the total spatial orientation of
the edges was calculated and found to be distributed in a fairly isotropic manner (Fig.
2).
Simulating the effects of heterogeneity on productivity
Model description and parameterization for the NEF

A three-dimensional adaptation of the ZELIG model (Weishampel and Urban in
press) was selected to simulate the southern transitional boreal forest zone found at
119
W +-0 -11---+ E
1\
I
o
] 3
z: 3
~a g c
a
Figure 2. Standard deviates of the first thousand I E I values> 0.5. The inset
~
shows the total sum of I E I values > 0.5 when grouped into the four cardinal
directions.
the NEF site. Like the original version (Smith and Urban 1988), this model operates
on a square grid of gap-sized (i.e., 10 x lO-m) plots. However, this model
incorporates the canopy structure used by Leemans and Prentice (1987) and the solar
geometry of later versions of ZELIG (Urban et al. 1991) adjusted for three-
dimensional space (Fig. 3). To implement this model, site characteristics (e.g., soil
water field capacity and wilting point, soil fertility and monthly values of
temperature, precipitation and radiation) and autecological parameters (e.g., tree age,
height and diameter maxima, growth and seedling establishment rates and light,
water, fertility and temperature requirements) were derived from empirical data and
published sources (e.g., Pastor and Post 1985) for the species of central Maine
(Levine at al. 1994). Although the soils in the study area are highly variable, only
parameters for a mesic soil (Knox et aI., unpublished manuscript) were used.
For this version of ZELIG, plot-to-plot interaction was strictly light-mediated;
other spatial processes, such as seed dispersal, were not considered. The radiation
environment at the NEF was predicted to consist of :=:55% direct-beam and 0:=45%
diffuse radiation, based on an algorithm that estimates radiation as a function of
latitude, slope, aspect and cloudiness (Bonan 1989). Given that NEF is a northern
latitude forest, the majority of direct-beam radiation is derived from the south (Fig.
Figure 3. Schematic of how trees are spatially depicted in a 5 x 5-grid plot

version of three-dimensional ZELIG (with each plot =10 x 10 m). Layers of
individual crowns may overlap. There is horizontal homogeneity at the plot
level.
a) b)
N N
0.15 100
NE NE
NW~5
50
5
E W E
SW SE
S S
Figure 4. (a) Distribution of proportion of direct-beam radiation to total

incoming radiation and (b) the average shadow length (m) around a 20-m-tall
tree at 45 intervals for the NEF.
121
4a). Over the course of the growing season, the characteristic sun angles associated
with each of these facets generate, on average, shadow lengths for a 20-m-tall tree as
shown in Figure 4b. The east-west direction has a lower average sun angle and
therefore casts a longer shadow. Diffuse radiation is determined by sampling an
isotropic sky for each 45 direction at several designated angles, whereas direct-beam
radiation impinging on a plot is a function of the estimated average solar incidence
angle for each of the eight primary compass directions. Radiation is attenuated as it
passes through the canopy according to Beer's Law and is calculated for each canopy
cell above a plot up to the height of the tallest tree at 1-m intervals.
Using the three-dimensional version of ZELIG (in a manner similar to that used
with a two-dimensional version of this model, as described by Weishampel et al.
1992), three-dimensional textural patterns were generated for the NEF site (Fig. 5).
In these results, there was an increase in texture as differences in plot-to-plot growth
rates and mortality (canopy gaps) increased over time. By 200 years, this increase
will, for the most part, level off. Textural dynamics for other plot-to-plot components,
such as the basal area of needl~leaf or broadleaf species, can also be derived.
However, since no underlying environmental heterogeneity was modeled, plot-to-
plot spatial variation was probably largely random, at least at scales greater than two
plots (Weishampel and Urban in press).
Modeling edge-related effects

The development of three-dimensional models permits the simulation of spatial
effects of landcover change on the structure and function of ecosystems. To address
questions concerning the effects of edge-associated heterogeneity on forest
productivity, a forest was simulated for a period of time, then a portion of the stand
was "harvested" to introduce an edge. The remaining forest was then "grown" under
the new ambient conditions. In a somewhat similar fashion, Ranney et al. (1981) used
a gap model to explore how forest edges undergo different successional dynamics
than and influence the species composition of the interior of forest islands in
Wisconsin. However, their model did not simulate forest edges and interiors as part
of a spatial continuum; thus, the extent of the edge influence could not be simulated.
With the three-dimensional gap model, light-mediated effects of different cut
orientations and the timing of harvesting on edge structure were simulated.
To model the effects of solar radiation on cut orientation, three scenarios were
used: cuts produced forest edges that faced north, south and east (Fig. 6). Because
thermal and evaporation regimes, normally associated with radiation loads, were not
modeled, east- and west-facing edge effects would be identical. To reduce the effects
of the ZELIG grid boundary, 15 transects of 15 cells each embedded in a 25 x 25-grid
plot run were considered in the analysis. In these simulations, cut regions were not
permitted to regenerate. In separate model runs, to mimic the effects of the timing of
harvesting, south-facing edges were induced at 0,40,80 and 120 years, representing
times before and after the transition from broadleaf to needleleaf species at 60 years
(Fig. 7). At the NEF, common broadleaf species are largely shade-intolerant and
needleleaf species are largely shade-tolerant. Simulations of each of the different
harvesting practices were run 15 times and averaged to reduce effects of model
a) 40-yr
10 1;1
III III
so - II;
.,..
if'
.. E .. e
'"
~ ,!
~~
'OJ
10
c) 120-yr
10 10
III I 1
-Ill
so .,..
... ""'"'.
III
"-
~ ~
'OJ
10
Figure 5. Simulated above-ground biomass textural dynamics for NEF from a 2S

x 2S-grid plot, three-dimensional ZELIG landscape.
stochasticity found in the regeneration, mortality and weather subroutines. For

illustration, we present here detailed results for edges created at 80 years.
Ten years after the 80-yr harvest, there was a significant edge effect for certain
cut orientations (Fig, 8), The east-facing edge plot exhibited a significantly higher
ranking for changes in stem density and biomass. The south-facing edge plot
exhibited a significantly higher ranking only for changes in stem density. The north-
facing edge plot exhibited no significantly higher ranking, The fact that the north-
facing edge revealed no significant difference was expected, since it is exposed only
to increases in diffuse radiation. However, edge effects on north-facing edges have
been observed in forests of the eastern United States (Wales 1972, Matlack 1994),
Significant edge effects were limited to the ZELIG plot adjacent to the edge, This
lack of light-mediated influence further into the intact forest suggests that the model
needs to be refined, possibly to incorporate additional microclimatic conditions such
as evaporative and thermal loadings associated with increased solar radiation.
When edge plots were compared with plots in a control, uncut ZELIG
simulation, they exhibited higher levels of above-ground biomass increase for the
123
a) N cut
41"" '.,,"dW.
Clearcut
I'"
Edg
If '!u
It
Il;, !
Fore t
',,,j! ",-"
I N
t
250-m or 25-ZEUG pIoII
b) S cut c) E cut
;IIJ
"
~ ;.~
l~
Figure 6. Different cutting regime/edge orientations as applied to three-

dimensional ZELIG simulations.
simulated 120-yr period following the cut, regardless of cut orientation (Fig. 9). Ten
years after the 80-yr cut, the average east- and north-facing increases were nearly
double the increase on the uncut plots. By 200 years, the average biomass increase
from the south-facing edge plots exceeded the increase on uncut plots by nearly a
third. While density for the uncut plots showed a decline from year 80, the edge plots
showed an initial increase in stem density as a result of the cutting. Stem density was
greatest for east-facing cuts followed by south-facing cuts. Approximately 40 years
after the cut, stem density decreased on the edge plots of both south- and north-facing
cuts and exceeded the stem density decline of the uncut plots. The stem density on the
edge plots of the east-facing cuts remained higher than that of the uncut plots for the
entire 120-yr period following the cut. This may be explained by the low sun
incidence angle, which allowed sunlight to reach the ground surface at the east-facing
edge and permitted the continued establishment of seedlings. The vast majority of
these seedlings, however, never received sufficient light to mature.
30
Needleleaf
1 ...............
...
25 ~ Broadleaf
. .. . -
("
~.----. . . .....
!
';:20
1
: 15
/ ..
~
~
.
ell
10
o +H
Vear
Figure 7. Simulated basal area trajectories for needle- and broadleaf species at
the NEF. Arrows repre~ent times when harvesting regimes were applied.
Edge
a)
6.5
5.5
5
~
C
<0
a: 45
Q)
())
6.5
b) N-tacing -
~ Solacing -<. -
Q)
-
> E-lac1ng .~
,,'
5.5
<"i .' !\
"
;.-
~
" ,.-
4.5
2 3 5 6 8 9 10
Cell Position
Figure 8. Average rankings of uncut transect cells for changes in (a) stem
density and (b) above-ground biomass at year 90, 10 years after an edge was
induced. Horizontal dotted lines represent 95 % confidence intervals based on a
null randomization.
125
a)
200
b) No cut
N-facing
S-facing
E-facing
80 100 '20 140 160 '80 200

Year
Figure 9. Average differences in (a) stem density and (b) above-ground biomass
for edge plots from the average of 80-yr plots from the uncut and differently
oriented cut scenarios.
To observe whether there was relaxation in the simulated edge effect, as well as
whether the timing of the cut influenced subsequent edge structure, edges were next
induced at 40-yr intervals in succession. In this case, all the simulated cuts in the edge
plots exhibited an increase in stem density shortly after the cut when compared with
control plots (Fig. 10). The higher density with the O-yr cut persisted longer, i.e., ==80
years, than that with the later cuts, whose higher density levels lasted ==30 years. Also,
fluctuations in density were not as great with the O-yr cut as with the later cuts. The
edge effect was not reduced with any of the cuts in the simulated time period. Most
simulated edge plots revealed higher levels of biomass than the control plots shortly
after the cut. However, the edge plots for the cut at 120 years required ==60 years to
exceed the biomass in the control plots. Again, the edge plots for the O-yr cut did not
exhibit as large an effect as found with the later cuts, and there was no indication of
reduced edge effect over time. More than likely, if the harvested region were
permitted to regenerate, the edge effect would attenuate more quickly.
a)
II
11
\ U
_---\~... n
.:'>-. -r:......
_j "U..:,\
Q' [] '..,.\
n, U
."j
Cul@Oyr -
b)
Cut@40 yr -0(>-' n
~ 2 Cul@80yr -0 - u
0 Cut@120yr - []
iii 1.5 D .&)..
E~
8> E 1
c c>
s::.
'" .l<
~OS
U
CI>
.1
""to
0 n
f
Qj
a: .(l.S
'~
I
1
20 40 GO 80 100 120 140 160 180 200
Year
Figure lO.Average differences in (a) stem density and (b) above-ground biomass
for south-facing edge plots since the time of cutting minus differences from the
uncut simulations for the same time periods.
Influence of edge on productivity at the NEF

To put the edge-related effects into perspective, we combined the products of the remote
sensing and modeling efforts. Although if given sufficient computer resources, the
model could run at the scale (i.e., resolution and extent) of the SAR image (Schwarz et
al. 1994), for this exercise we simply extrapolated the model predictions averaged from
the edge plots for the 15 x IS-grid plot simulations to the measured edge-orientation
classes from the scene. Hence, underlying these estimations are assumptions derived
from the model and the edge detection procedures. Some of the more gross assumptions
were that (i) edge effects were only light mediated, making east- and west-facing edges
equal; (ii) edge extent was only one plot deep; (iii) only mesic soils were present; (iv)
low-biomass areas represented clearcuts that occurred when the forest was 80 years old;
and (v) there was no regeneration in a cut area. Nonetheless, this approach provides a
mechanistic basis for the estimation of heterogeneity-related productivity effects that
are also being examined with new ground data.
The effects of edge-related heterogeneity on forest productivity were highly
scale dependent (Fig. 11). At the plot or pixel scale, when orientations were
considered separately (as shown in Figure 9), edge effects were fairly dramatic,
127
0' 40.0
C --0-- PloVPixel Scale
!
~ 30.0
----+-- Scene Scale
t.>
.E
u
~ 20.0
.~
=======-:~:...----_----
CD 10.0
Cl
~ 0.0.1
10 30 50 70 90 110
Years Since Harvesting
Figure 11. Edge-associated increases in percentage of above-ground biomass

productivity after cutting at 80, years for plot/pixel (i.e., 10 x 10-mlll.1 x 12.1-m)
and scene (i.e., 500 x 500 AIRSAR-pixel) scales.
ranging from ",,20% lower to 90% higher levels of productivity over a 120-yr span;
the overall average was ",,26% higher levels of productivity. At the scale of the SAR
scene, when the extent and orientation of edge pixels were considered, edge-related
productivity accounted for 7.5% to 4.9% higher levels from 10 to 110 years after
cutting, respectively. This increase, though substantially lower than the plot-scale
increase, still represents a significant level of enhanced productivity at the scene
scale. The scale differences were not surprising, since the low-biomass regions of the
NEF represent islands in a high-biomass matrix. As forests become more dissected
and practices such as stripcutting become more prevalent, edge-related influences
will undoubtedly play a more prominent role in the overall productivity of the
landscape.
Conclusions
Although spatial heterogeneity can be readily estimated from remotely sensed imagery,
this study attempted to further integrate it with a spatially explicit forest model. Thus,
the influence of environmental heterogeneity was not averaged out in the scaling
procedure. For the AIRSAR scene at the NEF, edge-related effects on productivity
were clearly substantial. This suggests that productivity in other regions undergoing
dramatic changes in land use patterns, such as the Brazilian Amazon, where harvesting
practices seemingly maximize the edge-to-forest interior ratio, or in the Pacific
Northwest, where the more checkerboard pattern of cutting has been used, may be
similarly affected. It remains to be seen whether the predicted productivity effects from
this study can be observed on the ground as well as in remotely sensed imagery.
The FED project is currently developing a modeling platform (Levine et aL
1993) that permits the interface of specialized, spatially explicit models of soil
(Bidlake et al. 1992), thermal (Smith and Goltz 1994) and forest physiology models
(Friend et al. 1993). Thus, the next step is to interactively link forest structural
dynamics with variation in the radiation, soil moisture and thermal environments.
These simulation results are to be compared with ground data from edges of different
ages and orientations from the NEF site.
Acknowledgments - This study was supported by a National Research Council (NRC)

Research Associateship and by funding for the Forest Ecosystems Dynamics (FED)
project from the NASA Ecological Processes and Modeling Program. The authors
would like to thank Elissa Levine for defining the soil characteristics, Dean Urban for
helping to develop species parameters for the NEF and Lara Prihodko and Guoqing
Sun for assisting with graphics and image processing.
References
Adair, M. and Guidon, B. 1990. Statistical edge detection operators for linear feature
extraction in SAR images. - Can. J. Rem. Sens. 16: 10-19.
Bidlake, WR, Campbell, G.S., Papendick, R.I. and Cullum, RF. 1992. Seed-zone temperature
and moisture conditions under conventional no-tillage in Alaska. - Soil Sci. Soc. Am. J. 56:
1904-1910.
Bonan, G.B. 1989. A computer model of the solar radiation, soil moisture, and soil thermal
regimes in boreal forests. - Eco!. Mode!. 45: 275-306.
Botkin, D.B. 1993. Forest dynamics: An ecological mode!. - Oxford University Press, Oxford,
UK. 309 pp.
Bradshaw, G.A. and Spies, TA. 1992. Characterizing canopy gap structure in forests using
wavelet analysis. - J. Eco!. 80: 205-215.
Busing, R.T. 1991. A spatial model of forest dynamics. - Vegetatio 92: 167-179.
Caldwell, M.M. and Pearcy, RW. 1994. Exploitation of environmental heterogeneity by
plants: Ecophysiological processes above- and belowground. -Academic Press, San Diego,
CA. 429 pp.
Chen, J., Franklin, J.F. and Spies, T.A. 1992. Vegetation responses to edge environments in old-
growth Douglas-fir forests. - Eco!. App!. 2: 387-396.
Chen, J., Franklin, J.F. and Spies, T.A. 1993. Contrasting microclimates among clearcut, edge,
and interior or old-growth Douglas-fir forest. - Agric. For. Meteor. 63: 219-237.
Clark, J.S. 1993. Scaling at the population level: Effects of species composition and population
structure. - In: Ehleringer, J.R and Field, C.B. (eds). Scaling Physiological Processes: Leaf
to Globe. Academic Press, San Diego, CA, pp. 255-285.
Cohen, W.B. and Spies, T.A. 1992. Estimating structural attributes of Douglas-fir/western
hemlock forest stands from Landsat and SPOT imagery. - Rem. Sens. Environ. 41: 1-17.
Cohen, WB., Spies, T.A. and Bradshaw, G.A. 1990. Semivariograms of digital imagery for
analysis of conifer canopy structure. - Rem. Sens. Environ. 34: 167-178.
Czaran, T and Bartha, S. 1992. Spatiotemporal dynamic models of plant populations and
communities. - Trends Eco!. Evo!. 7: 38-42.
Denslow, J.S. and Spies, TA. 1990. Canopy gaps in forest ecosystems: An introduction. - Can.
J. For. Res. 20: 619.
Deren, L. and Juliang, S. 1994. The wavelet and its application in image edge detection. -
ISPRS J. Photogramm. Rem. Sens. 49: 4-11.
Dougherty, E.R. 1992. An Introduction to Morphological Image Processing. - SPIE Press,
Bellingham, WA. 161 pp.
Esseen, P. 1994. Tree mortality patterns after experimental fragmentation of an old-growth
conifer forest. - Bio!. Conserv. 68: 19-28.
129
Fisher, R.A. 1925. Statistical Methods for Research Workers. - Oliver and Boyd, Edinburgh.
319 pp.
Fortin, M. 1994. Edge detection algorithms for two-dimensional ecological data. - Ecology
75: 956-965.
Friend, A.D., Running, S.W and Shugart, H.H. 1993. A physiology-based gap model of forest
dynamics. - Ecology 74: 792-797.
Ghiselin, J. 1977. Analyzing ecotones to predict biotic productivity. - Environ. Manage. 1:
235-238.
Hall, EG., Botkin, D.B., Strebel, D.E., Woods, K.D. and Goetz. S.1. 1991. Large-scale patterns
of forest succession as determined by remote sensing. - Ecology 72: 628-640.
Hansen, A.1. and di Castri, E 1992. Landscape Boundaries: Consequences for Biotic Diversity
and Ecological Flows. - Springer-Verlag, New York. 452 pp.
Harding, D.1., Blair, J.B., Garvin, J.B. and Lawrence, WT. 1994. Laser altimetry waveform
measurement of ve,getation canopy structure. - In: Proceedings of the IGARSS'94
Symposium, August 8-12, 1994, California Institute of Technology, Pasadena, CA, vol. 2,
pp. 1251-1253.
Held, D.N. et al. 1988. The NASA/JPL multifrequency, multipolarization airborne SAR
system. - In: Proceedings of tl"\e IGARSS'88 Symposium, September 12-16, 1988,
Edinburgh, pp. 317-322.
Holland, M., Risser, P.G. and Naiman, R.J. 1991. Ecotones: The Role of Landscape
Boundaries in the Management and Restoration of Changing Environments. - Chapman
and Hall, New York. 142 pp.
Holling, C.S. 1992. Cross-scale morphology, geometry, and dynamics of ecosystems. - Ecol.
Monogr. 62: 447-502.
Iverson, L.R, Graham, RL. and Cook, E.A. 1989. Applications of satellite remote sensing to
forested ecosystems. - Land. Ecol. 3: 131-143.
Johnston, C.A. and Bonde, J. 1989. Quantitative analysis of ecotones using a geographic
information system. - Photogramm. Engin. Rem. Sens. 55: 1643-1647.
Johnston, C.A, Pastor, J. and Pinay, G. 1992. Quantitative methods for studying landscape
boundaries. - In: Hansen, A.1. and di Castri, E (eds). Landscape Boundaries: Consequences
for Biotic Diversity and Ecological Flows. Springer-Verlag, New York, pp. 107-125.
Kachi, N., Yasuoka, Y., Totsuka, T. and Suzuki, K. 1986. A stochastic model for describing
revegetation following forest cutting: An application for remote sensing. - Ecol. Model. 32:
105-117.
King, AW 1991. Translating models across scales in the landscape. - In: Thmer, M.G. and
Gardner, R.H. (eds). Quantitative Methods in Landscape Ecology. Springer-Verlag, New
York, pp. 479-517.
Kolasa, J. and Pickett, S.T.A. 1991. Ecological Heterogeneity. - Springer-Verlag, New York.
332 pp.
Krummel, J.R. 1986. Landscape ecology: Spatial data and analytical approaches. - In: Dyer,
M.1. and Crossley, D.A, Jr. (eds). Coupling of Ecological Studies with Remote Sensing.
U.S. Man and the Biosphere Program, Department of State, Washington, DC, pp. 125-132.
Krummel, J.R, Gardner, RH., Sugihara, G., O'Neill, R.Y. and Coleman, P.R. 1987. Landscape
patterns in a disturbed environment. - Oikos 48: 321-324.
Kux, H.1.H. and Henebry, G.M. 1994. Evaluating anisotropy in SAR imagery using lacunarity
functions. - Int. Arch. Photogramm. Rem. Sens. 30: 141-145.
LaGro, 1., Jr. 1991. Assessing patch shape in landscape mosaics. - Photogramm. Engin. Rem.
Sens. 57: 285-293.
Lee, N.J. and Nakane, K. 1996. Forest vegetation classification and biomass estimation based
on Landsat TM data in a mountainous region of west Japan. - In: Gholz, HL, Nakane, K.
and Shimoda, H. (eds). The Use of Remote Sensing in the Modeling of Forest Productivity.
Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 159-171.
Leemans, R. and Prentice, I.e. 1987. Description and simulation of tree-layer composition and
size distributions in a primaeval Picea-Pinus forest. - Vegetatio 69: 147-156.
Levine, E.R., Ranson, K.J., Smith, J.A., Williams, D.L., Knox, R.G., Shugart, H.H., Urban,
D.L. and Lawrence, WT. 1993. Forest ecosystem dynamics: Linking forest succession, soil
process and radiation models. - Eco!. Mode!. 65: 199-219.
Levine, E.R., Knox, R.G. and Lawrence, WT. 1994. Relationships between soil properties and
vegetation at the Northern Experimental Forest, Howland, Maine. - Rem. Sens. Environ.
47: 231-241.
Liebermann, M., Liebermann, D. and Peralta, R. 1989. Forests are not just swiss cheese:
Canopy stereogeometry of non-gaps in tropical forests. - Ecology 70: 550-552.
Lopes, A., Nezry, E., Touzi, R. and Laur, H. 1993. Structure detection and statistical adaptive
speckle filtering in SAR images. - Int. J. Rem. Sens. 14: 1735-1758.
Major, J. 1951. A functional factorial approach to plant ecology. - Ecology 32: 392-412.
Martin, M.E. and Aber, }.D. 1996. Estimating forest canopy characteristics as inputs for
models of forest carbon exchange by high spectral resolution remote sensing. - In: Gholz,
H.L., Nakane, K. and Shimoda, R (eds). The Use of Remote Sensing in the Modeling of
Forest Productivity. Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 61-72.
Matlack, G.R. 1993. Microenvironment variation within and among forest edge sites in the
eastern United States. - Bio!. Conserv. 66: 185-194.
Matlack, G.R. 1994. Vegetation dynamics of the forest edge - Trends in space and
successional time. - J. Eco!. 82: 113-123.
McIntosh, R.P. 1985. The Background of Ecology: Concept and Theory. Cambridge
University Press, Cambridge, UK. 383 pp.
Miller, L.D. and Williams, D.L. 1978. Monitoring forest canopy alteration around the world
with digital analysis of Landsat imagery. - In: Proceedings of the I. U.ER.O. International
Symposium on Remote Sensing for Observation and Inventory of Earth Resources and the
Endangered Environment, July 2-8, 1978, Freiburg, Federal Republic of Germany, pp.
1721-1761.
Mooney, H.A. and Chapin, ES. III. 1994. Future directions of global change research in
terrestrial ecosystems. - Trends Eco!. Evo!. 9: 371-372.
Moran, E.E, Brondizo, E., Mausel, P. and Wu, Y. 1994. Integrating Amazonian vegetation, land
use and satellite data. - Bioscience 44: 329-338.
Musick, RB. and Grover, H.D. 1991. Image textural measures as indices of landscape pattern.
- In: Turner, M.G. and Gardner, R.H. (eds). Quantitative Methods in Landscape Ecology.
Springer-Verlag, New York, pp. 77-103.
Nelson, R., Horning, N. and Stone, T.A. 1987. Determining the rate of conversion in Mato
Grasso, Brazil, using Landsat MSS and AVHRR data. - Int. J. Rem. Sens. 8: 1767-1784.
O'Neill, R.Y., Krummel, J.R., Gardner, R.H., Sugihara, G., Jackson, B., DeAngelis, D.L.,
Milne, B.T., Turner, M.G., Zygmunt, B., Christensen, S.W, Dale, Y.R and Graham, R.L.
1988. Indices of landscape pattern. - Land. Eco!. 1: 153-162.
Pacala, S.W and Tilman, D. 1994. Limiting similarity in mechanistic and spatial models of
plant competition in heterogeneous environments. - Am. Nat. 143: 222-257.
Pacala, S.W., Canham, C.D. and Silander, J.A. 1993. Forest models defined by field
measurements. 1. The design of a northeastern forest simulator. - Can. J. For. Res. 23:
1980-1988.
131
Pastor, J. and Post, W.M. 1985. Development of a linked forest productivity-soil process
model. ORNUfM-95 19. - Oak Ridge National Lab, Oak Ridge, TN. 162 pp.
Pastor, J. and Broschart, M. 1990. The spatial pattern of a northern conifer-hardwood
landscape. - Land. Ecol. 4: 55-68.
Phillips, O.L. and Gentry, AH. 1994. Increasing turnover through time in tropical forests. -
Science 263: 954-958.
Platt, W.J. and Strong, D.R 1989. Gaps in forest ecology. - Ecology 70: 535.
Prihodko, L., Hammond, K., Knox, R.G., Kalb, V. and Levine, E.R. 1994. NASA/FED
welcome page. - Available from NASA Goddard Space Flight Center, Greenbelt, MD, via
the Internet: http://fedwww.gsfc.nasa.gov
Quattrochi, D.A. and Pelletier, R.E. 1991. Remote sensing for analysis of landscapes: An
introduction. - In: Turner, M.G. and Gardner, R.H. (eds). Quantitative Methods in
Landscape Ecology. Springer-Verlag, New York, pp. 51-76.
Ranney, J.W. 1977. Forest island edges - Their structure, development, and importance to
regional forest ecosystem dynamics. EDFB/IBP-77/1. - Oak Ridge National Lab,
Environmental Science Division Publication 1069, Oak Ridge, TN. 36 pp.
Ranney, J.W., Bruner, M.e. and Levenson, J.B. 1981. The importance of edge in the structure
and dynamics of forest islands. - ~n: Burgess, R.L. and Sharpe, D.M. (eds). Forest Island
Dynamics in Man-dominated Landscapes. Springer-Verlag, New York, pp. 67-95.
Ranson, K.J. and Williams, D.L. 1992. Remote sensing technology for forest ecosystem
analysis. - In: Shugart, H.H., Leemans, Rand Bonan, G.B. (eds). A Systems Analysis of
the Global Boreal Forest. Cambridge University Press, Cambridge, UK, pp. 267-290.
Ranson, K.J. and Sun, G. 1993. Retrieval of forest spatial pattern from SAR images. - In:
Proceedings of the IGARSS'93 Symposium, August 18-:-21, 1993, Tokyo, pp. 1213-1215.
Ranson, K.J. and Sun, G. 1994. Mapping biomass for a northern forest using multifrequency
SAR data. - IEEE Trans. Geosci. Rem. Sens. 32: 388-396.
Ranson, K.J. and Smith, J.A. 1995. Analyses of northern forest spatial patterns using fractal
dimensions. - IEEE Trans. Geosci. Rem. Sens. (in press).
Risser, P. 1993. Ecotones. - Ecol. Appl. 3: 367-368.
Running, S.W., Peterson, D.L., Spanner, M.A. and Teuber, K.B. 1986. Remote sensing of
coniferous forest leaf area. - Ecology 67: 273-276.
Running, S.W., Nemani, R.R, Peterson, D.L., Band, L.E., Potts, D.P., Pierce, L.L. and
Spanner, M.A 1989. Mapping regional forest evapotranspiration and photosynthesis by
Schlesinger, WH., Reynolds, J.P., Cunningham, G.L., Huenneke, L.P., Jarrell, WM., Virginia,
RA and Whitford, WG. 1990. Biological feedbacks in global desertification. - Science
247: 1043-1048.
Schwarz, P.A., Urban, D.L. and Weinstein, D.A 1994. Partitioning the importance of abiotic
constraints and biotic processes in generating vegetation patterns on landscapes. - In:
Program and Abstracts of the Ninth Annual U.S. Landscape Ecology Symposium, March
23-26,1994, Tucson, AZ, p. 116.
Shugart, H.H. 1984. A Theory of Forest Dynamics. - Springer-Verlag, New York. 278 pp.
Shugart, H.H., Smith, T.M. and Post, WM. 1992. The potential for application of individual-
based simulation models for assessing the effects of global change. - Ann. Rev. Ecol. Syst.
23: 15-38.
Skole, D.L. and Tucker, e.J. 1993. Tropical deforestation and habitat fragmentation in the
Amazon: Satellite data from 1978 to 1988. - Science 263: 954-958.
Smith, T.M. and Urban, D.L. 1988. Scale and resolution of forest structural pattern. - Vegetatio
74: 143-150.
Smith, J.A. and Goltz, S.M. 1994. Updated thermal model using simplified short-wave
radiosity calculations. - Rem. Sens. Environ. 47: 167-175.
Soule, M.E. and Kohm, K.A. 1989. Research Priorities for Conservation Biology. - Island
Press, Washington, DC. 97 pp.
Spanner, M., Johnson, L., Miller, J., McCreight, R, Freemantle, J., Runyon, J. and Gong, P.
1994. Remote sensing of seasonal leaf area index across the Oregon transect. - Ecol. Appl.
4: 258-271.
Spies, TA., Ripple, WJ. and Bradshaw, G.A. 1994. Dynamics and pattern of a managed
coniferous forest landscape in Oregon. - Ecol. Appl. 4: 555-568.
Steffen, WL., Walker, B.H., Ingram, J.S. and Koch, G.W 1992. IGBP Report 21. Global
change and terrestrial ecosystems: The operational plan. - International Geosphere-
Biosphere ProgramlInternational Council of Scientific Unions, Stockholm. 95 pp.
Syrjanen, K., Kalliola, R., Puolasmaa, A. and Mattsson, J. 1994. Landscape structure and
forest dynamics in subcontinental Russia European taiga. - Ann. Zool. Fenn. 31: 19-34.
Touzi, R, Lopes, A. and Bousquet, P. 1988. A statistical and geometrical edge detector for
SAR images. - IEEE Trans. Geosci. Rem. Sens. 26: 764-773.
Tucker, C.J., Townshend, J.R.G. and Goff, T.E. 1985. African land-cover classification using
satellite data. - Science:; 227: 369-375.
Ulaby, ET., Kouyate, F., Brisco, B. and Williams, TH.L. 1986. Textural information in SAR
images. - IEEE Trans. Geosci. Rem. Sens. 24: 235-245.
Urban, D.L., O'Neill, RY. and Shugart, H.H. 1987. Landscape ecology. - Bioscience 37:
119-127.
Urban, D.L., Bonan, G.B., Smith, TM. and Shugart, H.H. 1991. Spatial applications of gap
models. - For. Ecol. Manage. 42: 95-110.
Ustin, S.L., Wessman, C.A., Curtis, B., Kasischke, E., Way, J. and Vanderbilt, V.c. 1991.
Opportunities for using the EOS imaging spectrometers and synthetic aperture radar in
ecological models. - Ecology 72: 1934-1945.
Vitousek, P.M. 1994. Beyond global warming: Ecology and global change. - Ecology 75:
1861-1876.
Wales, B.A. 1967. Climate, microclimate and vegetation relationships on north and south
forest boundaries in New Jersey. - William L. Hutcheson Memorial Forest Bulletin 2: 1-60.
Wales, B.A. 1972. Vegetation analysis of north and south edges in a mature oak-hickory forest.
- Ecol. Monogr. 42: 451-471.
Walker, B.H. 1994. Landscape to regional-scale responses of terrestrial ecosystems to global
change. -Ambio 23: 67-73.
Wallin, D.O., Harmon, M.E., Cohen, WB., Fiorella, M. and Ferrell, WK. 1996. Use of remote
sensing to model land use effects on carbon flux in forests of the Pacific Northwest, USA.
- In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the
Waring, RH., Way, J., Hunt, E.R., Jr., Morrisey, L., Ranson, K.J., Weishampel, J., Oren, R. and
Franklin, S.E. 1995. Remote sensing with synthetic aperture radar in ecosystem studies. -
Bioscience (in press).
Way, J., Rignot, E.J.M., McDonald, K.e., Oren, R.; Kwok, R, Bonan, G., Dobson, M.e.,
Viereck, L.A. and Roth, J.E. 1994. Evaluating the type and state of Alaska taiga forests with
imaging radar for use in ecosystem models. - IEEE Trans. Geosci. Rem. Sens. 32:
353-370.
Weishampel, J.E and Urban, D.L. 1995. Coupling a spatially-explicit forest gap model with a
3-D solar routine to simulate latitudinal effects. - Ecol. Model. (in press).
133
Weishampel, J.F., Urban, D.L., Shugart, H.H. and Smith, J.B. 1992. Semivariograms from a
forest transect gap model compared with remotely sensed data. - J. Veg. Sci. 3: 521-526.
Weishampel, J.F., Sun, G., Ranson, K.J., Lejeune, K.D. and Shugart, H.H. 1994. Forest
textural properties from simulated microwave backscatter: The influence of spatial
resolution. -Rem. Sens. Environ. 47: 120-131.
Wessman, C.A. 1992. Spatial scales and global change: Bridging the gap from plots to GCM
grid cells. - Ann. Rev. Ecoi. Syst. 23: 175-200.
Wessman, c.A., Aber, J.D., Peterson, D.L. and Melillo, J.M. 1988. Remote sensing of canopy
chemistry and nitrogen cycling in temperate forest ecosystems. - Nature 335: 154-156.
Whittaker, R.H. 1953. A consideration of climax theory: The climax as a population and
pattern. - Ecoi. Monogr. 23: 41-78.
Wiens, J.A., Crawford, C.S. and Gosz, J.R. 1985. Boundary dynamics: A conceptual
framework for studying landscape systems. - Oikos 45: 421-427.
Williams-Linera, G. 1990. Vegetation structure and environmental conditions of forest edges
in Panama. - J. Ecoi. 78: 356-373.
Woodwell, G.M., Hobbie, J.E., Houghton, R.A., Melillo, J.M., Peterson, B.J., Shaver, G.R.,
Stone, T.A., Moore, B. and Park, A.B. 1983. Deforestation measured by Landsat: Steps
towards a method. - U.S. D~partment of Energy Publication DOE/EV/l0468-1,
Washington, DC. 62 pp.
Wu, Y.c. and Strahler, A.H. 1994. Remote estimation of crown size, stand density, and biomass
on the Oregon transect. - Ecoi. Appi. 4: 299-312.
Young, A. and Mitchell, N. 1994. Microclimate and vegetation edge effects in a fragmented
podocarp-broadleaf forest in New Zealand. - BioI. Conserv. 67: 63-72.
SIX
Combining Remote Sensing and Forest Ecosystem Modeling:

An Example Using the Regional HydroEcological Simulation
System (RHESSys)
Joseph C. Coughlan and Jennifer L. Dungan
Coughlan, J.C. and Dungan, J.L. 1996. Combining remote sensing and forest
ecosystem modeling: An example using the Regional HydroEcological Simulation
System (RHESSys). - In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use
of Remote Sensing in the Modeling of Forest Productivity. Kluwer Acad. Pub!.,
Images from airborne or satellite-based remote sensing systems are the only data
available for regional and global productivity studies that do not require interpolation
or extrapolation. Four categories of image use are identified: image classification,
model initialization, model input and model verification. Model initialization using
vegetation indices derived from images is discussed using a regional modeling
framework, the Regional HydroEcological Simulation System (RHESSys). In this
chapter, we illustrate RHESSys' sensitivity to soil moisture and the interrelationships
between the soil data theme and the vegetation and climate data themes. Improving
image transfer functions can increase the quality of vegetation estimates; however,
ancillary data (such as topography and soil data) are also needed at appropriate levels
of accuracy and precision. An example simulation is provided, which uses vegetation
data from two watersheds in western Montana. Results demonstrate the model's
sensitivity to soil data in a wet, dry climate, and indicate the importance of considering
the data collection process as an integrated effort guided by modeling requirements
and model sensitivity. Additional consideration must be made for validation and
collection of independent data for these purposes.
J.e. Coughlan and J.L. Dungan, JCWS Ames Operation, NASA, Ames Research
Center, Moffett Field, CA 94035, USA.
135
Introduction
Remote sensing data are collected at a large variety of scales, from pixels of a few
meters ranging over a few square kilometers to pixels of more than 1 km ranging over
several tens of degrees of latitude and longitude. Although the term "scale" has been
loosely applied in many ecological contexts, it actually incorporates two concepts: the
area or volume of the measurement unit or quantity being modeled (the "support"),
and the range or extent of the region being measured or modeled. The supports and
extents used as bases in current approaches to modeling vegetation structure and
function are often correlated, since their quotients must stay within the limits of
manageable parameterization and computation. For example, global models,
discussed in the third portion of this volume, most often use supports of 50 x 50, lOx
8 2
10 or, at most, 0.5 0 x 0.5 0 to cover the 1.513 x 10 km of the earth's land surface. At
the other extreme, the most detailed stand models, represented in the first portion of
this volume, may simulate forest carbon (C) dynamics at the level of 1 ha or less.
Images from airborne or satellite remote sensing systems are the only data available
that are collected at 'In appropriate scale for regional and global ecosystem
productivity studies. AU other measurements must be interpolated or extrapolated
(scaled up) in some way to apply to larger areas (Ehleringer and Field 1993, Foody and
Curran 1994). And although remote sensing information is often of lower quality than
direct field measurement, it is spatially exhaustive and of equal quality at all locations
(Burrough 1986). According to Graetz (1990), "global understanding is an impossible
task for the discipline of ecology to achieve without extensive and intensive use of
remotely sensed data."
Of course, earth-observing sensors do not collect direct information on vegetation
productivity. Sensor measurements in the optical region calibrated to the radiance of
the earth's surface are more closely related to vegetation structure and absorptance
than to production per se. To estimate production, models that simulate plant
processes are used with input data that are more easily described on a spatial basis.
There are many ways of utilizing remote sensing information with these models, and
much research is currently focused on the potential of these methods. The methods can
be thought of as belonging to one of four categories: partitioning of regions, estimation
of dynamic model variables, specification of the initial model boundary conditions
and, finally, confirmation of model results.
The first category, and the traditional use for remotely sensed data, includes
methods that classify images into regions of land cover. These regions are then
considered to be homogeneous with respect to certain properties or processes. When a
model is employed to simulate terrestrial processes, these regions may be used to
define different parameter sets. Loveland et al. (1991), Lathrop et al. (1994) and
Running et al. (1994) have discussed examples of these methods. Partitioning
alternatives are discussed in the latter part of this chapter.
Besides its use for stratification, spectral information can be used to provide a
process model with quantitative data. The latter three utilization methods, including
parameterization and confirmation of process models, all involve the use of an
empirically or theoretically derived transfer function between an independent variable
137
of interest and radiance or reflectance of the surface (the dependent variable). The
transfer function is applied to every scene element in order to produce an image
describing the spatial distribution of the independent variable. The scene element may
be a pixel or a pixel-aggregate, or a region defined in a classification. Transfer
functions used for the variables of interest range from simple linear transformations,
such as that of a near-infrared to red reflectance ratio to obtain a fractional value of
vegetation cover (Cary and Rosenswig 1987, Carlson et al. 1990, Leprieur et al. 1995),
or of derivative reflectance to derive foliar biochemical concentrations (Wessman et al.
1988, Gholz et al. 1996, Martin and Aber 1996), to geometric-optical model inversion
(Goel and Deering 1985, Rosema et al. 1992, Jupp and Walker 1996) to obtain
vegetation structure descriptions. The development of these transfer functions usually
relies on a calibration step, in which samples from the image are correlated with
vegetation variables measured on the ground.
Data for variables exogenous to a process model, or driving data, have been
estimated using transformed remotely sensed data. These are typically required for
each model time step or at some regular time interval. Examples of driving data for
models of forest productivity include climate variables such as air temperature and
solar radiation, and vegetation variables such as phytomass. Monitoring of foliar
biochemical concentrations as discussed by Martin and Aber (1996) and Gholz et al.
(1996) might be useful for driving models. At present, lack of available remotely
sensed data has limited the useful application of multitemporal data to models that
operate on at least biweekly time steps.
Model initialization with remote sensing data usually requires fewer, if any,
repetitive spectral measurements. Examples of initialization variables are leaf area
index (LAI), standing biomass and soil moisture status. Models may subsequently
modify the initial vegetation conditions via growth, litterfall or death processes. The
observed vegetation can be at a maximum value or a potential vegetation volume, or
may be in a disturbed state with a suboptimal volume.
The fourth category of utilization methods has been the least used. Model
confirmation with remote sensing data treats the data as independent and compares
them with a corresponding model prediction. For example, using the methods
described in Coughlan and Running (1989a), we spatially estimated plant water stress
and compared it with observed surface temperatures. Nemani et al. (1989, 1993a) used
the thermal channel on the NOAA Advanced Very High Resolution Radiometer
(AVHRR) sensor (Townshend 1994) to estimate surface temperatures over a forested
region as a means of indicating energy exchange; these estimates were then compared
with those for simulated water stress. Because a transpiring canopy has more latent
heat transfer than a water-stressed canopy, it appears relatively cool. Images taken in
the late summer showed a marked increase in mean surface temperature relative to
those taken in early summer, which appeared to confirm the simulations of water-
stressed vegetation.
In principle, any or all of these four functions could be fulfilled using remotely
sensed data. This chapter concerns the use of remote sensing data (the third method)
to specify the initial conditions of a forest ecosystem process model designed to
simulate daily photosynthesis and evapotranspiration.
Regardless of the method chosen, the scale of the ecological model used is defined
by at least two factors: the scale of the input data and the scale of the processes
described by the model equations. Ideally, all model input data, remotely sensed or
otherwise, should be relevant to the same support. Practically, this is nearly impossible
to achieve. The scale of the equations themselves may be even more elusive. Although
many are based on physical theory, and therefore may not be tied to particular spatial
dimensions, the homogeneity assumptions involved always imply error. As Schneider
(1993) pointed out, model equations should not be assumed to be without scale.
Integration of ecological modeling and remote sensing data requires that a model
be designed to accommodate the types of data obtainable from transformed images.
The challenge is to define a regional forest model in a way that captures basic system
behavior while maintaining enough simplicity to allow the model to be practically
initialized and executed. The simplicity of the models we illustrate here is related as
much to our ability to parameterize them as to our understanding of the physical
processes they represent.
Regional HydroEcological Simulation System (RHESSys)

A variety of approaches may be taken to the application of a terrestrial model in a
spatial context. Many recent examples have been summarized and discussed in
Goodchild et al. (1993). Each approach is dependent upon the research or application
objective expressed in the model (e.g., Dickenson et al. 1986, Sellers et al. 1987,
Parton et al. 1987, Bonan 1991, McMurtrie et al. 1992, Lee et al. 1993). Our regional
modeling method is centered around the application of remotely sensed LAI to
quantify water and C fluxes in complex terrain. We employ a distributed modeling
framework in a specialized GIS data set that includes an ecosystem model, a spatial
climatic data simulator and a topographically based water routing model. The system,
including models and data processing, is collectively called the Regional
HydroEcological Simulation System, or RHESSys (Nemani et al. 1993c).
RHESSys evolved from the objective of quantifying regional ecosystem C, water
and nitrogen (N) fluxes by executing an ecosystem model in a spatial context.
RHESSys helps transform spatial data from a variety of sources into model-ready
inputs, interfaces models, and performs data processing, all to facilitate computation
of spatial fields of water, C and N fluxes and processes. Without RHESSys's
automation, defects and errors can be easily introduced into the labor-intensive data
processing, simulation and analysis processes. RHESSys encapsulates and documents
the complexity inherent in organizing, manually processing and transforming raw data
into a finished product, like a map of annual forest net primary production (NPP).
RHESSys processes data as mean areal values and can generate results for model
inputs organized as polygons or rasters.
Spatial data
RHESSys spatial input data are organized into three categories: topographic, soil and
vegetation data. The data layers are stored as binary raster files, but the model input
data RHESSys generates are stored in a flat ASCII database called a cartridge file
139
(CRT) (Table 1). RHESSys was built with a distinct intermediate step between the
processing of individual spatial data layers and their union into a model input set. An
ASCII flat database was selected as the input format to facilitate viewing and analysis
of model inputs using a data viewing program or common computer language like awk
(Dougherty 1990) to screen illogical model inputs. Since the data layers come from
various sources, data may not overlay properly or, if overlaid, may include
combinations of site values that are illogical as model inputs. For example, for a
typical climate of the Pacific Northwest, an LAI of 10.0 and a soil water-holding
capacity of 5 cm are both reasonable as individual values; however, they are not
plausible as combined values for the same site. A model run resulting from these
values would produce a negative annual C balance and extreme water stress for a
substantial portion of the summer. Identification of suspect entries may indicate errors
in registration or individual data layers; these errors may be corrected or deleted or
may justify the reregistration of some data layers.
Topography
The CRT input database in Table 1 contains several fields generated from topographic
data for a large basin in western Montana. Elevation data sampled at 30-m and 100-m
intervals are available digitally from the USGS and Defense Mapping Agency (USGS
1990). From either of these two resolutions, slope, aspect and elevation are
constructed at supports compatible with those ofTM- or AVHRR-derived LAI layers,
respectively. Computation of slope and aspect is particularly important for accurate
estimates of incoming short-wave radiation and its effect on the timing of snowmelt
(Coughlan and Running in press). Site elevation is needed for temperature and
precipitation extrapolation along elevation gradients using local knowledge of
precipitation and temperature isohyets. The topographic digital elevation model
(DEM) is also used to define the upslope drainage area and gradient for each cell;
these measurements are then used to compute the topographic soil moisture index (<I
for TOPMODEL.
Soil
Soil data define a site's capacity to retain and gradually release water. A soil's water-
holding capacity is essentially its field capacity, while a soil water retention curve
defines how a soil releases water and how much water it contains at "wilting point."
RHESSys uses a spatial definition of soil water-holding capacity derived from digital
maps of soil properties. Soil information is commonly stored in maps that have been
digitized into vector format (Burrough 1986). Soils maps are usually
pseudoquantitative, mapping the means while ignoring the within-unit spatial
variation; they also include the effects of human interpretation. Soils have been
mapped by point sampling and by assuming concurrence with both observed
vegetation and hillslope patterns. Maps on which soil boundaries follow hillslope
contours and abruptly end at a ridge or stream are commonly seen; in flatter areas,
soils are mapped according to vegetation patterns observed in aerial photographs or on
other vegetation maps. Soils in mountainous watersheds are known to be shallower
......
.t>-
Table 1. A forest watershed was processed in RHESSys using a 30m resolution (Fig. 1), and the resulting simulation input o
records are stored in a CRT file in tabular form. Shaded entries indicate the fields necessary for running RHESSys with
;J
TOPMODEL, while unshaded entries are required for running RHESSys with a one-dimensional soil model. ell has a range of ~
14.5 (wet) to 1.5 (dry). ~

~
Aspect Elevation Gradient Total LAI Soil water Cell Area Topographic soil moisture ~
:::tl
~
(0) (m) (0) storage capacity (km2) index (<I ;:s
0
(m) ...
~
Mean Intervals V:l

~
::l
S
'"
212.14 1353.28 15.08 5.08 0.51 1 95 7.81 7 Ot)
76.46 1169.62 2.26 5.65 0.51 2 68 8.35 8 S

230.69 1834.89 6.43 5.21 0.51 3 26010 8.14 14 So
~
30.8 1853.86 6.08 5.93 0.51 4 21859 7.94 13 ~

>:l.
247.17 1857.43 4.9 5.38 0.51 5 82538 8.01 15 ~
63.51 1903.73 2.87 5.14 0.51 6 171334 8 15 =

Ot)
236.19 1158 3.98 5.01 0.51 7 6 14.78 2
~
353.16 1215.5 1.17 5.83 0.51 8 452 12.33 12 ~
266.21 1310 1.54 5.36 0.51 9 949 12.2 13 ~
...'"'"t:I
Cl
>:l.
l::
'"l
...
~.
~.
141
along ridges and on steep gradients and deeper in downslope areas like valleys. The
inability to properly characterize soil heterogeneity with digitized soils classes is
being addressed with inference techniques that modify soils maps based on
topographic features (Zhu and Band 1994).
Vegetation
In RHESSys, LAI defines not only the exchange surface area for photosynthesis and
transpiration but also other forest variables (e.g., respiring stem, root and leaf biomass)
that are assumed to be proportional to LA!. There is no method to remotely sense root
mass and respiring stem mass, and since both variables are broadly correlated with
LA! (Running and Gower 1991), in RHESSys the two variables are derived from LA!.
The LAI-root-stem mass relationship is an empirical ratio constructed from past
applications of the FOREST-BGC model in western Montana (Donner and Running
1986, McLeod and Running 1987, Running et al. 1989). The assumed proportional
relationship between biomass and LA! requires that only LA! data be used for model
input.
Comprehensive spatial sampling of LAI across a landscape is rare. Calibrations to
develop empirical transfer functions of visible and infrared images are often based on
a small number of measurements that are not spatially representative. Our applications
of RHESSys have been based on a linear regression equation established from stand
measurements in an experimental forest within the region being modeled (see
RHESSys example below, Table 1) and Bands 4 (760--900 nm) and 3 (630--690 nm)
from the Landsat Thematic Mapper (TM). This approach is limited by at least two
factors: error inherent in the regression and the fact that the area used to develop the
regression may not be representative of the landscape. For example, relationships
between LAI and spectral reflectance for conifer forests would overestimate LAI for
broadleaf forests in the same area, which in this region are common in riparian zones
and after recent disturbances. In an attempt to correct for this problem and the
additional effects of a varying understory, Nemani et al. (1993b) offered an alternative
transfer function involving TM Band 5 (1550--1750 nm) in addition to Bands 4 and 3.
They found maximum differences of 2 Mg ha- r yr"l in annual net canopy C gain after
adding the middle infrared band to the LAI estimator, representing a change of
approximately 20%. Improvement of these regression approaches will depend on
increased attention to spatially representative, statistically sound ground sampling
schemes. Nonregression approaches (Rosema et al. 1992, Dungan et al. 1994) may
also depend on improved sampling schemes, improved scene models or both.
RHESSys
RHESSys consists of three core models: FOREST-BGC, MTCLIM and TOPMODEL
(Fig. 1). FOREST-BGC is the ecosystem process model diagrammed in Figures 2 and
3 (Running and Coughlan 1988, Running and Gower 1991). MTCLIM is a climatic
simulator built to generate remote site microclimate data for FOREST-BGC when it is
run for mountainous terrain (Running et al. 1987, Hungerford et al. 1989). Beven and
Kirkby's (1979) TOPMODEL, a model for the horizontal redistribution of water
....-
.j:>
Regional HydroEcological Simulation System (RHESSys) N
Data Source Derived inputs GIS processing Model input files Models Model outputs
~
~
Climate NWS Radiation

GOES Temperature ~
~
Humidity
Rainfall Registration ~
:::tl
~
Vegetation TM Biome type - . . . - Drainage area
MSS leaf area index ~
AVHRR Stream network ~
Cartridge 1-' MT.lIM ET ~
land unit ;:s
Topography USGS Elevation

--.. Topography-soils
-El FOREST-BGC PSN
'"
~.
DEM Aspect index ,:111'" .:.. .::::. S
DMA Slope """'
Registration 111"" Means and III"'''TOPMODEl Discharge So
~
frequency
Soils SCS Texture, depth. distributions
water-holding - - . (optional)
capacity
~
~
~
NWS- National Weather Service MTCLIM - Mountain Microclimate Simulator ~
GOES- Geostationary Operational Environ. Satellite FOREST-BGC - Forest Ecosystem Simulator ~
TM- landsat Thematic Mapper TOPMODEl - Hydrologic Routing Simulator ....
'"
MSS- landsat Multispectral Scanner "'tl
AVHRR - NOAA/Advanced Very High Resolution Radiometer ET- Evapotranspiration <::l
USGS- United States Geological Survey PSN- Net photosynthesis l}
C"\
DEM- Digital Elevation Model Discharge - Stream discharge ....
~.
DMA- Defense Mapping Agency Cartridge - land unit parameterization
SCS- Soil Conservation Service Distribution - ~.
Withir.-Iand-unit parameterization
Figure 1. Diagram of the Regional HydroEcological Simulation System (RHESSys) showing the data transformation from raw
sources to final products.
143
within a hillslope, is used to refine the soil water component of FOREST-BGC (Band
et al. 1993). Each model's data requirements and assumptions are listed in the papers
cited above; however, some variables are spatially dynamic, as previously described.
FOREST-BGC
FOREST-BGC is a nonlinear process-level ecosystem simulation model that
calculates the cycling of C, water and N through forest ecosystems (Running and
Coughlan 1988, Running and Gower 1991). One key attribute facilitating the regional
application of FOREST-BGC is its quantification of vegetation using LA!. FOREST-
BGC computes a daily water balance including precipitation, evaporation, snow
accumulation, snowmelt, runoff and transpiration (Fig. 3a). Carbon dynamics include
daily canopy net photosynthesis and maintenance respiration (MR), annual
photosynthate allocation, tissue growth, growth respiration, litterfall and litter
decomposition (Fig. 3b). The time step for all N processes in the model is annual,
including mineralization, deposition, uptake and allocation to canopy and N losses
from the ecosystem.
Two of the major flux processes strongly influenced by plant responses to abiotic
conditions are transpiration and photosynthesis. Transpiration is computed using the
Penman-Monteith approach (Eq. 1, Fig. 3a) on a per-LAI basis, with canopy
conductance determined by air temperature and reductions in conductance for
freezing nights, high atmospheric vapor deficits, suboptimal solar radiation loads and
negative leaf water potentials.
(1)
where:
A. =latent heat of vaporization (J m'3)
E = transpiration (m day'l)
Q = incoming short-wave radiation (W m'2)
cp =specific heat of air (J ki l e l)
l
'Y = psychrometric constant (Pa e )
p = density of air (kg m'3)
VPD = canopy vapor pressure deficit (Pa)
ra = aerodynamic resistance (s m'l)
re = canopy resistance (s m'l)
o = slope of the vapor pressure curve (Pa (lei)
T sec = effective duration of daylight (sec day").
Wind velocity in a coniferous canopy is fixed in the model at 5 m S'I simply because
that value produces a good approximation of ra ; more exact methods require measured
wind velocity data, which are difficult to obtain (Campbell 1977). Soil water potential,
'lis' a critical determinant of re, is defined with a simple reciprocal relationship between
......
.j:.
.j:.
Forest-BGe
;i
<1>
Daily Yearly ~
<1>
PPT Meteorological data Growth ~

Air temperature Leaf ::tl
respiration <1>
- Radiation ::l
Precipltalion
7 re' c
Annual sum: i ...
<1>
- Humidity - Photosynthesis
- Evapotranspiration ~
;:s
- Respiration '"
~.
s
So
<1>
~
~
~
~
Annual allocation: ~
Carbon. Nitrogen
Leaf (LAI) ...'"~
"tI
Stem
~;~':ll (3
Root
~
...,
~.
~.
Figure 2. Compartment and flow diagram of the FORESTBGC ecosystem model listing the major water, C and N components
and mass flows (from Coughlan 1994).
145
soil water content and soil water-holding capacity. Formulations of water retention
based on 'II, and dependent on soil type, such as those of Saxton et al. (1986), can
approximate specific soil matrix potentials. Gross canopy photosynthesis (Pg ) (kg CO 2
2
m day"') is computed using a simple CO 2 diffusion gradient and mesophyll
conductance in the formulation by Lohammar et al. (1980) (Fig. 3b).
(2)
where:
~ = leaf conductances for water (m SI)
K m = CO 2 mesophyll conductance (m SI)
8C02 = CO 2 gradient (kg m- 3)
Canopy night MR and daily stem and root respiration are calculated with a Q IO
temperature response per unit mass and deducted from P g The remaining accumulated
photosynthate is annually allocatfjd to tissue growth. A mean P g LAr1sec 1 is computed
using canopy average radiation illumination and temperature; the P g rate is then
multiplied by total LAI and time for a daily total. Radiation is attenuated with the
Beer-Lambert Law for light penetration in a canopy, assuming an extinction
coefficient of -0.5. Maximum and minimum ambient air temperatures are used to
compute a mean daily canopy air temperature. Optimal mesophyll conductance is
reduced by scaler multipliers for suboptimal conditions of temperature and solar
radiation.
MTCLIM
MTCLIM is an extrapolation and simulation model that generates remote site-specific
microclimate data from central base station measurements. The input variables for
MTCLIM are daily maximum and minimum temperatures and precipitation. The site-
specific microclimate outputs used for input into FOREST-BGC are maximum and
minimum temperatures, dew-point temperature, precipitation and incoming short-
wave radiation (Q). Temperature and precipitation are lapse-rate dependent, covarying
with elevation, and are set to match observed rates for the location. Air temperature is
also empirically adjusted according to site-specific amounts of daily Q and LA!.
Relative humidity is derived from the dew-point temperature, which is assumed to
equal the minimum daily air temperature. Q is not routinely available but is a critical
variable for estimating ET and Pg and must be simulated if it cannot be measured.
Potential Q for any day and surface is computed using sun-earth geometry, following
Garnier and Ohmura (1968) and Buffo et al. (1972). Bristol and Campbell's (1984)
atmospheric transmissivity algorithm is used to reduce potential Q to incident Q.
Bristol and Campbell's (1984) algorithm replaces the more commonly reported but
less quantitative percent-cloud-cover index (0.0-1.0) as a means to quantify the
opacity of the atmosphere. The transmissivity computation assumes a stable air mass
with a minimum nighttime and a maximum daytime temperature.
.j::>.
0'1
a WATER
-
l
~~0 .IGl"oJ ~
0t.~ ... u. ~
. ----I --...~ ... (3 "'4ttn",.cI'II.h
G).tt U
, . . _.... K.. CIItt,
~
~
@.., ~.t . .,
t
.---....lJj @ ... ".,.~ ...
~
l (3)_1 " .. ::>;,
0,-...... .... ~
;:;
~
~
;:s
to,
~"
S
S-
~
.......
~
~.
~
~
''0'. ~
... , ___ ~
'
A~_
...-.r-.a- .til
~
"tl
,_ ... .
o C3
0"1 _. .,,~._. ~
:;::
-..".1:& .......--..1;'
l l <:>
~1
~
q"
Figure 3. Computational flow diagram of the water and C fluxes in FOREST-BGC. Variables are directed by arrows toward
operations contained in boxes. Terminal nodes in the tree are model inputs or constants. (a) Hydrological calculations for water
flux. (b) Calculations for C flux.
147
z
o
[C
IX:
~
o
TOPMODEL
In a mountainous watershed, the spatial variation in soil moisture is approximated
with TOPMODEL, which computes the redistribution of water due to saturated and
overland flow (Beven and Kirkby 1979). TOPMODEL augments the one-dimensional
soil bucket model defined from the soil water-holding capacity field in the RHESSys
input (Table 1). The saturated flow from each raster cell of a watershed is computed
as:
(3)
where:
2 l
qj = saturated flow (m dai )
2
to = lateral transmissivity (m da/)
~ = local slope angle (0)
Sj = soil moisture deficit below saturation (m)
M = exponential decay rate of transmissivity with depth (m).
The two main assumptions in Eq. 3 are that (i) the water table is parallel to the surface,
which can be approximated by ~, and (ii) the downhill saturated flow of each cell, qj'
has an exponential relationship to soil moisture deficit, Sj' When Sj = Om, the soil is
saturated and precipitation inputs to the soil are routed as overland flow. s; is
dependent on the mean soil water deficit of the entire watershed through the mean
watershed topographic index, <1>,
(4)
where:
S = mean watershed storage deficit (m)
2
aj = area (m)
 = areal average of the In(a/tan ).
The basic assumption in TOPMODEL is that the hydrologic behavior ofthe watershed
is adequately described by <1>. All cells with the same will have the same soil
moisture deficit. This assumption allows watershed data to be summarized by 
intervals and not by individual raster elements. For RHESSys, the intervals become
an organizing layer to define a CRT database. Instead of generating a CRT data set for
the entire raster population, the mean data attributes for each interval can be used
(Table 1), with the shaded columns representing the interval and the rest of the
variables representing the mean value for that interval. Soil water defined in
FOREST-BOC is updated based on the net flow into and out of the interval. Water
is also extracted and added to the soil compartment by FOREST-BOC.
149
Landscape aggregation
Digital data sets, especially those constructed from remote sensing variables, are often
so large that they consume prohibitive amounts of simulation time and resources on
modern computer workstations. For example, a small 17- krn2 watershed sampled with
TM data generates approximately 20,000 CRT entries and takes 3-24 hours of
computer time for one seasonal run, depending on the workstation. These large data
sets contain redundancy because they are generated by the resolution of the data
sampling method without regard for the data information content. Aggregation
reduces the input data set size while maintaining the information content of the data
set. Resulting land surface aggregates should have low internal variance and high
between-aggregate variance. For RHESSys, aggregated input CRT data contain the
mean properties of some aggregation scheme in place of the individual raster
elements. An example of such a representation is the -based aggregation depicted in
Table 1. Mean model inputs are generated for each class. Data aggregation methods
can employ numerical, geographical, hydrological and/or ecological principles to
simplify the data inputs (Coughlan. and Running 1989).
Geographic adjacency is the simplest criterion for merging cells into larger units.
This method assumes a geographic heuristic - adjacent land units are more similar
than distant land units. The raster data are merged into large cells and then the cell
means are used as model inputs. This method is common for distributed spatial
models, as have been used in SHE (Abbott et al. 1986) and DSVM (Wigmosta et al.
1994) to simplify gridded data into larger grids for the sake of reducing computational
complexity.
Hydrologic land units or hillslopes may be correlated with vegetation patterns and
processes (Band et al. 1991) and form the basis of a landscape aggregation scheme.
Band (1986) defined an automated method to identify hillslope polygons from digital
elevation models. The hillslope units are used as organizing overlays and their internal
variance is assumed to be less than that among adjacent hillslopes. Aspect is one
nonlinearly averaging variable that is by definition fairly uniform within a hillslope.
Model runs are made on the mean hillslope values, which are represented as CRT file
records (Table 1).
An ecological similarity index can form the basis of a landscape organization
scheme. The index is the organizing data layer for which data are averaged for a
simulation. Coughlan and Running (1989b) describe EDiS, a knowledge-based
method for constructing an index for purposes of aggregating data for RHESSys.
Operating on data from the CRT file, the EDiS model infers site water balance and
ecological conditions and classifies model inputs based on ecologically inferred
limiting factors. EDiS is the automated equivalent of an expert predicting model
performance from model inputs and organizing model inputs based on model
sensitivity. The classes are a layer for which mean data values are computed, just as
they might be for hillslopes. As a RHESSys preprocessor, EDiS functions about 103
times faster than a simulation with the combined MTCLIM and FOREST-BGC
models (Coughlan 1991), which can significantly reduce overall simulation time if
EDiS is used to aggregate data prior to simulation. For example, a 20,000-cell
watershed (see RHESSys example below) was reduced to 21 distinct simulation

classes.
Database techniques can also organize and reduce data. In such techniques, spatial
data layers are aggregated into discrete intervals and the model is executed on all
possible combinations of these intervals (Coughlan 1991). Ideally, this method should
maintain more detail, that is, define smaller intervals, for critical model variables.
Simulation with data intervals assumes that there is too much precision in the data
layers and that reducing the precision by aggregating into intervals will maintain
sufficient information content. An input CRT file contains all data interval
combinations and simulation results are weighted by the frequency of each interval
permutation.
One of the more complex aggregation methods used in RHESSys is a two-step
approach in which hillslopes are first used to divide the landscape into polygons with
similar slope and aspect values, then grouped into simulation units by using intervals
of <P values. A mean LAI value is determined for each <P interval, but the hillslope's
topographic characteristics are shared by all <P intervals. The hillslopes contain
relatively uniform aspect, a nonlinearly averaged variable. The averaged aspect and
slope within a hillslope produces less variance and error in simulated Q. For each
hillslope, a single climatic data set is simulated with MTCLIM and then a daily
FOREST-BGC simulation is executed on every <P interval within that hillslope unit. At
the end of each simulation time step, soil water is redistributed within the hillslope
among all the <P intervals.
A second database contains the vegetation and soil data for the within-hillslope <P
intervals (Table 2). These <P intervals are linked to the hillslope units in Table 1 by the
cell number and the shaded intervals field, enumerating the n intervals contained
within each hillslope. The second database, called FREQ, contains the hillslope
identification number and its n <P entries. The weighted mean of the FREQ entries for
a hillslope matches its mean LAI, area and <P in Table 1. The simulations are
performed on the FREQ file entries and are summed to obtain a mean hillslope result.
No optimal aggregation strategy has yet been defined for RHESSys. The
appropriate aggregation method is determined partially by the objectives and the
available data. For a watershed in western Montana represented with 30-m data, Band
(1993) compared several RHESSys aggregation methods, some with and some
without TOPMODEL. A 17-km2 watershed, segmented into six hillslope aggregations
with <P intervals, most nearly matched results made on the entire 20,000-cell database.
In flatter areas, hillslopes may be poorly formed and the hillslope aggregation method
will be ineffective. Band (1993) reported that spatial variations in soil moisture
resulted from different aggregation schemes and that this simulated soil moisture
heterogeneity had measurable effects on watershed photosynthesis and
evapotranspiration, especially during drying periods in the late summer.
151
Table 2. FREQ database defines the within-hillslope landunits. Each entry in
the CRT database has at least one corresponding entry in the FREQ file (Table
2). Table 2 is a partial listing showing entries for the 1 and 2 hillslopes in Table 1.
Cell (hillslope unit) TSI interval Area Total LAI Soil K,

2
(km ) (m 2 m- 2) (em sec-I)
1 4.5 1 5.84 0.54

1 5.5 31 5.92 0.61
1 6.5 38 4.21 0.65
1 7.5 19 4.99 0.69
1 8.5 1 7.97 0.74
1 13.5 4 6.49 1
1 14.5 1 4.61 1.06
2 4.5 3 3.48 0.54
2 5,5 10 6.08 0.6
2 6.5 19 5.67 0.65
2 7.5 11 5.87 0.7
2 8.5 14 5.7 0.75
2 9.5 6 5.28 0.8
2 10.5 4 5.68 0.85
2 11.5 1 6.82 0.88
3 1.5 1 5.26 0.39
RHESSys example
Band's (1993) results suggested that efforts to increase the overall accuracy of a
regional estimate may depend neither on decreasing support nor on refining the
accuracy of the transformation functions that drive vegetation variables. To increase
the precision of remotely sensed variables, concomitant increases should be made in
the precision of other spatial data themes, since these layers affect interpretation of
remotely sensed vegetation data via a model. To illustrate the significance of soil data
on interpretion of remotely sensed vegetation data, we ran RHESSys with and without
soil limitations for both a xeric and a mesic watershed. Eliminating the soil data theme
is, of course, an extreme treatment, but its removal illustrates one of the boundaries of
RHESSys's behavior.
In this case, RHESSys was run to estimate seasonally accumulated net
photosynthate (PSN) and MR for two different 17-km 2 watersheds in western
Montana. The C that can be allocated for growth (PSN) is the difference between the
gains from gross photosynthesis (Pg) and losses from MR for any given time period
(PSN = Pg - MR). Seasonally accumulated PSN and MR were chosen as summary
variables because they are sensitive to interactions between the spatially dynamic
variables quantifying site microclimate, soil moisture, plant physiologic responses and
plant biomass. Both soil moisture and climate can alter the balance of PSN and MR
and can affect the watershed's annual C balance, transforming the ecosystem from a
net C source into a sink when all CO 2 flux processes are considered. Forest MR is
dependent on the amount of respiring mass, its N content (spatially and temporally
invariant in our example) and ambient air temperature (Ryan 1991). PSN is also
controlled by leaf N concentrations and leaf biomass (represented as LA!) in addition
to light absorption, water availability and vapor pressure deficit (VPD).
Photosynthesis can be limited by summer drought and high VPD, which cause
stomatal closure and limit CO 2 uptake. However, MR is unaffected by stomatal
closure, which means forests can have a negative C balance when their stomates are
closed for prolonged periods of time.
The two watersheds used in our example are located along a regional north-south
precipitation gradient in western Montana and are approximately 200 km apart. The
north site, Soup Creek, receives 80 cm of precipitation annually, nearly twice the
amount of the southern site, and has a mean LAI of 7.1. The southern site, Elk Creek,
has a mean LAI of 3.0 and receives about 40 cm of precipitation annually. Both areas
were represented at 30-m resolution with LANDSAT TM data and digital elevation
model data and with soil maps rasterized to 30 m. There are approximately 20,000
cells in each watershed. Input databases were produced for a RHESSys simulation
without TOPMODEL (see unshaded fields in Table 1), which was disabled because its
operation on 20,000 units would have demanded a prohibitive expenditure of
computer resources.
RHESSys was executed on both watersheds - first with a simulated soil moisture
content, which allows for water stress induced by soil matrix potential, and then with
an unlimited soil moisture content to prevent soil-induced water stress. Both stressed
and unstressed simulation sets were sensitive to leaf stomatal closure induced by high
VPDs, freezing night temperatures and low light intensity. These and other abiotic
conditions controlling plant physiological responses varied temporally and spatially
within the watershed.
Figure 4 shows the relationship between PSN and MR at every tenth cell. A
comparison of MR between watersheds shows that the MR in Elk Creek has more
variance than that in Soup Creek. The greater variation is predominately explained by
the greater variance in LAI and forest biomass in Elk Creek. MR for both the
simulated soil water and the unlimited soil water RHESSys runs was identical,
because soil moisture content does not affect the model's computation of MR (Fig.
3b). Differences in PSN between the stressed and unstressed plots are due to
reductions in Pg caused by water stress induced by low soil matric potentials (Fig. 3a)
and stomatal closure (Fig. 3b).
Figure 4 shows that when soil moisture is not limited, increases in MR are
accompanied by an increase in PSN, although the PSN increases appear asymptotic at
high MR values. The asymptotic response of PSN to increased MR is primarily due to
the fact that PSN reaches a threshold with leaf biomass, while MR is linearly related
to biomass. Pg is driven by Q (Eq. 2) and FOREST-BGC uses a Beer's Law attenuation
of Q (Fig. 3b) in which radiation is asymptotically absorbed with increasing LAI
(Running and Coughlan 1988). At higher LAIs, the total amount of Q absorbed per
ground area increases, but absorption per unit of leaf area decreases, so that the
North Fork Elk Creek, water-limited case
North Fork Elk Creek, non-waterlimited case
2,0,10 4 Soup Creek, water-lim~ed case
Soup Creek, non-water-lIm~ed case
>. 1.5'104
Cll
J:= ..
Cl
~
(/)
'iii
~ 1.0.104
C
>.
(/)
.9 Ii 2.lp"ti~........... rT:lr~~-"
0 ~ ...
J:=
~ 5,0'103
z
1000 2000 3000 4000 5000 6000 7000 8000

Maintenance Respiration (kg ha" y(l)
Figure 4, Plots of annual maintenance respiration and net photosynthesis with and without water limitations for two western
Montana watersheds. Each is represented at 90 m2 with every tenth cell plotted. UI
l..>J
-
benefits of additional LAI grow increasingly marginal. As LAI increases, MR

increases linearly but P g increases asymptotically. This relationship defines a
theoretical limit to the amount of LAI that can be supported even without water
limitations. In the unstressed plots, no point of inflection (maximum value) for MR
was observed, indicating that Q does not limit LAI and that with a wetter climate,
additional LAI could be supported.
RHESSys simulations incorporating soil water limitations show increases in MR
(and biomass) eventually leading to decreases in PSN.Benefits of increasing LAI and
per-unit-ground-area Pg are offset by increased losses resulting from the maintenance
of additional leaf mass. The MR points of inflection are at 300 (kg ha-' yr-') in Elk
Creek and at 550 (kg ha- 1 y{') in Soup Creek. Differences in MR inflection values
between watersheds are primarily explained by the wetting and drying properties of
the climate. Soup Creek receives twice the annual precipitation of Elk Creek and also
has a slightly longer winter season. These factors reduce annual evaporative demand
which, despite cooler temperatures, can increase seasonal PSN (Running 1984). With
annual variations in climate, a shifting can occur along the MR and PSN axes as
complex biotic and abiotic interactions affect PSN and MR in nonlinear and unequal
directions.
Soil water storage can buffer the effects of annual climatic variations. It postpones
the onset of water stress in coniferous forests caused by periodic or seasonal summer
drought (Waring and Franklin 1979). Soil moisture content is primarily determined by
soil water-holding capacity and factors regulating its inputs and outputs, including
snow hydrology and forest cover (Coughlan and Running in press). Soil moisture must
be replenished annually so that the net change in storage over time is zero; this usually
occurs in the winter and spring.
Conclusions
The ability to obtain and accurately summarize major data themes (e.g., climate and
soils) is important to forest modeling and prediction. In this chapter, we illustrated
RHESSys' sensitivity to soil moisture and the interrelationships between that data
theme and the data themes of vegetation and climate. Extensive efforts in one domain
(such as development of a transfer function for remotely sensed data) to improve a
single data layer without concomitant increases in accuracy and precision in the
supporting data layers may be futile. For this reason, it is important to consider the
data collection process as an integrated effort. The unifying strategy should be
summarized by models that are the integrating tools through which data are
transformed into a geographically referenced result. Model sensitivity and limitations
in both ecosystem knowledge and software engineering should guide data collection
efforts. RHESSys is an example of one system that can be used to define which data
are needed and to determine the level of precision necessary in the collection process.
Many challenges are encountered in the use of remote sensing to aid the spatial
prediction of forest processes. As summarized here, there are many current alternative
approaches, none of which is without flaws. While research in methods of
parameterization continues, including development of more accurate transfer
155
functions and improved aggregation methods, there is less emphasis on creative means
of validating models on scales that cannot be measured directly. While validation may
be a difficult goal, progress will likely be made through increased attention to
consistency in modeling decisions and to the development of new approaches to the
description of uncertainty associated with the accuracy of spatial predictions.
Acknowledgments - We thank Ramakrishna Nemani, David Peterson, Lawrence Band

and Steve Running. Diane Wickland of NASA's Ecological Processes and Modeling
Branch supported this work.
References
Abbott, M.B., Bathurst, J.e., Cunge, J.A., O'Connell, P.E. and Rasmussen, J. 1986. An
introduction to the European hydrological system - Systeme Hydrologique Europeen,
"SHE": History and philosophy of a physically-based distributed modeling system. - J.
Hydro!. 87: 45-59.
Band, L.E. 1986. Topographic partition of watershed with digital analysis of models. - Water
Res. Res. 22: 15-24. .
Band, L.E. 1993. Effect ofland surface representation on forest water and carbon budgets. - J.
Hydro!. 150: 749-772.
Band, L.E., Running, S.w., Peterson, D.L., Lammers, R., Dungan, J.L. and Nemani, R.R.
1991. Forest ecosystem processes at the watershed scale: Basis for a distributed mode!. -
Eco!. Mode!. 56: 171-196.
Band, L.E., Patterson, P., Nemani, R.R. and Running, S.w. 1993. Forest ecosystem processes
at the watershed scale - Incorporating hillslope hydrology. - Agric. For. Meteoro!. 63:
93-126.
Beven, K.J. and Kirkby, M.J. 1979. A physically based, variable contributing area model of
basin hydrology. - Hydro!. Sci. Bull. 24: 43-69.
Bonan, O.B. 1991. Atmosphere-biosphere exchange of carbon dioxide in boreal forests. - J.
Oeophys. Res. 96: 7301-7312.
Bristol, K.L. and Campbell, O.S. 1984. On the relationship between incoming solar radiation
and daily maximum and minimum air temperature. - Agric. For. Meteoro!. 31: 159-166.
Buffo, J., Fritschen, L. and Murphy, J. 1972. Direct solar radiation on various slopes from 0
to 60 north latitude. - USDA Forest Service, Research Paper PNW-142, Pacific Northwest
Forest and Range Experiment Station, Portland, OR. 74 pp.
Burrough, P.A. 1986. Digital elevation models. - In: Principles of GIS for Land Resources
Assessments. Oxford University Press, Oxford, UK, pp. 39-56.
Campbell,O.S. 1977. An Introduction to Environmental Biophysics. - Springer-Verlag, New
York. 159 pp.
Carlson, T.N., Perry, E.M. and Schmugge, TJ. 1990. Remote estimation of soil moisture
availability and fractional vegetation cover for agricultural fields. - Agric. For. Meteoro!.
52: 45-69.
Cary, E. and Rosenzweig, e. 1987. Determination of vegetated fraction of surface from
satellite measurements. - Adv. Space Res. 7: 77-80.
Coughlan, J.C. 1991. "Biophysical aggregations of a forested landscape." - Ph.D. dissertation,
University of Montana, Missoula.
Coughlan, J.C. and Running, S. W. 1989a. Variable landscape aggregation for large scale
watershed evaporation estimates. - In: Symposia Proceedings on Headwaters Hydrology,
June 27-30, 1989, Missoula, MT. AWRA, Bethesda, MD, pp. 75-82.
Coughlan, J.e. and Running, S.w. 1989b. An expert system to aggregate biophysical attributes
of a forested landscape within a geographic information system. AI Applications. - Nat.
Res. Manage. 3: 35---43.
Coughlan, J.C. and Running, S.w. 1995. Regional ecosystem simulation: A general model for
simulating snow accumulation and melt in mountainous terrain. - Land. Ecol. (in press).
Dickenson, R.E., Henderson-Sellers, A., Kennedy, P.l. and Wilson, M.F. 1986.
Biosphere-Atmosphere Transfer Scheme for the NCAR community climate model. -
NCAR, Technical Note NCARffN-275+STR, Boulder, CO. 72 pp.
Donner, B.L. and Running, S.W. 1986. Water stress response after thinning Pinus canlarla
stands in Montana. - For. Sci. 32(3): 614-625.
Dougherty, D. 1990. sed & awk. - O'Reilly & Associates, Inc., Sebastopol, CA. 414 pp.
Dungan, J.L., Peterson, D.L. and Curran, P.J. 1994. Alternative approaches for mapping
vegetation quantities using ground and image data. - In: Michener, W., Brunt, J. and
Stafford, S. (eds). Environmental Information Management and Analysis: Ecosystem to
Global Scales. Taylor & Francis, London, pp. 237-261.
Ehleringer, J.R and Field, C.B. (eds). 1993. Scaling Physiological Processes: Leaf to Globe.-
Academic Press, San Diego, CA. 388 pp.
Foody, G. and Curran, P. (eds). 1994. Environmental Remote Sensing From Regional to Global
Scales. - J. Wiley and Sons, Chichester, UK. 238 pp.
Gamier, B.J. and Ohmura, A. 1968. A method of calculating the direct shortwave radiation
income of slopes. - J. Appl. Meteorol. 7: 796---800.
Gholz, H.L., Curran, P.I., Kupiec, J.A. and Smith, G.M. 1996. Assessing leaf area and canopy
biochemistry of Florida pine plantations using remote sensing. - In: Gholz, H.L., Nakane,
K. and Shimoda, H. (eds). The Use of Remote Sensing in the Modeling of Forest
Productivity. Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 3-22.
Goel, N.S. and Deering, D.W. 1985. Evaluation of a canopy reflectance model for LAI
estimation through its inversion. - IEEE Trans. Geosci. Rem. Sens. 23: 674-684.
Goodchild, M.E, Parks, B.O. and Steyaert, L.T. 1993. Environmental Modeling with GIS. -
Oxford University Press, London. 488 pp.
Graetz, D. 1990. Remote sensing of terrestrial ecosystem structure: An ecologist's pragmatic
view. - In: Hobbs, R.I. and Mooney, H.A. (eds). Remote Sensing of Biosphere Functioning.
Springer-Verlag, New York, pp. 5-30.
Hungerford, RD., Nemani, RR, Running, S.w. and Coughlan, J.e. 1989. MTCLIM - A
mountain microclimate simulation model. - USDA Forest Service, Research Paper INT-
414, Intermountain Research Station, Ogden, UT. 52 pp.
forests using geometric optical modelling. - In: Gholz, H.L., Nakane, K. and Shimoda, H.
(eds). The Use of Remote Sensing in the Modeling of Forest Productivity. Kluwer
Lathrop, RG.I., Aber, J.D., Bognar, JA, Ollinger, S.Y., Casset, S. and Ellis, J.M. 1994. GIS
development to support regional simulation modelling of north-eastern (USA) forest
ecosystems. - In: Michener, W., Brunt, J. and Stafford, S. (eds). Environmental Information
Management and Analysis: Ecosystem to Global Scales. Taylor & Francis, London, pp.
431---451.
157
Lee, T.J., Pielkie, R.A., Kittel, TG.F. and Weaver, J.F. 1993. Atmospheric modeling and its
spatial representation of land surface characteristics. - In: Goodchild, M.F., Parks, B.O. and
Steyaert, L.T (eds). Environmental Modeling with GIS. Oxford Press, New York, pp.
108-122.
Leprieur, c., Verstraete, M.M., Pinty, B. and Chehbouni, A. 1995. NOAAlAVHRR vegetation
indices: Suitability for monitoring fractional vegetation cover of the terrestrial biosphere. -
In: Proceedings of the Sixth International ISPRS Symposium on Physical Measurements
and Signatures in Remote Sensing, January 17-21, 1995, Val d'Isere, France, pp.
1103-1110.
Lohammar, T, Larsson, S., Linder, S. and Falk, S.O. 1980. FAST - Simulation models of
gaseous exchange in Scots pine. - Ecol. Bull. (Stockholm) 32: 505-523.
Loveland, TR., Merchant, J.w., Ohlen, D.O. and Brown, J.F. 1991. Development of a land
cover characteristics database for the conterminous U.S. - Photogramm. Engin. Rem. Sens.
57: 1453-1463.
Martin, M.E. and Aber, J.D. 1996. Estimating forest canopy characteristics as inputs for
H.L., Nakane, K. and Shimoda H. (eds). The Use of Remote Sensing in the Modeling of
McLeod, S.D. and Running, S.w. '1987. Comparing site quality indices and productivity in
ponderosa pine stands of western Montana. - Can. 1. For. Res. 18: 346-352.
McMurtrie, R.E., Leuning, R., Thompson, w.A. and Wheeler, A.M. 1992. A model of canopy
photosynthesis and water use incorporating a mechanistic formulation of leaf CO 2
exchange. - For. Ecol. Manage. 52: 261-278.
Nemani, R.R. and Running, S. W. 1989. Estimation of regional surface resistance to
evapotranspiration from NDVI and thermal-IR AVHRR data. - J. Appl. Meteorol. 28:
276-284.
Nemani, R.R., Pierce, L.L., Running, S.W. and Goward, S.N. 1993a. Developing
satellite-derived estimates of surface moisture status. - J. App!. Meteor. 32: 548"':557.
Nemani, R.R., Pierce, L.L., Running, S.W. and Band, L.E. 1993b. Forest ecosystem processes
at the watershed scale: Sensitivity to remotely-sensed leaf area index estimates. - Int. J.
Rem. Sens. 14: 2519-2534.
Nemani, R.R., Band, L.E., Running, S.W. and Peterson, D.L. 1993c. Regional
HydroEcological Simulation System: An illustration of the integration of ecosystem
models in a GIS. - In: Goodchild, M.F., Parks, B.O. and Steyaert, L.T. (eds). Environmental
Modeling with GIS. Oxford University Press, New York, pp. 296-304.
Parton, w.J., Schimel, D.S., Cole, c.v. and Ojima, D.S. 1987. Analysis of factors controlling
soil organic levels in Great Plains grasslands. - Soil Sci. Soc. Am. 1. 51: 1173-1179.
Rosema, A., Verhoef, W., Noorbergen, H. and Borgesius, J.J. 1992. A new forest light
interaction model in support of forest monitoring. - Rem. Sens. Environ. 42: 23-41.
Running, S.W. 1984. Microclimate control of forest productivity: Analysis by computer
simulation of annual transpiration and photosynthesis balance in differing environments. -
Agric. For. Meteorol. 23: 267-288.
Running, S.W. and Coughlan, J.C. 1988. A general model of forest ecosystem processes for
regional applications. 1. Hydrological balance, canopy gas exchange and primary
production processes. - Ecol. Model. 42: 125-154.
Running, S.w. and Gower, S.T 1991. FOREST-BGC, a general model of forest ecosystem
processes for regional applications. 2. Dynamic carbon and nitrogen budgets. - Tree
Physiol. 9: 147-160.
Running, S.W., Nemani, R.R. and Hungerford, R.D. 1987. Extrapolation of synoptic
meteorological data in mountainous terrain, and its use for simulating forest
evapotranspiration and photosynthesis. - Can. J. For. Res. 17: 472-483.
Running, S.w., Nemani, RR., Peterson, D.L., Band, L.E., Potts, D.F., Pierce, L.L. and
coupling satellite data with ecosystem simulation. - Ecology 70(4): 1090-1101.
Running, S.w., Loveland, T.R. and Pierce, L.L. 1994. A vegetation classification logic based
on remote sensing for use in global biogeochemical models. - Ambio 23: 77-81.
Ryan, M.G. 1991. Effects of climate change on plant respiration. - Ecol. Appl. I: 157-167.
Saxton, K.E., Rawls, w.J., Romberger, J.S. and Papendick, R.I. 1986. Estimating generalized
soil-water characteristics from texture. - Soil Sci. Soc. Am. J. 50: 1031-1036.
Schneider, D.C. 1993. Quantitative Ecology: Spatial and Temporal Scaling. - Academic Press,
San Diego, CA. 395 pp.
Sellers, P.F. and Dorman, J.L. 1987. Testing the Simple Biosphere model (SiB) using point
micrometeorological and biophysical data. - J. Clim. Appl. Meteorol. 26: 622-651.
Townshend, J.RG. 1994. Global data sets for land applications from the Advanced Very High
Resolution Radiometer: An introduction. -Int. J. Rem. Sens. 15(17): 3319-3332.
USGS. 1990. Digital Elevation Models. National Program Technical Instructions, Data Users
Guide 5. - U.S. Geologic Survey, Reston, VA. 40 pp.
Waring, RH. and Franklin, J.F. 1979. Evergreen coniferous forests of the Pacific Northwest. -
Science 204: 1380-1386.
Wessman, c.A., Aber, J.D., Peterson, D.L. and Melillo, J.M. 1988. Remote sensing of canopy
Wigmosta, M.S., Vail, L.w. and Lettenmaier, D.P. 1994. A distributed hydrology-vegetation
model for complex terrain. - Water Res. Res. 30: 1665-1679.
Zhu, A. and Band, L.E. 1994. A knowledge-based approach to data integration for soil
mapping. - Can. J. Rem. Sens. 20(4): 408-418.
SEVEN
Forest Vegetation Classification and Biomass Estimation

Based on Landsat TM Data in a Mountainous Region of West
Japan
Nam J. Lee and Kaneyuki Nakane
Lee, N.J. and Nakane, K. 1996. Forest vegetation classification and biomass
estimation based on Landsat TM data in a mountainous region of west Japan. - In:
Modeling of Forest Productivity. Kluwer Acad. Pub!., Dordrecht, The Netherlands,
pp. 159-171.
Landsat Thematic Mapper (TM) data corrected topographically with the aid of digital
terrain data were applied to the classification and mapping of forest vegetation and the
estimation of its biomass in a mountainous region of Hiroshima Prefecture in west
Japan.
Topographic correction was made based on the relationship between cos r (the
solar incidence angle relative to the local terrain slope) and practical radiance response
(digital number of TM). The forest vegetation was well classified into three forest
types, including deciduous broadleaf forest, pine forest and Japanese cedar plantation,
and into two nonforest types, c1earcut area and cultivated land area, using the decision
tree classification method based on corrected TM data. Classification accuracies for
each vegetation type increased by 8-11 % when corrected TM data were used instead
of uncorrected data. Four vegetation indices were evaluated. Linear relationships were
observed between two vegetation indices and forest biomass. However, the coefficient
values of these relationships were not identical among the vegetation types. The
correlation coefficient (r) between the Normalized Difference Vegetation Index
(NDVI) and biomass for the pine forest was 0.85; correlation coeffi-
cients between the Differential Vegetation Index (DVI, Band 5 - Band 7) and biomass
for the Japanese cedar plantation and the deciduous broadleaf forest were -0.83 and
0.80, respectively. Based on the linear relationships, above-ground biomass for all
vegetation types was estimated and mapped. Mean biomass for the pine, Japanese
cedar and deciduous broadleaf forests wa<; estimated to be about 143, 135 and 121 t
ha- l, respectively, and the mean and total biomass of forest vegetation within the study
area (2040 ha) were estimated to be about 133 t hal and 275.0 x 103 t, respectively.
N. J. Lee and K. Nakane, Department ofEnvironmental Studies, Faculty ofIntegrated

Arts and Sciences, Hiroshima University, Higashi-Hiroshima 724, Japan.
159
Introduction
Landsat Thematic Mapper (TM) data have been applied successfully to the
classification and mapping of structural and functional characteristics of forest
vegetation. However, few studies have reported the estimation and classification of
forest biomass for mountainous regions (Senoo et al. 1983).
There have been some attempts at determining canopy types, heights and densities
of forest stands by remote sensing methods (Nelson et al. 1988). Since canopy
structure in a given forest stand is related to other vegetation parameters such as
biomass, age and density (Rock et al. 1986, Peterson et al. 1987), the spectral
information of a canopy may provide correlative assessments of structural and
functional features of the stand. However, forest vegetation radiance responses
recorded in Landsat data vary with such environmental factors as terrain, weather and
vegetation distribution patterns. Thus, spectral responses of various vegetation types
should be corrected on the basis of not only topographic and atmospheric conditions,
but also surrounding factors that may affect the relationship between the sensor and
the target (Duggin 1983). Spectral properties in mountainous regions, in particular,
generally require topographic correction because of the variation in radiance from
inclined surfaces compared with radiance from horizontal surfaces (Holben and
Justice 1981, Walsh 1987). These topographic corrections can decrease the probability
of misclassification, which also has been well demonstrated (Eliason et al. 1981,
Justice et al. 1981, Horler and Ahern 1986, Cavayas 1987, Chavez 1988, 1989).
Topographic correction using DTM (Digital Terrain Model) data, based on the
practical relationship between slope data and radiance responses to forest vegetation,
has rarely been reported. Thus, in this study we obtained not only theoretical but also
practical relationships between slope data and radiance responses and corrected TM
data topographically.
We derived two vegetation indices, the Normalized Difference Vegetation Index
(NDVI) and the Differential Vegetation Index (DVI), as possible indicators for
classifying forest vegetation and estimating forest biomass. These indices are based on
the fact that the chlorophyll pigments in vegetation canopies absorb red light and
reflect infrared light (Sellers 1985, 1987, Lillesand and Kiefer 1987, Peterson et al.
1988). The ratio between infrared and red radiation is a sensitive indicator of green
biomass (Tucker 1979). By exploiting the vegetation indices, forest vegetation can
potentially be quantitatively estimated and mapped.
In this study, we evaluated the validity and effectiveness of classifying forest
vegetation using topographically corrected TM data in a mountainous forest zone of
west Japan. We then attempted to estimate and map forest biomass based on the
relationship between the vegetation indices and biomass data obtained on the ground.
161
Study site and methods
Description of the study area

The study area was about 24 km 2 in and around Mt. Ohomine, located about 20 km
southwest of Hiroshima City in west Japan (Fig. 1). Mt. Ohomine, which stands in the
central part of the study area, has an altitude of 1040 m. The southern slope of Mt.
Ohomine is steep, with a mean gradient of about 30. During the last decade, the
annual rainfall and mean annual temperature in the study area ranged from 1500 to
2000 mrn yr- 1 and from 8 to 12C, respectively (Hiroshima Prefecture 1989).
The study area was within a mid-temperate forest province that featured four main
types of land use: residential area, cultivated land, pasture and forest. However, most
of the study area was covered with forest vegetation, including the plantation of
Japanese cedar (Cryptomeria japonica) and cypress (Chamaecyparis obtusa). The
principal forest vegetation types were deciduous broadleaf forest, pine (Pinus
densiflora) forest and Japanese cedar plantation. The dominant floristic components
of the canopy in the deciduous broadleaf forest were Quercus serrata, Castanea
crenata and Carpinus laxiflora.
40 N+--t--t---i
Sludyare
Figure 1. Map of the study area. The topographic map below provides a
magnified view of the study area of Hiroshima Prefecture above.
Basic concepts and methods for topographic correction

The radiance (L) of any wavelength recorded by a satellite sensor is expressed by
L = T x A x R x cos r + Lp + Ln , (1)
where T is the atmospheric transmittance, A is the total downwelling radiance and R is
the reflectance of the target. L p is the atmospheric path radiance. L n is the reflectance
from the surrounding area of the object. The value of L n may be small and difficult to
measure because the variables are subject to surrounding effects. Therefore, L n may be
omitted. Cos r is the solar incidence angle relative to the local terrain angle. The
radiance (L) may be regarded as changing only with R, assuming that the values of T,
A, Lp> and cos r remain constant. Eq. (1) can be generally used as the radiance recorded
in Landsat TM data. This is the equation used for topographic correction of Landsat
TM data in mountainous regions, based on the Lambertian surface (Senoo et al. 1983,
Sjoberg and Horn 1983, Kawata et al. 1988).
However, cos r values are not constant in mountainous regions. The reflectance
observed in mountainqus regions tends to follow Minnaert's law based only on the
slope gradient of the target, rather than Lambert's law (Kawata et al. 1991). Path
radiance (Lp ) also may vary depending on atmospheric conditions. Thus, atmospheric
path radiances (Lp ) should be required for each band; these can be calculated using the
regression method. The offset values on each band in the study area may be considered
nearly equal to those of path radiance (Mather 1987, Lillesand and Kiefer 1987,
Chavez 1988, 1989). Therefore, by eliminating the effects of cos r and atmospheric
path radiance (Lp ) in mountainous regions, the corrected radiance (L) of the target can
be derived as follows:
(2)
where L is the observed radiance of a target pixel. Thus, the corrected reflectance (L)
can be derived from Eq. (2), if the cos r value or slope gradient in the mountainous
region is known. Eq. (2) does not take multiple reflection and diffuse reflectance by
the surrounding area into consideration.
Image data
The image data used for this study were Landsat TM data collected on 8 May 1987
(path 112, row 36). The study area extracted from the TM data was composed of 230
lines by 165 pixels. The thermal band (Band 6) was not used because of its poor spatial
resolution and the low contrast in the forest area. The elevation and azimuth of the sun
were 58 and 117, respectively.
Ancillary data
For topographic analysis of such variables as elevation, slope and direction, DEM
(Digital Elevation Model) data were produced based on the 1/25,000 topographic map.
The grid size of the DEM was 28.5 m x 28.5 m, corresponding to one pixel of the
Landsat TM image. The vegetation map based on ground measurements and aerial
photographs was used to discriminate among the vegetation types in each training area.
163
A D1M was produced from DEM data using elevations of the corresponding pixel
locations. The values of cos r were calculated from the solar incidence angle relative
to the local terrain angle, based on the altitude and azimuth of the sun.
Flow of study
Registration. The Landsat 1M image data were geometrically corrected by means of
an Affine transformation method, using 18 ground control points. The
root-mean-square error was within one pixel. Resampling for registering the Landsat
1M image and the DEM data was carried out by nearest-neighbor analysis.
Training area selection. Several training areas were selected for each vegetation
type: 15 for the deciduous broadleaf forest, nine for the pine forest, nine for the
Japanese cedar plantation, six for the clearcut area and five for the cultivated land area.
The size of each training area was represented by a mean digital number designating
either a 3-x-3 or a 4-x-4 matrix of pixels. The vegetation type of each training area was
confirmed in the field.
Analysis technique and classification accuracy. Classification was accomplished
by means of either a binary decision tree method, a threshold method or both.
Unknown pixels were classified using one or more decision functions (Swain and
Hauska 1977). The statistics of the training data sets were computed (i.e., the mean
and covariance matrix were plotted for each training class).
Classification accuracy was estimated by means of the contingency table as a
proportion of the sum of the nondiagonal and diagonal entries for each class (Belward
and DeHoyos 1987).
Plant biomass measurement. The height (H) and dbh (diameter at breast height) of
all trees (dbh >1.0 cm) were measured on plots selected within each training area that
were 15 m x 15 m or 20 m x 20 m. The age of the forest stand was also measured
dendrochronologically using trees at the canopy level. Biomass and forest stand age
were estimated on eight plots in the deciduous broadleaf forest, nine plots in the pine
forest and nine plots in the Japanese cedar plantation. Estimation of forest biomass
was carried out using allometric relationships (Ando et al. 1968, Nakane et al. 1984,
Miura 1992) based on Hand dbh data.
Computation of vegetation indices (VIs). Vegetation indices (VIs) used here were
calculated from the relationships between bands as follows:
NDVI = (Band 4 - Band 3) / (Band 4 + Band 3),
RVI = (Band 4 - Band 3), (3)
DVI =(Band 5 - Band 7),
ND = (Band 7 - Band 5) / (Band 7 + Band 5).
Estimation and mapping of forest biomass. Above-ground biomass was estimated
from the relationship between VIs and biomass measured in the vegetation plots.
Estimated forest biomass was mapped at intervals of 50 t ha-'.
Results and discussion
Topographic correction
Figure 2 shows the relationships between radiance response (L) and cos r for each
vegetation type and band. In this figure, digital number (DN) was used instead of
radiance response. As Figure 2 illustrates, there is a linear relationship between
radiance response and cos r; this was predicted by Eq: (2). Hence, the y intercept
corresponds to Lp expressed in DN values. The correlation coefficients are high, except
the one for the broadleaf forest in Band 4, as shown in Table 1. The fact that Lps
estimated for different vegetation types coincide strongly supports the effectiveness of
Eq. (2). Bands 4 and 5 were greatly affected by cos r, while the other bands changed
only slightly. From these results, it can be seen that Bands 4 and 5 must be corrected
topographically. DN changed most significantly with cos r in the deciduous broadleaf
forest, but less in the Japanese cedar plantation, which may be due to the differing
optical properties of the leaves. As can be seen in Table 1, the correlation coefficient
in
(r) values were higher the pine forest than in the broadleaf forest. The relatively low
correlation coefficient in the deciduous broadleaf forest may be due to structural
variations in the forest canopy.
120
Band 3 Band 4 .. .
....
.. 0\.-
/ ..
f./4
.....
8 V ..
/" _.. _.I-
~:~--
,. ~-
40
~
Lp
.,
~ Lp ----
~
E 0
::l
z 120
~
'c;,
a
Band 5 .........
fl' Band 7
80 .. ~
.?'~ 0
40 /'
"'-----

-_ ........
~
0 0
0 0
~ .. ' -
Lp
00 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
cos .., cos 'Y
Figure 2. Relationships between cosine r and digital number (DN) on Bands 3, 4,

5 and 7 in the training areas of each forest vegetation type. Landsat TM data used
in this study were obtained on 8 May 1987. L p is the atmospheric path radiance.
A- - - - -A Deciduous broadleaf forest 0-------0 Pine forest
- - - - - - - Japanese cedar plantation
165
Table 1. Correlation coefficients for the linear regressions between cos rand DN
for Bands 3, 4, 5 and 7 obtained in the training areas of each forest type.
Bands
Forest types Band 3 Band 4 Band 5 Band 7
Deciduous 0.60 0.42 0.82* 0.61

Pine 0.91 ** 0.91 ** 0.81* 0.72
Japanese cedar 0.84* 0.83* 0.93** 0.85*
* Indicates significance at the p < 0.05 level.

** Indicates significance at the p < 0.01 level.
Forest vegetation classification

The original data were corrected1opographically according to Eq. (2). A decision tree
classification method was used to classify forest vegetation as follows. Forest
vegetation types distributed in areas lower than 800 m above sea level (a.s.l.) were
separated from those in the nonforest area by the threshold of 35 on Band 3. The
threshold of 60 on corrected Band 5 divided the forest vegetation into deciduous and
coniferous forest; coniferous forest was further divided into Japanese cedar plantation
and pine forest by the threshold of 32 on corrected Band 5 (Fig. 3). It was easy to
separate the Japanese cedar plantation from the pine forest, possibly because the pine
forest had less transpiration (Keil et al. 1990) or less leaf biomass than did the
Japanese cedar plantation.
I
I
Clearcut area Cultivated
land area Corrected Band 5 > 32
"
Pine "
Japanese cedar
Figure 3. Vegetation types classified using the decision tree classification method
based on topographically corrected Landsat TM data. - - - : YES - - - - : NO
Classification accuracy
Classification accuracies for forest vegetation types based on the topographically
corrected TM data increased for nearly all classification types, compared with those
based on the uncorrected data (Table 2). Classification accuracies increased from
81.3% to 87.1 % for the deciduous broadleaf forest, from 74.3% to 85.4% for the
Japanese cedar plantation and from 73.4% to 82.0% for the pine forest. The overall
classification accuracies for the study area improved from 75.1 % to 81.2%, indicating
the effectiveness of topographically correcting TM data on vegetation classification in
mountainous regions. Although misclassification frequently occurred in the deciduous
broadleaf forest and the pine forest, it was mainly due to the mixed characteristics of
the forest canopy. Seasonal changes in the TM data also tended to decrease
classification accuracy (Hudson 1987); the incidence of misclassification may be
reduced further by including an analysis of phenologies of the two forest types
(Nelson et al. 1984, Mather 1987).
Table 2. Classification accuracy of each vegetation type based on corrected TM

data obtained in the training areas.
Japanese Clearcut Cultivated

Classification Broadleaf Pine cedar area land area Total Accuracy (%)
Broadleaf 209 16 2 13 0 240 87.1 (81.3)

Pine 18 105 5 0 0 128 82.0 (73.4)
Japanese cedar 0 21 123 0 0 144 85.4 (74.3)
Clearcut area I I 0 64 30 96 66.7 (61.5)
Cultivated land area 0 0 0 12 68 80 85.0 (85.0)
Total 228 143 130 89 98 688 81.2 (75.1)
Note: Figures in parenthesis show the percentage of classification accuracy based

on the uncorrected TM data obtained in the training areas.
Distribution and mapping of forest vegetation

The deciduous broadleaf forest was mainly distributed over the portion of the
mountain area higher than 800 m a.s.l., which occupied about 24% of the study area.
The Japanese cedar plantation was distributed in competition with the pine forest in
the area less than 800 m a.s.l., but the pine forest was more widely distributed in the
lower area than was the Japanese cedar plantation.
According to the vegetation map, the distribution areas occupied by the pine forest,
the deciduous broadleaf forest and the Japanese cedar plantation made up about 40%,
25.9% and 22% of the study area, respectively. The cultivated land area and clearcut
area occupied about 3% and 10% of the study area, respectively (Plate 1).
Vegetation index (VI) and forest biomass

Various VIs were calculated from the spectral responses ofTM data extracted from the
training areas, and relationships between VIs and forest biomass were obtained (Figs.
4, 5 and 6). The correlation coefficients between them are also given in Table 3. As
shown in Figure 4 and Table 3, NDVI and RVI were good indicators of pine forest
167
biomass, with correlation coefficient values of 0.85 and 0.85, respectively. The
biomass of the deciduous broadleaf forest and the Japanese cedar plantation had less
significant correlations with either NDVI or RVI, but relatively high correlations with
DVI (r = 0.80 and -0.83, respectively). The pine forest was composed of relatively
similar age classes and generally was fully stocked with uniform canopy heights,
crown types and tree diameters compared with the deciduous broadleaf forest, which
resulted in relatively high sensitivity to NDVI (Sader et al. 1989). Although the
Japanese cedar plantation had relatively similar age classes and uniform canopy
heights and tree diameters, the radiance response was more scattered in the Japanese
cedar plantation than in the pine forest, owing to the latter's sharp triangular crown
type. DVI regions are more sensitive to vegetation density than to leaf moisture
content and color, especially in the early stages of regeneration (Horler and Ahern
1986). In the Japanese cedar plantation, changes in DVI with biomass possibly can be
attributed to the influences of shadowing and leaf biomass. They also might be
considered the results of negative correlation.
Table 3. Correlation coefficients (r) for the linear regression between vegetation
indices and the biomass of each forest type measured in the training areas.
Vegetation indices (VIs)
Forest biomass NDVI RVI ND DVI
Broadleaf -0.32 -0.32 0.54 0.80

Pine 0.85 0.85 -0.58 -0.65
Japanese cedar -0.49 -0.50 0.51 -0.83
NDVI: (Band 4 - Band 3)/(Band 4 + Band 3); RVI: (Band 4 / Band 3)

ND: (Band 7 - Band 5)/(Band 7 + Band 5); DVI: (Band 5 - Band 7)
Linear relationships between VIs and above-ground biomass were derived as

follows:
Pine forest biomass = 3.72 x NDVI - 3.4, (4)
Deciduous broadleaf forest biomass = 7.70 x DVI - 213.6, (5)
Japanese cedar biomass =-11.26 x DVI + 541.7. (6)
Estimated forest biomass

An overall forest biomass for the study area was estimated based on the relationships
(Figs. 4, 5 and 6) between VIs and forest biomass measured in the training areas (Table
4). Pine forest biomass ranged from 50 to 250 t ha'l. The mean and total areal biomass
were estimated to be about 143 t ha'l and 135.0 x 103 t, respectively. Pine forest
biomass was mainly distributed from 100 to 200 t ha", occupying about 96% of the
pine forest area (Plate 2(b)). This indicated that the pine forest was composed of
200

::; 150
.c

"'"''"
E
~ 100
Y = 3.72X - 3.4
r = 0.85
35 55
NDVI
Figure 4. The relationship between NDVI and pine forest biomass.
300

::l 200
'"
E
o
iii
Y =-11.26X 541.7
= -0.83
100
r

25 30 35 40
DVI
Figure 5. The relationship between DVI and the biomass of the Japanese cedar
plantation. DVI: (Band 5 - Band 7).
200

'"
.c

-; 100
E
~
o
iii
Y = 7.70X - 213.6
r = 0.80

30 35 40 45 50 55
DVI
Figure 6. The relationship between DVI and the biomass of the deciduous
broadleaf forest. DVI: (Band 5 - Band 7).
169
relatively even-aged crowns and tree diameters. Biomass for the deciduous broadleaf
forest was evenly distributed from 50 to 250 t ha- 1. The mean and total biomass of the
deciduous broadleaf forest were estimated to be about 121 t ha- 1 and 69.1 x 103 t,
respectively. In the Japanese cedar plantation, the mean and total biomass were
estimated to be approximately 135 t ha- 1 and 71.0 x 103 t, respectively. The biomass of
the deciduous broadleaf forest and the Japanese cedar plantation was evenly
distributed from 50 to 250 t ha 1. The mean and total biomass of the forest area (2040
ha) within the study area were estimated to be 133 t ha 1 and 275.0 x 103 t, respectively
(Table 4).
Table 4. Number of test pixels of each forest type and biomass in the study area
estimated from the relationships between vegetation indices and the biomass of
each vegetation type.
Number of pixels
Biomass classes (t ha") Average Total
biomass biomass
Forest types 0-50 -100 -150 -200 -250 -300 Total (tha") (xlOJ t)
Broadleaf 2,036 1,698 2,105 1,506 1,699 95 9,137 121 1.11

Pine 67 515 8,628 5,884 22 15,116 143 2.17
Japanese cedar 2,061 920 1,284 1,553 2,578 2 8,398 135 1.14
Total 4,164 3,131 12,017 8,943 4,299 97 32,651 133 4.41
Estimated forest biomass for each of the vegetation types was then mapped (Plate
2) using six biomass classes ranging from 0 to 300 t ha'l.
Conclusions
Using Landsat TM data corrected topographically by DTM data in a mountainous
region in the temperate forest zone, forest vegetation was well classified and mapped.
Above-ground biomass was also estimated and mapped based on relationships
between vegetation indices (NDVI or DVI) and biomass data obtained on the ground.
Topographic correction between cos r and radiance response was used for
classifying the forest vegetation. Topographic correction was clearly effective on
Bands 4 and 5. The correlation coefficient value was higher in the pine forest than in
the broadleaf forest. Such differences are often caused by type variations within the
canopy structure and by topographic orientation. Classification accuracies for the
corrected TM data increased by approximately 8-11 % for all classification types,
compared with those based on uncorrected data.
Forest vegetation was classified into three forest types, including deciduous
broadleaf forest, pine forest and Japanese cedar plantation, and into two nonforest
types, clearcut area and cultivated land area. Forest and nonforest area were clearly
divided by the threshold values of Band 3. Corrected Band 5 was successfully used to
differentiate the deciduous broadleaf forest from the pine forest, and the pine forest
from the Japanese cedar plantation.
Relationships between vegetation indices and biomass data were useful for
estimating forest biomass. NDVI, in particular, had a good linear relationship with
pine forest biomass, demonstrating that it may be a good indicator for estimating pine
forest biomass. On the other hand, DVI had close relationships with both Japanese
cedar plantation and deciduous broadleaf forest biomass.
The mean biomass of the pine, Japanese cedar and deciduous broadleaf forests was
estimated to be about 143, 135 and 121 t ha-', respectively. Total biomass for the forest
area (2040 ha) was estimated to be about 275.0 x 103 t. These results indicate that
forest biomass in a temperate forest zone can be quantitatively and continuously
estimated and mapped based on relationships between vegetation indices (VIs)
calculated from Landsat TM data and forest biomass data obtained on the ground.
Acknowledgments - The authors are grateful to Mr. Y. Kimura for his critical and
valuable suggestions. We also thank the members of the Laboratory of Ecology and
Environmental Sciences at Hiroshima University for their assistance in the fieldwork
associated with this stud!'.
References
Ando, T., Hatiya, K., Doi, K., Kataoka, K., Kato, Y. and Sakaguchi, K. 1968. Studies on the
system of density control of Cryptomeria japonica stand. Reprint. - Government Forest
Experiment Station, Bulletin 209, Tokyo. 76 pp.
Belward, A.S. and DeHoyos, A. 1987. A comparison of supervised maximum likelihood and
decision tree classification for crop cover estimation from multi temporal Landsat MSS
data. - Int. J. Rem. Sens. 8: 229-235.
Cavayas, F. 1987. Modeling and correction of topographic effect using multitemporal satellite
images. - Can. J. Rem. Sens. 13: 49-67.
Chavez, P.S. 1988. An improved dark-object subtraction technique for atmospheric scattering
correction of multispectral data. - Rem. Sens. Environ. 24: 459--479.
Chavez, P.S. 1989. Radiometric calibration of Landsat Thematic multispectral images.-
Photogramm. Engin. Rem. Sens. 55: 1289-1294.
Duggin, M.J. 1983. The effect of irradiation on vegetation assessment. - Int. J. Rem. Sens. 4:
601-608.
Eliason, PT., Soderblom, L.A. and Chavez, P.S. 1981. Extraction of topographic and spectral
albedo information from multispectral images. - Photogramm. Engin. Rem. Sens. 48:
1571-1579.
Hiroshima Prefecture, 1989. The Guide of Environmental Use in Hiroshima Prefecture. -
Department of Environmental Preservation, Hiroshima Prefecture, pp. 13-27.
Holben, B.N. and Justice, C. O. 1981. An examination of spectral band ratioing to reduce the
topographic effect on remotely sensed data. - Int. 1. Rem. Sens. 2: 115-133.
Horler, D.N.H. and Ahern, F.J. 1986. Forest information content of Thematic Mapper data. -
Int. J. Rem. Sens. 7: 405-428.
Hudson, W.D. 1987. Evaluating Landsat classification accuracy from forest cover-types maps.
- Can. J. Rem. Sens. 12: 39-42.
Justice, C.O., Wharton, S.w. and Holben, B.N. 1981. Application of digital terrain data to
quantify and reduce the topographic effect on Landsat data. - Int. J. Rem. Sens. 2: 213-230.
Kawata, Y., Ueno, S. and Kusaka, T. 1988. Radiometric correction for atmospheric and
topographic effects on Landsat MSS images. - Int. J. Rem. Sens. 9: 729-748.
171
Kawata, Y, Ohtani, A. and Kusaka, T. 1991. Analytical correction method for atmospheric and
topographic effects on rugged terrain image data. - Trans. Soc. Instr. Contr. Engin. 27:
386-393.
Keil, M., Schardt, M., Schurek, A. and Winter, R. 1990. Forest mapping using satellite
imagery. - Photogramm. Rem. Sens. 45: 33-46.
Lillesand, TM. and Kiefer, R. W 1987. Remote sensing and image interpretation. 2d ed. - J.
Wiley and Sons, New York. 721 pp.
Mather, P.M. 1987. Computer processing of remotely-sensed images. - J. Wiley and Sons,
New York. 329 pp.
Miura, M. 1992. Studies on the allometry in deciduous broad-leaved forest. Research on the
estimated method of resources in deciduous broad-leaved forest. - Bulletin of the Ministry
of Education, Research Report 02660155, Tokyo. pp. 47-54.
Nakane, K., Yamamoto, M. and Tsubota, H. 1984. Cycling of soil carbon in a Japanese red pine
forest. 1. Before a clear-felling. - Bot. Mag. Tokyo 97: 39-60.
Nelson, R.F., Latty, R.S. and Mott, G. 1984. Classifying northern forest using Thematic
Mapper simulation data. - Photogramm. Engin. Rem. Sens. 50: 607-617.
Nelson, R., Krabill, Wand Tonelli, J. 1988. Estimating forest biomass and volume using
airborne laser data. - Rem. Sens, Environ. 24: 247-267.
Peterson, D.L., Spanner, M.A., Running, S.W. and Teuber, K.B. 1987. Relationship of
Thematic Mapper simulator data to leaf area index of temperate coniferous forests. - Rem.
Peterson, DL, Aber, J.D., Matson, P.A., Card, D.H., Swanberg, N., Wessman, C. and Spanner,
M.A. 1988. Remote sensing of forest canopy and leaf biochemical contents. - Rem. Sens.
Environ. 24: 85-108.
Rock, B.N., Vogelmann, J.E., Williams, D.L., Vogelmann, A.F. and Hoshizaki, T 1986.
Remote detection of forest damage. - Bioscience 36: 439-445.
Sader, S.A., Waide, R.B., Lawrence, WT. and Joyce, A.T 1989. Tropical forest biomass and
successional age class relationships to a vegetation index derived from Landsat TM data. -
Sellers, PJ. 1985. Canopy reflectance, photosynthesis and transpiration. - Int. J. Rem. Sens. 6:
1335-1372.
Sellers, P.J. 1987. Canopy reflectance, photosynthesis and transpiration. 2. The role of
biophysics in the linearity of their interdependence. - Rem. Sens. Environ. 21: 143-183.
Senoo, T, Iwanami, E., Tanaka, S. and Sugimura, T 1983. Forest type classification in broad
mountainous area by two seasonal Landsat MSS data after ratioing. - Rem. Sens. Soc.
Japan 3: 55-64.
Sjoberg, R.W and Horn, B.P. 1983. Atmospheric effects in satellite imaging of mountainous
terrain. -App!. Opt. 22: 1702-1716.
Swain, P.H. and Hauska, H. 1977. The decision tree classifier: Design and potentia!. - Trans.
Tucker, c.J. 1979. Red and photographic infrared linear combinations for monitoring
vegetation. - Rem. Sens. Environ. 8: 127-150.
Walsh, SJ. 1987. Variability of Landsat MSS spectral responses of forest in relation to stand
and site characteristics. - Int. 1. Rem. Sens. 8: 1289-1299.
E IG HT
Forest Structure and Productivity along the Oregon Transect

David L. Peterson
Peterson, D.L. 1996. Forest structure and productivity along the Oregon transect. - In:
Modeling of Forest Productivity. Kluwer Acad. Publ., Dordrecht, The Netherlands,
pp. 173-218.
The Oregon Transect Ecosystem Research (OTTER) Project, conducted from 1990 to
1992, was an investigation of regional and seasonal variations in forest ecosystem
processes involving carbon (C), nitrogen (N) and water. Methods of field ecology,
surface meteorology, computer simulation and remote sensing were applied to
the study of six primary coniferous forest sites and three fertilization-treatment sites
along an environmental gradient across west central Oregon. The objective of the
OTTER Project was to address two main questions: (i) can generalized ecosystem
models, designed to use mainly variables to be derived from remote sensing data,
explain the variation in ecosystem functioning found across the environmentally
variable landscape of Oregon? and (ii) do good relationships exist between the
regional variation in these driving variables and remotely sensed data? A large team of
scientists supported by airborne remote sensing efforts collected a very wide range of
ecological, climatological, biophysical and biochemical variables relating to net
primary production (NPP), photosynthesis, evapotranspiration and nutrient cycling in
these forests. The team used both a light-use efficiency model and a mechanistic
ecosystem process model to predict NPP across the transect. The driving variables of
each model formed the basis for the remote sensing studies. Correlative analyses and
radiative transfer models were used to study the relationships between LAI, specific
leaf area, standing biomass, foliar biomass, foliar chemistry, canopy temperature,
relative humidity, vapor pressure deficit, incident and fraction of absorbed
photosynthetically active radiation (PAR) and spectral reflectance from a wide variety
of remotely sensed data.
D.L. Peterson, Ecosystem Science and Technology Branch, NASA, Ames Research
Center, Moffett Field, CA 94035, USA.
173
Introduction
Understanding regional variation in forest structure and productivity is an important
step toward understanding differences between local and global-scale findings. By
studying ecosystem processes across large geographic regions of environmental
variability in time and space, we can determine whether the principal factors
controlling these processes have been captured in our algorithms and models. One
method of conducting such studies is to maintain the sampling support (spatial unit of
measurement) used in site-specific or local studies, comparing predictions of a broad
range of process rates made across a variety of sites. If we have succeeded in a linear
explanation of this regional variation, then we might have confidence in extrapolating
our knowledge to other regions with similar characteristics or in interpolating our
results across the landscape within the range of environmental driving factors for
which we have validated our findings.
This extrapolation or interpolation, however, brings up the very real problem of
acquiring the variables required to operate models synoptically across different
landscapes or over time in a changing landscape. It is clearly impossible to use
conventional field methods while maintaining the same support throughout,
necessitating some form of statistical surface fitting. Interpretation of aerial
photography is synoptic, but the information content is often too limited for
quantitative modeling and the process too laborious to repeat for changing landscapes.
Only digital remote sensing data, acquired from either satellite or aircraft sensors,
offer the potential for consistent replication and for the range of ecological and
environmental variables required in models.
By keeping the support equal to local scales, thereby gaining knowledge from these
studies but restricting our focus to the study of processes varying substantially at
regional scales, we can validate the combined use of computer simulation and remote
sensing. On the one hand, the combination might be used to establish a spatially
distributed datum to examine large- (fine-) scale ecosystem processes such as
productivity within a watershed or across a disturbed landscape, in which comparisons
can be made to highly detailed local observations. On the other hand, since
comparable logic has been incorporated into global biospheric models that use
smaller- (coarse-) scale remote sensing data with similar, although greatly limited
spectral properties and other variables, the regional approach might be used to validate
and interpret global simulations. Regional models, although simplified from the
locally based perspective, are often more complex and demanding in driving variables
than are global models. Eventually, a suite of nested models might emerge that allows
interpretation and prediction across a range of scales. In any case, validated regional
models can be applied to many ecosystem analysis problems that occur at a regional
scale. This chapter reports results of a regional-scale test called the Oregon Transect
Ecosystem Research (OTTER) Project (Peterson and Waring 1994) conducted in the
forests of the state of Oregon, USA.
A steep environmental gradient within a single biome facilitates the study of
regional variations in ecosystem processes by means of transects. In Oregon, the
combination of mountainous topography and a consistent climatic pattern has
175
produced a well-defined distribution of coniferous forests that have been successfully

studied for decades using transects (e.g., Grier and Running 1977, Gholz 1982). In
Oregon's northwest region, two mountain ranges (one low and coastal, the other
inland and higher) lie parallel to each other and normal to the prevailing storm fronts
originating over the Pacific Ocean. Thus, two steep climatic gradients occur due to the
orographic effects of these ranges on the on-shore weather systems, producing a wide
range of annual precipitation, seasonal temperature and relative humidity values. Solar
insolation is roughly constant across this transect, varying mainly with cloudiness and
terrain. The region is dominated by a single coniferous biome, containing both very
large, rapidly growing trees in dense stands well adapted to the moderate climate of
the coastal side and short, hardy trees, sparsely distributed and well suited to the
extremely dry conditions of the interior desert in the rain shadow of the two mountain
ranges (Fig. 1, Waring and Franklin 1979). A typical east-west transect representing
the entire climatic range covers only about 250 lan. Within this short distance, the
variation in ecosystem responses to climate fluctuation permits studies in which the
scale length of climate variation (tens of kilometers) corresponds to a large distributed
variation in ecosystem structure and function. The transect experimental method
permits an examination of the factors controlling regional variations in ecosystem
response, a scale intermediate between local- and global-scale variation.
At the same time, a compact transect provides logistical advantages for carrying out
a seasonal investigation in which each study site along the transect is given
approximately equal attention. A transect is also ideally suited for an aircraft-based
remote sensing study for the obvious reason of aircraft operation efficiency, and for the
Coast Washington
Range
-46'
Malor Vegetation Zones

-45
PlCea silchensis
Pacific
ocean
- 44'
- 43
Idaho
- 42'
Nevada
I I I I I I
124' 121 120' II C)' liS' 117'
Figure 1. Map of the study area in Oregon showing site locations and major
vegetation zones (after Franklin and Dyrness 1973, Gholz 1982; see Table 1 for
site descriptions and notation).
less obvious reason of obtaining critical ground observations contemporaneous with

aircraft overflights. Research on remote sensing' data acquired from advanced airborne
sensors, data suitable only for regional- to local-scale studies, indicates that many
more relevant variables can be extracted than can be obtained from satellite data alone.
This finding offers the potential to retain more of the local detailed information and to
improve the links between regional and local modeling. Aircraft platforms are
currently carrying state-of-the-art sensors with an extensive range of spectral, spatial,
angular, temporal and radiometric characteristics. This kind of research anticipates the
day when satellite-borne sensors of comparable performance will be available for
global coverage. Research establishing relationships between ecosystem variables and
both airborne and satellite remote sensing data through correlative techniques and
radiative transfer models is needed in the design and prelaunch stages of satellite
sensors. With these considerations in mind, the NASA-sponsored OTTER Project
selected sites following the transect described in Figure l.
The OTTER Project was focused on predicting major C, N and water ecosystem
fluxes, and on examini,"!-g the factors that dynamically regulate these fluxes. The study
combined computer simulation and experimental and theoretical remote sensing
methods, as well as field and laboratory techniques. This chapter will present the
results of studies centered on two main questions: (i) can generalized ecosystem
models, designed to use primarily variables derived from remote sensing data, explain
variations in ecosystem functioning found across the environmentally variable
landscape of Oregon? and (ii) do good relationships exist between the regional
variation in these driving variables and remotely sensed data?
Background
Much of our understanding of the coniferous forest biome of the Pacific Northwest is
derived from studies of Oregon and Washington forests initiated two decades ago as
part of the International Biological Programme (IBP) (Waring and Franklin 1979,
Reichle 1981). The IBP collected intensive data on climatic, ecologic and disturbance
variables from what is now the National Science Foundation (NSF)-sponsored
Long-Term-Ecological-Research site at the H.J. Andrews Experimental Forest in
Oregon. Transect studies across Oregon reported the large annual covariation in
ecosystem function (above-ground net primary production, or ANPP), climatic indices
(site water balance) and structure (standing biomass, LAI) (Grier and Running 1977,
Waring et al. 1978, Waring 1980, 1983, Gholz 1982). Studies of a variety of vegetation
communities including forests have shown that climate constrains NPP by limiting the
utilization of intercepted photosynthetically active radiation (IPAR, Monteith 1977,
Landsberg 1986, Goward 1989, Landsberg et al. 1996). Photosynthesis in conifer
forests is partially or completely limited by drought, extreme humidity deficits and
frost (Waring and Schlesinger 1985).
When nutrient availability is limited, C processes can also be constrained. Studies
in forests by several authors have shown that the foliar pools of C, N and phosphorus
compounds reflect physiological activity. For example, during periods of rapid
growth, concentrations of storage compounds such as starch decrease quickly (Tromp
1970, Chapin and Kedrowski 1983, Pate 1983, Chapin et al. 1986). Similarly,
177
chlorophyll concentration is at a maximum during periods of high photosynthetic and

growth rates and at a minimum during the coldest periods (Linder 1980). Foliar
chemicals may also provide information on the factors controlling growth:
fertilization effects on N compounds (Van den Driessche and Webber 1977, Matson
and Waring 1984), nutrient limitations on soluble nonstructural carbohydrates (Meyer
and Splittstoesser 1971, Chapin 1980, Birk and Matson 1986) and water limitations on
amino acids (Pate 1983). These foliar concentrations have been correlated to whole
plant and ecosystem processes as well. For example, the relationship between foliar N
concentration and ANPP is strong both within and across species (Cole and Rapp
1981, Van Cleve et al. 1983). The turnover of N (net annual mineralization and
decomposition rate) has also been related to foliar chemistry (Melillo et aI. 1982,
Meentemeyer and Berg 1986, Aber et al. 1989a). More recently in Oregon, Myrold et
al. (1989) reported large annual variations in soil (microbial biomass, respiration) and
nutrient (N turnover rate, N mineralization) characteristics covarying with ANPP and
LA! across Oregon in the same sites studied by Gholz (1982). While a large database
has been created from these widely distributed studies, no data existed prior to the
OTTER Project in which climatic, ecologic, chemical and structural information was
collected together throughout the year to examine seasonal and regional dynamics
within one biome.
In the last decade, much of our understanding of how ecosystems function has been
formalized into mathematical models. Mechanistic ecosystem models were
formulated to match observations made on rates of growth, litterfall production, litter
decomposition, evapotranspiration, photosynthesis and streamflow (Running et al.
1975, Waring and Running 1976, Sollins et al. 1981). These models have been
validated at a site-specific level. However, there are several problems associated with
extrapolating such models to large contiguous regions. These models are complex,
involving many variables, many of which can be measured only in the field. In general,
these data requirements are well beyond the capacity of remote sensing. Nevertheless,
remote sensing data in conjunction with other geographic data can satisfy some of the
key data requirements. The models must be simplified but can still make reliable
regional predictions if most of the site-specific, species-specific and fine-scale
information is retained. One such model called FOREST-BGC (Running and
Coughlan 1988, Running and Gower 1991) was derived from very site-specific
heritage models but was expressly designed to use only those variables believed to be
accessible by means of remote sensing (detail on FOREST-BGC can be found in
Coughlan and Dungan 1996). The full model was validated in the OTTER Project
based on timing of phenomena (predawn leaf water potentials), annual integrations
(ANPP, hydrologic, C and photosynthesis) and equilibrium conditions at a stand age
of 100 years (LAI and biomass).
Using Airborne Thematic Mapper (Landsat 4) Simulator data with a spatial support
of 30 m to match ground plot sizes, Peterson et al. (1987) found a strong direct
relationship between regional variations in LAI and the simple ratio (SR) of the
infrared-to-red reflectance measurements of the sensor (Running et al. 1987a). We
also studied the effects on these relationships of additive radiance from atmospheric
scattering and variable transmittance from atmospheric absorption, finding that the
atmospheric effect was magnified by altitude variations across this mountainous

terrain and must be removed from the data prior to calculating the ratio for consistency
across space and time (Spanner et al. 1984, Peterson et al. 1987). Following these
studies, the same experiments were repeated using the coarser spatial (l-krn) but
similar spectral band data of the Advanced Very High Resolution Radiometer
(AVHRR) data of NOAA, achieving comparable results but also suggesting that the
time variation in these data matches the seasonal variation in LAI (Spanner et al.
1990a). Another index calculated from AVHRR data is the Normalized Difference
Vegetation Index (NDVI = (I-SR)/(l+SR)), which is related to the capacity of
vegetation to intercept and absorb photosynthetically active radiation (IPAR and
APAR, respectively, e.g., Sellers 1985). While studies of LAI had been conducted in
Oregon forests, no regional dynamic studies of APAR or IPAR had been undertaken.
One of the objectives of the OTTER Project was to relate the seasonal variations in
LAI, APAR and related variables to remote sensing data through the use of correlative
techniques and radiative transfer (RT) models.
In the mid-1980s, Peterson et al. (1988) introduced the possibility of deriving
chemical variables related to the forest canopy from analysis of high-spectral-
resolution reflectance data. Initial studies in Oregon and several other states suggested
that spectroradiometric measurements of plant canopies can be correlated with
variations in canopy chemistry, primarily involving lignin and possibly N
concentrations (Wessman et al. 1988, Aber et al. 1989a, Peterson and Running 1989,
Gholz et al. 1996, Martin and Aber 1996). Determining seasonal variations in canopy
chemical concentrations and their potential relationship to both nutrient and C cycling
processes and imaging spectrometer data was another objective of the OTTER Project.
Standing stocks of woody biomass or timber volume in Oregon coniferous forests
can be estimated using remote sensing data with the methods of multistage sampling
(Peterson and Card 1977), but these methods do not provide spatially continuous data
and require extensive ground measurements. Some studies have shown that forest
conditions such as stand density, age class, crown closure and cover type can be
discriminated by properly calibrated spectral clustering and classification of Landsat
imagery (Fox and Mayer 1980, Strahler et al. 1981, Spanner et al. 1990b, Wallin et al.
1996, among many other studies). These variables can be used to estimate biomass on
a continuous spatial basis. In recent years, several promising techniques have emerged
to estimate biomass directly from remote sensing data. Li and Strahler (1986)
developed a geometric-optics model based on the illumination of conical shapes in a
field of conifer trees to predict biomass and other stand variables, while Durden et al.
(1989) developed a radar backscattering model based on the proportion and
orientation of biomass components. In the OTTER Project, we examined the extent to
which radiative transfer methods could be extended to the large biomass range of
Oregon forests.
Finally, climatic conditions that constrain production (e.g., canopy and air
temperature, relative humidity, soil moisture, incident solar radiation fluxes) were also
evaluated. In addition, cloud cover, which affects the fraction of solar radiation
reaching ecosystems as well as the ratio of direct to diffuse radiation, was estimated in
the OTTER Project after methods developed by Eck and Dye (1991). At night,
179
minimum canopy temperatures approach dew point, and Running et aI. (1987b) have
shown that these temperatures can be used to calculate absolute humidity. Further, if
canopy temperatures reach the freezing point, photosynthesis will be constrained the
following day because of stomatal closure. By invoking various assumptions about
vertical mixing in the boundary layer as amplified by spatially variable surface heating
and temperatures over an array of land conditions, we can obtain estimates of relative
humidity and vapor pressure deficits near the surface (Jedlovec 1990). This technique
relies on two thermal wavelength observations provided by AVHRR. To estimate air
temperature and environmental moisture conditions,the AVHRR's surface
temperature fields can be combined with NDVI measurements (plant coverage). The
combined measurements can be used to calculate drought levels (Goward et al. 1985,
Running and Nemani 1988, Goward and Hope 1989, Goward and Dye 1996). The
large differences in heat capacitance between foliage and soils result in large
variations in surface temperature (Ts). When soils are wet, the differences can be
small; when they are dry, the differences are large. The slope of the Ts-NDVI
relationship has been found to vary with soil wetness (Nemani et al. 1993). The
challenge to remote sensing is to extract these variables for complex mountainous
terrain with enough precision to drive ecosystem models.
Methods
The design and methods of the OTTER Project are described in detail, primarily in
two special journal issues (Ecol. Appl. 4(2): 210-343 and Rem. Sens. Environ. 47(2):
107-108,154-166,190-203 and 216-230). A summary list of variables and processes
measured, measurement methodes) used (e.g., simulated or estimated from either
ecosystem or RT modeling) and references is presented in Table 1.
Six study sites were selected from forest ecosystems across the climatic gradients
of west central Oregon (Fig. 1). Details of the sites, their characteristics and
measurements and ecosystem simulation results are contained in Peterson et al.
(1987), Runyon et al. (1994), Matson et al. (1994), Myrold (1994) and Running
(1994). At three sites, additional forest stands were studied. An Alnus rubra (red alder)
stand was included near the coastal site (1) to compare an N-fixing deciduous species
with the coastal conifers. An old-growth Tsuga heterophylla (western hemlock) stand
was also measured near the coastal site. Additionally, three fertilization treatments
were studied: singly and multiply fertilized stands of Pseudotsuga menzeisii
(Douglas-fir) at the Scio site (3) in the western Cascade Mountains, and a
urea-fertilized stand (5) of Pinus ponderosa (Ponderosa pine) on the east side of the
Cascades. In the Ponderosa pine stands, the overstory trees had been removed,
releasing suppressed and young Ponderosa pines and a significant understory canopy.
Intensive field campaigns were conducted over the four seasons. The timing of
these campaigns was based on the following seasonal criteria:
1. February-March: prebudbreak at all sites, when leaf area is at a minimum
and low temperatures are likely to be constraining many ecosystem
processes;
00
Table 1. Ecosystem and environmental variables and processes measured. * = field measurements taken at site 3 only 0
Variable or Measurement Reference Simulated and remotely estimated ;;l

~
processes method to method variables
~
~
Meteorological: ~
Incident shortwave meteorological station Campbell Scientific Instruments, MT-CLIM (Glassy and Running 1994 ;:.;,
~
solar radiation located within 1-15 kIn Inc., Logan, Utah, USA and Running et al. 1987b) ~
0
Precipitation of each site, airport data plus NOAA MT-CLIM and slope of AVHRR ~
(site 6) and AVHRR temperature bands ~
;::
Humidity thermal IR data Runyon et al. 1994 Minimum NDVI vs Ts extrapolation
Air temperature Price 1984 MT-CLIM
'"S
OQ
Soil temperature Direct at site S
Cloudiness TOMS UV attenuation Eck and Dye 1991 ;;.
~
Optical depth of Sunphotometer and Wrigley et al. 1992
the atmosphere white reflectance panel ~
l}
Ecophysiological and nutrient cycling: 7
OQ
Predawn xylem Pressure bomb at Scholander et al. 1965, Slope ofNDVI-Ts, Goward 1989
water potentials monthly intervals Waring and Cleary 1967 FOREST-BGC, Running 1994 ~
Photosynthesis Yoder 1992 FOREST-BGC, Running 1994 ~
Gas exchange <;1
Conductance Gas exchange FOREST-BGC, Running 1994 ...'"
Litterfall 50 x 50-cm littertraps* Matson et al. 1994 ""1::1
Cl
Net N-mineralization Resin core method* DiStefano and Gholz 1986 l:l..
:;::
ammonification Resin core method* DiStefano and Gholz 1986 ~
~.
nitrification Resin core method* DiStefano and Gholz 1986 ~.
Soil water storage I-m cores, pressure volume Waring and Major 1964
Root biomass 20-cm root cores* David D. Myrold (pers. comm.)
Soil respiration Head-space chambers* Pamela A. Matson (pers. comm.)
Maintenance respira- 50% of root growth Ryan 1991 FOREST-BGC, Running 1994
tion
Table J. (continued)
Variable or Measurement Reference Simulated and remotely estimated

Ecophysiological and nutrient cycling: (continued)

Wood growth Five-yr rings of stem cores Runyon et al. 1994 FOREST-BGC, Running 1994
Foliar growth First-yr leaveslbranch
Root growth Ingrowth to root cores* f (litterfall), Raich and
Nadelhoffer 1989
Growth respiration FOREST-BGC, Running 1994
Decomposition rate FOREST-BGC, Running 1994
Evapotranspiration FOREST-BGC, Running 1994
Canopy and tree biophysical characteristics:
Spectral reflectance Spectroradiometers
of components: Spectron SE 590 Williams et al. 1984
plants 45, 0 observations Abuelgasim and Strahler 1994
soil/litter Undisturbed soil/litter Goward et al. 1994, "
flat fields Averaged readings/campaign Matson et al. 1994
Intercepted PAR Ceptometer-transmitted Runyon et al. 1994
radiation
LAI Leaf: LI-COR LAI-3100 Runyon et al. 1994 Running 1994
meter
overstory Ceptometer-transmitted Pierce and Running 1988 NIRlred ratios, Spanner et al. 1994
radiation
understory LI-COR leaf-area meter Welles and Norman 1991 Geometric optics model, Li and
Strahler 1986
Allometric relation (dbh) Gholz et al. 1979 Reflectance model, Johnson et al.
1994, Gong et al. 1992
00
Weight-to-area ratio Law and Waring 1994 Law and Waring 1994 --
.....
00
Table 1. (continued). N
Variable or Measurement Reference Simulated and remotely estimated ~

~
~
~
Canopy and tree biophysical characteristics (continued): <Q,

Specific leaf area Five branches, weight-to Matson et al. 1994 ~
~
area ratio ;:;l
~
dbh Diameter tape ~
Standing biomass dbh, height allometry Gholz et al. 1979 Wu and Strahler 1994, Moghaddam ~
;:s
et al. 1994, Running 1994
S
""
Crown radius, average Steel tape, ocular Abuelgasim and Strahler 1994 OQ
Stand height Clinometer Abuelgasim and Strahler 1994 s

Height to base of Clinometer Abuelgasim and Strahler 1994 So
~
crown
Stand density Field cruise, >5 cm dbh Runyon et al. 1994 Geometric optics model (Li and
Strahler 1986) ~
Tree spatial distribution' Point-to-tree distance, azimuth Wu and Strahler 1994 ~
Hemispherical albedo ASAS bidirectional data <Q,
Canopy temperature AirTherm sensor MT-CLIM, Glassy and Running 1994 ~
<ii
Leaf and canopy chemical concentrations and contents: ....
""
Nitrogen Colorimetry, flow analyzer Technicon Instrument Corp. 1977 FOREST-BGC lOO-yr equilibrium, '"1::1
Cl
Running 1994 f}
Chlorophyll Extract and spectrophoto- Derivative spectroscopy, Johnson et al. "....
~.
metry 1994 ~.
Lignin Permanganate ADF Van Soest and Wine 1968
Cellulose Ashing of residue Van Soest and Wine 1968
Starch Glucose oxidase Matson and Waring 1984
Amino acids Colorimetry of extract Lee and Takahashi 1966
183
2. May-June: budburst period, when soils are near saturation and maximum
rates of many ecosystem processes should be observed;
3. August: full expansion of foliage at all sites, with a high probability of
drought; and
4. September-October: period during which foliage of deciduous trees and
shrubs has senesced and rainfall has normally begun to recharge soils;
freezing temperatures are unlikely to extend throughout the day.
Acquisition of the remotely sensed data (Table 2) was coordinated with the field
campaigns, the former being entirely dependent upon suitably clear weather. GOES
and AVHRR satellite data as well as data from the ultralight aircraft were collected
more frequently. In addition to the airborne and satellite sensors, spectroradiometers
and sunphotometers were used in the field to calibrate and correct the remote sensing
data and to acquire data directly.
The OTTER Project involved scores of scientists from throughout the United
States, Canada and Australia, and data were shared among all team members. The
project was coordinated jointly by Dr. Richard H. Waring of Oregon State University
and this author at NASA Ames Research Center. More than 15 gigabytes of data were
collected and managed in two data management systems: the Pilot Land Data System
of Ames and the LTER Data Base at Oregon State University (Skiles and Angelici
1993). Skiles, Angelici and Popovici of Ames have created a series of five CD-ROMs
containing about 4 gigabytes of representative data sets of all sites (available from the
Data Active Archive Center, Oak Ridge National Laboratory, Tennessee).
Field results
Three methods of estimating overstory LA! on the ground (all values on an all-sided
basis) were tested: (i) measuring sapwood basal area, compensating for tree taper and
converting to LAI with allometric equations (Waring et al. 1982); (ii) measuring PAR
transmittance under cloudless conditions and converting to LAI using the extinction
coefficient from the Beer-Lambert Law of light extinction (Pierce and Running 1988);
and (iii) measuring gap fraction of the canopy by sensing the attenuation of diffuse
blue light under cloudy conditions at five zenith angles (Welles and Norman 1991).
LA! estimated from method (ii) varied from 0.5 for the western juniper of the dry
desert interior site (6) to 10.6 for the Douglas-fir/western hemlock forests ofthe low-
elevation Cascade Mountain site (3). The values obtained using the other two methods
were consistent with these values but somewhat lower; all other analyses involving
LA! discussed in the remainder of this chapter are based on the LAI values of method
(ii). The range is consistent but more conservative than are the results of Gholz (1982)
and Peterson et al. (1987). Law and Waring (1994) estimated the understory LAI of a
thinned Ponderosa pine stand at site 5 at 1.3, increasing the overall LAI at this site to
about 2.0. The LAI of the understory vegetation at other sites was not included since
these were closed-canopy sites.
Midsummer ceptometer measurements (from method ii) were used to calculate the
fraction of intercepted PAR (f;PAR). Annual values ranged from 22% to 99.5%,
......
00
Table 2. Remote sensing instruments, satellite and aircraft platforms and characteristics. .j:>.
OrbitJ Platform Sensor Time of dayI Spectral Spatial Spectral Viewing Other ~
~
Altitude campaign resolution resolution range angle character-

(nm) istics
s:
~
<Q,
~
Polar orbit GOES (NOAA) Total Ozone Mapping Hourly/ Broad-band UV 50-170km 370 Global ~
;;l
Spectrometer (TOMS) continuous Cl
...
~
Polar orbit Nimbus-7 (NOAA) Advanced Very High Resolu- 2:00 pm/all loo-nm visible l.lkm 500-12,000 Nadir to 5 bands V)
~
tion Radiometer (AVHRR) clear days I J.ITIR 30 :::s
'"
S
OQ
20,000 m ER-2 aircraft Advanced Visible Infrared Solar noon! -lOnm 23 m 400-2450 Nadir 224 bands
Imaging Spectrometer (AVIRIS) 8, 10/90; 5/91
S
(NASA)
Thematic Mapper Simulator Solar noon! 5D-loonm 25m 450-2300 Nadir 12 bands
So
~
(TMS) all ~
Cl
Thermal Infrared Multi- Solar noon! 1000nm 30m 8000-12000 Nadir l:l..
~
channel Scanner (TIMS) all s:

OQ
Color infrared aerial Solar noon! Panchromatic 2m Photo IR Nadir Large
photography (CIR) all format <Q,
~
0
1O,600m DC-8 Synthetic Aperture Radar Solar noon! 20MHz 6.7 x 12 m 68,24,6 em 40-45 Multipolari-
(NASA) (AIRSAR) 6,8/90 P,L,C incidence zation ...'"~
"'tI
bands angle C3
l:l..
:;::
0 ,...,
4500m C-130 Advanced Solid-state Array Solar noon, 15nm 4.25 x 2 m 465-871 every 15 Principal ...:;::.
(NASA) Spectrometer (ASAS) 9:00 am and/or 45 and per- ~.
NSool (a TMS) 3:0D-4:oo pm/ fore/aft pendicular
TIMS 6,8/90 solar planes
CIR
Sun-tracking sunphotometer all IOnm - vis-NIR Solar disk 10 channels
Table 2. (continued)
Orbit! Platform Sensor Time of day/ Spectral Spatial Spectral Viewing Other
Altitude campaign resolution resolution range angle character-
1,400-1,800 m Cessna light Compact Airborne Spectro- Solar noon! 3-{j nm 2x9m 425-925 Nadir Spectral
aircraft (Borstad graphic Imager (CASI) 8/90; 5/91 mode
Associates, U.S.) Spatial mode
600m Apache light Fluorescence Line Solar noon! 1.4nm 5x4m 430-800 Nadir Spectral
aircraft (Monotec Imager (FLI) 6/90 mode
Ltd., Canada) Spatial mode
150-450 m Ultralight Spectron SE 590 Solar noon 10-12nm 1.5 x 380-1100 Nadir Point data
aircraft 1.5-17 m
(OSU) AirTherm Thermal IR sensor Mornings Broad-band TIR Variable TIR Nadir
Video camera Afternoons Visible -2m Visible Nadir
6m Truck-mounted Spectron SE 590 Midday 10-12nm -2m
boom (OSU) Barnes MultiModular lOOnm -I m 500-900 Nadir LandsatMSS
Radiometer (MMR) bands
Ground Field-portable Spectron SE 590 Midday 10-12nm
instruments Spectron SE 393 Midday 50nm 10 FOV 1100-2500 Nadir
ALEXA spectroradiometer
SIRIS spectroradiometer
LI-COR LI-200Z pyranometer
BarnesMMR >(Meterological
Decagon ceptometers station)
Reagan sunphotometer
Digital camera
35 mm with fish-eye lens ... 00
VI
Campaigns: February, June, August and October 1990, March and May 1991
-
assuming J;PAR remained constant throughout the year. Total annual incident PAR
ranged from 2735 MJ m- 2 to 1934 MJ m- 2 , although all of the sites occurred at
approximately the same latitude. The wide variation was due to large differences in
cloud and fog cover. Annual intercepted PAR varied from 602 MJ ha- 1y{lto 2248 MJ
ha-1yr- l .
A variety of field methods were used to estimate growth and biomass values. Total
biomass varied from about 10 Mg ha- 1 to>700 Mg ha- 1for an old-growth coastal Sitka
spruce stand (site 10). NPP varied from 3 Mg ha- 1yr"l for sites 5 and 6 to 25.7 Mg hal
yr"l for the fertilized Douglas-fir site, 3F. Foliar biomass and litterfall followed the
same trend, increasing east to west, with foliar biomass ranging from <2 Mg ha- ' at
1
site 5 to >15 Mg ha- at site 3. Below-ground NPP (estimated using the method of
Raich and Nadelhoffer (1989)) represented from 53% to 60% of total NPP for the
eastern stands and from 20% to 32% for the western stands. Based on intercepted PAR
and measured total NPP, the light conversion efficiency varied from 0.18 to 0.92 but
there was only a weak relationship between IPAR and NPP (Fig. 2, Runyon et al.
1994). These results may be somewhat surprising given the linear relationship
between LA! andANPP found by Gholz (1982). However, Gholz's methods could not
account for light use or for daily environmental constraints on PAR utilization. In
the OTTER Project, a simplified light-use efficiency model (in addition to
FOREST-BGC) was evaluated to examine environmental constraints on production
(see the next section).
Soil moisture for 20-cm cores (David D. Myrold, personal communication 1994)
from March 1990 to June 1991 for sites 3 and 5 ranged from <10% to almost 50% for
site 5, and from about 20% to almost 50% for site 3. Myrold estimated that the upper
and lower limits to available soil water to I-m depth (assuming uniform soils with
depth) were 27 cm at site 3 and about 23 cm at site 5. Myrold (1994) also reported
cumulative litterfall mass at site lA (the red alder site) from June to October 1990 as
3.13 Mg ha- 1 yr- 1 and annuallitterfall at sites 3, 3F and 3IF as 5.75, 6.83 and 6.61 Mg
ha'! y{l, respectively. The litterfall mass at the alder site was comparable to the foliar
biomass (4.3 Mg ha- 1) but higher than litterfall (1.9 Mg ha- 1) estimated by Runyon et
el. (1994). The data at site 3 reflected the effect of fertilization. A comparison of root
dynamics between the control and the intensively fertilized site 3 stands suggested that
the control stand produced about 40% more roots than the intensively fertilized stand
and that root inputs made up only about 15-25% oflitterfall (Myrold 1994).
Concentrations of total N, total chlorophyll and free amino acids in foliage varied
both seasonally and among sites (Matson et al. 1994, Johnson et al. 1994, Table 3).
Total N and total chlorophyll tended to be highest in the conifer sites in August and
October, lowest overall at site 4, the subalpine forest, and elevated for the fertilized
sites 3 and 5. The proportion of total N to total chlorophyll was much greater during
August and October than in March, consistent with other published evergreen studies
for seasonal climates (Linder 1980). The coastal sites did not show significant
seasonal variation in this ratio, suggesting that the decline in winter concentrations in
the eastern sites may have more to do with cold temperatures than with low radiation
levels. Highest total N was observed in the alder site (1A). Even though the alder site's
chlorophyll concentration was lower on a weight basis in May and October, the
Table 3. Seasonal data on foliar chemistry and leaf area. SLA =specific leaf area. All surface areas are expressed on an all-sided
basis.
Site Date Total nitrogen Chlorophyll Lignin Starch SLA LAI

designation MMIYY
mgg'! mgcm2 kg ha'! mgg'! mgcm2 kgha'! mg g'! mg cm2 kg ha'! mgg'! mg cm 2 kg ha'! cm 2 g'! m2 m2
1-0G 8/90 10.57 0.140 92.4 1.89 0.025 16.5 181.9 2.41 1591 40.66 0.539 355.7 75.5 6.6
2 8/90 12.02 0.226 128.8 2.51 0.047 26.8 166.5 3.13 1784 3.72 0.070 39.9 53.2 5.7
3-C 8/90 13.46 0.181 191.9 4.95 0.066 70.0 148.7 2.00 2120 9.83 0.132 139.9 74.5 10.6
3-F 8/90 15.96 0.203 215.2 6.11 0.078 82.7 161.1 2.05 2173 1.22 0.016 17.0 78.7 10.6
4 8/90 10.7 0.321 61.0 2.88 0.086 16.3 119.1 3.58 680 5.77 0.173 32.9 33.3 1.9
5-C 8/90 12.77 0.370 25.9 2.7 0.078 5.5 117.1 3.39 237 3.4 0.099 6.9 34.5 0.7
5-F 8/90 14.17 0.401 28.1 2.83 0.080 5.6 130.0 3.68 258 17.77 0.503 35.2 35.3 0.7
6 8/90 12.67 0.722 36.1 1.49 0.087 4.4 76.0 4.32 216 13.14 0.764 38.2 17.6 0.5
I-A 10/90 24.99 0.298 137.1 4.1 0.049 22.5 150.0 1.79 823.4 0.99 0.012 5.5 83.9 4.6
1-0G 10/90 11.28 0.119 78.5 4.15 0.044 29.0 190.3 2.00 1320 0.0 0.0 0.0 95.0 6.6
2 10/90 9.98 0.170 96.9 2.64 0.045 25.7 119.0 2.02 1151 16.07 0.273 115.6 58.8 5.7
3-C 10/90 13.45 0.195 206.7 5.12 0.074 78.4 133.2 1.93 2046 1.99 0.029 30.7 69.1 10.6
3-F 10/90 12.37 0.220 233.2 4.5 0.080 84.8 191.0 3.40 3604 11.62 0.207 219.4 56.1 10.6
4 10/90 10.59 0.341 64.8 2.92 0.094 17.9 798.5 6.38 1212 0.0 0.0 0.0 31.1 1.9
5-C 10/90 10.85 0.248 17.4 1.99 0.045 3.2 142.6 3.26 228 0.32 0.007 0.5 43.8 0.7
5-F 10/90 12.29 0.313 21.9 2.32 0.059 4.1 131.3 3.34 233 0.0 0.0 0.0 39.3 0.7
6 10/90 10.55 0.584 29.2 1.35 0.075 3.8 101.3 5.60 280 4.06 0.225 11.3 18.1 0.5
I-A 5-6/91 24.39 0.216 99.4 3.26 0.029 13.3 n.a. n.a. n.a. 74.3 0.658 302.7 113.0 4.6
1-0G 5-6/91 9.91 0.134 64.3 3.01 0.041 19.7 n.a. n.a. n.a. 148.4 2.00 960.0 74.2 4.8
2 5-6/91 12.05 0.244 97.6 2.39 0.048 19.2 n.a. n.a. n.a. 137.1 2.78 1112.0 49.4 4.0
3-C 5-6/91 11.15 0.191 141.3 3.7 0.068 46.6 n.a. n.a. n.a. 173.3 2.96 2190.4 58.3 7.4
3-F 5-6/91 13.22 0.278 205.8 3.46 0.073 54.0 n.a. n.a. n.a. 192.7 4.09 3026.6 47.4 7.4
5-C 5-6/91 9.54 0.283 14.2 1.99 0.060 3.0 m.a. n.a. n.a. 87.1 2.60 130.0 33.6 0.5
5-F 5-6/91 12.89 0.377 18.9 2.22 0.064 3.2 n.a. n.a. n.a. 91.9 2.69 134.5 34.2 0.5
6 5-6/91 10.03 0.677 20.0 1.23 0.082 2.3 n.a. n.a. n.a. 88.1 5.86 175.8 15.0 0.3 00
-.I
chlorophyll concentration was not different than that of the conifers on a mass per unit
leaf area basis. This suggests that, at least with respect to photosynthesis, the alder
leaves are not functionally different early and late in the growing season than in
midseason (Matson et al. 1994). Lignin and cellulose concentrations did not vary
between August and October, but there were sizable differences between sites. The
most obvious seasonal pattern was found for starch, with elevated concentrations (up
to 19%) in May and June, just prior to and during budbreak.
Total canopy contents tended to vary as total foliar biomass. The prebudbreak
buildup of starch apparently occurs because photosynthesis precedes initiation of
growth in the spring; once growth begins, starch reserves are depleted. Further, the
sites with fertilization had the highest starch concentrations and site 3F lost the
greatest amount of starch once growth began. Among canopy chemical variables,
canopy N content (averaged for all dates or for August alone) was the most strongly
correlated with ANPP. Studies of N mineralization showed that among conifer sites,
the values were very low and increasing with fertilization and that, as expected, values
for the alder stand were very high by comparison (Myrold 1994).
Modeling results
Two approaches to modeling NPP were evaluated. In one method, no actual computer
simulation model was developed, but the principles regarding constraints on net
photosynthesis and C assimilation were used in a modeling construct comparable to
the light-use efficiency models used in many global C models (Runyon et al. 1994,
Goward and Dye 1996).
The other approach involved the use of FOREST-BGC, driven by high-resolution
data from meteorological stations and initialized from measurements taken on the
ground. The simulation results of this model were evaluated by comparison with field
measurements (Running 1994).
Environmental constraints on the light-use efficiency model

Various studies have shown that a direct relationship exists between intercepted
photosynthetically active radiation (IPAR) and the annual growth of vegetation (see
review by Landsberg et al. 1996). Remote sensing studies have established that IPAR
is directly related to NDVI. Recent studies have shown a direct correlation between the
integration of seasonal NDVI values and annual NPP for ecosystems worldwide
(Goward and Dye 1987, Goward et al. 1987, Running and Nemani 1988). However,
the field-measured values of ANPP in the OTTER Project did not show a constant
light-conversion efficiency when plotted against IPAR (Fig. 2).
In this region, at least three climatic factors constrain photosynthesis: freezing
temperature, soil drought, and high vapor pressure deficits. Thus the simple light-use
efficiency model was reconstructed to reduce IPAR when any of these conditions
prevailed. Meteorological data from the stations coupled with direct measurements of
predawn vapor pressure deficits were used to make this calculation according to a set
of rules (Waring and Franklin 1979, Runyon et al. 1994).
189
30T""""""-----------~
- >- 20
~
';"
ell
J:
-
C)
~
c. 10
c.
z<[
1000 2000 3000

IPAR u (MJ m'2 y(l)
Figure 2. Variation of above-ground net primary production (ANPP) with

increasing interception of photosynthetically active radiation (IPAR) across the
Oregon transect (from Runyon et al. 1994).
When the reduced values of IPAR (or utilizable IPAR, IPAR u) were used to
calculate light-use efficiency, the range in lOu varied much less and the relationships
between ANPP, NPP and IPAR were linearalized (Fig. 3), with the exception of the
old-growth western hemlock stand (1G). Annual reductions from IPAR to IPAR u
ranged from 8% to 77%, with the importance of each climatic factor varying across the
transect (Fig. 4).
These results suggest that even monthly monitoring of variables that may be
correlated with predawn water potential, such as canopy temperature, relative
humidity and soil dryness, might be sufficient to implement this approach regionally,
provided the rules for IPAR reduction remain the same.
Mechanistic ecosystem modeling

Rather than reducing IPAR, FOREST-BGC simulates changes in stomatal
conductance by maintaining a site water balance, beginning with saturated soils in the
springtime, and calculating daily soil water inputs and outputs. Model variables were
initialized from field data, published data and previous experience. Stomatal
conductance was assumed to be 0.0010 mm Sl for all sites based on literature
reports (Knight et al. 1981, Baldocchi et al. 1988, Ryan 1991); all other model
parameterizations are discussed in Running (1994). Thus, the same constraints upon
ecosystem production used in the light-use efficiency model are also incorporated into
the logic of FOREST-BGC, although they are used in a mechanistic manner. In
addition, each of the environmental constraints used to calculate the reductions of
IPAR above is needed. A surface climatology model called MT-CLIM (Glassy and
Running 1994) was also used and evaluated in the OTfER Project (see also Coughlan
and Dungan 1996) to extrapolate meteorological station data to other locations.
FOREST-BGC also requires soils data, such as soil water-holding capacity (SWHC),
which remain very difficult to acquire over large regions and are generally not yet
available by remote sensing analysis.
JO JOr-----------,
,1.099
ro Y.OOI25.
->
- 20
III
~
3f
_>
III
20
~
01
01
a..
a..
z
10 8: 10
<t Z
<t
0
0 1000 2000 1000 2000
IPAR. (MJ m yr ') IPAR. (MJ m' yr ')
Figure 3. (a) Linear correlation between the amount of PAR utilized (IPAR) and
ANPP; (b) total NPP (calculating below-ground allocation based on the method
of Raich and Nadelhoffer (1989) and assuming 50% as NPP) as correlated with
IPARu (from Runyon et a1. 1994).
100
VPD
a:
g o Drought
nOlo
80 Freezing
ii
~
c
c
< 60
.E
c
0
ti
~
40
'2
a:
'EII 20
~
et
0
1A 2 3 4 5 6
SIt
Figure 4. The fractions of annual intercepted PAR that could not be utilized by
the various forest stands because of climate constraints on production (from
Runyon et a1. 1994).
191
Model simulations consisted fIrst of fIve-day output of the hydrologic and canopy
process variables for one year, using the existing LA! estimates for sites 4 and 5, to
validate the "fast-response" hydrologic and physiological processes interannually
across the transect. Then a 100-yr simulation was performed, using LA! for control
sites 4 and 5 and 1990 climate data. The initial stem C was set at 10% of the current
value to represent a young sapling stand just reaching maximum LA!. These results
were used to evaluate time-integrating processes like stem biomass development,
equilibrium conditions in LAI at 100 years, leaf N concentrations as the N cycle
developed and decomposition rates after soil C stabilized. Since the OTIER Project
stands are not even-aged, this simulation only approximated near-equilibrium
conditions.
Model validations: Hydrologic balance

Only two sites, 2 and 5, experienced signifIcant seasonal canopy water stress. Both the
timing and the magnitude of water potentials are well correlated with observations
(Fig. 5, for site 2), implying a ryasonable hydrologic balance calculation, given the
assumed (and perhaps conservative) value of SWHC of 22.6 cm. The date of the
simulated onset of increasing water stress is late, which might be attributed to an
inaccurate measurement of SWHC. Simulations for site 3 predicted a seasonal
summer water stress, consistent with many observations of stress in these forests
(Zobel et al. 1976), but no water stress was actually observed at the site. By increasing
soil available water to 50 cm (equivalent to a rooting depth of>2 m), seasonal water
stress is eliminated in the simulations. This indicates a key limitation in the model
when default values are used.
ANPP
Since FOREST-BGC was conceived as a generic conifer forest model, its performance
for the deciduous alder stand (site lA) may be in error. The model overestimated
ANPP for the alder stand by nearly one-half (Fig. 6a). Running (1994) attributed this
to the lack of an N limitation on growth and probable underestimation of respiration
losses and/or allocation of C to the below-ground components, including N-fIxing root
symbionts. On the other hand, the simulated effects of seasonal water stress for site 3,
which has nearly a year-round photosynthesis period, apparently were not signifIcant
at an annual level.
Treating LAI as a dependent variable tests the model's logic concerning climate
controls on vegetation development. FOREST-BGC considers not only site water
balance but also other limits on canopy development, such as N availability. The
100-yr simulations (Fig. 6b) indicated that the simulated equilibrium LAI at 100 years
for site 3 (4.7) was signifIcantly less than its value at 20 years (6.5). In this case, N
limitation appeared to reduce the maximum value from the measured value of 10.6.
The model predicted that fertilization would allow an LAI of 6.6. However, Matson et
al. (1994) found that N may not be limiting growth on the fertilized site, implicating
the spurious water stress effect in the model. The simulated LAI of the alder stand (site
lA) was significantly higher than the observed value. In this moderate climate with no
N limitation, the limits on alder leaf development are apparently not well described by
the model. The model also overestimated ANPP for alder in a magnitude comparable
to the overestimation of LAI, implying either that simulated respiration losses are too
low or that above-groundlbelow-ground allocation patterns are in error.
Stem biomass
The accumulation of stem biomass as simulated at 100 years for mature, unperturbed
forest stands (with data from Gholz (1982) used for site 5) was consistent with the
measured field values (Fig. 6c). Even though ANPP and LAI were overestimated for
site IA (alder), stem biomass was not. The model predicted rapid early accumulation
for this site, peaking at 30 years (at 132 Mg ha'\ Respiration losses (and mortality,
which is not explicitly treated in the model) reduced stem biomass after 30 years, and
the 100-yr value of 103 Mg ha'l closely matched the observed value. Even though the
a
-2
(i field data
Q. - aGe simulated
~ -1.6
iii
E
Gl
.
1.2
~
~ -0.8

~

-0.4
0 50 100 150 200 250 300 350 400
Day of Year
0.35 140
b - - photosynlhesis
120
~
0.3 - - - transpiration ;,
"c 100 III
or:
0.25
E
,.
0.2
80 "~
Cl
l:
.g 60
Ul
'iii
~ 0.15 Gl
or:
'Q. 40
III i:
l: 0.1 >.
C'CI 20 III
0
F '0
0.05 0 or:
Q.
0 20
0 50 100 150 200 250 300 350 400
Day of Year
Figure 5. Seasonal water relations and photosynthesis for site 2: (a) predicted
(from FOREST-BGC model) and observed predawn leaf water potentials for
1990; (b) simulated seasonal canopy processes illustrating the influence of
drought (from Running 1994).
193
pattern is consistent with observations, the inconsistencies in ANPP and LA! imply
that the model is failing to capture alder dynamics, which, in this case, may require
more explicit inclusion of recruitment, regeneration and mortality values. The conifer
site most poorly predicted was the subalpine site 4, for which stem biomass was
significantly underestimated even though ANPP and LA! were not. Very cold winter
temperatures may affect C allocation patterns at this site.
Nitrogen concentrations
The lOO-yr simulations of leaf N are based on an annual N budget calculation that
includes soil and litter decomposition, leaf N retranslocation and atmospheric
additions stoichiometrically partitioned to leaf, stem and root growth. In this way, the
model controls allocation patterns and limits leaf development. This approach
produced leaf N values well correlated with observed values (Fig. 6d) but with a
substantial bias when the fertilized and alder sites were included. The source of this
bias needs further research and possibly improved logic in the soil N compartments,
using mechanistic methods such,as those of Rastetter et al. (1991).
__ 20..-----------""'" 6r----------....,....-----,
~ Yc 1.10X+O.47 "._,,-'
-~ 16 RZ=Q.82 ..-"/,,,- 1:1
co
!. 12
.JJ
"..",
~
z -,,-,;,'
,.,.,.
!!!
J (b)
en 00~:.-..,4,..----=-6----:,'="2---"16::---'-:!2O 10
Measured NPP (Mg ha ot yr"')
O.S ,,:.- c
(e) (dl
0" 0.5 1 1.5 2 2.5 3 3,5 4
100 200 300 400 500 600 700
Me...ure<t Stem 81oma. . IMg I\a ') Me..ul'Od Loll N ('llo Dry M...)
Figure 6. Validation of the FOREST-BGC model predictions of (a) ANPP against

observed field measurements for climate year 1991 (fertilized sites 3F and SF
excluded) (from Running 1994); 100-yr simulations of LA! using the climate of
1991 and measured LA! (from Runyon et aI. 1994) (adapted from Running
1994); (c) 100-yr accumulation of stem biomass for 1991 climate year compared
with measured stem biomass by Runyon et aI. (1994) (from Running 1994); and
(d) simulated leaf N concentration at the end of a 100-yr simulation, using 1991
climate year compared with values measured in March 1990 by Matson et a1.
(1994) (from Running 1994).
Remote sensing results
Surface climate variables

Incident Photosynthetically Active Radiation (PAR). Incident PAR was estimated from
a physical model of potential PAR for a cloudless atmosphere reduced by the
attenuation estimated from analysis of TOMS UV data. The calculation was applied to
large cells of about 1 latitude by 1.25 longitude. Three cells covered all of the
0 0
OTTER Project sites. Attenuation was estimated by assuming a low surface

reflectance in the ultraviolet of 2-5%, considering that Rayleigh scattering and
absorption in the ultraviolet is produced by the same atmospheric features that
attenuate PAR. This mean monthly TOMS value (taken at noontime) was used to
calculate incident PAR for the OTTER Project, with excellent correspondence with
2
field data (R = .96), showing no bias across all sites and a small underestimate for the
coastal site.
Fraction of incident PM intercepted by the canopy (f;PAR). f[PAR as estimated on the
ground using the ceptdmeter ignores the radiation reflected from the canopy, which
introduces an error of about 2-10% in the visible spectrum. Field measurements were
also made at solar noon at midseason and the value applied to the full year, assuming
that (i) the seasonal capacity of the canopy to absorb PAR varies little when the
radiation is high and (ii) wintertime variation is insignificant because radiation loads
are low and zenith angles are steep.
AVHRR NDVI data were also used to estimate f[PAR' using results from other
studies. The capacity of canopy leaf area to absorb PAR has an approximately
exponential relationship with LAI. Several studies in Oregon, including those of
Spanner et al. (1990a) and Law and Waring (1994), have shown that the SR of near-
infrared-to-red reflectance is roughly linear with LAI and asymptotic with hPAR'
Transforming the SR to NDVI tends to linearalize the relationship between hPAR and
NDVI, making NDVI a more sensitive estimator offlPAR'
Although the reflectance of conifer canopies tends to be low relative to broadleaf
plants of comparable LAI (i.e., the additive atmospheric effects result in lower NDVI
for equivalent LAI), with proper calibration NDVI can be used to estimate hPAR
regionally. The effect of the atmosphere is to suppress the maximum value of AVHRR
NDVI, which we corrected for by calibrating values against NDVI taken by a
spectroradiometer mounted on an ultralight aircraft flying at 500 m. Then calibrated
AVHRR NDVI values were used to estimate IPAR for midsummer conditions (Fig. 7).
The one outlier is site 5, the harvested Ponderosa pine stand, likely because of
underestimation of hPAR in the field due to sampling difficulties at low LAI and
clumped tree distributions. In addition, NDVI of the understory is higher than for the
conifers for the equivalent LAI, so that the composite NDVI is higher than an equal
LAI of conifers alone. Exposed sandy soils and grasses at this site would also
contribute to a higher NDVI.
Finally, site data from Figure 7 were multiplied by incident PAR to estimate
satellite-derived IPAR (Fig. 8). As would be expected from the previous analysis, the
satellite method overestimates IPAR for site 5, while the relationship for the remaining
195
100+---_t_~-_t_~-_t_~-_+_~~_+
%IPAR =(NOVI121) - 4.0

r" =0.99, excluding Site 5
80
a::
~
~
'C 60 ; _..__ _- --,<81
f
:::l
III
m
E
-b 40
c::
:::l
E!
Cl
0.2 0.4 0.6 0.8

July AVHRR NOVI
Figure 7. Fraction of annual IPAR intercepted by vegetation from ground-

measurement techniques compared with NDVI calculated from atmospherically
adjusted AVHRR data (from Goward et aI. 1994).
2200
.
"',.,
1800
':'E ])
...
~
a:: 1400
~
'C
!OJ
E 1000
:;;
Gl
~
~
rn
2oo-f---+---t-----t----t---+
200 600 1000 1400 1800 2200
Ground-estimated IPAR (MJ m-2 yr")
Figure 8. Satellite vs ground estimates of annual total IPAR (from Goward et aI.
1994). Ground estimates are from measurements under the canopy of percent
PAR transmitted and from pyranometer observations. Satellite estimates are
based on the relationship in Figure 7 and the estimates of incident PAR derived
from TOMS data (Goward et aI. 1994).
sites tends to be linear although biased, requiring further examination. One potential
source of this bias is the calibration of the AVHRR NDVI using the ultralight data.
Even for a flight altitude of 500 m, the ultralight NDVI values still include a significant
additive atmospheric effect, which is more intense closer to the ground. While a
ground-stationed sunphotometer was used during the aircraft campaigns to measure
the optical depth of the atmosphere to correct for atmospheric effects, use of such an
instrument was impractical for the numerous AVHRR dates over the entire transect,
especially since many of the scenes were retrieved from the archives after the
conclusion of the field campaigns.
Absolute humidity. The relationship between the absolute humidity measured at the
meteorological stations and that estimated by the slope of the plot between the two
thermal channels of AVHRR data is shown in Figure 9. As absolute humidity
increases, greater atmospheric attenuation in the longer wavelength channel of
AVHRR produces a lower observed brightness temperature, and a reduced slope. The
results are generally consistent with more humid conditions at the western end of the
transect. Scatter in the relationship can be attributed to several sources. First, the
humidity sensors on the ground began to read too low at midseason. Second,
cloudiness obscured the satellite data for a number of dates, while the precision in the
thermal temperatures reported for the AVHRR data was only 0.5c. Excluding the
outliers, the RMSE error is about 2.6 g m 3, with a coefficient of variation of 20-40%.
These results are insufficient for the purposes of the light-use efficiency model, even
though there is a relationship and seasonal patterns look reasonable. Further research
is required to improve this approach for regional applications.
Air temperature. The relationship between satellite estimates of air temperature and
ground measurements was linear, with little bias, uniform scatter and an RMSE of
5.4C. This RMSE value is substantially higher than that found for ocean surface
temperatures (l.OC, McClain et al. 1985) and three times the RMSE found for
agricultural sites on flat terrain (Cooper and Asrar 1989). All of the OTTER Project
temperatures were for midday and the satellite data include some clouds and snow
effects. The scatter may also result from unknown variations in surface emissivities in
these heavily disturbed landscapes. Using a different approach, Running et al. (l987b)
compared nighttime minimum brightness temperatures obtained from the 0200
overpass of AVHRR with dew-point temperatures measured on the ground for conifer
forest sites in Montana. The added potential benefit of their analysis is that absolute
humidity can be calculated and converted to relative humidity using daytime air
temperatures.
Vapor pressure deficit. Absolute humidity and air temperature can be used to calculate
atmospheric vapor pressure using the formula of Rosenberg et al. (1983)
X=eaMmJRT
where X = absolute humidity, ea = vapor pressure, M m= molecular mass of water, R =
universal gas constant, and T =absolute air temperature. Vapor pressure deficit (VPD)
is then the difference between (e.) and saturation vapor pressure as a function of T
alone. Although scatter in the OTTER Project data (Fig. 10) is high (RMSE = 0.7 kPa),
197
16 v
0
0
oo~
, ._--
uc ..... 0
0
0 e
._- .- . e 0
00
00 e
r-.
rt._
oeQ>
.. 0 cloud.
0
)( 6P
.. $~
~
..
)( )(
- -- --
0 .... 1
o lIlte2
""+ .. .j-
)(
.. .. 0
o -.,
X ..... 0 0 e
.. lIlte,
+
o
0.7 0.8 0.8 1.1 1.2
Slope of Relation Between T. llIld T.
Figure 9. Estimates of absolute humidity from analysis of temperature data of

the AVHRR satellite vs field measurements (from Goward et al. 1994).
6
( 061"1
=
e 61.. 2
061 3
)( S'''.
_. -- -- .
~ .. SI.. 6
; 3 - r 2 0.54
e
)(
i
x
i
+
:>
2
,
I
x
11'I O
)
b .' _. ___1._
0 I
0
-1 F--+---+---+---1--~--t
-1 0 1 2 3 6
Vapor Pres.ure Deflcl1 (kPlI)
Eatimated from S8te1 rementa
Figure 10. Vapor pressure deficit at the land surface, as measured at the hour of
the satellite overpass from meteorological stations, compared with computed
estimates from analysis of satellite data (from Goward et a1. 1994).
there is little bias. An RMSE of about 0.1 kPa is preferred for modeling. Given the
error in both Tair and, the error in VPD is not surprising, implying that improvements
in estimating temperatures or better knowledge of surface emissivities could reduce
this error significantly.
Soil drought. Drought was estimated using changes in the slope of the NDVI and T s
relationship over an array of 9 x 9 pixels surrounding each site. The correspondence of
this index to field-measured values of predawn leaf water potentials was poor (see Fig.
16 in Goward et al. 1994). The satellite index may respond more to conditions of the
surface soils than to the underlying soil conditions experienced by most roots, leading
to a very premature sensing of soil dryness. While further research may improve this
technique, these results cannot be used to predict drought conditions. Since drought
accounted for a very sizable proportion of the reduction in utilized PAR in the
light-use efficiency model (Runyon et al. 1994), the satellite estimates of drought
found in the OTTER Project are not practical at this time.
Biophysical and structural variables
Leaf Area Index (LAI)

Atmospheric corrections. Data from a variety of broad-band and high-spectral-
resolution sensors (AVIRIS, TMS, CASI, SE590 and ASAS, Table 2) were studied for
spectral relationships with LAI (Spanner et al. 1994). To make comparisons across
such a wide range in sensor performance characteristics and flight altitudes required
correction to top-of-canopy reflected radiance. The at-sensor radiance measurements
for all but the SE590 sensor were corrected for additive atmospheric effects using the
data from the sunphotometers located on the ground. The method of Kneizys et al.
(1989), called the LOWTRAN-7 code, was used. Aerosol optical depths varied by a
factor of three across all dates, with the highest values recorded in August due to
smoke from wildfires and agricultural burning. For other dates, the low optical depths
are characteristic of very clear atmospheric conditions. Because the SE590 data were
taken frequently at about 100 m above the ground, simultaneous sunphotometer
measurements were not possible. Instead, a halon white reflectance panel was
mounted on board beneath the sensor, the aircraft oriented toward the sun and
reference readings taken at a variety of flight altitudes. Regression of these readings to
the elevation of the particular study site provided correction to top-of-canopy values.
Broad-band TMS results. Previous research has shown that the SR of near-infrared-to-
red band reflectance is sensitive to a broader range of LAI than are other indices
(Peterson et al. 1987). In this study, the SR was calculated using the band closest to the
680 nm chlorophyll absorption maximum and the near-infrared band taken from the
780-790 nm region. The results for the AVIRIS, CASI, TMS and SE590 data are
presented in Figure 11. The relationship for the TMS data is clearly asymptotic to LAI
= 7-8; a logarithmic curve fits the data well and yields a standard error of estimation
of 0.47. These data include observations for August at peak LAI and those for March
at minimum LA!. Although the range of LAIs is different, the relationship is extremely
consistent. The recent work of Myneni et al. (1994) has shown that the SR and most
199
other indices used to monitor vegetation amount are primarily sensitive to the amount
and activity of absorbers in the canopy. Not only is the LAI at a minimum in March,
but winter levels of chlorophyll in the leaves are still suppressed for the eastern sites.
This TMS relationship is significantly lower than our previously published results for
Oregon forests (Peterson et al. 1987). In those studies, we estimated the atmospheric
correction by regressing reflectance across a range of brightness targets measured
from a helicopter. Those studies clearly overcorrected the red band data, making the
SR too high. Since canopy reflectance in the red band is very low (about 2%) and most
of the at-sensor radiance arises from the atmosphere, the vegetation indices are very
sensitive to variations in the red band, the denominator.
High-spectraL-resoLution sensors. While a logarithmic relationship was also fitted to
the data from the other (high-spectral-resolution) sensors for purposes of comparison,
the results appear to be almost linear, although the scatter in the data is higher. In
addition, the ratios are all significantly higher, nearly twice those of the TMS, and the
results more consistent among the sensors. The red bandwidth of the TMS is 60 nm
(630-690 nm), whereas those of the other sensors are from 6 to 10 nm. Since the TMS
band tends to integrate higher values on either side of the chlorophyll peak, the value
should be somewhat higher, leading to lower values of the SR. Tests of this effect
showed that only about 10% of the difference could be explained in this way. The only
other possible source of this difference is radiometric calibration, as noted before by
20
o TMS
6 SE590 6
16 AVIRIS
CASI
0
~ 12
a:
"..
~ 8 0
~
0
0 2 4 6 8 10 12
Leaf Area Index
Figure IJ. Maximum and minimum LAI (from two seasons) compared with
atmospherically corrected simple ratio (SR) data from broad-band (TMS) and
narrow-band (AVIRIS, Spectron, CASI) airborne remote sensing instruments
(from Spanner et al. 1994). For all instruments, August SR was selected to
correspond to the seasonal maximum, while March SR was used for the seasonal
minimum except for the CASI data (May). A logarithmic equation was fitted to
each instrument data set to facilitate comparison.
Wrigley et al. (1992). While the standard error of estimation is sufficient for
ecosystem modeling purposes, this error is known to increase for more open canopies
having large background reflectance variations (Spanner et al. 1990b).
Seasonal LAI: TMS results. Tracking the SR across all dates of the campaign showed
that some seasonal variation is observed. However, the patterns are only partially
consistent with seasonal variations in LAI. A variety of other radiometric factors
appear to be more responsible. At low sun angles in the spring and fall, red upwelling
radiance is typically very low and unstable, especially following correction for
atmospheric effects. This leads to increased uncertainty in the SR during low sun angle
conditions. In general, the SR was consistently higher during the summer at maximum
LAI display than for other seasons. While seasonal variation in LAI explains some of
the variation observed here and in Spanner et al. (1990a) for AVHRR data, solar
illumination angle, sensor view angle and atmospheric correction errors are likely to
be responsible for the lack of consistency in sensing seasonal patterns of LA! for all
dates, despite the favorable comparison between March and August.
Off-nadir view angle effects. One of the biotic factors thought to be responsible for
difficulties in sensing seasonal changes in overstory LA! is the seasonal variation in
the background, particularly in more open forest stands. Spanner et al. (1990b)
showed that when the canopy is seen from the nadir view, background variations can
have a pronounced effect on both SR and NDVI. One potential solution to this
problem is off-nadir viewing, particularly in the retrodirection, i.e., along the angle of
solar illumination. At this angle, all shadows are minimized and the view is dominated
by only illuminated objects. Further, the tall architecture of conifer canopies would
tend to cover the background when the latter is viewed obliquely. Johnson et al. (1994)
analyzed the data from the Advanced Solid-state Array Spectrometer (ASAS), which
acquires a sequence of images of the canopy from seven different view angles along
the flight path of the aircraft. When the aircraft flight path is aligned with the solar
illumination angle, data recorded looking forward from the aircraft represent
backscattered radiation near the retrodirection, while data obtained looking backward
represent forward scattered radiation. The theory of bidirectional reflectance
distribution predicts that the highest reflectance values are sensed in the retrodirection.
Johnson et al. compared data for June and August, during which large variations in
senescence of understory vegetation typically are observed. Their results indicated
that while the asymptotic behavior of the SR:LAI relationship is a reliable predictor
for off-nadir view angles, off-nadir backscatter relationships are significantly better
for predicting one month using the other month's data. These results suggest that
variance caused by changes in the background reflectance is minimized when the
sensor views the canopy at an off-nadir angle.
Understory LAI studies. Thinning of the two Ponderosa pine stands at sites 5 and 5F
released the understory plants; these consisted of bitterbrush (Purshia tridentata) on
both sites, with Idaho fescue (Festuca idahoensis) in one stand and greenleaf
manzanita (Arctostaphlos patula) in the other. These plants represent 46% of the total
vegetation covering 18% of the ground area. Artificial canopies having a broad range
of LAI were created from freshly cut branches of the two shrubs, arranged in a small
201
measurement area in the field to mimic the actual plants, and reflectance and IPAR
were measured using a spectroradiometer and ceptometer, respectively (Law and
Waring 1994). The f!pAR was logarithmically related to LAI, as expected, with good
results. Various spectral indices were tested and it was determined that/!PAR was best
estimated by NDVI, while SR best estimated LAI (Fig. 12). In this case, the SR:LAI
relationship was linear up to an LAI of 8, a result very similar to that obtained using
the same spectroradiometer from the ultralight to estimate canopy LAI. These data
were used to calculate the contribution of the understory vegetation to ANPP, > 40%
(115 g m'2 yr'!) of the total ANPP (265 g m'2 yr,I). This total ANPP is very similar to
the undisturbed ANPP of a nearby Ponderosa pine stand as measured by Gholz (1982,
220 g m'2 y r'\
1.0
Manzanita
.. 10
Manzanita
(e)
0
..
(a)
0.8 0
8
.0
a: 0.6 6
~ Ci
<;:. ...J
0.4 4
0.2 2
y= -0.126 + 1.316x RZ = 0.86 y= -2.132 + 1.129x R 2 =0.86
0.0 0
0.0 0.2 0.4 0.8 0.8 0 2 4 6 8 10
NOVI SR(NIRIR)
1.0 10
Bilterbrush Bltterbrush
(b) (d)
0.6 6
0.6 6
a: Ci
.....
a.
<;:. 0.4 4
0.2 2
Y = -0.109 + 1.332x R Z= 0.83 Y= -2.022 + 1.294x RZ =0.74
0.0 0
0.0 0.2 0.4 0.6 0.8 a 2 4 6 8 10
NOVI SR (NIRIR)
Figure 12. (a, b) Relationships between fractional IPAR for simulated

understories of bitterbrush and manzanita at site 5 and NDVI (calculated from
data taken from Spectron SE590 field spectroradiometer); and (c, d)
relationships between understory LAI and the simple ratio (SR) (from Law and
Waring 1994).
SpectraL derivatives ofhigh-spectraL-resoLution data. Gong et al. (1992) estimated the

2
variation in overstory LAI (0.87-2.72 m2 m ) for points throughout site 5 using the
hyperspectral data of the Compact Airborne Spectrographic Imager (CASI) operating
in spectral mode. First- and second-order derivatives of reflectance spectra were used
to suppress the effects of soil background on the forest spectral reflectances. Spectral
derivatives increase the correlations between LAI and derivative spectra of CASI data
more than those between LAI and the reflectance spectra of CASI data when
environmental variability such as background soil and atmospheric effects vary at a
lower rate than signal spectra. For instance, the highest coefficient of determination
obtained for single-channel LAI prediction using reflectance data is 0.681 with a
standard error of estimation of 0.345. These values are considerably improved to 0.904
and 0.189 for wavelength 472 nm, and to 0.898 and 0.195 after taking the first- and
second-order derivatives, respectively.
A similar analysis was performed for AVIRIS data for all sites and dates by Johnson
et al. (1994). They found that AVIRIS Reflectance Factor (RF) was negatively
correlated to LAI throughout the visible and short-wave infrared regions, and
positively correlated in the near-infrared. This pattern is mainly due to absorption in
the visible and short-wave infrared regions by chlorophyll and the accessory pigments,
and by leaf water, respectively. There is a strong positive correlation with LAI in the
near-infrared due to leaf scattering, consistent with the results of Spanner et al.
(1990b) across conifer forests throughout the western United States, but in contrast
with the findings of Peterson et al. (1987) and Gong et al. (1992). Spanner et al.'s
findings were for closed canopy forests only, whereas Gong et al.'s study was for only
the Ponderosa pine sites. In the latter case, the high spatial resolution of the CASI data
would introduce spectral understory variation due to canopy openness, which can
confuse the SR:LAI relationships, as was found in Spanner et al. (1990b) and was
probably true for Peterson et al.'s findings as well. For the first-order derivative of RF
(RF'), LAI is strongly correlated at 752 nm (81 % of variance explained), the red edge
(Fig. 13). Here, as in Gong et aI., the derivative of RF performs better for single
channels than does the RF alone. A multiple stepwise regression was performed for
both the AVIRIS and the CASI data against LA!. Johnson et al. (1994) found that a
three-term equation explained 97% of variance with RF' wavelength selections of 752
nm, 1280 nm and 1165 nm, whereas Gong et al. found that a two-term equation
explained 96.3% of variance with selected RF' at 472 and 719 nm.
Foliar biomass estimation by RT modeling. An entirely different approach to
estimation of LAI and foliar biomass was taken by Wu and Strahler (1994). They used
the sites and data of Peterson et al. (1987) rather than the OTTER Project sites because
of previous extensive field spectral measurements, but the results were nevertheless
entirely consistent with those of the OTTER Project. The geometric-optics model of
Li and Strahler (1986) was calibrated from illuminated and shaded radiances of the
scene components together with data on tree spacing distribution functions. The model
was then inverted from observed canopy reflectance to estimate tree density and
average crown radius. The model was statistically fitted to the spatial pattern of
reflectance from an array of pixels surrounding the central point of interest. The two
estimation variables, density and crown radius, were sufficient to estimate foliar
203
reflectance
factor
-C
QI
'u
first
derivative
~o
o 0
c
.2
tii
..
~
o
o
Wavelength (nm)
Figure 13. Correlogram between LAI and AVIRIS data across all sites and all
dates except March 1990 (LAI from Runyon et aI. 1994). Solid line represents
correlogram of LAI and reflectance factor data after correction of AVIRIS data
for atmospheric and cosine effects. Dashed line represents correlogram between
LAI and the first derivative of the reflectance factor data (from Johnson et aI.
1994).
biomass and standing woody biomass from allometric relations. Crown radius was
linearly related to tree bole diameter measured at breast height (dbh) and distributed
spatially according to a lognormal function. Having estimated mean crown radius and
thus mean dbh, Wu and Strahler (1994) estimated other tree properties from allometric
relationships. The results for foliar biomass indicate this approach succeeded
remarkably well (Fig. 14a). The main problems encountered were in fitting the model
to the data from the most dense conifer stands, through which little if any of the
background reflectance was visible. Foliar biomass was easily converted to LAI using
specific leaf area. Estimation of standing biomass is discussed below.
Specific leaf area (SLA) relationships. Pierce et al. (1994) examined the relationships
2
between LAI and specific leaf area (SLA, cm g-l) and leafN content. They found that
SLA tends to decrease with increasing aridity of the forest site, leading to increased
water-use efficiency. Accordingly, SLA for the moderate, wet coastal stands is highest
in Oregon and declines toward the eastern sites, with the lowest values found for the
western juniper stands in the high desert. Thus, SLA is correlated with LA!. This leads
to the intriguing question of just what the spectral indices are actually sensing. Recent
radiative transfer analyses by Myneni et al. (1994) showed that each of the various
indices used in remote sensing can be reduced in the limit of a continuous function to
a derivative of the red edge in plants, the sharp change in reflectance between the
strong chlorophyll absorption region in the red band and the highly reflective
scattering region in the near-infrared. Myneni et al. concluded that NDVI and SR are
related to the amount and activity of the absorbers in the canopy leaves, or chlorophyll
per unit leaf area. Across the conifer species of Oregon, N concentration (in mg ii,
roughly proportional to chlorophyll concentration) is fairly constant; thus chlorophyll
per unit leaf area tends to increase across the Oregon transect (west to east). Since
variations in SLA across Oregon significantly contribute to the range of LAI by
altering the allocation of foliar biomass to leaf surface area, the OlTER Project data
appear to be consistent with Myneni et al.'s theory. Pierce et al. (1994) suggested that
LAI estimated by sensor data may be used to estimate SLA when, in fact, the converse
may be true provided that an estimate of foliar biomass can be attained, as in Wu and
Strahler (1994).
Standing biomass
Estimation using RT modeling of optical data. Inversion of the geometric-optics
model to crown radius and tree density leads to allometric estimates of standing
woody biomass. By adding foliar biomass, total above-ground standing biomass is
produced, which is predicted reasonably well up to values of about 500 Mg ha- 1 (Fig.
14b). The model significantly overestimates the highest biomass stand (2.7 Mg ha- 1 vs
1.1 Mg ha- 1 observed). For these dense canopy situations, mutual shading and no
reference to the background soil/understory reflectance reduces the performance of
the model (Li and Strahler 1992). While the model is in error for this very high-
biomass, old-growth Douglas-fir stand, the maximum stand biomass measured for
OlTER Project sites was only about 700 Mg ha- ' , and the range is more typical of the
second-growth forests that dominate the current landscape of Oregon. The method of
Wu and Strahler (1994) is thus potentially well suited to the continuous geographic
estimation of biomass over large regions because the method depends only upon the
image data itself once the model is calibrated. Prior to the development of this
approach, standing biomass was estimated from remote sensing data using the
methods of multistage sampling (Peterson and Card 1977). The multistage inventory
approach, however accurate, suffers from a lack of geographic specificity: only larger
aggregated segments of the landscape can be estimated with success. The model ofWu
and Strahler (1994) needs further refinement to reduce the problems encountered with
dense canopies and to simplify the requirements for calibration so that the model is
more accessible for use.
While the approach used by Wu and Strahler (1994) considered the spatial variance
in a neighborhood of pixels surrounding the central point of interest, the model can
also be fitted to the angular variance of a single pixel, the bidirectional reflectance
distribution function (BRDF), as measured by such sensors as ASAS. Abuelgasim and
Strahler (1994) performed experiments to test the fit between ASAS data and the
mutual shadowing version of the geometric-optics model (Li and Strahler 1992) and
found that the basic features of directional reflectance are captured by the model. For
most sites, the model correctly determines the strong reflectance asymmetry in the
solar principal plane with the viewing angle. These results are encouraging and
suggest various improvements in the model. The most significant is an improvement
of the measurement technique for component signatures, both illuminated and shaded
(see Abuelgasim and Strahler 1994). Second, canopy transmittance and multiple
205
:c
Gi
30000
i.
~
(a)
m
.c
Cl
C.
III
III
m
E
0
iii
Gl
Cl
.~
"0
u.
5000 10000 15000 20000 25000 30000
Foliage Biomass (kg ha- 1) (Remote Sensing)
~
2000
'C (b)
Gi
~
~
";"
1500
m
.c
Cl
~ 1000

III
III
m

E
0
iii
500
-
Cii
{2
00

500 1000 1500 2000 2500 3000
Total Biomass (Mg ha1) (Remote Sensing)
Figure 14. (a) Estimated foliar biomass of nine test stands (Peterson et aI. 1987)
using inversion of the geometric-optics model compared with foliar biomass
calculated using allometric relationships based on ground measurements of tree
dbh; the coefficient of determination of the regression is 0.75 (from Wu and
Strahler 1994); (b) total standing biomass of nine test stands (Peterson et al.
1987) as estimated by inversion of the geometric-optics model compared with
estimated total standing biomass from ground dbh measurements; coefficient of
determination of regression is 0.63 (from Wu and Strahler 1994).
scattering among crowns and the ground might be accommodated (Nilson and Ross
1996). The model is not yet ready for inversion to test for its capability to estimate
biomass from the angular variance measurements. The authors suggest that full
characterization of spatial variance, angular variance and the relationship between the
two induced by the nature of the objects in the scene and their illumination conditions
may lead ultimately to an accurate inversion procedure (Jupp and Strahler 1991, Jupp
and Walker 1996).
Estimation using RT modeling of radar data. The recent interest in using

multifrequency, polarimetric synthetic aperture radar (SAR) measurements to
characterize forest properties stems from the fact that, in contrast to optical remote
sensing methods, SAR generates microwaves that can penetrate cloud and smoke and
are independent of daylight and temperature (Moghaddam et al. 1994). Studies in low-
biomass ecosystems have indicated that it may be possible to obtain information about
trunk structure (height, radius, dielectric constant), understory structure (shape and
size of components, and their dielectric constant and orientation) and the forest floor
(roughness and permitivity), if longer wavelength microwaves are used (Kasischke et
al. 1985, Cimino et al. 1986, Sun and Simonett 1988, Hess et al. 1990, Way et aI. 1990,
Le Toan et al. 1992, Dobson et al. 1992). Theoretical modeling of the forward-
scattering problem (Durden et al. 1988, 1989, 1991) has indicated that, in forested
regions as diverse as those in Oregon, various scattering mechanisms may dominate
depending upon the combination of forest components and their condition.
Combinations of these quantities may be used to determine forest biomass, but given
the biomass, one cannot use it to calculate the total backscatter power, even though the
calculation can be performed given the components. The reason is that a given value
of biomass can correspond to several combinations of these component parameters,
each combination resulting in a different backscatter. While previous research had
been limited to biomass values up to about 200 Mg ha", the greater range found in
Oregon (8-960 Mg ha'l) provided a rigorous test of the resolving power of these data
to determine standing biomass.
All polarizations of the radar data at all three wavelengths saturate as biomass
increases to around 200 Mg ha'l, regardless of radar beam incidence angle (for
example, Fig. 15). Comparable results are found for other forest parameters, density,
basal area and trunk height. Further, for smaller values, the change in radar backscatter
is nonuniform and unsmooth, but this might be attributed to the small sample size. A
sensitivity analysis was carried out using the forest scattering model, where the
variations in radar cross section were calculated as several model parameters were
changed. Here also there is no unique dependence of radar backscatter with biomass,
although the model does show unique and sensitive relationships between radar
backscatter and dielectric constant, soil properties and density below the saturation
point. Algorithms would have to be developed to estimate these parameters from radar
data, and their combination may help provide inputs to ecosystem models. Because of
the complexity of the present scattering model, it is almost impossible to formulate an
inversion/estimation algorithm. Hence, simplification of the model is needed to reduce
the number of unknowns, given the limited number of measurements available.
Resorting to an empirical regression approach, as suggested by some researchers (e.g.,
Le Toan et al. 1992), may suffice for the range and specific arrangement of forest
components for which the corresponding data are obtained, but ultimately, inversion
algorithms based on theoretical scattering models will be needed to provide more
versatility and applicability to larger forested regions (Moghaddam et al. 1994).
207
0.--------------, 0,-------------,
(a) (b)
iii' -5
iii' .5 ~
~
c c
o o
U ~
.
Jl 10 (/)
co
co
e
e u
~ 15 pw > PHV
> ...... LYY :I: .25 ..... LHY
CW C-HV
.3O+--........---.---..---........----l
200 400 600 800 1000 o 200 400 600 800 1000
Woody Biomass (I ha") Woody Biomass (I ha"')
Or------------,
(c)
iii'
~c 5 _~ .. - -0'
.2 , ~. ...................
~ 10 :. ~:);:.~:.~ .
e :".,-
U
b
:I: -15
PHH
:I:
..... LHH
C-HH
........-~
20~--:-!":--r---...,....-
o 200 400 600 800 1000
Woody Biomass (I ha ')
Figure J5. Relationships between various Airborne Synthetic Aperture Radar

(ASAR) variables and total standing woody biomass: (a) VV-radar-scattering
cross section; (b) HV-radar-scattering cross section; and (c) HH-radar-scattering
cross section. po, L- and C-band lines were calculated from the forest model of
Durden et al. (1988, 1989, 1991) (from Moghaddam et al. 1994).
Canopy biochemical variables

The influence of leaf biochemicals on leaf reflectance is due to absorption throughout
the infrared region from various stretching modes of organic bonds. Because
biochemical compounds share many of the same chemical bonds (those of the light
atoms C, H, 0, N) responsible for absorption, the challenge is to differentiate the
contribution of one compound without undue interference from other compounds
having an overlapping absorption spectrum. These studies require high-spectral-
resolution data with a high signal-to-noise characteristic (Gholz et al. 1996).
The only airborne sensor with such characteristics available at the time of the
OTfER Project was the Airborne Visible-Infrared Imaging Spectrometer (AVIRIS),
with a spectral resolution of about 10 nm continuous from 400-2400 nm and a signal-
to-noise ratio (SNR) of several hundred. The CASI sensor has an even higher spectral
resolution, 3-6 nm, but is limited to the 425-925 nm region. The CASI data were
studied to determine relationships to the chlorophyll red edge. The relationships

between canopy biochemical content, measured as total canopy content (kg ha- l ),
concentration by weight (mg g"l) and concentration per unit leaf area (mg cm- 2), and
the canopy reflectance factor (RF) and its first-order derivative (RF'), were analyzed
by correlogram by wavelength to look for regions of strong correspondence and by
stepwise multiple regression to develop multiterm estimation equations relating
chemical content to reflectance and its derivative (Johnson et al. 1994, Matson et al.
1994). Selection of estimation or correlative wavelength regions was interpreted using
published tables of absorption from organic bonds (Barton and Windham 1988,
Barton and Himmelsbach 1991). In a separate analysis reported in Matson et al.
(1994), an inverted Gaussian model (Miller et al. 1990) was fitted to the red-edge
feature in the RF spectrum, resulting in mean values of the red-edge spectral position
or inflection point, and the chlorophyll position as well as the NDVI. The results of
this model were related to foliar biomass, chlorophyll and N content:
Total canopy contents (kg ha-'j

Total canopy nitrogen (total N), chlorophyll and lignin are strongly correlated with
LAI because of the large regional variations in LAI relative to the more slowly varying
chemical concentrations. Since RF is dominated by variations in LAI, only the
derivative RF results (RF') are reported here. The RF' at 752 nm is the strongest
wavelength associated with total N and chlorophyll as well as with LAI, explaining
76%, 67% and 81 % of variance, respectively. This is not too surprising given the
strong intercorrelation of these three attributes. The feature at 752 nm is the maximum
slope of the red edge (Gholz et al. 1996). Given the analysis of Myneni et al. (1994),
this selection would seem to correspond to the amount and activity of chlorophyll, the
strongest absorber in the region. The other two wavelengths selected for a three-term
2
total N estimation equation were 1593 and 2283 nm (R = 0.90). While each selection
improved the correlation, only the selection at 2283 nm had a direct interpretation. It
occurred midway between two N-H combination bands at 2274 and 2294 nm. The
other wavelengths selected for chlorophyll cannot be explained in this way. When the
spectral range was limited to only the 1500-1800 nm region, canopy total N, biomass
and lignin content all included 1772 nm in their estimation equations, a spectral
feature associated with foliar water. Once this partial correlation at 1772 nm was
included, the remaining two selections for total N were 1545 and 1554 nm, near a
major N-H absorption feature centered at 1510 nm.
Lignin content was estimated by a three-term estimation equation involving RF'
at 1297, 1261 and 2353 nm. The 752-nm band was not selected despite the strong
intercorrelation with LAI (91 %). The first two wavelengths occur in a region of
strong C-H second overtones near 1230 nm, but are high for this feature. This
correlation may also be influenced by the cross-correlation with LA!. The third
term, RF' at 2353 nm, is located between two major lignin C-H combination bands
centered at 2332 and 2380 nm. For the analysis confined to the 1500-1800 nm
region, the wavelengths selected were 1644, 1663 and 1772 nm, the latter the same
as for foliar biomass. Several lignin absorption features have been identified in this
region at 1674 and 1696 nm.
209
Correlations between starch content and RF' were not significant, perhaps due to
the spectral confusion with cellulose, which varies with time differently from starch
reserves.
Concentration per unit leaf area

While the theoretical study of spectral derivatives by Myneni et al. (1994)
concentrated on the red edge feature in plants, the arguments are equally compelling
for absorber amount and activity for the other biochemical fractions. Myneni et al.
implied that chemical content per unit leaf area should be related directly to the
spectral derivatives. Chemical concentration measured as g cm-2 is poorly correlated
with LA! (-0.65 for total N, -0.63 for lignin, -0.15 for chlorophyll and -0.18 for starch),
unlike canopy content. Thus foliar amount should not bias these correlations. The first
total N selection against RF' is 552 nm, the "green peak" of vegetation reflectance.
While other investigators have correlated chlorophyll concentration (approximately
proportional to total N) with RF at 540 (Horler et al. 1983) and 580 nm (Card et al.
1988), the RF' should go to zer9 in the first-order derivative precisely at the green
peak. The result found here implies miscalibration of the sensor data, variation in the
precise location of the green peak in different species (Goward et al. 1994) or possible
interactive effects with other constituents. The other two selections at 1184 and 2164
nm occur off peak near several protein absorptance features at 1188 and 2188 nm.
The selected RF' wavelengths for chlorophyll seem to have little physical
significance when the red edge is not selected. However, Curran et al. (1990) showed
how the red edge predicts leaf chlorophyll well but is obscured at the canopy level in
Pinus elliottii forests because of background variations. The first selection at 1069 nm
is off the peak N-H feature at 1060 nm. A similar situation prevails for lignin, where
the first two selected wavelengths occur at absorption minima, according to Barton
and Himmelsbach (1991). The lignin selections are also inconsistent with earlier work
of Wessman et al. (1988). Again, starch concentration is not significantly correlated.
Despite the good fit between measured and predicted total N, chlorophyll and lignin
for both canopy content and for concentration per unit leaf area, these results are
somewhat inconclusive. In about half the RF' wavelengths selected, plausible
explanations can be advanced about the bond feature responsible, which is
encouraging. In the remaining cases, an explanation could not be postulated, either
because the selected wavelength occurs a bit too far from a bond known to be
associated with a particular biochemical, or because of possible interactive effects
with absorption by other covarying chemicals. Further, it remains unclear just what
chemical property is being sensed, the total content or the concentration. Multiple
scattering within leaves and within canopies, coupled with diffuse transmittance by
leaves in the infrared, complicates interpretation of what material property is being
sensed. Some intermediate property determined by the mean free path of photons
through leaves and canopies, coupled with the distribution of chemical compounds,
some entity between total content and concentration, is being sensed. Clearly, only
advanced radiative transfer models, which treat leaf reflectance explicitly for specific
absorptivity by biochemicals, can begin to answer these questions. On the other hand,
some of the uncertainties encountered in this research may arise from the sensors
themselves. Field difficulties in the adequate characterization of chemical properties

for the very large canopies of Oregon forests could also contribute.
Summary and conclusions

Provided that accurate measurements of the driving variables of the two ecosystem
models tested can be made, each can make good predictions of ANPP that are linear
across a regional gradient. Most important, these models account for the constraints on
production imposed by climatic conditions. Sensing the onset of soil drought was
found to have the greatest overall effect on performance of the light-use efficiency
model; unfortunately, this parameter is very difficult to measure remotely. The
mechanistic model (FOREST-BGC) is difficult to parameterize, especially for soil
water-holding capacity and rooting depth. At the present time, estimation of these two
variables by remote sensing analyses is not available.
We suggest that LAI, accumulation of standing biomass and leaf N concentration
could be predicted across the diverse landscape of Oregon with such a model, which
ignores species, lateral water movement, community life histories and compositional
strategies, and the like.
In the OTTER Project, the remote sensing effort was concentrated on intact forest
ecosystems (except site 5) to validate the logic of modeling routines, identify
questions pertaining to ecosystem model parameterization, uncover difficulties with
applying knowledge gained in other ecosystems to the mountainous terrain occupied
by Oregon forests and test the information extraction methods across a broad and
diverse range of forests and sensing systems. Although the results are promising, the
study revealed that much more research and refinement will be needed to make
systematic use of satellite data for important environmental driving variables. Overall,
the remote sensing analyses confirmed that, in most cases, good relationships do exist
between ecosystem variables and remote sensing data, whether analyzed through
statistical means or through radiative transfer models. And, since the analyses covered
a very broad range in sensor resolutions with similar results, the study suggests that a
basis for extrapolation to larger regions may be possible. Since much of the OTTER
Project was devoted to a few well-characterized sites of intact forests, while much of
the Oregon landscape is highly fragmented and disturbed (Wallin et al. 1996), much
remains to be done to understand how these results behave when extended and
interpolated over this diverse landscape. Because the OTTER Project was not a scaling
exercise, questions about how radiometric and ecologic conditions and relationships
are maintained across a fragmented successional landscape still need careful
exploration and development.
Acknowledgments - The author wishes to express his deep appreciation to Dr. Richard
H. Waring for all his contributions to the OTTER Project. The project involved many
scientists, graduate students and technical specialists from universities, NASA and
research institutions throughout the United States and Canada, whose hard work and
dedication are only hinted at here. The OTTER Project was sponsored principally by
two NASA programs, the Terrestrial Ecosystems Research Program (headed by Dr.
Diane Wickland of the Office of Mission to Planet Earth) and the Biospheric Research
211
Program (headed by Dr. Maurice Averner of the Office of Life and Microgravity
Science and Applications).
References
Aber, J.D., Wessman, C.A., Peterson, D.L., Melillo, J.M. and Fownes, J. 1989a. Remote
sensing of litter and soil organic matter decomposition in forest ecosystems. - In: Hobbs,
RJ. and Mooney, H.A (eds). Remote Sensing of Biosphere Functioning. Springer-Verlag,
Abuelgasim, AA and Strahler, AH. 1994. Modeling bidirectional radiance measurements
collected by the Advanced Solid-state Array Spectrometer (ASAS) over Oregon transect
conifer forests. - Rem. Sens. Environ. 47(2): 261-275.
Baldocchi, D.D., Hicks, B.B. and Meyers, T.P. 1988. Measuring biosphere-atmosphere
exchanges of biologically related gases with micrometeorological methods. - Ecology 69:
1331-1340.
Barton, EE., II and Windham, W.R. 1988. Determination of acid-detergent fiber and crude
protein in forages by near-infrared reflectance spectroscopy: Collaborative study. - J.
Assoc. Offic. Anal. Chern. 71: 620-651.
Barton, EE., II and Himmelsbach, D:S. 1991. Near-infrared reflectance spectroscopy and other
spectral analyses. - In: Davis, AM.C. and Creaser, C.S. (eds). Analytical Applications of
Spectroscopy II. The Royal Chemistry Society, Thomas Grahm House, Cambridge, UK,
pp.240-247.
Birk, E.M. and Matson, P.A 1986. Site fertility affects seasonal carbon reserves in loblolly
pine. - Tree Physiol. 2: 17-27.
Card, D.H., Peterson, D.L., Matson, P.A and Aber, J.D. 1988. Prediction of leaf chemistry by
the use of visible and near-infrared reflectance spectroscopy. - Rem. Sens. Environ. 26:
123-147.
Chapin, ES., II. 1980. The mineral nutrition of wild plants. - Ann. Rev. Ecol. System. 11:
233-260.
Chapin, ES., II and Kedrowski, R.A 1983. Seasonal changes in nitrogen and phosphorus
fractions and autumn retranslocation in evergreen and deciduous taiga trees. - Ecology 64:
376-391.
Chapin, ES., II, McKendrick, J.D. and Johnson, D.A. 1986. Seasonal changes in carbon
fractions in Alaskan tundra plants of differing growth form: Implications for herbivory. - J.
Ecol. 76: 707-732.
Cimino, J.B., Brandini, A, Casey, D., Rabassa, J. and Wall, S. 1986. Multiple incidence angle
SIR-B experiment over Argentina: Mapping of forest units. - IEEE Trans. Geosci. Rem.
Sens. 24: 498-509.
Cole, D.W. and Rapp, M. 1981. Element cycling in forest ecosystems. - In: Reichle, D.E. (ed).
Dynamic Properties of Forest Ecosystems. Cambridge University Press, Cambridge, UK,
pp. 341-409.
Cooper, D.L. and Asrar, G. 1989. Evaluating atmospheric correction models for retrieving
surface temperatures from the AVHRR over a tallgrass prairie. - Rem. Sens. Environ. 27:
93-102.
Coughlan, J.c. and Dungan, lL. 1996. Combining remote sensing and forest ecosystem
modeling: An example using the Regional HydroEcological Simulation System (RHESSys).
- In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the
Modeling of Forest Productivity. Kluwer Academic Publishers, Dordrecht, The Netherlands,
pp. 135-158.
Curran, P.J., Dungan, J.L. and Gholz, H.L. 1990. Exploring the relationship between
reflectance red edge and chlorophyll content in slash pine. - Tree Physiol. 7: 33-48.
DiStefano, J.E and Gholz, H.L. 1986. A proposed use of ion exchange resins to measure
mineralization and nitrification in intact soil cores. - In: Commun. Soil Sci. Plant Anal. 17:
989-998.
Dobson, M.e., Ulaby, ET., Le Toan, T., Beaudoin, A., Kasischke, EJ. and Christensen, N.
1992. Dependence of radar backscatter on conifer forest biomass. - IEEE Trans. Geosci.
Rem. Sens. 30: 412-415.
Durden, S.L., Zebker, H.A. and van Zyl, J.J. 1988. Application of radar polarimetry to forestry.
- In: Proceedings of the IGARSS'88 Symposium, September 12-16, 1988, Edinburgh, pp.
1003-1004.
Durden, S.L., van Zyl, J.J.. and Zebker, H.A. 1989. Modeling and observations of the radar
polarization signature of forested areas. - IEEE Trans. Geosci. Rem. Sens. GE28: 290-301.
Durden, S.L., Klein, J.D. and Zebker, H.A. 1991. Polarimetric radar measurements of a
forested area near Mt. Shasta. - IEEE Trans. Geosci. Rem. Sens. 29: 444-450.
Eck, T. and Dye, D.G. 1991. Satellite estimation of photosynthetically active radiation at the
Earth's surface. - Rem. Sens. Environ. 38: 135-146.
Fox, L., III and Mayer, K.E. 1980. Forest Resource Classification of the McCloud Ranger
District, Mt. Shasta, California, Using Landsat Digital Data. - USDA Forest Service, Final
Report EM-7145-2, Washington, De. 77 pp.
Franklin, J.E and Dyrness, e.T. 1973. Natural vegetation of Oregon and Washington. - USDA
Forest Service, Pacific Northwest Forest and Range Experiment Station, General Technical
Report PNW-8, Portland, OR. 417 pp.
Gholz, H.L. 1982. Environmental limits on aboveground net primary production, leaf area, and
biomass in vegetation zones of the Pacific Northwest. - Ecology 63: 469-481.
Gholz, H.L., Grier, e.e., Campbell, A.G. and Brown, A.T. 1979. Equations for estimating
biomass and leaf area of plants in the Pacific Northwest. - Oregon State University, Forest
Research Lab, Research Paper 41, Corvallis, OR. 39 pp.
Gholz, H.L., Curran, PJ., Kupiec, J.A. and Smith, G.M. 1996. Assessing leaf area and canopy
biochemistry of Florida pine plantations using remote sensing. - In: Gholz, H.L., Nakane,
Productivity. Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 3-22.
Glassy, J.M. and Running, S.w. 1994. Validating diurnal climatology logic of the MT-CLIM
model across a climatic gradient in Oregon. - Ecol. Appl. 4(2): 248-257.
Gong, P., Pu, R. and Miller, J.R. 1992. Correlating leaf area index of Ponderosa pine with
hyperspectral CASI data. - Can. J. Rem. Sens. 18(4): 275-282.
Goward, S.N. 1989. Satellite bioclimatology. - J. Clim. 7: 710-720.
Goward, S.N. and Dye, D.G. 1987. Evaluating North American net primary productivity with
satellite observations. - Adv. Space Res. 9: 239-249.
Goward, S.N. and Hope, A.S. 1989. Evapotranspiration from combined reflected solar and
emitted terrestrial radiation: Preliminary FIFE results from AVHRR data. - Adv. Space Res.
9: 239-249.
pp.241-272.
Goward, S.N., Cruikshanks, G.e. and Hope, A.S. 1985. Observed relation between thermal
emissions and reflected spectral reflectance from a complex vegetated landscape. - Rem.
213
Goward, S.N., Kerber, A., Dye, D.G. and Kalb, V. 1987. Comparison of North and South
American biomes from AVHRR observations. - Geocarto 2: 27-40.
Goward, S.N., Waring, RH., Dye, D.G. and Wang, J. 1994. Ecological remote sensing at
OTIER: Satellite macroscale observations. - Ecol. Appl. 4(2): 322-343.
Grier, C.e. and Running, S.W. 1977. Leaf area of mature northwestern coniferous forest:
Relation to site water balance. - Ecology 58: 893-899.
Hess, L.L., Melack, J.M. and Simonett, D.S. 1990. Radar detection of flooding beneath the
forest canopy: A review. - Int. J. Rem. Sens. II: 1313-1325.
Horler, D.N.H., Dockray, M. and Barber, J. 1983. The red edge of plant leaf reflectance. - Int.
J. Rem. Sens. 4: 273-288.
Jedlovec, G.J. 1990. Precipitable water estimation from high-resolution split window radiance
measurements. - J. Appl. Meteorol. 29: 863-877.
Johnson, L.F., Hlavka, C.A. and Peterson, D.L. 1994. Multivariate analysis of AVIRIS data for
canopy biochemical estimation along the Oregon transect. - Rem. Sens. Environ. 47:
216-230.
Jupp, D.L.B. and Strahler, A.H. 1991. A hotspot model for leaf canopies. - Rem. Sens.
Environ. 38: 193-210.
forests: The challenge for remote sensing and the role of geometric-optical modelling. - In:
pp.75-108.
Kasischke, E.S., Clinthrone, J.e. and Dobson, M.e. 1985. Analysis of a southern pine forest
using X-, C- and L-band dual-polarized SAR imagery. - In: Proceedings of the IGARSS
'85 Symposium. IEEE, New York, p. 174.
Kneizys, F., Shettle, E., Anderson, G., Abrew, L., Chetwynd, J., Shelby, J. and Gallery, W.
1989. Atmospheric transmittance/radiance: Computer code LOWTRAN7. - Hanscom Air
Force Base, Air Force Geophysics Lab, Publication AFGL-TR-88-0177, Environmental
Research Papers 1/1010, Hanscom, MA. 137 pp.
Knight, D.H., Fahey, T.J., Running, S.w., Harrison, A.T. and Wallace, L.L. 1981. Transpiration
from 100-year-old lodgepole pine forests estimated with whole tree photometers. -
Ecology 62: 717-726.
Landsberg, J.J. 1986. Physiological Ecology of Forest Production. - Academic Press, New
York. 198 pp.
Landsberg, J.J., Prince, S.D., Jarvis, P.G., McMurtrie, RE., Luxmoore, Rand Medlyn, B.E.
1996. Energy conversion and use in forests: An analysis of forest production in terms of
radiation utilisation efficiency (e). - In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds).
The Use of Remote Sensing in the Modeling of Forest Productivity. Kluwer Academic
Publishers, Dordrecht, The Netherlands, pp. 273-298.
Law, B.E. and Waring, RH. 1994. Remote sensing of leaf area index and radiation intercepted
by understory vegetation. - Ecol. Appl. 4(2): 272-279.
Le Toan, T., Beaudoin, A., Riom, J. and Guyon, D. 1992. Relating forest biomass to SAR data.
- IEEE Trans. Geosci. Rem. Sens. 30: 403-411.
Lee, L.P. and Takahashi, T. 1966. An improved colorimetric determination of amino acids with
the use of ninhydrin. - Anal. Biochem. 14: 71-77.
Li, X. and Strahler, A.H. 1986. Geometric-optical bidirectional reflectance modeling of a
coniferous forest canopy. - IEEE Trans. Geosci. Rem. Sens. GE24: 906-919.
Li, X. and Strahler, A.H. 1992. Geometric-optical bidirectional reflectance modeling of the
discrete-crown vegetation canopy: Effect of crown shape and mutual shadowing. - IEEE
Trans. Geosci. Rem. Sens. 30: 276-292.
Linder, S. 1980. Chlorophyll as an indicator of the nitrogen status of coniferous seedlings. -
New Zeal. J. For. Sci. 10: 166-175.
Martin, M.E. and Aber, J.D. 1996. Estimating forest canopy characteristics as inputs for
Matson, P.A. and Waring, R.H. 1984. Effects of nutrient and light limitation on mountain
hemlock: Susceptibility to laminated root rot. - Ecology 65: 1517-1524.
Matson, P.A., Johnson, L.E, Billow, T., Miller, J. and Pu, R 1994. Seasonal patterns and
remote spectral estimation of canopy chemistry across the Oregon transect. - Ecol. Appl.
4(2): 280-298.
McClain, E.P., Pickel, W.G. and Walton, c.c. 1985. Comparative performance of
AVHRR-based multichannel sea surface temperatures. - 1. Geophys. Res. 90: 11587-11601.
Meentemeyer, V. and Berg, B. 1986. Regional variation in rate of mass-loss of Scots pine
needle litter in Swedish pine forests as influenced by climate and litter quality. - Scand. J.
For. Res. 1: 167-180.
Melillo, J.M., Aber, J.D. and Muratore, J.M. 1982. Nitrogen and lignin control of hardwood
leaf litter decomposition. - Ecology 63: 62-626.
Meyer, M.M. and Splittstoesser, W.E. 1971. The utilization of carbohydrate and nitrogen
reserves by Trows during its spring growth period. - Physiol. Plant. 24: 306-314.
Miller, J.R, Hare, E.W. and Wu, J. 1990. Quantitative characterization of the vegetation red
edge reflectance. 1. An inverted-Gaussian reflectance model. - Int. J. Rem. Sens. 11:
1755-1773.
Moghaddam, M., Durden, S. and Zebker, H. 1994. Radar measurement of forested areas
during OTTER. - Rem. Sens. Environ. 47(2): 154-166.
Montieth, J.L. 1977. Climate and efficiency of crop production in Britain. - Phil. Trans. Roy.
Soc. (Series B) 281: 277-294.
Myneni, RB., Hall, EG., Sellers, P.J. and Marshak, A.L. 1994. The meaning of spectral
vegetation indices. - In: Proceedings of the IGARSS '94 Symposium, August 8-12, 1994,
California Institute of Technology, Pasadena, CA, 18 pp.
Myrold, D.O. 1994. Nitrogen and carbon cycling along the Oregon transect. - NASA, Ames
Research Center, Final Report, Grant NAG2-719, Moffett Field, CA. 5 pp.
Myrold, D.O., Matson, P.A. and Peterson, D.L. 1989. Relationships between soil microbial
properties and aboveground stand characteristics of conifer forests in Oregon. -
Biogeochem. 8: 265-281.
Nemani, R.R., Pierce, L.L., Running, S.w. and Goward, S.N. 1993. Developing satellite-derived
estimates of surface moisture status. - J. Appl. Meteorol. 32: 548-557.
Pate, J.S. 1983. Patterns of nitrogen metabolism in higher plants. - In: Lee, J.A., McNeill, S.
and Robinson, I.H. (eds). Nitrogen as an Ecological Factor. Blackwell, Oxford, UK, pp.
225-255.
215
Peterson, D.L. and Card, D.H. 1977. Issues arising from the demonstration of Landsat-based
technologies and mapping of the forest ecosystems of the Pacific Northwest states. - In:
Shahrohki, F. (ed). Remote Sensing of Earth Resources. Vol. 6. University of Tennessee
Space Institute, Tullahoma, TN, pp. 65-100.
Peterson, D.L. and Running, S.W. 1989. Applications in forest science and management. - In:
Asrar, G. (ed). Theory and Applications of Optical Remote Sensing. J. Wiley and Sons,
Peterson, D.L. and Waring, R.H. 1994. Overview of the Oregon Transect Ecosystem Research
Project. - Ecol. Appl. 4(2): 211-215.
Peterson, D.L., Spanner, M.A., Running, S.W. and Teuber, K.B. 1987. The relationship of
Thematic Mapper Simulator data to leaf area index of temperate coniferous forests. - Rem.
Peterson, D.L., Aber, J.D., Matson, P.A., Card, D.H., Swanberg, N.A., Wessman, C.A. and
Spanner, M.A. 1988. Remote sensing of forest canopy and leaf biochemical contents. -
Pierce, L.L. and Running, S.w. 1988. Rapid estimation of coniferous leaf area index using a
portable integrating radiometer. - Ecology 69: 1762-1767.
Pierce, L.L., Running, S.w. and Walker, J. 1994. Regional scale relationships ofleaf area index
to specific leaf area and leaf nitrogen content. - Ecol. Appl. 4(2): 313-321.
Price, J.S. 1984. Land surface temperature measurements from the split window channels of
the NOAA-7 Advanced Very High Resolution Radiometer. - J. Geophys. Res. 89(D5):
7231-7237.
Raich, J.W. and Nadelhoffer, K.J. 1989. Belowground carbon allocation in forest ecosystems:
Global trends. - Ecology 70: 1346--1354.
Rastetter, E.B., Ryan, M.G., Shaver, GR, Melillo, J.M., Nadelhoffer, K., Hobbie, J.E. and
Aber, J.D. 1991. A general biogeochemical model describing the responses of C and N
cycles in terrestrial ecosystems to changes in CO 2 , climate and N deposition. - Tree
Physiol. 9: 101-126.
Reichle, D.E. (ed). 1981. Dynamic Properties of Forest Ecosystems. - Cambridge University
Press, Cambridge, UK. 683 pp.
Rosenberg, N.J., Blad, B.L. and Verma, S.B. 1983. Microclimate: The Biological Environment.
2d ed. - J. Wiley and Sons, New York. 495 pp.
Running, S.W. 1994. Testing FOREST-BGC ecosystem process simulations across a climatic
gradient. - Ecol. Appl. 4(2): 238-247.
Running, S.W. and Coughlan, J.C. 1988. A general model of forest ecosystem processes for
regional applications. 1. Hydrologic balance, canopy gas exchange and primary production
processes. - Ecol. Model. 42: 125-154.
Running, S.W. and Nemani, R.R. 1988. Relating the seasonal pattern of the AVHRR
normalized difference vegetation index to simulated photosynthesis and transpiration of
forests in different climates. - Rem. Sens. Environ. 17: 472-483.
Running, S.w. and Gower, S.T. 1991. FOREST-BGC, a general model of forest ecosystem
processes for regional applications. 2. Dynamic carbon allocation and nitrogen budgets. -
Tree Physiol. 9: 147-160.
Running, S.w., Waring, R.H. and Rydell, R.A. 1975. Physiological control of water flux in
conifers: A computer simulation model. - Oecologia 18: 1-16.
Running, S.W., Peterson, D.L., Spanner, M.A. and Teuber, K.B. 1987a. Remote sensing of
forest leaf area index. - Ecology 67: 273-275.
Running, S.W., Nemani, R.R. and Hungerford, R.D. 1987b. Extrapolation of synoptic
meteorological data in mountainous terrain and its use for simulating forest evapotranspiration
and photosynthesis. - Can. J. For. Res. 17: 472-483.
Runyon, J., Waring, RH., Goward, S.N. and Welles, J.M. 1994. Environmental limits on net
primary production and light-use efficiency across the Oregon transect. - Ecol. Appl. 4(2):
226-237.
Ryan, M.G. 1991. A simple model for estimating gross carbon budgets for vegetation in forest
ecosystems. - Tree Physiol. 9: 255-266.
Scho1ander, P.E, Hammel, HT., Bradstreet, E.D. and Hemmingsen, E.A. 1965. Sap pressure in
vascular plants. - Science 148: 339-346.
Sellers, P.J. 1985. Canopy reflectance, photosynthesis and transpiration. -lnt. J. Rem. Sens. 6:
1335-1271.
Skiles, J.W. and Angelici, G.L. 1993. Data management for support of the Oregon Transect
Ecosystem Research (OTTER) Project. - NASA, Ames Research Center, NASA
Contractor Report 4557, Moffett Field, CA.
Sollins, P., Goldstein, RA., Mankin, J.B., Murphy, C.E. and Swartzman, G.L. 1981. Analysis
of forest growth and water balance using complex ecosystem models. - In: Reichle, D.E.
(ed). Dynamic PropeI;ties of Forest Ecosystems. Cambridge University Press, Cambridge,
UK, pp. 537-566.
Spanner, M.A., Acevedo, W, Teuber, K.B., Running, S.W, Peterson, D.L., Card, D.H. and
Mouat, D.A. 1984. Remote sensing of leaf area index of temperate coniferous forests. - In:
Proceedings of Symposium on Machine Processing of Remotely Sensed Data, Purdue
University, West Lafayette, IN, pp. 362-370.
Spanner, M.A., Pierce, L.L., Running, S.W. and Peterson D.L. 1990a. Seasonal trends of
Spanner, M.A., Pierce, L.L., Peterson, D.L. and Running, S.W I990b. Remote sensing of
coniferous forest leaf area index: The influence of canopy closure, understory vegetation
and background reflectance. - Int. J. Rem. Sens. 11: 95-111.
Spanner, M.A., Johnson, L., Miller, J., McCreight, R, Freemantle, J., Runyon, J. and Gong, P.
1994. Remote sensing of seasonal leaf area index across the Oregon transect. - Ecol. Appl.
4(2): 258-271.
Strahler, A.H., Franklin, J., Woodcock, C.E. and Logan, T.L. 1981. FOCIS: A forest
classification and inventory system using Landsat and digital terrain data. - USDA Forest
Service, Nationwide Forestry Applications Program, Report NFAP-255, Salt Lake City,
UT.60pp.
Sun, G.Q. and Simonett, D.S. 1988. Simulation ofL-band HH microwave backscattering from
coniferous forest stands: A comparison with SIR-B data. - Int. J. Rem. Sens. 9: 907-925.
Technicon Instrument Corporation. 1977. Individual/simultaneous determinations of nitrogen
and/or phosphorus in BD acid digests. Industrial Method Number 329-74W - Technicon
Instrument Corporation, Tarrytown, NY. 9 pp.
Tromp, J. 1970. Storage and mobilization of nitrogenous compounds in apple trees with
special reference to arginine. - In: Luckwell, L.c. and Cutting, c.v. (eds). Physiology of
Tree Crops. Academic Press, New York, pp. 143-159.
Van Cleve, K., Oliver, L., Schlenter, R, Viereck, L.A. and Dyrness, c.T. 1983. Production and
nutrient cycling in taiga forest ecosystems. - Can. J. For. Res. 13: 747-766.
Van den Driessche, R. and Webber, J.E. 1977. Variation in total and soluble nitrogen
concentrations in response to fertilization of Douglas-fir. - For. Sci. 23: 135-142.
217
Van Soest, PJ. and Wine, R.R 1968. Determination of lignin and cellulose in acid-detergent
fiber with permanganate. - J. Assoc. Offic. Anal. Chern. 51: 780-785.
Wallin, D.O., Harmon, M.E., Cohen, WB., Fiorella, M. and Ferrell, WK. 1996. Use of remote
sensing to model land use effects on carbon flux in forests of the Pacific Northwest, USA. -
In: Gholz, RL., Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in the
Modeling of Forest Productivity. KIuwer Academic Publishers, Dordrecht, The Netherlands,
pp.219-237.
Waring, R.R 1980. Site, leaf area, and phytomass production in trees. - In: Benecke, U. and
Davis, M.R (eds). Mountain Environments and Subalpine Tree Growth. New Zealand
Forest Service, Technical Paper 70, Wellington, New Zealand, pp. 125-135.
Waring, RH. 1983. Estimating forest growth and efficiency in relation to canopy leaf area. -
Adv. Ecol. Res. 13: 327-354.
Waring, R.R and Major, J. 1964. Some vegetation of the California coastal redwood in relation
to gradients of moisture, nutrients, light, and temperature. - Ecol. Monogr. 34: 167-215.
Waring, R.H. and Cleary, B.D. 1967. Plant moisture stress: Evaluation by pressure bomb. -
Science 155: 1248-1254.
Waring, RH. and Running, S.W 1976. Water uptake, storage and transpiration by conifers: A
physiological model. - In: Lange, 0., Kappen, L. and Schulze, E.D. (eds). Water and Plant
Life - Problems and Modem Approaches. Springer-Verlag, Berlin, pp. 182-202.
Waring, RH. and Franklin, J.F. 1979. Evergreen coniferous forests of the Pacific Northwest. -
Science 204: 1380-1386.
Waring, R.H. and Schlesinger, W.H. 1985. Forest Ecosystems: Concepts and Management. -
Academic Press, Orlando, FL. 340 pp.
Waring, RH., Emmingham, WH., Gholz, H.L. and Grier, c.c. 1978. Variation in maximum
leaf area in Oregon and its ecological significance. - For. Sci. 24: 131-140.
Waring, RR, Schroeder, P.E. and Oren, R. 1982. Application of the pipe model theory to
predict canopy leaf area. - Can. J. For. Res. 12: 556-560.
Way, J.B., Paris, J., Kasischke, E., Slaughter, c., Viereck, L., Christensen, N., Dobson, M.,
Ulaby, F., Richards, J., Milne, A., Sieber, A., Ahern, F., Simonett, D., Hoffer, R., Imhoff, M.
and Weber, J. 1990. The effect of changing environmental conditions on microwave
signatures of forest ecosystems: Preliminary results of the March 1988 Alaska SAR
experiment. - Int. J. Rem. Sens. 11: 1119-1144.
Welles, J.M. and Norman, J.M. 1991. Instrument for indirect measurement of canopy
architecture. -Agron. J. 83: 818-825.
Wessman, c.A., Aber, J.D., Peterson, DL and Melillo, J.M. 1988. Remote sensing of canopy
Williams, DL, Goward, S.N. and Walthall, c.L. 1984. Collection of in situ forest canopy
spectra using a helicopter: A discussion and preliminary results. - In: Proceedings of the
Tenth International Symposium on Machine Processing of Remotely Sensed Data, Purdue
University, West Lafayette, IN. IEEE, New York, pp. 94-106.
Wrigley, RC., Spanner, M.A., Slye, RE., Peuschel, R. and Aggarwal, H.R. 1992. Atmospheric
correction of remotely sensed image data by a simplified model. - J. Geophys. Res.
97(Dl7): 18797-18814.
Wu, Y. and Strahler, A.H. 1994. Remote estimation of crown size, stand density, and biomass
on the Oregon transect. - Ecol. Appl. 4(2): 299-312.
Yoder, B. 1992. "Photosynthesis of conifers: Influential factors and potential for remote
sensing." - Ph.D. dissertation, Oregon State University, Corvallis, OR. 127 pp.
Zobel, D.8., McKee, W.A., Hawk, G.M. and Dymess, C.T. 1976. Relationships of environment
to composition, structure and diversity of forest communities of the central western
Cascades of Oregon. - Eco!. Monogr. 46: 135-156.
NINE
Use of Remote Sensing to Model Land Use Effects on Carbon

Flux in Forests of the Pacific Northwest, USA
David O. Wallin, Mark E. Harmon, Warren B. Cohen, Maria Fiorella and William K.
Ferrell
Wallin, D.O., Harmon, M.E., Cohen, W.B., Fiorella, M. and Ferrell, w.K. 1996. Use
of remote sensing to model land use effects on carbon flux in forests of the
Pacific Northwest, USA. - In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The
Use of Remote Sensing in the Modeling of Forest Productivity. Kluwer Acad. Publ.,
Reducing the uncertainty in the global carbon (C) budget will require better
information on regional C budgets. We discuss the use of a simple "metarnodel," in
2
conjunction with satellite data, to quantify C flux from a l2,OOO-km forestland study
area in Oregon. The model tracks C storage in living, detrital and forest products
pools. Between 1972 and 1991, total C flux from this study area to the atmosphere was
estimated to average 1.13 Mg ha'l yr't, with values ranging from -4.7 to + 15.8 Mg ha'l
y(l. This spatial variability was related to site quality, land use and historical factors.
These results are used to illustrate the natural and anthropogenic sources of
heterogeneity that can influence C budgets at the regional scale and to demonstrate
how remotely sensed data can be used to help quantify this heterogeneity.
D.O. Wallin, Center for Environmental Sciences, MS-9181, Western Washington

University, Bellingham, WA 98225-9181, USA. M.E. Harmon and WK. Ferrell,
Department of Forest Science, Forestry Sciences Laboratory, Rm. 020, Oregon State
University, Corvallis, OR 97331, USA. WE. Cohen and M. Fiorella, USDA Forest
Service, Pacific Northwest Research Station, Forestry Sciences Laboratory, 3200 SW
Jefferson Way, Corvallis, OR 97331, USA.
219
Introduction
The potential for substantial climate change resulting from increasing atmospheric
concentrations of radiatively active trace gases has motivated efforts to identify natural
and anthropogenic sources and sinks for these gases (Houghton and Woodwell 1989,
Post et al. 1990). Among these radiatively active trace gases, carbon dioxide (C0 2) has
the greatest potential to affect global climate; hence, the global carbon (C) budget has
received special attention (IPCC 1990). Recent global-scale estimates (Sundquist
1993, Houghton et al. 1992) suggest that emissions of CO2 to the atmosphere from the
combustion of fossil fuels and land use account for 5.4 0.5 and 1.6 1.0 Pg C yr",
respectively, Dixon et al. (1994) recently reduced the estimate ofland use emissions to
0.9 0.4 Pg C yr"'. The oceans are thought to be a sink for 2.0 0.8 Pg C Yr"', while
observed increases in atmospheric CO 2 concentrations account for about 3.2 0,1 Pg
C yr". This observed increase and the known sources leave a "missing sink" in the
biosphere of about 1.1 1.0 Pg C yr"'. The search for this missing sink is at the core
of much of the current work on global C budgets.
The certainty in the measurements of various global C pools and fluxes varies
widely. There have been economic incentives to carefully track the global
consumption of fossil fuels; hence, there is more confidence in the estimates of fossil
fuel emissions than for the other sources and sinks. Solomon et al. (1993) pointed out
that the ocean and atmosphere systems are reasonably well mixed and that the
measurement and modeling of C pools and fluxes between them is a tractable problem
in physical chemistry and fluid dynamics. For this reason, the estimates of C flux
between the ocean and atmosphere systems have remained within a fairly narrow
(although perhaps not accurate) range of values over the past 20 years (Keeling 1973,
Tans et al. 1990, Orr 1993). Solomon et al. (1993) went on to point out that the
terrestrial system is fundamentally different from those of the ocean and the
atmosphere in the sense that it is not at all well mixed with respect to C. In the
terrestrial system, heterogeneity in C pools and fluxes exists at all spatial and temporal
scales and results from a complex interaction of natural and anthropogenic factors.
This heterogeneity has resulted in a wide range of estimates of terrestrial C sources
and sinks.
The relative certainty in the other C pools and fluxes points towards a terrestrial C
sink. Several lines of evidence suggest that the missing C sink is most likely to be
found in northern, mid-latitude forests (Tans et al. 1990, Kauppi et al. 1992, Taylor
and Lloyd 1992). Dixon et al. (1994) calculated that, if these forests are to account for
the missing C in the global budget, their net accumulation rate must be about 1.5 Mg
C yr'l, or about 2-3% of their present standing stock. They pointed out that although
this is within the range of values observed at selected sites, it is considerably higher
than the most recent estimate for the continental United States (0.4 Mg C ha" yr"l,
Turner et al. 1995) and up to four times higher than the observed globally averaged
rate of accumulation for northern, temperate forests (reviewed by Dixon et al. 1994).
Efforts to reduce the uncertainty in the global terrestrial C budget will require better
information about the spatial and temporal heterogeneity of C storage and C flux in
various regions throughout the world. In this chapter, we discuss some of the natural
221
and anthropogenic sources of heterogeneity in forested ecosystems and demonstrate

how remotely sensed data can be used to quantify this heterogeneity. Our discussion
centers on forests because they contain about 60% of the global terrestrial C stocks
(Waring and Schlesinger 1985). To provide a framework for discussion, we focus on
our efforts to develop a C budget for the Pacific Northwest (PNW) forests of the
United States. We describe a new model called LANDCARB that enables us to
integrate a simplified stand-level C model with satellite imagery and spatial
information on climate and soils. Classified satellite imagery is used to provide initial
forest conditions and to track changes in forest structure resulting from succession and
both natural and anthropogenic disturbance. Model output includes maps of C storage
and C flux. Model results that quantify the C budget from 1972 to 1991 for a
12,000-km2 study area in the central Oregon Cascade Mountain Range (Cohen et al.
in review) are reviewed and used to illustrate our discussion. Our results are not
consistent with the most recent reviews (Dixon et al. 1994, Turner et al. 1995), since
we conclude that our study area was a large net source of C to the atmosphere between
1972 and 1991.
PNW forests
The forests of the PNW are among the most productive in the world and contain many
tree species that attain great ages and substantial stature (Waring and Franklin 1979).
Several tree species achieve ages of 500-1000 years in natural forests that may contain
individual trees ~ 100-200 cm in diameter at breast height (dbh) with heights of
60-80 m. These forests have the capacity to store very large quantities of C. An
intensively studied, 450-yr-old Douglas-fir (Pseudotsuga menziesii) stand on
moderately productive (site-class 3) land in the central Oregon Cascade Mountains
contained 611 Mg C ha- 1 in above- and below-ground living and detrital pools and in
the mineral soil (Grier and Logan 1977, Harmon et al. 1986). Sixty-yr-old stands on
comparable land contain between 259 and 274 Mg C ha- 1 in these same pools (Harmon
et al. 1990b).
Harmon et al. (1990b) showed that harvesting these old-growth forests and
replacing them with young plantations would result in a large net release of C to the
atmosphere, even when storage of C in forest products is considered. Although young
plantations have a higher net annual rate of C uptake than old forests, the total amount
of C stored in young plantations is minimal compared with that of old-growth stands.
Harvesting an old-growth stand results in a large increase in the amount of dead wood
on the site in the form of tops, branches and roots. Even if the decay rate remains
constant after harvesting, the amount of C released to the atmosphere from this very
large detrital pool is substantial during the first several decades after harvesting.
Similarly, much of the C removed from the site by timber harvesting is quickly
released to the atmosphere during primary and secondary manufacturing processes
and by incineration and decomposition of short-lived forest products. Many decades
are required before the net C accumulation rate by regenerating trees in plantation
stands exceeds the net C emission rate to the atmosphere by the detrital pools and the
forest products sector. It may take 200 years before the total amount of C stored by the
stand approaches preharvest levels.
Although these stand-level dynamics are relatively well documented, developing a

regional C budget requires the integration of these results in both the time and space
domains. Integration in the spatial domain requires information on the distribution of
stand ages, species, site productivity and management techniques. Integration in the
time domain requires spatially explicit information on changes in stand age in
response to timber harvesting, wildfire, succession and changes in management
practices and forest product utilization standards. Assembling this information is a
challenging interdisciplinary problem in ecology, silviculture, economics, history and
social science.
Study area
Our study area is in western Oregon on the western slopes of the Cascade Range. It
extends from an elevation of about 300 m on the edge of the Willamette Valley in the
west to elevations of over 2000 m at the crest of the Cascade Mountain Range in the
east (Plate 1). In general, the climate is characterized by mild, wet winters and warm,
dry summers (Waring and Franklin 1979). However, the large elevation range and the
steep, highly dissected terrain produces high variability in microclimate and forest
productivity. Our study area lies within the Western Cascades and High Cascades
physiographic provinces and mostly within the western hemlock (Tsuga heterophylla)
vegetation zone (Franklin and Dyrness 1973). Dominant tree species include western
hemlock, Douglas-fir and western red cedar (Thuja pUcata). The area includes a wide
variety of natural and managed stand types and ages.
Our study area includes 811,788 hectares of forested lands in a mixture of public
and private ownership (Plate 1). The USDA Forest Service manages 53.5% of these
forested lands including 7% in Congressionally designated wilderness areas. The
Bureau of Land Management (BLM) administers 8.7% of the forested lands and
37.1 % is in private ownership. Nearly all the private land is held by large corporations
and is intensively managed for timber production.
Historical context
In the late nineteenth century, timber harvesting began throughout the PNW on the
most productive lands at low elevations near the coast; in the early twentieth century,
it began at lower elevations on the western slopes of the Cascade Range (Harris 1984,
Robbins 1988). Timber harvesting on public lands throughout the region began in the
1940s. Harvest levels were modest until the 1960s and 1970s. During the period for
which satellite data are available (1972 onward), virtually all of the cutting on public
lands in our study area was in primary (previously uncut) forest and the cutting on
private lands involved a mixture of primary and secondary forest (Bassett and Choate
1974, Gedney 1982, Gedney et al. 1987).
The primary land use for forest lands in this region is timber production, and the
PNW has provided a significant share of the global timber supply for many decades
(Waddell et al. 1989). Unlike timber harvesting in some parts of the world, this activity
in the PNW is not the initial step in a permanent or semipermanent conversion from
223
one cover type to another (i.e., conversion of forest land to agriculture or pasture).
Douglas-fir is the dominant tree species in both primary and secondary forests.
Timber harvesting influences the regional C budget by altering the overall
frequency distribution of stand age classes and their spatial distribution on the
landscape. Under presettlement conditions, wildfire was the primary factor controlling
the age-class distribution of these forests (Agee 1991, 1993). Although reconstructing
the presettlement forest age-class distribution is technically problematic and
politically contentious, it is clear that stands ;::: 200 years old were not uncommon
(Spies and Franklin 1988, Booth 1991). In contrast, private lands throughout the
region are currently managed on rotation lengths of about 50 years while federal lands
have, until recently, been managed for rotation lengths of about 100 years (USDA
Forest Service 1990, Spies et al. 1994). In future years, the public forest land will
likely be managed on much longer rotations (FEMAT 1993). Since the private lands
are located at lower elevations on the most productive land (site classes 1, 2 and 3,
Plate 1), they have the largest capacity to store C. In contrast, the public lands are
located at higher elevations on less productive land (site classes 3, 4 and 5, Plate 1) and
have a lower capacity to store C:Rotation lengths currently in use have the effect of
maintaining the private lands much farther below their maximum potential C storage
levels than is the case for public lands (Krankina and Harmon 1994).
Changing forest practices over the past several decades must be considered when
developing a C budget. Prior to the 1960s, there was greater reliance on natural
regeneration after timber harvesting; this resulted in slower rates of C accumulation
following a harvest. Currently, on both public and private lands throughout the region,
stands are manually replanted with native tree species. On private lands, there tends to
be a greater emphasis on replanting the more desirable commercial species and on the
use of genetically "improved" varieties. On public lands, replanting includes a wider
variety of species from locally collected seed sources. Timber harvesting and site
preparation prior to replanting have varied considerably over the past 50 years. One of
the most notable trends has been a reduction in the amount of woody debris left on the
site after harvesting. During the 1910s, up to 75% of the woody material was left on
the site after harvesting, compared with 36-54% during the 1980s (Harmon et al. in
press). Additional changes on federal lands since the mid-1980s include reductions
both in the use of burning prior to planting and in the use of herbicides to control
shrubs and deciduous tree competition for the young conifers.
Over the last century, there have also been important changes in the ways harvested
C is used in the PNW (Harmon et al. in review). These include changes in the
disposition of saw and veneer mill waste as well as in the disposal of paper and wood
waste. For example, until the 1940s most mill waste was incinerated without energy
recovery. Since that time, an increasing share of mill waste (3D-40%) has been used
for paper production. Perhaps more significant have been the changes in paper waste
disposal. Although paper in sanitary landfills decays very slowly, most paper waste
was incinerated deliberately or accidentally in open dumps until the mid-1970s. Only
since that time have landfills been a significant C sink for forest products (Harmon et
al. in review).
The LANDCARB model

2 2 6 2
The development ofC flux models for use at landscape (10 km ) to regional (10 km )
scales has involved several different approaches. One approach for coverage of large
areas involves the use of complex, process-based models that operate in large grid
cells (Running et al. 1989, Turner and Marks 1993). This approach is attractive in the
sense that the process-based foundation makes it practical to evaluate interannual
variation and ecosystem responses to climate change scenarios. The use of large grid
cells is necessitated by the computational complexity of such models. In some cases,
however, critical elements of spatial heterogeneity in the system may not be
adequately represented by large grid cells. The use of smaller grid cells can alleviate
this problem, but computational limitations will then reduce the spatial extent of the
area that can be modeled. Another approach involves the use of a detailed model to
generate a series of "look-up tables." Values from these look-up tables can then be
used to produce maps of C storage for very large areas (B urke et al. 1990, 1991). We
have chosen a "metamodel" approach: the linkage of a detailed stand-level model to a
less detailed model that mimics its behavior (Law and Kelton 1991).
In our case, a detailed, stand-level C model (Harmon et al. 1990b) was used to
generate "data" to parameterize simple functions that described the mass of C in living
and dead pools of a forest. The original model, called Disturbed Forest Carbon (DFC)
(Harmon et al. 1990b), simulates the accumulation of C in living pools (leaves,
branches, boles, fine roots and coarse roots) following a disturbance and tracks this C
as it cascades through various detrital pools (forest floor, fine woody debris, coarse
woody debris and dead roots). The live portion of DFC was parameterized to mimic
commonly used yield tables for Douglas-fir (Curtis et al. 1982). The input of detritus
and its decay was parameterized from published (Harmon et al. 1986, Harmon and
Chen 1991, Harmon et al. 1990a) and unpublished data (unpublished data, Oregon
State University Forest Science Data Bank, Corvallis, OR). The functions that were
parameterized by means of the DFC model were used to develop a less detailed and
less computationally intensive metamodel that could be applied at landscape to
regional scales using small grid cells. This model, called LANDCARB, was
specifically designed to examine the effects of wildfire and land use on C budgets. The
model was not designed to evaluate the direct (C02 fertilization) or indirect (climatic)
effects of changes in atmospheric CO 2concentrations on C budgets. Hence, the model
evaluates the effects of land use under constant climate conditions with no CO 2
fertilization. LANDCARB was designed to be spatially explicit and to utilize input
from satellite imagery and other spatial databases. The model is described in detail by
Wallin et al. (unpublished manuscript); key features are briefly discussed in the
following section.
Living carbon pools

A Chapman-Richards function was used in LANDCARB to describe the change in
total live C as a function of time since disturbance:
LIVE =LIVEMAX * O_e(oBI*AGE1)B2
225
6OOr--------------------....,
Total
.;"500
ltl
.c
~400
CD
/
/
~300
o /
,j
.
Ci5
c: 200
.8
<3 100
~ ...
/ ............ _ FO<llSI Products
100 300 400
Figure J. Change in C storage at the stand level on site-class 3 lands. Curves

illustrate response following the c1earcut harvesting of an old-growth stand at
time zero. Output is from the LANDCARB model. See text for details.
where LIVE is the total live e stores in Mg e ha- , LIVEMAX is the maximum live e
I
stores, B I is the rate that determines how quickly live e approaches the maximum and
B2 determines how long plant production lags behind the maximum rate (Fig. I). AGE
is the number of years since the last disturbance. Parameter values are presented in
Table 1.
Table J. Parameter values for the Chapman-Richards live C function in

LANDCARB.
Site LIVEMAX
index Mg e ha- 1 B1 B2
1 650 0.021 1.98

2 570 0.021 1.97
3 460 0.021 1.96
4 310 0.020 1.92
5 230 0.020 1.88
Labile dead carbon pools

We did not include mineral soil e in our dead pool, because this pool is thought to
change little fol1owing disturbance (Johnson 1992). Since we are primarily interested
in e pools that are likely to change in response to wildfire or land use, we model only
the labile dead e (forest floor, fine woody debris, coarse woody debris and dead
roots). "Total dead en could be obtained by adding an estimate of mineral soil e
derived from the STATSGO database (Soil Conservation Service 1991).
The rate of change for the dead C pool is a function of the inputs from the living
pool and harvesting disturbance minus the losses from decomposition and site
preparation (i.e., residue burning). Both decomposition and input rates represent
aggregated values for all components (leaves, small branches, boles, roots) of the dead
or living pools, respectively (Table 2). At any given time step, the decay rate is the
proportion of the dead labile pool that is lost to decomposition. Similarly, the litterfall
rate is the proportion of the living pool that is added to the dead labile pool. With
increasing site index (decreasing site productivity), leaf area index (LA!) changes very
little, but bole and large root volume - and, therefore, total C storage - declines
markedly (Table 1). This means that the more recalcitrant components of the dead
labile pool (boles and large roots) make up a smaller proportion ofthe total dead labile
pool. Although decay rates for each component of this pool generally decrease with
increasing site index, the smaller volume of boles and large roots in the dead labile
pool means that the aggregate decay rate actually increases slightly. With declining
site productivity, boles and large roots also account for a declining proportion of the
living C pool. Since th~ turnover of leaves, small branches and fine roots changes very
little with site quality, the aggregate litterfall rate also increases with increasing site
index. We assume that harvesting removes 50% of the live C and adds the remainder
to the dead pool (Harmon et al. in press). Finally, for these simulations we assume that
fire was not used for site preparation, so no reductions were made to the dead pool
after harvesting.
Table 2. Parameter values of the asymptotic decay function used in modeling

labile dead C stores in LANDCARB.
Site Decay Litterfall

-I
index rate yr rate yr- I
1 0.030 0.005
2 0.030 0.005
3 0.030 0.006
4 0.031 0.007
5 0.031 0.008
We initialized the dead labile C pool with estimates for stands with a range of ages,
site classes and management histories (Table 3). Values used for this initialization
were developed using the preceding equations and by assuming that stands :2: 140
years old (in 1972) originated after a wildfire in an old-growth forest and stands :s; 60
years old (in 1972) originated from clearcut logging of old-growth stands. Rates of C
removal during harvesting are consistent with historical data (Harmon et al. in press).
Values in Table 3 were used only to assign the initial dead labile C pool sizes for 1972.
For all subsequent dates, dead labile C storage was calculated based on the model
framework described previously and the observed changes in stand age.
227
Table 3. Harvest removal rates and initial C stores in the labile dead C pool for
various forest ages and site classes used in parameterizing LANDCARB.
l
Initial stores (Mg C ha- )
Harvest Site class

Stand removal
age (yr) (%) 1 2 3 4 5
5 50 349 308 252 178 137

15 50 258 228 187 140 112
25 50 194 172 141 116 97
60 40 104 92 79 78 75
140 0 94 82 78 73 62
200+ 0 103 91 88 70 59
Forest products
The bole C that is harvested and removed from a site is tracked in LANDCARB
through the forest products sector using an approach similar to that used by Harmon
et al. (1990b). Overall, we assume that 40% of the harvested bole C is injected into the
atmosphere immediately after harvesting. This is to account for losses during primary
and secondary manufacturing and incineration of short-lived forest products. The
remaining 60% is added to the forest products inventory and is subjected to an annual
loss of 2% to account for decay of buildings, fences and other wooden structures as
well as incidental losses of paper and other forest products.
LANDCARB behavior at the stand level

The behavior of the LANDCARB model for a single stand on site-class 3 land is
illustrated in Figure 1. After the harvesting of an old-growth stand at time zero, there
is a large increase in the size of the dead labile pool resulting from the large input of
logging slash. The size of the dead labile pool then declines rapidly because of decay
and the minimallitterfall inputs from the very small postharvest living pool. The dead
labile pool drops to a minimum value at year 90 before climbing back to the
equilibrium value as litterfall inputs from the living pool increase between years 100
and 200. The living C pool drops to zero at the time of harvesting and returns to 90%
of the preharvest level (LIVEMAX) in year 140. The forest products pool is initialized
at time zero with 30% of the C from the preharvest living pool (50% of the living pool
is removed from the site during harvesting, 60% of which enters long-term forest
products). The decay rate reduces this initial forest products pool by 50% within 34
years and by 90% within 112 years. Total C storage (live + dead + forest products)
returns to 90% of the preharvest level in 136 years. Similar total C dynamics occur for
stands in each of the other site classes (Fig. 2).
800
":lll 700
.c
0>600
5'1.3
~
~500
III
.9400 Stt
C/)
300
--- ---- ------------------
St185
D
~200
100
0 100 200 300
Year
Figure 2. Change in total C storage for stands on site-class 1-5 lands. Curves
illustrate response following the cleareut harvesting of an old-growth stand at
time zero. Output is ftom the LANDCARB model. See text for details.
Model integration with spatial data

The equations describing C stores are dependent upon climate, soils and stand age.
Calculations for an entire landscape require climate, soils and stand-age data for the
entire study area. Constructing a C budget for a landscape requires information on
changes in C storage between two or more points in time. Over the time scales
considered here (19 years), this requires data on changes in stand age resulting from
succession, timber harvesting and wildfire. Climate and soils are considered to remain
constant over this period. We relied on the use of a site-class map (Isaac 1949) to
integrate the influence of climate and soils on tree growth and decomposition rates
(Plate 1). This map provides coverage for the entire region and corresponds well with
observed patterns of forest productivity.
Natural and anthropogenic sources of heterogeneity
Ma~mumporenffmcarbonsroroge
PNW forests provide an extreme example of how regional C densities can differ from
the average continental-scale values used in global C budgets. For example, in their
recent review of global terrestrial C budgets, Dixon et al. (1994) represented the
continental United States using an average value of 170 Mg C ha'l for the total C
storage in all above- and below-ground living and detrital pools, including the 0
horizon and mineral soil (to a depth of 1 m). In contrast, old-growth forests on the
most productive (site-class 1) lands in the PNW store an average of 753 Mg C ha'l,
exclusive of the mineral soil (Tables 1 and 3). Furthermore, even stands on the least
productive (site-class 5) sites contain a minimum of 123 Mg C ha'l, exclusive of the
mineral soil (Fig. 2). The top I m of mineral soil on these site-class 5 lands would add
an additional 104 Mg C ha,l to total C storage (P. Homann, unpublished; derived from
229
Soil Conservation Service 1991). Thus, there is considerable potential for error
propagation resulting from this regional- and local-scale variability.
Using satellite data to monitor changes in stand age

Much of the uncertainty in the C budget for forest systems results from inadequate
knowledge of the forest age-class distribution and uncertainty as to the changes in this
distribution over time. This uncertainty may exist even when forest inventory data are
present. For example, a recent C budget for the continental United States (Turner et al.
1995) used inventory data to derive forest age-class distributions. For some land
management categories in some parts of the country, such as USDA Forest Service
lands in the Rocky Mountain region and throughout the eastern United States, Turner
et al. (1995) assumed that the age-class distribution was similar to that on adjacent
private lands. Since the age-class structure on private lands is often quite different
from that on USDA Forest Service lands, this could lead to a significant error in the
overall estimates of stores and fluxes. In another example, a C budget was constr':lcted
using data from an inventory of <;':anadian forests and another independent inventory
of Canadian peatlands (Kurz et at. 1992). These data were not spatially explicit and the
degree of overlap in forested peatlands could not be determined; this uncertainty could
also have led to significant error.
The shortcomings of inventory data can potentially be overcome using satellite
data. For example, within our study area, Cohen and Spies (1992) and Cohen et al.
(1995) demonstrated that forest composition, stand age and structure can be mapped
using Thematic Mapper (TM) data. Initially, they used unsupervised classification to
separate four forest-cover classes: open 30%), semiopen (30-85%), closed-canopy
mixed conifer-hardwood (> 85%) and closed-canopy conifer (> 85%). Then,
statistical models were developed for distinguishing among successional stages within
the closed-canopy conifer class. Three successional stages were identified: young
80 years old), mature (80-200 years old) and old-growth (> 200 years old).
Accuracy of predictions for the three age classes was 75%, whereas an overall
accuracy of 82% was achieved for all six forest-cover classes. All class boundaries
were selected to represent ecologically meaningful transitions that could also be
distinguished with a high degree of accuracy. We used these structural classes to
represent "structural" age classes (Table 4). On average, these structural ages
correspond to time since last disturbance; however, on suboptimal sites, regeneration
may proceed slowly following a disturbance. In these cases, the structural age may lag
behind the chronological age. These six structural age classes were then used to
construct C storage maps. The number of years represented by these classes varied
somewhat, with the highest resolution in the three youngest age classes. Fortuitously,
these classes provide the highest temporal resolution in the portion of the successional
trajectory where total C storage is changing most rapidly (Fig. 1).
Table 4. Percentage of area by forest age class for 811,788 hectares of forested
lands from 1972 to 1991. Age classes represent "structural age" (after Cohen et
al. 1994); see text for explanation of structural age classes.
Age
Structure class
class (yr) 1972 1976 1984 1988 1991
Open 0-10 9.1 11.5 11.0 11.6 9.1

Semiopen 11-20 11.0 11.2 17.3 19.7 24.4
Closed-mix 21-30 6.8 6.7 5.7 5.9 5.8
Young 31-80 21.0 20.7 20.1 19.4 18.9
Mature 81-200 17.7 16.8 14.9 13.7 13.1
Old-growth 200+ 34.4 33.1 31.1 29.9 28.8
Satellite data for our study area were acquired for 1972, 1976, 1984, 1988 and
1991. Imagery for the first two years came from the Landsat Multispectral Scanner
(MSS), while imagery for the latter three years came from the Landsat TM sensor. The
Cohen et a1. (1995) age-class map for our study area was developed using the 1988
image. A second age-class map was developed independently using the 1972 MSS
data. A change detection algorithm was used to develop maps of the area harvested
during each of the four periods. Wildfires within the study area during the period from
1972 to 1991 were minimal and are ignored in this analysis. The 1972 and 1988 age-
class maps and the timber-harvest maps were then used to derive age-class maps for
1976, 1984 and 1991 (described in Cohen et a1. in review).
The largest declines in area between 1972 and 1991 were for the mature and
old-growth age classes (Table 4). Since these are the age classes with the largest total
C storage, harvesting of these sites results in the largest net release of C to the
atmosphere (Fig. 2). Net loss of mature and old-growth forest area averaged 4618 ha'l
y{l (0.57% of the study area) during the period from 1972 to 1991.
Total carbon flux: Mass balance considerations

The LANDCARB model was used to produce a single map of total C storage for each
of five years (1972, 1976, 1984, 1988 and 1991). For each of these years separate
maps were also produced to illustrate storage patterns in the living, detrital and forest
products C pools. Differencing these data layers over any two periods yields a map of
C flux for the relevant pool. These C flux values are presented in Figure 3 and a map
of total C flux for the study area over the period from 1972 to 1991 is presented in
Plate 2.
Our primary interest was to quantify the net exchange of C between the atmosphere
and all components of the terrestrial ecosystem (live, detrital and forest products). This
total C flux over the period from 1972 to 1991 averaged a release (source) of 1.13 Mg
C ha'lyr'l from the terrestrial system to the atmosphere (Fig. 3). Many earlier C budget
studies (Armentano and Ralston 1980, Moulton and Richards 1990, Kauppi et a1.
231
2 source 10 atmosph ra
15
ell
.c
Cl
~
~ 05
Ii:
c:
~ -05
U
,_ Il188 1!lll8 1991 11172" 1991

Years
Fo< Prodoc1s TD
Figure 3. Carbon budget for 811,789 hectares offorested land between 1972 and
1991. Total C flux from this area over this period was 17.5 Tg or 1.13 Mg ha"1 yr"!.
1992) did not adequately deal with the detrital flux or with the decomposition and
incineration of forest products. Doing so involves ignoring mass balance and
calculating ecosystem C flux as live flux (net growth) minus harvest. For our study
area over the period from 1972 to 1991, such a calculation yields an average sink of
0.068 Mg C ha'! yr'!. This figure does not accurately represent the total net C flux for
the terrestrial system. The failure of recent studies to adequately consider all
components of the terrestrial C budget has added much confusion to an already
challenging problem.
Temporal variability in carbon flux

Total C flux values for our study area ranged from a sink of 4.7 Mg C ha'! yr'! to a
source of 15.8 Mg C ha,l yr'! (Table 5). The variability was determined primarily by
which portion of a site's successional trajectory (Fig. 2) was captured during our
1972-1991 sampling period. Since young and mature stands store less ethan
old-growth stands, harvesting them results in a somewhat smaller release of C to the
atmosphere. Over our 19-yr study interval, the largest C sources were old-growth
stands on the most productive sites (site classes 1, 2 and 3) that were harvested
between 1972 and 1976. Regardless of site class, harvested stands undergo their
largest net loss of C (live, detrital and forest products) during the first 25-30 years
after harvesting (Fig. 2). Hence, stands harvested at the beginning of our 19-yr study
were near their minimum C storage levels in 1991. Within the next five to ten years,
the net loss of C from these stands should stop. These stands will then begin to
accumulate C, although it will require an additional 200 years for them to approach
preharvest C storage levels. Stands harvested towards the end of our study were also
sources but have not yet had time to release as much C to the atmosphere. Stands
harvested prior to 1972 were somewhat smaller sources since they had already lost
tv
V)
Table 5. Overall (a) frequency distribution of C flux values and distributions by ownership (b) and site index (c). Values in the tv
I I
first line of data represent C flux in absolute amounts (Mg C ha- yr- ). All other values in panels (a), (b) and (c) are expressed as
percentages of the 811,788ha forested study area. * ~
l'l>
~
l'l>
(a) C flux, 1972-1991 (Mg C ha'l yr'l)
<Q.,
::I:l
Sinks Sources l'l>
< > < > ;:s
C
....
l'l>
-5.0 -4.0 -2.0 -0.25 0.25 2.0 4.0 8.0 16.0 Total
~
;::l
--I 1 I 1 I I I I 1
S
'"
ClQ
Overall 2.7 14.7 15.1 27.6 4.0 15.9 13.6 6.4 100.0 S
So
l'l>
(b) By ownership
~
BLM 0.3 1.8 1.2 2.0 0.4 1.4 1.2 0.5 8.7 l}
Forest Service 1.1 4.6 7.4 16.6 1.3 7.0 5.8 2.6 46.5 ~
ClQ
Private 1.3 7.9 3.9 5.2 2.2 6.9 6.5 3.2 37.1 <Q.,
Wilderness 0.0 0.3 2.5 3.6 0.0 0.4 0.1 0.0 7.0 ~
~
~
(c) By site index class "'tl
c:l
l::>..
:;::
Site 1 0.6 0.0 0.4 0.7 0.2 0.2 0.6 0.3 3.0 <"'l
....
~.
Site 2 2.1 0.0 1.2 2.4 0.4 0.4 2.4 0.7 9.7
~.
Site 3 0.0 14.7 9.4 18.8 3.1 13.2 9.9 5.3 74.4
Site 4 0.0 0.0 3.1 3.7 0.2 1.3 0.6 0.1 9.0
Site 5 0.0 0.0 1.1 1.9 0.1 0.8 0.2 0.0 4.0
* Note: Last columns in panels (b) and (c) do not total] 00% due to round-off error.
233
some of their original C prior to the beginning of our study. Stands harvested more
than about 30 years before 1972 were the biggest C sinks in our study area. These
stands were near their minimum total C storage levels in 1972 (Figs. I and 2) and then
rapidly accumulated C throughout our study period.
Effect of management on carbon flux

C flux is strongly influenced by land management (Table 5). Nearly all of the
wilderness areas in our study were either simulated to be in equilibrium or were small
sinks for atmospheric C. The small source areas in the wilderness appear to have
resulted either from natural disturbances or from windthrow or slash fires that burned
into a wilderness from adjacent land. A majority of the USDA Forest
Service-managed lands were also either in equilibrium or were small sinks for C.
USDA Forest Service-managed lands also included some area as either large sources
or large sinks; however, there was a higher proportion of BLM-administered and
privately held lands in these C flux classes. These results are consistent with the
observation that the conversion from primary to secondary forest was occurring more
rapidly on the BLM-managed and privately held lands than on USDA Forest
Service-managed lands (Spies et al. 1994, Cohen et al. 1995).
Conclusions
The results presented here illustrate how remotely sensed data, used in conjunction
with other spatial data and a metamodel, can be used to quantify the exchange of C
between the atmosphere and terrestrial ecosystems. The spatial and temporal
heterogeneity in the terrestrial system makes it necessary to independently model C
flux at a very large number of points on the ground (Solomon et al. 1993). This
requires detailed data on forest condition and change. Remote sensing offers the only
practical means to systematically map and monitor changes in forest condition over
large areas and across various land-ownership categories.
The use of simple metamodels can provide a powerful means to quantify ecosystem
behavior at large spatial scales. The development of complex models can often serve
as a useful heuristic device and complexity is often necessary to adequately capture
the dynamics of small reference stands. At larger spatial scales, the idiosyncratic
behavior of individual stands may become less important as the goal shifts to
describing the behavior of the larger system. If new driving factors do not come into
play at this larger spatial scale (O'Neill et al. 1986), the average behavior of a large
number of stands can be described by means of a simplified metamodel. The
computational simplicity of a metamodel makes it possible to examine the response of
a very large area using relatively small grid cells.
Model validation always presents significant challenges and is often particularly
problematic when the aim is to model the response of large geographic areas over
several decades. Although we have discussed the problems that may result from the
use of inventory data to drive ecosystem C models. these data can be invaluable for
model validation. We are compiling timber-harvest data for use in model validation.
Data on the volume of timber harvested annually on both public and private lands are
routinely compiled at the county level. These data are also compiled for subcounty
administrative units on public lands. These harvest records provide an independent

check of our estimates of harvested C, which, in turn, will provide an indirect
assessment of the accuracy of living and detrital pools.
We conclude that our study area has been a substantial source for atmospheric CO 2
during the period from 1972 to 1991 (Fig. 3, Plate 2 and Table 5). This conclusion is
not consistent with continental- and global-scale studies (the latter reviewed in Dixon
et al. 1994), which conclude that northern, temperate forests have been C sinks over
this period. Additional work will be required to determine whether the results from our
study area are representative of the region as a whole and to predict the C budget for
this region in the future. Our work demonstrates that forest management practices can
have a major impact on regional C budgets. If management plans currently under
consideration for public lands (FEMAT 1993) are implemented, little primary forest
will be harvested during the coming years. On federally managed public lands where
harvesting will continue, rotation lengths approaching 200 years will frequently be
used. Proposed harvesting guidelines will result in reductions in the amount of C
removed from the harvest site through retention of live trees and coarse woody debris.
This practice also has been increasingly emphasized on private lands in recent years.
Improved silvicultural practices on both public and private lands also are resulting in
better regeneration success. With respect to forest products, improved utilization
standards both in the manufacturing sector and in the marketplace are likely to reduce
the proportion of harvested C that goes into short-term storage pools. Although
considerable work remains to be done, these trends suggest that the amount of C
released from PNW forests may be substantially reduced in the coming years.
Acknowledgments - Funding for this work was provided by the Ecology, Biology and
Atmospheric Chemistry Branch of the Terrestrial Ecology Program at NASA
(W-18,020), and by the National Science Foundation-sponsored H.J. Andrews
Experimental Forest LTER Program (BSR 90-11663), the Global Change Research
Program and the Inventory and Economics Program of the PNW Research Station,
part of the USDA Forest Service.
References
Agee, J.K. 1991. Fire history of Douglas-fir forests in the Pacific Northwest. - In: Ruggiero,
L.P., Aubry, K.B., Carey, A.B. and Huff, M.H. (eds). Wildlife Habitat Relationships in
Old-growth Douglas-fir Forests. USDA Forest Service, General Technical Report
PNW-GTR-285, Pacific Northwest Forest and Range Experiment Station, Portland, OR,
pp.25-34.
Agee, J.K. 1993. Fire Ecology of Pacific Northwest Forests. - Island Press, Washington, DC.
493 pp.
Armentano, T.Y. and Ralston, C.w. 1980. The role of temperate zone forest in the global
carbon cycle. - Can. J. For. Res. 10: 53-60.
Bassett, P.M. and Choate, G.A. 1974. Timber resource statistics for Oregon. - USDA Forest
Service, Resource Bulletin PNW-RB-56, Pacific Northwest Forest and Range Experiment
Station, Portland, OR. 55 pp.
235
Booth, D.E. 1991. Estimating prelogging old-growth in the Pacific Northwest. - J. For. 89:
25-30.
Burke, I.e., Schimel, D.S., Yonker. e.M., Parton. WJ. Joyce. L.A. and Lauenroth. WK. 1990.
Regional modeling of grassland geochemistry using GIS. - Land. Ecol. 4: 45-54.
Burke, I.e.. Kittel, TG.F.. Lauenroth, WK. Snook, P., Yonker, C.M. and Parton. WJ. 1991.
Regional analysis of the central great plains. - Bioscience 41: 685-692.
Cohen, W.B. and Spies, TA. 1992. Estimating structural attributes of Douglas-fir/western
hemlock forest stands from Landsat and SPOT imagery. - Rem. Sens. Environ. 41: 1-17.
Cohen. WB., Spies, TA. and Fiorella, M. 1995. Estimating the age and structure of forests in
a multiownership landscape of western Oregon, USA. - Int. J. Rem. Sens. 16: 721-746.
Cohen, W.B., Wallin, D.O. Harmon, M.E. and Fiorella. M. (in review). Estimated carbon flux
between 1972 and 1991 from forests of the Pacific Northwest region of the United States.
Curtis. R.O., Clendenen. G.W.. Reukema, D.L. and DeMars, DJ. 1982. Yield tables for
managed stands of coast Douglas-fir. - USDA Forest Service. General Technical Report
GTR-PNW-135, Pacific Northwest Forest and Range Experiment Station. Portland. OR.
182 pp.
Dixon. R.K. Brown, S., Houghton, R.A., Solomon, A.M., Trexler. M.e. and Wisniewski, J.
1994. Carbon pools and flux of global forest ecosystems. - Science 263: 185-190.
FEMAT. 1993. Forest Ecosystem Management: An Ecological, Economic. and Social
Assessment. - Report of Forest Ecosystem Management Assessment Team. July 1993,
Portland, OR. 836 pp.
Franklin, J.F. and Dyrness. e.T. 1973. Natural Vegetation of Oregon and Washington. -
Oregon State University Press, Corvallis, OR. 452 pp.
Gedney, D.R. 1982. Timber resources of western Oregon: Highlights and statistics. - USDA
Forest Service, Resource Bulletin PNW-RB-97, Pacific Northwest Forest and Range
Experiment Station. Portland. OR. 84 pp.
Gedney, D.R., Bassett, P.M. and Mei, M.A. 1987. Timber resource statistics for non-federal
forest land in west-central Oregon. - USDA Forest Service, Resource Bulletin
PNW-RB-143. Pacit1c Northwest Forest and Range Experiment Station. Portland. OR. 26
pp.
Grier, e.C. and Logan, R.S. 1977. Old-growth Pselldotsuga menziesii communities of a
western Oregon watershed: Biomass distribution and production budgets. - Ecol. Monogr.
47: 373-400.
Harmon. M.E. and Chen, H. 1991. Coarse woody debris dynamics in two old-growth
ecosystems. - Bioscience 41: 604-610.
Harmon, M.E. Franklin, J.F., Swanson, FJ., Sollins, P., Gregory, S.Y., Lattin. J.D. Anderson.
N.H . Cline. S.P., Aumen, N.G. Sedell. J.R.. Lienkaemper. G.W, Cromack. K. Jr. and
Cummins, K. W 1986. Ecology of coarse woody debris in temperate ecosystems. - Rec.
Adv. Ecol. Res. 15: 133-302.
Harmon. M.E.. Baker. G.A., Spycher, G. and Greene, S.E. 1990a. Leaf-litter decomposition in
the Picea/fsll[?a forests of Olympic National Park, Washington. USA. - For. Ecol. Manage.
31: 55-66.
Harmon, M.E. Ferrell. WK. and Franklin, J.F. 1990b. Effects on carbon storage of conversion
of old-growth forests to young forests. - Science 247: 699-702.
Harmon, M.E., Garman. S.L. and Ferrell. WK. 1995. Modeling the historical patterns of tree
utilization in the Pacit1c Northwest: Implications for carbon sequestration. - Ecol. Appl. (in
press).
Harmon, M.E.. Harmon. J.M .. Ferrell. WK. and Brooks, D. (in review). Modeling carbon
stores in Oregon and Washington forest products: 1900-1992.
Harris, L.D. 1984. The Fragmented Forest. - University of Chicago Press, Chicago. 211 pp.
Houghton, R.A and Woodwell, G.M. 1989. Global climate change. - Sci. Am. 260: 36-44.
Houghton, R.A., Callander, B.A and Varney, S.K. (eds). 1992. Climate Change 1992. -
Cambridge University Press, Cambridge, UK. 150 pp.
IPCe. 1990. Climate Change: The Intergovernmental Panel on Climate Change Scientific
Assessment. Houghton, J.T, Jenkins, G.J. and Ephraums, J.J. (eds). - Cambridge
University Press, Cambridge, UK. 365 pp.
Isaac, L.A 1949. Better Douglas-fir Forests from Better Seed. - University of Washington
Press, Seattle, WA. 64 pp.
Johnson, D.W 1992. Effects of forest management on soil carbon storage. - Water Air Soil
Poll. 64: 83-120.
Kauppi, P.E., Mielikainen, K. and Kuusela, K. 1992. Biomass and carbon budget of European
forests, 1971 to 1990. - Science 256: 70-74.
Keeling, e.D. 1973. Industrial production of carbon dioxide from fossil fuels and limestone. -
Tellus 25: 174-198.
Krankina, O.N. and Harmon, M.E. 1994. The impact of intensive forest management on
carbon stores in forest ecosystems. - World Res. Res. 6: 161-177.
Kurz, WA., Apps, M.I., Webb, TM. and McNamee, P.J. 1992. The carbon budget of the
Canadian forest sector: Phase I. - Forestry Canada Information Report NOR-X-326. 104
pp.
Law, AM. and Kelton, WD. 1991. Simulation Modeling and Analysis. - McGraw-Hili Book
Company, New York. 759 pp.
Moulton, R.I. and Richards, K.R. 1990. Costs of sequestering carbon through tree planting and
forest management in the United States. - USDA Forest Service, General Technical Report
GTR-WO-58, Washington, De. 49 pp.
O'Neill, RY., DeAngelis, D.L., Waide, J.B. and Allen, T.F.H. 1986. A Hierarchical Concept of
Ecosystems. - Princeton University Press, Princeton, NJ. 253 pp.
Orr, J.e. 1993. Accord between ocean models predicting uptake of anthropogenic CO 2, -
Water Air Soil Poll. 70: 465-482.
Post, WM., Peng, T-H., Emanuel, WR., King, AW, Dale, V.H. and DeAngelis, D.L. 1990.
The global carbon cycle. - Am. Sci. 78: 310-326.
Robbins, WG. 1988. Hard Times in Paradise: Coos Bay, Oregon, 1850-1986. - University of
Washington Press, Seattle, WA. 194 pp.
Running, S.W., Nemani, RR, Peterson, D.L., Band, L.E., Potts, D.F., Pierce, L.L. and
Soil Conservation Service. 1991. State Soil Geographic Data Base (STATSGO) Data Users
Guide. - USDA Soil Conservation Service, Miscellaneous Publication 1492, Washington,
De. 88 pp.
Solomon, A.M., Prentice, I.e., Leemans, R. and Cramer, WP. 1993. The interaction of climate
and land use in future terrestrial carbon storage and release. - Water Air Soil Poll. 70:
595-614.
Spies, TA. and Franklin, J.F. 1988. Old-growth and forest dynamics in the Douglas-fir region
of western Oregon and Washington. - Nat. Areas J. 8: 190-201.
Spies, TA., Ripple, W.I. and Bradshaw, G.A 1994. Dynamics and pattern of a managed
coniferous forest landscape. - Ecol. Appl. 4: 555-568.
Sundquist, E.T 1993. The global carbon dioxide budget. - Science 259: 934-941.
Tans, P.P., Fung, I.y. and Takahashi, T 1990. Observational constraints on the global
atmospheric CO 2 budget. - Science 247: 1431-1438.
237
Taylor, J.A. and Lloyd, J. 1992. Sources and sinks of atmospheric CO 2 , - Aust. J. Bot. 40:
407-418.
Turner, DL and Marks, D. 1993. Application of topographically distributed models of energy,
water and carbon balance over the Columbia River basin: A framework for simulating
potential climate change effects at the regional scale. - In: Proceedings of the 32nd Hanford
Symposium on Health and the Environment (Regional Impact of Global Climate Change:
Assessing Change and Response at the Scales that Matter), October 18-21, 1993, Battelle
Pacific Northwest Labs, Richland, WA.
Turner, D.L., Koerper, G.J., Harmon, M.E. and Lee, J.J. 1995. A carbon budget for forests of
the conterminous United States. - Ecol. Appl. 5: 421-436.
USDA Forest Service. 1990. Land and Resource Management Plan: Willamette National
Forest. - USDA Forest Service, Pacific Northwest Region, Portland, OR. 715 pp.
Waddell, K.L., Oswald, D.O. and Powell, D.S. 1989. Forest statistics of the United States,
1987. - USDA Forest Service, Resource Bulletin PNW-RB-168, Pacific Northwest Forest
and Range Experiment Station, Portland, OR. 106 pp.
Waring, R.H. and Franklin, J.E 1979. Evergreen coniferous forests of the Pacific Northwest.-
Science 204: 1380-1386.
Waring, R.H. and Schlesinger, W.H. 1985. Forest Ecosystems: Concepts and Management. -
Academic Press, New York. 339 pp.
SECTION THREE
Global-Level Analyses
TEN
Global Biospheric Monitoring with Remote Sensing

Samuel N. Goward and Dennis G. Dye
Goward, S.N. and Dye, D.G. 1996. Global biospheric monitoring with remote
sensing. - In: Gholz, H.L., Nakane, K. and Shimoda, H. (eds). The Use of Remote
Sensing in the Modeling of Forest Productivity. Kluwer Acad. Pub!., Dordrecht, The
Current understanding of global-scale patterns of terrestrial biospheric processes is

relatively poor. A better understanding of these patterns is needed to evaluate global
environmental changes that may occur in the near future. Satellite remote sensing
offers significant promise for providing the information needed to improve this
understanding, but uncertainty concerning remote sensing signal information still
hinders its usage. This chapter explores selected examples of satellite-derived
information based on current understanding of the observations. Primary emphasis is
given to the use of Advanced Very High Resolution Radiometer (AVHRR)
observations to parameterize primary production efficiency models. Specifically, we
consider the bases for extracting information on incident photosynthetically active
radiation (PAR), the fraction of PAR captured by plant canopies, air temperature,
atmospheric vapor pressure deficits, surface soil moisture and above-ground biomass
from satellite observations. Utilizing the potential of satellite remote sensing - the
record of terrestrial vegetation and related environmental spatiotemporal patterns -
in global-scale ecological research now appears feasible. Further significant progress
will be achieved not only with better sensors but more importantly by the
development of new conceptual perspectives that consider the linkages between what
is observed by the sensors and what occurs in the terrestrial biosphere.
S.N. Goward, Laboratory for Global Remote Sensing Studies, Department of

Geography, University of Maryland, College Park, MD 20742, USA. D.G. Dye,
Global Engineering Laboratory, Institute of Industrial Science, University of Tokyo,
7-22-1 Roppongi, Minato-ku, Tokyo 106, Japan.
241
Introduction
Vegetation processes, particularly photosynthesis, are essential to all life forms on
earth; they also determine the environment in which that life is sustained (Budyko
1974, Ehrlich and Roughgarden 1987). Vegetation serves as a primary absorber of
incident solar radiation as well as the determinant of how this energy is redistributed
into the atmosphere (Sellers 1965, Rosenzweig and Dickerson 1986). Further
biospheric processes influence the geochemistry of the planet by modulating
atmospheric chemical composition, especially concentrations of CO 2 and Hp, which
in turn modulates climate (Bolin 1977, Hansen et al. 1981, Post et al. 1990, Houghton
et al. 1990, Aber 1992). The relationship between photosynthesis and planetary life
has changed rapidly over the last century. Human beings have increasingly employed
technology to extract and utilize terrestrial resources, particularly from the biosphere
(Thomas 1966, Vitousek et al. 1986, Grigg 1987, Turner 1989, Turner et al. 1990).
Given our relatively limited knowledge of the global-scale character of vegetation
patterns and processes, it is not yet possible to know whether these significant human
interventions are altering the capacity of the planet to sustain life.
The need to better understand global-scale biospheric activity has been
highlighted by the inability to balance the global carbon (C) budget (Duvigneaud et
al. 1979, Post et al. 1990, Tans et al. 1990). Currently, there is considerable
discrepancy between known patterns of annual C emission and uptake by oceans and
land areas. Observed atmospheric CO 2 concentrations are approximately 30% too
low when emissions and uptake are compared. This uncertainty clearly makes
assessment of possible future atmospheric composition, and therefore climate,
exceptionally difficult. Much attention has been drawn to the role played by the
terrestrial biosphere, particularly forests, as the primary missing sink, as well as a
major store of planetary C (Scientific Committee on Problems of the Environment
1984, Tans et al. 1990).
Given the critical role the terrestrial biosphere plays in the earth system, our
knowledge of this biosphere, particularly at global scales, is exceptionally poor. A
comparison of various attempts to map global land cover reveals a variation of more
than a factor of 2 in the estimated proportion of land covered by forests, grasslands
and deserts (Townshend et al. 1991, DeFries and Townshend 1994). Errors increase
in proportion to the level of detail required. Answers to more sophisticated questions,
such as how to determine the age distribution of forests, how much biomass exists in
living tissue, or how growth processes vary from year to year or from decade to
decade, simply do not exist.
Satellite remote sensing

One means to improve understanding of the global patterns and dynamics of
terrestrial vegetation systems is through analysis of satellite-acquired remotely
sensed observations (National Research Council 1986, Mooney and Hobbs 1990,
Goward and Kazantsev 1991). Since the 1960s, satellite observatories have provided
the consistent and repetitive global observations needed to examine the regional
activity and seasonal dynamics of the biosphere.
243
While these satellite observatories may not be capable of representing the fine
detail observed at local field sites, they do provide what cannot be obtained in any
other way: comprehensive but spatially discrete and consistent observations, repeated
systematically over hours, days, weeks, years, decades and, potentially, centuries
(Yates et al. 1986). Some observatories, such as the National Oceanic and
Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer
(AVHRR), are able to collect daily data of the entire planetary surface at spatial
resolutions as fine as 1 km 2 (Townshend and Tucker 1981, Schneider and McGinnis
1982). Sensors on other satellites reduce temporal repeat frequency but increase
spatial resolution to 30 m or better (e.g., Landsat, SPOT and IRIS, Freden and
Gorden 1983, Cushnie 1988). Still other satellites in high-altitude, geostationary
orbits remain positioned over the same area of the earth, providing observations of
the visible hemisphere on at least an hourly basis (Tarpley 1979). The capability to
repeatedly observe all regions of the earth using the same measurement system is
unique in the history of earth sciences. Thus, satellite remote sensing observatories
provide, for the first time, a consistent and instantaneous means to examine the earth
as a unified system of time-vary,{ng, dynamic physical and biological processes.
Remote sensing measurements

As discussed throughout this volume, interpretation of remote sensing measurements
relative to vegetation attributes and processes is a complex problem. In general,
multiple factors within vegetation canopies, in the adjacent landscape and often in the
intervening atmosphere determine the signal recorded by the satellite sensor (Colwell
1974, Swain and Davis 1978, Asrar 1989). Determining the influence of a single
factor in a particular signal is difficult locally and even more so at global scales. For
example, the attempt to relate visible and near-infrared spectral vegetation indices
(SVIs) to canopy leaf area index (LAI) may be successful for a single species or
growth form (Curran and Milton 1983, Asrar et al. 1984, Spanner et al. 1994).
However, quite different relationships are found between diverse vegetation types
(Curran 1983, Baret and Guyot 1991). This multiple-factor interpretation problem
affects all remote sensing measurements, independent of observation wavelength or
other sensor attributes. As the size of the area considered expands from regions to
continents and, finally, to the globe, the interpretation process becomes more difficult
because each of the multiple factors takes on a wider range of values.
Evaluation of global-scale biospheric activity from remotely sensed measurements
requires not only scaling of locally relevant data to regions but also consideration of
possible new ecosystem variables defined by the integrated character of the
remotely sensed observation itself (Mooney and Field 1988, Townshend and Justice
1990, Field 1991, Ehleringer and Field 1993). Thus, interpretation of remotely
sensed observations may at first appear daunting. However, inspection of available
satellite observations has revealed that satellite sensors record spatiotemporal
patterns bearing a strong relationship to known patterns of terrestrial biospheric
variability in space and time (Justice et al. 1985, Koomanoff 1989, Townshend et al.
1991). This discrepancy between the physical complexity of remotely sensed
measurements and the apparent simplicity of observed global-scale measurement
patterns suggests that the ecosystem variables observed are simple but robust
descriptors of terrestrial vegetation patterns. Although we may not yet fully
understand the physiological or environmental significance of remotely sensed
measurements, we can certainly appreciate their value in assessing the patterns and
dynamics of the terrestrial biosphere at global scales.
Global remote sensing possibilities

With the foregoing discussion in mind, we consider the biospheric monitoring
potential of two existing sensors: the TOMS (Total Ozone Mapping Spectrometer)
and the AVHRR (Advanced Very High Resolution Radiometer). This narrow focus on
two of the many earth satellite sensors reflects our own long-term experience with
these observation records. Further, recent production by NOAA and NASA of a new,
high-quality summary of the historical AVHRR record (NOAAJNASA AVHRR Land
Pathfinder data set) presents a major new opportunity to explore application of
remote sensing to global-scale monitoring of the biosphere (James and Kullari 1994).
More than 10 years ago, we began exploration of the AVHRR observation record
to assess the value of visible/near-infrared spectral vegetation index measurements in
biospheric monitoring (Goward et al. 1985b). This research led us to recognize the
need for complementary information on related phenomena (Goward and Dye 1987,
Goward 1989, Goward and Hope 1989, Goward et al. 1994a). Our focus has
remained primarily on the AVHRR observations, because once we were familiar with
the data structure it was easier to proceed within the context of that data set. Our first
departure from this focus was reflected in a study of the TOMS observations to
estimate incident photosynthetically active radiation, through collaboration with an
experienced TOMS researcher (Eck and Dye 1991).
Our work with the TOMS and AVHRR sensors has revealed that these
observations contain much more information about the terrestrial biospheric
environment than was anticipated during mission planning. We believe that this
thorough consideration of specific sensor records, performed with a particular
objective in mind, has produced significant advances in the scientific value of these
observatories. Based on this experience, we also briefly consider what measurement
advances may result from studies of data obtained by more sophisticated
observatories designed to improve the AVHRR observation data set.
None of this emphasis on the TOMS and, particularly, the AVHRR sensor, is
meant to imply that AVHRR-type sensors are the only remote sensing instruments of
value to global-scale terrestrial research. We recognize the complementary value of
observations from the many other available and planned sensor systems but have not
yet explored these additional sources of global observations (Rasool 1987, Gurney et
al. 1993, Committee on Earth Observation Satellites 1995). Indeed, one of the
shortcomings of contemporary applications of remote sensing to global-scale studies
is that so few studies have attempted to exploit the diversity of measurements
possible, even from one sensor system such as the AVHRR. We find that the primary
constraints for using multiple sources of remotely sensed data include i) large data
volumes, ii) arcane and incompatible data formats and iii) a requirement for
specialized knowledge. We believe that all existing and planned satellite remote
245
sensing observatories are potentially rich sources of terrestrial environmental

information that await exploitation by knowledgeable earth scientists and data
technicians. We present our experiences with the TOMS and AVHRR sensors as
examples of the scientific progress that can be achieved.
Much of the inspiration for this exploration of AVHRR and TOMS observations
can be found in the context of the energy conversion, or production efficiency,
concept discussed elsewhere in this volume (Landsberg et al. 1996). The generalized
nature of this PEM (production efficiency model) concept appears well suited to the
simple but robust character of satellite remote sensing observations. Based on current
understanding of the concept (Monteith 1977, Jarvis and Leverenz 1983, Landsberg
1986, Prince 1991, Prince et al. 1994, Runyon et al. 1994), the PEM concept requires,
at a minimum, measurements of incident PAR (photosynthetically active radiation),
fractional PAR absorbed by vegetation canopies, environmental constraints including
air temperature, vapor pressure deficit and low soil moisture (drought), living
biomass (for calculating respiration) and C3 , C4 and CAM vegetation types.
To date, we have achieved some success in deriving all but the last of these
variables from either the TOMS or the AVHRR sensor. Future versions of the
AVHRR and other sensors may assist in derivation of the latter, photosynthetic
pathway assessment. This success hints at the potential for studying the earth's
biosphere using the full range of available and planned satellite remote sensing
observatories.
Photosynthetically active radiation

Photosynthesis and net primary production (NPP) are driven by energy from solar
radiation at wavelengths between approximately 400 and 700 nm. The amount of
PAR received at the earth's surface per unit time is thus a key variable in many
physically based models ofNPP and related biospheric processes (e.g., Heimann and
Keeling 1989, Prince 1991, Sellers 1991, Potter et al. 1993). Measurements of PAR
are more rare. Global patterns of PAR have previously been based only on estimates
derived from full-wavelength (300-3000 nm), short-wave (SW) irradiance data
collected at a small number of ground-based measurement stations around the world
(e.g., L6f et al. 1966, Yefimova 1971). The introduction of satellite remote sensing
methods has provided a practical means for improving the accuracy of measurements
of spatial and temporal dynamics of PAR at the earth's surface. Such methods
provide an important new data source for evaluating the dynamics of energy flow and
mass exchange in the biosphere (Dye and Goward 1993).
Satellite techniques for estimating solar irradiance on a global basis were initially
developed to provide data that would support modeling of the energy balance of the
earth-atmosphere system, and were therefore directed toward estimating and
mapping broadband SW irradiance (e.g., Schmetz 1991, Darnell et al. 1992, Pinker
and Laszlo 1992a). In recent years, considerable success has been achieved in
estimating global PAR by combining satellite measurements of cloud cover with
atmospheric radiation transfer models parameterized for the 400--700 nm region.
Satellite-based methods for estimating PAR on a global basis have been introduced
by Eck and Dye (1991) and Pinker and Laszlo (1992b). These methods are
conceptually similar and differ primarily in the complexity of the radiation models
and spectral regions employed.
The methods described by Eck and Dye (1991) and Pinker and Laszlo (1992b)
employ narrow-band satellite reflectances to account for the effects of clouds and
aerosols on a predicted clear-sky or top-of-the-atmosphere irradiance. The Pinker and
Laszlo (1992b) method employs visible radiance data from the International Satellite
Cloud Climatology Project (lSCCP) Cl data product in a detailed radiation model to
estimate PAR over the globe. The ISCCP Cl product is a global, 2.5-resolution
gridded data set containing information on cloud, atmosphere and surface
characteristics derived in part from the combined observations of polar-orbiting and
geostationary satellite sensors (Schiffer and Rossow 1983). A possible advantage of
the ISCCP data employed by Pinker and Laszlo (1992b) is that the geostationary
observations, for the portions of the globe they cover, potentially provide a more
reliable representation of diurnal variations in cloud conditions than that achieved by
daily observations from a polar-orbiting sensor alone. The method described by Eck
and Dye (1991, see also Peterson 1996) combines a comparatively simple radiation
model with ultraviole,t (UV, 370-nm) reflectance data from the polar-orbiting
Nimbus-7 Total Ozone Mapping Spectrometer (TOMS) (Plate 1). PAR backscatter is
inferred from the observed 370-nm reflectivity based on the spectrally flat reflectivity
of clouds reported across near-UV and PAR wavelengths (Novoseltsev 1964). The
advantage of near-UV data is that the background surface reflectivity is generally
lower and less variable (Eck et al. 1987) in this spectral region than in the visible
region, which enables greater precision in accounting for the PAR backscatter
attributable to clouds and aerosols. The accuracy of the TOMS estimates of monthly
PAR was observed to be between 4% and 6% (root-mean-square differences) in
comparisons with estimates made on the ground for several climatologically diverse,
midlatitude locations (Eck and Dye 1991). While reliable time series of PAR
measurements made on the ground and representing a broader sample of global
climate regimes are required for a more thorough evaluation of the absolute and
relative accuracies of PAR estimates from existing methods, the validation results
from Eck and Dye (1991) are encouraging.
Fractional PAR absorbed in canopies

A critical aspect of the conversion of sunlight into plant matter is the capture of
incident PAR by the photosynthetically active canopy components, primarily green
leaves. At first, it would seem that this fraction could be derived simply by measuring
observed surface reflectance in the PAR wavelengths, similar to the estimation of
incident PAR. This might be possible if both background surface reflectance and
woody elements of the canopy displayed constant and preferably perfect (100%)
reflectance. Unfortunately, soil, litter and bark reflectances vary widely in the PAR
region, from almost 0% to more than 50% (Stoner and Baumgardner 1981, Goward
et al. 1994a). If snow is also considered, this range increases to almost 100%.
One of the most intriguing aspects of satellite remote sensing is the information
contained in the combination of visible and near-infrared spectral reflectance
measurements (Tucker 1979, Jackson 1983, Daughtry et al. 1992). SVI measurements
247
appear to be uniquely associated with the presence of photosynthetically active

pigments in vegetation. These pigments, particularly chlorophyll but also carotenes,
xanthophylls and anthocyanins, absorb strongly in PAR wavelengths but do not
absorb in near-infrared wavelengths (Gates et al. 1965, Allen and Richardson 1968,
Wooley 1971). Further, the internal structure of living green tissue causes strong
scattering which, because of low absorptance in the near-infrared, results in strong
reflectance in near-infrared wavelengths (Nilson and Ross 1996). Thus, leaf structure
and leaf pigments produce a unique visible/near-infrared "signature" with low visible
and generally high near-infrared reflectance.
This unique spectral reflectance of vegetation has been studied for more than half
a century (Ives 1939, Krinov 1947, Billings and Morris 1951, Colwell 1956, Jordan
1969). However, it was only with the advent of satellite remote sensing, particularly
the Landsat Multispectral Scanner (MSS) and Thematic Mapper (TM) sensors, that
the pervasive character of this attribute of vegetation was fully recognized (Deering
et al. 1975, Blair and Baumgardner 1977, Graetz and Gentle 1982, Foran 1987).
Much effort has gone into explaining the information content of SVI measurements.
What began as a search for unique spectral "signatures" of vegetation species or types
soon became a consideration of SVI measurements as a generalized estimate of the
pigment-bearing elements of the canopy (Kauth and Thomas 1976, Kettig and
Landgrebe 1976, Jayroe 1978, Bauer et al. 1981, Badhwar 1984). This perspective
was further encouraged by the availability of global-scale AVHRR observations (Fig.
1). For the first time, frequent, repeated, global observations of SVI were available
for inspection. It was found that these measurements are strongly correlated with the
seasonality and annual growth patterns (e.g., NPP) of vegetation at continental and
global scales (Goward et al. 1985b, Justice et al. 1985, Tucker et al. 1985, Holben
1986, Townshend and Justice 1986, Goward and Dye 1987) (Fig. 2).
A full explanation of the observed correlations between SVI measurements and
vegetation activity is still to be achieved. Early field and Landsat MSS studies in
grasslands and agricultural areas suggested strong relationships with green LAI,
green biomass, total biomass and percentage of ground cover (Deering et al. 1975,
Holben et al. 1980, McDonald and Hall 1980, Bauer et al. 1981, Tucker et al. 1981).
Studies with forest canopies have shown similar success relative to green LAI and
ground cover, but not with standing woody biomass (Dethier 1974, Peterson et al.
1987). Comparison of empirical results among vegetation types (e.g., grasslands,
broadleaf forests and needleleaf forests) has shown that quite different relationships
occur as a function of vegetation type (Curran 1981, Curran 1983, Williams 1991). In
addition, most of these relationships have been shown to be quite nonlinear,
saturating at high values of green LAlor green biomass. Thus, scaling between
measurements made at different spatial scales is difficult.
The variability in relationships between plant canopy variables and SVI
measurements is explained, in part, by the multiplicity of factors that produce
observed spectral reflectance patterns (Colwell 1974). Theoretical radiative transfer
studies have demonstrated the need to consider at least the following landscape
variables: (i) LAI, (ii) leaf angle distribution, (iii) spectral optics of the green foliage
and woody components and (iv) the spectral reflectance of the background surface
Figure I. Estimate of the fraction of incident PAR (fpar ) absorbed by green plant
canopies (white, "" 100%; gray, "" 50%; black, "" 0%) for the period 1-10 June
1987, based on AVHRR 8-km NOAAlNASA Pathfinder data. A single, linear
relationship is assumed between NDVI (normalized difference vegetation index)
measurements and f par The NDVI value has been adjusted for water vapor
attenuation based on Figure 4. An average global aerosol attenuation
adjustment is also included (Goward et aI. 1991). No attempt was made to adjust
for possible background soil and litter variations.
(e.g., soil, litter, snow and water) (Suits 1972, Kimes and Kirchner 1982, Verhoeff
1984, Nilson and Ross 1996). Even within a particular vegetation type, variations in
leaf angle distribution or leaf optics affect SVIs when LAI and percentage of ground
cover remain constant. Such effects are most clearly observed in marginal
environments where lack of moisture or severe cold limit growth. In desert regions,
many species are found to have leaves in which the optical properties of
photosynthetically active pigments are masked by surface coatings (e.g., those of
pubescence or waxy epidermis) that reduce midday light absorption (Ehleringer and
Mooney 1978, Ehleringer and Werk 1986). Similar effects are achieved by more
erectophile leaf angle distributions or more dynamic, solar-tracking leaf orientations
(Forseth and Ehleringer 1983). Curiously, similar reductions in observed canopy
pigment absorption occur at high latitudes (and altitudes) where lichen and moss
species replace vascular plants (Tieszen and Johnson 1968, Petzold and Goward
1988). These variations in canopy physiognomic attributes reduce the observed SVI
even when LAI and percentage of ground cover remain similar to those of less
marginal ecosystems.
Difficulties in finding a generalized explanation of SVI measurements in terms of
LAI led to consideration of PAR absorption in canopies. Early work with grasslands
249
1600
~ 1400
"1
1200
l::
.,
.51
<.> 1000
. . .+...- : +.. ._.

:3
e
"0
.Coniferous ...L.-..... 0.5 g/MJ
800
ll<
t' ;
I.._-_1____
I _+_0
b0 ~C~~~~~i:land_.._..
os
o
~ G~ssland
600
.@ Tund;a-Conif .
.,
ll< 400
Q)
Z
!
.~ .
200
o-~-I-'-~-!-'--'----'--1i--'-'--'--!-~-'+~+~--t-'~+
o 200 400 600 800 1000 1200 1400 1600

Absorbed PAR (MJ m")
Figure 2. Observed correlation between estimated annual APAR (tPA~*fpar)

and reported above-ground NPP for North American biomes (Goward et al.
1985b, Goward and Dye 1987, Goward et al. 1987, Koomanoff 1989). NPP
measurements are from Whittaker and Likens (1975). The dispersion in the
pattern suggests that other factors besides APAR determine biome variations in
NPP.
and crops suggested that the fractional PAR (fpar) absorbed in canopies has an almost
linear relationship with SVI (Kumar and Monteith 1981, Daughtry et al. 1982, Asrar
et al. 1984, Hatfield et al. 1984). Similar results have also been shown for forests
(Peterson and Waring 1994). This linearity means that measurements scale from local
to regional. These empirical results have been confirmed primarily in theoretical
studies, which show a consistency in the relationship over a wide range of leaf optical
properties, leaf angle distributions and illumination and viewing geometries (Sellers
1985, Choudhury 1986, Baret and Guyot 1991, Goward and Huemrnrich 1992).
Thus, the fpar measured using SVIs is being widely employed to examine global-scale
biospheric activity (Goward and Dye 1987, Prince 1991, Potter et al. 1993, Ruimy et
al. 1994, Prince and Goward 1996).
One unresolved difficulty in estimating fpar from SVIs is the effect of variable
background spectral reflectance (Choudhury 1986, Huete et al. 1986, Huete and
Tucker 1991). Several modifications to SVI calculations have been proposed to
minimize this effect in semi-arid regions (Huete 1988, Major et al. 1990). At global
scales, however, the most serious aspects of this problem occur where background
soil is covered with litter or where both are covered with snow. Litter spectral
reflectance falls between spectral reflectances of green vegetation and soils,
confusing interpretation of f par (Goward et al. 1994a). For example, AVHRR
(:I)
(b)
251
Figure 3. (opposite page) (a) Winter NDVI measurements for North America
recorded 15-30 February 1986, based on NOAA Global Vegetation Index data
reprocessed at the University of Maryland (Goward et a1. 1994a). NDVI
measurements between 0.2 and 0.4 were recorded in this winter "leaf-off"
period in deciduous forest regions of the eastern United States as well as in
mixed forests of southern Canada. These values may be compared with the low
(0.1) NDVI measurements from southwestern U.S. desert regions and the
intensely cultivated southern Mississippi River valley region. Discrepancies in
these values may be the result of variations between litter reflectance on the
forest floor and on the bare soil of agricultural fields and deserts. Low NDVI
values recorded adjacent to the border between the United States and Canada
result primarily from snow cover, as shown in (b). The visible reflectance image
for the same observation period shows a pattern of high-reflectance snow cover
(Goward and Huemmrich 1992). It is interesting to note that the effects of snow
cover are not as apparent north of this region. Whether this results from an
absence of snow or an overstory of forest trees in this region is not known. These
observed winter regional variations in NDVI suggest that a simple "green
foliage" explanation of spectral vegetation indices is insufficient. In terms of
global NPP estimation, however, this "error" may not be significant.
Temperatures are typically near or below freezing during this time of year and
the incident PAR is relatively low, which results in minimal NPP in any case.
observations record higher SVI measurements for deciduous forest regions in winter
than for adjacent agricultural areas, even though there is little green foliage present in
either landscape (Fig. 3). This may be due to the contrast between litter background
in forested areas and bare soil in agricultural areas. A converse problem occurs when
snow is present. Snow typically has higher visible than near-infrared reflectance,
which contracts the basic SVI signal. Thus, where snow is present the observed
landscape SVI drops sharply, even in evergreen forests. This can best be seen in
boreal forests regions of the Northern Hemisphere, where the forests seem to
"disappear" in winter observations (Fig. 3).
Although background reflectance appears to have rather dramatic effects on SVI,
enough perhaps to preclude winter use of SVIs to estimate f par ' this limitation needs
to be evaluated in context. If what is sought is an estimate of f par needed to sustain
vegetation growth, then loss of the SVI signal during winter, low-radiation periods
may not be important. Within the context of the PEM concept, photosynthesis cannot
take place at air temperatures below OC and incident PAR radiation is relatively low
during this period in any case. Inability to accurately estimate winter f par ' therefore, is
either irrelevant or of minor importance. This is a good example of the type of
integrated, phenomenological thinking that needs to be brought to the interpretation
of remotely sensed observations for global-scale biospheric monitoring. Although it
may soon be possible to improve the specificity of such ecosystem assessments using
more sophisticated sensors, signals from current observatories appear to contain most
of the information needed to evaluate seasonal and interannual growth dynamics.
Environmental constraints
Estimation of environmental constraints on photosynthetic activity (extreme air
temperature, high air saturation vapor deficit, low soil moisture) presents a more
complex remote sensing interpretation problem than that encountered with incident
PAR and canopy fpar ' Inference of land surface environmental conditions is a major
objective in terrestrial remote sensing research (Goward and Oliver 1977, Yates et al.
1986, Goward 1989, Pinker 1990). In particular, thermal infrared (TIR)
measurements appear to have significant potential in this regard because they are
directly related to the kinetic energy of the observed surface (Wolfe 1965, Heilman et
al. 1976, Price 1980). For certain environmental variables, especially soil moisture,
passive microwave observations also show potential (Njoku 1982, Choudhury 1989).
Progress in application of TIR observations to environmental assessments has
been slow. This is primarily because land surface energy (and therefore temperature)
is a complex function depending on multiple variables, many of which (e.g., incident
solar radiation) vary over short time periods (Watson 1973, Smith et al. 1985, Price
1989). Most studies have developed methods requiring coincident ground
observations of such conditions as air temperature to resolve specific environmental
constraints (Jackson et al. 1981, Seguin and Itier 1983, Carlson and Buffum 1989).
Essentially, at the single pixel level, inference of land surface environmental
conditions based on TIR observations is an indeterminant problem because too many
variables are needed to evaluate the biophysical significance of a particular TIR
measurement.
Various means are available to increase the dimensionality of the TIR data, for
example, collection of measurements in additional spectral bands, one of the
procedures used for atmospheric sounding (Susskind et al. 1984). Atmospheric
sounding of air temperature and water vapor profiles, using instruments such as the
HIRSIMSU in the Tiros Operational Vertical Sounder (TOVS), appears to be quite
relevant to the estimation of variables needed in biospheric monitoring. However,
aside from introducing another data product into the analysis process, these
observations are not well designed to estimate environmental conditions at the near-
surface levels in the atmosphere that most directly affect biospheric activity. For
example, the air temperature estimates generally represent an average value for the
atmosphere, ranging from 500 mb to 1000 mb. Nevertheless, there may be creative
ways to interpret these sounder observations that would complement the estimates we
derive from the AVHRR sensor.
An alternative approach to increasing TIR measurement dimensionality is to
examine a spatial array of observations. This "contextual" analysis approach, which
is particularly relevant to the AVHRR and other land surface sensors, increases
measurement dimensionality by sacrificing observational spatial resolution.
Applications of contextual analysis to global-scale biophysical remote sensing have
been quite limited so far. The most extensive use has been in cloud detection research,
where image variance is used to determine cloud presence (Coakley and Bretherton
253
1982, Gutman et al. 1987. Saunders and Kriebel 1988, Stowe et al. 1991). A similar
conceptual evolution is developing in inference of atmospheric properties (Jedlovec
1990, Kleespies and McMillin 1990, Harris and Mason 1992, Tame et al. 1992).
Contextual analysis has also begun to attract attention as a way to better exploit
thermal infrared observations for inference of land surface environmental conditions
(Goward et al. 1985a, Goward and Hope 1989, Nemani and Running 1989, Carlson
et al. 1990, Price 1990, Nemani et al. 1993, Carlson et al. 1994).
Atmospheric humidity
Inference of atmospheric humidity takes advantage of multispectral TIR
observations. For example, the afternoon AVHRR sensor collects two wavelengths of
TIR (channel 4: 10.3-11.2 mm and channel 5: 11.5-12.5 mm). This "split-window"
TIR observation set was originally deployed on the AVHRR to improve estimates of
ocean surface temperatures (Prabhakara et al. 1974). Atmospheric water vapor
between the sensor and the surface differentially attenuates TIR signals in the two
channels. The differential signal is diagnostic of the water vapor effect, allowing the
true surface radiant emissions (and temperature) to be estimated (McClain et al.
1985).
This differential signal should also be capable of providing estimates of the
amount of water vapor in the intervening atmosphere. Procedures that exploit this
opportunity have been developed for ocean areas (Dalu 1986) and use of these
procedures is being considered for land areas (Justice et al. 1991). The difticulty
encountered in land areas is that the difference between the two TIR signals varies
due to both atmospheric water vapor and the magnitude of the TIR radiation source
(i.e., the temperature of the land surface) (Kleespies and McMillin 1990, Harris and
Mason 1992). Daytime land surface temperatures are highly variable: contrasts in
excess of 60C have been recorded in semi-arid landscapes (Seguin and Itier 1983).
To account for the strength of the radiation source, it has been proposed that two or
more observations from targets with different temperatures (i.e., a contextual sample)
be used (Kleespies and McMillin 1990, Jedlovec 1990, Harris and Mason 1992).
Preliminary experiments with this split-window, contextual TIR approach to
estimation of atmospheric water vapor have proven successful in the FIFE (First
International Satellite Land Surface Climatology Project Field Experiment) and
OTTER (Oregon Transect Ecosystems Research) studies (Goward et al. 1994b,
Peterson 1996). This success has led to a global application of these procedures using
the NASA/NOAA AVHRR Pathtinder data set (Prince and Goward 1996). The results
of this application are encouraging but will require further validation to contirm the
derived water vapor patterns (Fig. 4).
Air temperature and soil moisture limitations

Both air temperature and a soil drought index may be derived by examining the
correlation between SVI measurements and surface temperature (TJ measurements
in a contextual array of observations (Fig. 5). This correlation has been widely
observed in diverse regions of the globe (Goward et al. 1985a, Goward and Hope
Figure 4. Distribution of atmospheric water vapor (g cm- 2 ) derived from

AVHRR split-window thermal infrared measurements (channels 4 and 5)
obtained during the period 1-10 June 1987, based on AVHRR 8-km
NOAAINASA Pathfinder data (see Prince and Goward 1996 for technique).
(Gray tone key: white, > 5 g cm ; light gray, == 2.5 g cm2 ; dark gray, < 1 g cm2 ;
2
black, no value). One of two factors resulted in no solution: (i) sensor saturation
above 320 K (major desert regions) or (ii) method failure between 290 K and 294
K. Solutions to the latter problem are currently under investigation. The broad
regional patterns correspond to expectations. Confirmation of more local,
continental-scale variations awaits further analysis. Atmospheric saturation
vapor deficit is derived from this observation as well as from air temperature
measurements derived as shown in Figures 5 and 6a.
1989, Nemani and Running 1989, Carlson et al. 1990, Price 1990, Smith and
Choudhury 1991, Hope and McDowell 1992, Nemani et al. 1993). It has been
proposed as a means to assess surface moisture conditions in relation to
evapotranspiration and canopy conductance (Hope et al. 1987, Nemani and Running
1989, Carlson et al. 1990). The conceptual and theoretical bases for this inference are
still under investigation.
Research to date indicates that the observed midday Normalized Difference
Vegetation Index (NDVI)ffs correlation results from contrasts between sunlit soil
temperatures and vegetation canopy temperatures (Goward and Hope 1989, Prihodko
1992, Carlson et al. 1994, Goward et al. 1994b). Surface soil temperatures vary
widely as a function of surface thermal inertia, net radiation balance and fluxes of
sensible and latent heat (Geiger 1965, Monteith and Unsworth 1990). Because
canopies consist of foliage with a small mass per unit volume (nearly equal to that of
air), canopy temperatures rarely deviate from ambient air temperatures by more than
255
50 A
f!
0
,15 ~ 00
~
U
10
~
"
'i:
i!

....._.
.._- c:
..
.. ..,
....
...
::l
35
y 1
1::, C>
....-
....
.... ::
Ol
ell
0. ::r
E :lO
ell
Eo<
T
25 "
20
n,a 0.4 0,5 0,6 07 (Ul (J.9
ormalized DilTerenee V gcl.<llion Index
l. D\'1>
Figure 5. Example of negative correlations observed between NDVI and surface

temperature in a contextual array of AVHRR data. Measurements are derived
from l-km local area coverage data collected for the NASA FIFE experiment in
Kansas, USA. Air temperature is inferred from surface temperature for an
"infinitely thick" or completely closed canopy. It can be seen that there is little
change in air temperature between the two days. The surface soil moisture
condition is derived from the slope of the relationship, adjusted for the solar
zenith at the time of the observation (Prince and Goward 1996). Here, the solar
zenith changes little between the two days, suggesting wetter surface conditions
on 4 July 1987.
2.0C, whether or not the vegetation is actively transpiring (Geiger 1965, Aston and
van Bavel 1972, Gates 1980).
On the other hand, soils generally have relatively high thermal inertia (Van Wijk
and De Vries 1963). When soils are dry and directly exposed to sunlight, soil
temperatures can exceed air temperatures by as much as 3O-60C during the summer
under midday, clear-sky conditions. When soils are wet, soil thermal inertia increases
but the majority of absorbed radiant energy is then converted to latent heat through
evaporation. Wet soil temperatures approach air temperature with strong evaporation.
These considerations lead to the conclusion that, for midday TIR observations,
canopy T s measurements should approximate measurements of air temperature, and
the canopy/soil T s contrast should vary with soil wetness. Since SVI measurements
are diagnostic of canopy foliage presence, observed T s should, in general, be
inversely correlated with SVI measurements. Extrapolation of the observed NDVIITs
relationship to an NDVI of an infinitely thick vegetation canopy should provide an
estimate of air temperature, and evaluation of the slope of the NDVIlTs relationship
should be diagnostic of surface wetness. Based on tests of this contextual
methodology for FIFE, OTTER and the continental United States (Prihodko 1992,
Nemani et al. 1993, Goward et al. 1994b), this approach has recently been applied to
the global AVHRR Pathfinder data set (Prince and Goward 1996). The results reveal
expected patterns of higher air temperatures and larger droughts in tropical deserts
and lower temperatures with little drought in midlatitude forests (Fig. 6). As with the
water vapor assessment, further validation of this analytical approach is needed;
however, these preliminary results clearly demonstrate the possibility of global-scale
biospheric monitoring of daytime air temperature and surface moisture conditions.
Biomass estimation
Above-ground standing biomass is one of the more interesting vegetation variables
that might lend itself to remote sensing evaluation. This attribute is highly variable
across the earth's surface; where human presence is strong, it can also be highly
variable in time (e.g., in deforested areas of the Amazon). Much of the remote sensing
effort to date has focused on the use of active microwave (radar) to detect biomass
(Hussin et al. 1991, Dobson et al. 1992). Although radar observations show
considerable promise, .the absence of truly global observations with appropriate
frequencies (wavelengths) and temporal resolution has so far limited global-scale
interpretations.
An optical wavelength approach to biomass estimation may be possible, as
indicated by several field studies carried out over the last several decades (Stanhill
1970, Otterman 1984, Nilson and Peterson 1994). Researchers have found that the
visible surface reflectance of vegetation canopies decreases significantly as the
woody biomass increases (Li and Strahler 1985, Hellden 1987, Jasinski and Eagleson
1990). This appears to occur because of the increased shadowing within the canopy
volume. These results suggest that a visible brightness image of land areas might
indicate the amount of woody biomass present.
Recently, we compiled a six-month minimum brightness image of the globe using
visible band (channell) reflectance data from the AVHRR Pathfinder data set (Plate
2). One of the values of the Pathfinder processing was that careful attention was given
to sensor calibration (James and Kullari 1994). The visible/near-infrared
measurements reported in the Pathfinder record are based on calibrated planetary
reflectance. Such calibrations eliminate the varying effects of sensor drift and
incident radiation, leaving only variations that occur as a result of atmospheric
attenuation and changes in surface reflectance (Jasinski and Eagleson 1990). We
chose to compute a six-month minimum brightness to reduce the variable effects of
aerosol and cloud attenuation that occur within individual daily views of the earth.
For each hemisphere, we processed the summer six-month period. This was done
under the assumption that the lowest reflectance observation from this six-month
period represents both the clearest atmosphere and the largest canopy (i.e., biomass)
light attenuation that occurred during this summer growing season.
Comparison of these reflectance data with known standing biomass figures
produces a highly nonlinear but coherent relationship (Fig. 7). The results suggest a
near-linear relationship up to approximately 60 kg m2 and then saturation with little
change in reflectance with increasing biomass. For many forests, as reported in field
studies, this saturation point coincides with a stand age of approximately 30 years
257
(a)
(hI
Figure 6. (a) Global patterns of air temperature inferred from the NDVIITs
relationship described in Figure 5. (Gray tone key: white, no solution; light gray,
> 315K; dark gray, = 300K; black, < 290K). (b) Current surface soil moisture
conditions were inferred from the slope of the NDVIIT s relationship.
Observations were made during the period 1-10 June 1987, based on AVHRR
8-km NOAAINASA Pathfinder data. (Gray tone key: white, wet; light gray,
moist; dark gray, wet; black, wet or no value). The large areas of "no value" in
deserts are due to the fact that no solution is possible in this setting, i.e., there is
no vegetation. The splotchy pattern of no value in the soil moisture analysis
occurs where the observed solution is outside the expected range.
,.....
':' 80
S a.
bO 70 0
e Ul 60
en
til
50
....S
~
0
40
y = 184.14 x 2 I
-----J
"0
s:: 30
::l
0
~ 20
,
III
> 10
0
~ 0
0 2 4 6 8 10 12
ear- urface Vi ible Refiectance (%)
,.....
':'
S 80
bO
b.
e 70
0
Ul
en 60 0
0
til
S 50 0 0
.
T..
P5
0
40 ()
0
I y = 716.61 X'
26
j .:l...
"0
s:: 30
::l
0
~ 20 0
d>
>
0 10
.a
<: -'- , , 10
5 10 15 20 25 30
At- atellite Vi ible Reflectance (%)
Figure 7. Comparison of field site biomass and visible reflectance: (a) near-
surface observations from western Oregon OTTER sites (Runyon et al. 1994,
AC observations from R. McCrieght); (b) global-scale patterns from ground
measurements compiled by Box (1988) and Box et al. (1989), compared with
minimum brightness values for the reported locations in Plate 2. The OTTER
observations are highly nonlinear. A similar nonlinear pattern may be fitted to
the global-scale measurements. The differences in the reflectance range
(1.8-11 % for the OTTER site vs 4-13% for global-scale measurements) are
probably the result of atmospheric light scatter, which is strongly apparent in
AVHRR observations but little evident in the low-altitude observations from
OTTER.
259
(Nilson and Peterson 1994). The patterns revealed in this image suggest a good
approximation of the expected distribution of global-scale standing biomass values.
The nonlinearity of the derived relationship presents interpretation problems across
scales; however, at least in maintenance respiration (MR) assessments obtained from
these values, it appears that saturation occurs at about the same biomass values found
here (Hunt 1994). More bothersome is the low accuracy of observed patterns in low-
biomass areas due to the variable brightness of background soils, as noted earlier.
Estimation of the woody biomass content of vegetation canopies at global scales is
clearly still at an early stage of development, but marked improvement may be
possible, for example, through the use of both microwave and optical measurements.
Another possible alternative is to increase the number of optical wavelengths used in
the analysis. In any case, the basic potential of biomass estimation is apparent in this
simple analysis, which demonstrates the considerable depth of information available
in even a relatively simple observation set such as that provided by the AVHRR
sensor.
Photosynthetic pathways
The significant difference in dry matter yield of absorbed PAR (i.e., production
efficiency) for vegetation of differing photosynthetic pathways strongly suggests the
need to differentiate among various vegetation types. This is particularly true where
seasonal shifts from C3 to C4 plants occur. In fact, the environmental information
discussed earlier may provide the insight needed to make this determination (Terri
and Stone 1976, Ehleringer and Bjorkman 1977, Ehleringer 1978, Eickmeier 1978,
Long 1983). On the other hand, it might be particularly useful to determine these
patterns independently of the environmental information. Some hope of
accomplishing this has been generated by detailed studies of the short-wave infrared
(SWIR-1.2 to 2.4 Jlll1) reflectance patterns of agricultural crops (Goward 1985). Not
enough information on the optical properties of various vegetation types in SWIR
wavelengths is available to substantiate this possibility. However, the current
availability of SWIR in Landsat TM observations and the potential availability of
these measurements in a future AVHRR-type sensor encourage the expectation that
such assessments may soon be possible at global scales.
Future prospects
Over the next decade, nations throughout the world are planning to launch over 80
satellite earth-observing missions carrying a total of more than 200 sensors
(Committee on Earth Observation Satellites 1995). Many of the observations
obtained from these missions will be of immediate relevance to biospheric
monitoring, including RADARSAT (Canada), ADEOS and ADEO (Japan) and
ENVISAT (USA). There will also be several AVHRR-heritage instruments, which
will offer additional capabilities not available with the current AVHRR sensor; these
include VEGETATION (France), VIRSR (USA-NOAA) and MODIS (USA-NASA).
A low-spatial-resolution version of the SPOT solid-state sensor system known as
VEGETATION is in preparation for a proposed launch in 1997 (Saint 1994). This
sensor would provide the same temporal coverage and spatial resolution as the
AVHRR sensor, but would use a considerably improved radiometer with 12-bit
precision and spectral measurements in the blue (0.43-0.47 J.lII1), red (0.61-0.68 J.lII1),
NIR (0.78-0.89 11m) and SWIR (1.58-1.75 11m) portions of the spectrum. The
improved radiometry would refine the precision of derived variables such as f par The
restricted NIR spectral range would avoid the greatest problems with water vapor
attenuation, and the combination of blue and red wavelengths would provide a
potential to estimate aerosol attenuation effects (Kaufman and Tanre 1992). These
same wavelengths might improve assessment of seasonal variations in foliage
pigment as well as allow refined assessments of variables such as standing biomass
to be made, since it would be more possible to distinguish between background
materials and pigmented foliage. The SWIR measurements would provide a first
opportunity to attempt to discriminate C3 and C4 vegetation types at global scales.
The use of solid-state sensors also has the advantage that pixel sizes would be less
variable across an image. Thus, collection of "true" 1-krn observations would occur
(i.e., they would a1way~ be near 1 km).
NOAA is planning to launch a new version of the AVHRRsensor, VIRSR (Visible
Infrared Scanning Radiometer), by the year 2000 (J.D. Tarpley, personal communication).
The radiometer would be improved to 16-bit precision, a new SWIR spectral band
(1.65-1.75 11m) would be added and the near-infrared spectral band would be
restricted to the 0.72-0.90 region to avoid the attenuation effects of ozone and water
vapor in the 0.9-1.1 11m spectral region. Measurements obtained using this sensor
would be similar to but of a higher precision than those from the VEGETATION
sensor and would also include thermal infrared measurements. It could be quite
interesting to combine the 10:30 a.m. VEGETATION reflectance measurements with
those of the VIRSR 2:30 p.m. TIR.
By far the most ambitious sensor planned for the near future for global-scale
vegetation research is the Earth Observing System (EOS) MODIS (Moderate
Resolution Imaging Spectrometer) (Asrar and Dokken 1993, Running et al. 1994).
This instrument would be a highly sophisticated follow-on to the AVHRR
observatory. It would include 36 spectral wavelength observations, some 20 of which
would be used for terrestrial environmental monitoring. Radiometric data would be
captured with 16-bit precision and two spatial resolutions, 1 krn and 250 m, would be
acquired simultaneously for visible, red and near-infrared wavelengths. The spectral
coverage would include 20 wavelengths in the solar spectral region (0.4-3.0 11m) and
16 wavelengths in the TIR (3.0-15.0 11m) spectral region.
The MODIS sensor has been designed to significantly refine the suite of
measurements noted previously, as well as to enable the measurement of more
sophisticated variables. For example, six of the TIR spectral wavelengths are directed
toward separating the variable influence of atmospheric water vapor and cirrus ice
crystals from measurements of surface emissivity and kinetic temperature. Thus,
multiple remotely sensed measurements could provide the multiple degrees of
freedom needed to separate the signal sources. Similar multifactor perspectives are
the inspiration for MODIS sensor design in other wavelength regions. The MODIS
instrument would fly in concert with several other sensors, notably MISER and
261
ASTER, whose respective missions would be to conduct atmospheric analyses and

high-spatial-resolution Landsat SPOT observations in spectral regions similar to
those of the MODIS instrument.
Summary and conclusions

Satellite-based remotely sensed observations provide a unique new capacity to
inspect and monitor the global-scale patterns and seasonal dynamics of the terrestrial
biosphere. Within the context of the PEM concept and with the use of the AVHRR
and TOMS sensors, we have extracted most of the variables needed to monitor the
seasonal dynamics and interannual variability of terrestrial primary production. The
data collected using these two sensors represent at least 10 years of coincident
observations (1981-1990). These data provide a first step toward understanding the
bioclimatology of biospheric activity at global scales.
While we recognize the potential for using observations from the many other
earth-observing sensors either currently operating or planned for the near future, we
have not yet addressed this complex multiple observation analysis problem. The very
qualities that make these observations so attractive for biospheric research, their
spatial and temporal resolution, result in the production of large volumes of data. For
example, a l2-yr compilation of processed AVHRR Pathfinder data has produced a
file of over 100 gigabytes (I gigabyte = 1000 megabytes). Currently available
computers are barely able to handle the data from a single sensor. Moving to multiple
sensors is nearly impossible. Although computer technology continues to advance
rapidly, earth scientists interested in working with satellite remote sensing data must
understand the magnitude of the data handling problems associated with this work.
These comments, relative to the AVHRR and TOMS sensors, serve only to introduce
problems that will be encountered with VEGETATION, AMIR and most
emphatically EOS. Data rates of gigabytes per second are expected to flow from the
Earth Observing System. Highly creative technical and scientific approaches to
processing and analyzing this massive amount of information must be developed to
enable these earth-observing missions to succeed.
In addition to the technical aspects of remote sensing and data processing, there is
still much research needed to capture the value of such observations. At the regional
to global scales at which satellite-based terrestrial sensors typically operate, our
understanding of biospheric patterns and processes is still poor. Most current
vegetation research is concentrated at the leaf, plant and local community level.
Repetitive satellite observations, as shown here, describe what occurs at regional to
global scales. Clearly, structures and processes observed in the laboratory and at the
Held site must be combined to explain observed regional and global patterns. Local
phenomena, in turn, must operate in the context of what is observed at the coarser
scale. Somehow the step across scales must be made if we are to understand the links
between local field studies and satellite observations. The problems presented by
geographical scale cannot be underestimated (Rosswall et al. 1988). Efforts to
address geographical scaling carried out in field programs such as FIFE, OTTER,
HAPEX and BOREAS represent a first attempt to link satellite remote sensing
measurements with known surface phenomena (Sellers et al. 1988, Prince et al. 1995,
Waring and Peterson 1994, Sellers et al. 1995). This scaling work must be given
greater emphasis.
Challenges also exist in translating observed global-scale patterns and processes
into principles and concepts that describe how regional ecosystems and global
biomes are defined by, interact with and determine the earth's environmental
systems. For vegetation scientists nonplused by the prospect of using remote sensing
technology, the intellectual challenges presented by satellite observations are still
accessible. Well-posed questions are the only means of achieving progress in our
understanding of these observations. The potential for global-scale climate and
environmental change mandates a more thorough understanding of global-scale
biospheric dynamics as a major factor in the earth's environmental system. Satellite
remote sensing, even in its current undeveloped form, provides a significant
contribution to this intellectual challenge.
Acknowledgments - The work presented in this chapter has been supported in part by
NASA Grants NAG 9-503 (FIFE) and NAGW 1152 (HQ-TE), and by National
Science Foundation Grant BSR-8905278. D. Dye's contributions while at the
University of Tokyo were supported by a research grant from the Toyota Motor
Corporation of Japan. The AVHRR 8-km data set was supplied by the NOAAlNASA
AVHRR Land Pathfinder Project. The GVI 15-km observations were supplied by
Dan Tarpley of NOAA. Dr. Stephen Prince, Mr. Scott Turner and Ms. Shelley
Thawley provided significant assistance in preparation of the data sets and images.
References
Aber, J.D. 1992. Terrestrial ecosystems. - In: Trenberth, K.E. (ed). Climate Systems Modeling.
Cambridge University Press, Cambridge, UK, pp. 173-200.
Allen, W.A. and Richardson, AJ. 1968. Interaction of light with a plant canopy. - Appl. Opt.
58: 372-376.
Asrar, G. (ed). 1989. Theory and Applications of Optical Remote Sensing. - J. Wiley and Sons,
New York. 734 pp.
Asrar, G. and Dokken, OJ. 1993. 1993 EOS Reference Handbook. - NASA Earth Science
Support Office, Goddard Space Flight Center, NP-202, Greenbelt, MD. 145 pp.
Asrar, G., Fuchs, M., Kanemasu, E.T. and Hatfield, J.L. 1984. Estimating absorbed
photosynthetic radiation and leaf area index from spectral reflectance in wheat. - Agron. J.
76: 300-306.
Aston, A.R. and van Bavel, C. H.M. 1972. Soil surface water depletion and leaf temperature. -
Agron. J. 64: 368-373.
Badhwar, G.D. 1984. Automatic corn-soybeans feature extraction and classification. - Rem.
Baret, F. and Guyot, G. 1991. Potential and limits of vegetation indices for LAI and APAR
assessment. - Rem. Sens. Environ. 35: 161-173.
Bauer, M.E., Daughtry, C.ST. and Vanderbilt, vc. 1981. Spectral-agronomic relations of
corn, soybeans and wheat canopies. - LARS Purdue University, Technical Report 091281,
West Lafayette, IN. 204 pp.
263
Billings, W.D. and Morris, RJ. 1951. Reflection of visible and infrared radiation from leaves
of different ecological groups. - Am. J. Bot. 38: 327-331.
Blair. B.O. and Baumgardner. M.E 1977. Detection of green and brown wave in hardwood
canopy covers using multi-date. multispectral data from Landsat-I. - Agron. J. 69:
808-811.
Bolin, B. 1977. Changes of land biota and their importance for the carbon cycle. - Science 196:
613-615.
Box, E.O. 1988. Estimating the seasonal carbon source-sink geography of a natural,
steady-state terrestrial biosphere. - J. Appl. Meteorol. 27: 1109-1123.
Box. E.O.. Holben. B.N. and Kalb. V. 1989. Accuracy of the AVHRR vegetation index as a
predictor of biomass, primary productivity and net CO 2 flux. - Vegetatio 80: 71-89.
Budyko. M.1. 1974. Climate and Life. - Academic Press. New York. 508 pp.
Budyko. M.1. 1980. Global Ecology. - Progress Publishers, Moscow. 322 pp.
Carlson, T.N. and Buffum, MJ. 1989. On estimating total daily evaporation from remote
surface temperature measurements. - Rem. Sens. Environ. 29: 197-207.
Carlson. T.N., Perry, E.M. and Schmugge, TJ. 1990. Remote estimation of soil moisture
availability and fractional vegetation cover for agricultural fields. - Agric. For. Meteorol.
52: 45-69.
Carlson. T.N., Gilles, R.R. and Perry, E.M. 1994. A method to make use of thermal infrared
temperature and NDVI measurements to infer surface soil water content and fractional
vegetation cover. - Rem. Sens. Rev. 9: 161-173.
Choudhury. BJ. 1986. An analysis of observed linear correlation between net photosynthesis
and a canopy-temperature-based plant water stress index. - Agric. For. Meteorol. 36:
323-333.
Choudhury. BJ. 1989. Monitoring global land surface using NIMBUS-7 37 GHz polarization
difference. -Int. J. Rem. Sens. 10: 1579-1605.
Coakley. J.A.. Jr. and Bretherton, EP. 1982. Cloud cover from high-resolution scanner data:
Detecting and allowing for partially filled fields of view. - J. Geophys. Res. 87(C7):
4917-4932.
Colwell. J.E. 1974. Vegetation canopy ret1ectance. - Rem. Sens. Environ. 3: 174-183.
Colwell, R.N. 1956. Determining the prevalence of certain cereal crop diseases by means of
aerial photography. - Hildgardia 26(5): 223-286.
Committee on Earth Observation Satellites. 1995. Coordination for the next decade. -In: 1995
CEOS Yearbook. 133 pp.
Curran. PJ. 1981. Multispectral remote sensing for estimating biomass and productivity. - In:
Smith. H. (cd). Plants in the Daylight Spectrum. Academic Press. London, pp. 65-99.
Curran. PJ. 1983. Multispectral remote sensing for estimation of green leaf area index. - Phil.
Trans. Roy. Soc. London (Series A) 309: 257-270.
Curran, PJ. and Milton, EJ. 1983. The relationship between the chlorophyll concentration.
LAI and ret1ectance of a simple vegetation canopy. -Int. J. Rem. Sens. 4(2): 247-255.
Cushnie. J. 1988. The acquisition of SPOT-I HRV imagery over southern Britain and northern
France, May 1986-May 1987. -Int. J. Rem. Sens. 9: 159-167.
Dalu. G. 1986. Satellite remote sensing of atmospheric water vapor. - Int. J. Rem. Sens. 7:
1089-1097.
Darnell. W.L.. Staylor. W.E. Gupta, S.K.. Ritchey. N.A. and Wilber. A.C. 1992. Seasonal
variation of surface radiation budget from International Satellite Cloud Climatology Project
CI data. - J. Geophys. Res. 97(014): 15741-15760.
Daughtry, e.ST., Gallo, K.D. and Bauer, M.E. 1982. Spectral estimates of solar radiation
intercepted by corn canopies. - Purdue University, AgRISTARS Technical Report
sr-PZ-04236, West Lafayette, IN. 82 pp.
Daughtry, e.S.T., Gallo, K.D., Goward, S.N., Prince, S.D. and Kustas, WP. 1992. Spectral
estimates of absorbed radiation and phytomass in plant canopies. - Rem. Sens. Environ. 39:
141-152.
Deering, D.W, Rouse, J.J.W, Haas, R.H. and Schell, 1.S. 1975. Measuring forage production
of grazing units from Landsat MSS data. - In: Proceedings of the Tenth International
Symposium on Remote Sensing of the Environment, Environmental Research Institute of
Michigan, Ann Arbor, MI, pp. 1169-1178.
DeFries, R.S. and Townshend, 1.R.G. 1994. Global land cover: Comparison of ground-based
data sets to classification with AVHRR data. - In: Foody, G. and Curran, P. (eds).
Environmental Remote Sensing from Regional to Global Scales. 1. Wiley and Sons, New
York, pp. 84-110.
Dethier, B.E. 1974. Phenology Satellite Experiment, Final Report. - Cornell University,
Contract NAS 5-21781, Ithaca, NY. 758 pp.
Dobson, M.e., Ulaby, ET. and LeToan, T. 1992. Dependence of RADAR backscatter on
conifer forest biomass.- IEEE Trans. Geosci. Rem. Sens. 30: 412-431.
Duvigneaud, P., Bolin, B., Degens, E.T. and Kempe, S. 1979. The global bio-geo-chemical
carbon cycle. - In: Bolin, B., Degens, E.T., Kempe, S. and Ketner, P. (eds). The Global
Carbon Cycle. 1. Wiley and Sons, New York, pp. I-56.
Dye, D.G. and Goward, S.N. 1993. Photosynthetically active radiation absorbed by global land
vegetation during August 1984. - Int. 1. Rem. Sens. 14: 3361-3364.
Eck, T.E. and Dye, D.G. 1991. Satellite estimation of incident photosynthetically active
radiation using ultraviolet reflectance. - Rem. Sens. Environ. 38: 135-146.
Eck, T.E., Bhartia, E, Hwang, P.K. and Stowe, L.L. 1987. Reflectivity of the Earth's surface
and clouds in ultraviolet wavelengths from satellite observations. - 1. Geophys. Res. 92:
4287-4296.
Ehleringer,l.R. 1978. Implications of quantum yield differences on the distributions of CJ and
C4 grasses. - Oecologia 31: 255-267.
Ehleringer, 1.R. and Bjorkman, O. 1977. Quantum yields for CO2 uptake in CJ and C4 plants.-
Plant Physiol. 59: 86-90.
Ehleringer, 1.R. and Mooney, H.A. 1978. Leaf hairs: Effects on physiological activity and
adaptive value to a desert shrub. - Oecologia 37: 183-200.
Ehleringer, 1.R. and Werk, K.S. 1986. Modifications of solar-radiation absorption patterns and
implications for carbon gain at the leaf level. - In: Givnish, T.l. (ed). On the Economy of
Plant Form and Function. Cambridge University Press, Cambridge, UK, pp. 57-80.
Ehleringer, 1.R. and Field, e.B. (eds). 1993. Scaling Ecological Processes: Leaf to Globe. -
Academic Press, San Diego, CA. 388 pp.
Ehrlich, P.R. and Roughgarden, J. 1987. The Science of Ecology. - MacMillan, New York. 710
pp.
Eickmeier, WG. 1978. Photosynthetic pathway distributions along an aridity gradient in Big
Bend National Park, and implications for enhanced resource partitioning. - Photosynthetica
12: 290-297.
Field, C.B. 1991. Ecological scaling of carbon gain to stress and resource availability. - In:
Mooney, H.A., Winner, W.E. and Pell, E.J. (eds). Responses of Plants to Multiple Stresses.
Academic Press, New York, pp. 35-65.
Foran, B.D. 1987. Detection of yearly cover change with Landsat MSS on pastoral landscapes
in Central Australia. - Rem. Sens. Environ. 23: 333-350.
265
Forseth. LN. and Ehleringer. J.R. 1983. Ecophysiology of two solar tracking desert annuals 4.
Effects of leaf orientation on calculated daily carbon gain and water usc emciency. -
Oecologia 58: 10-18.
Freden. S.e. and Gorden. E. Jr. 1983. Landsat Satellites. - In: Colwell. R.N. (cd). Manual of
Remote Sensing. 2d ed. American Society of Photogrammetry. Falls Church. VA. pp.
517-570.
Gates. D.M. 1980. Biophysical Ecology. - Springer-Verlag. New York. 611 pp.
Gates. D.M . Keegan. H.J. Schleter. J.e. and Weidner. Y.R. 1965. Spectral properties of plants.
- AppL Opt. 4: 11-20.
Geiger, R. 1965. The Climate Near the Ground. - Harvard University Press. Cambridge. MA.
611 pp.
Goward. S.N. 1985. Shortwave infrared detection of vegetation. - Adv. Space Res. 5(5):
41-50.
Goward. S.N. 1989. Satel1ite bioclimatology. - 1. Clim. 7(2): 710-720.
Goward. S.N. and Oliver, J.E. 1977. The Application of Remote Sensing Techniques in
Microscale Climatology. - Indiana Academy of Science, Indianapolis. IN. pp. 326-337.
Goward. S.N. and Dye. D.G. 1987. Evaluating North American net primary productivity with
satel1ite observations. - Adv. Space Res. 7( 11): 165-174.
Goward. S.N. and Hope, A.S. 1989. Evapotranspiration from combined reflected solar and
emitted terrestrial radiation: Preliminary FIFE results from AVHRR data. - Adv. Space Res.
9(7): 239-249.
Goward. S.N. and Kazantsev. N.N. 1991. Monitoring environmental change: Geographic
information systems and remote sensing. - In: Mather. J.R. and Sdasyuk, G.Y. (cds). Global
Change: Geographical Approaches. University of Arizona Press. Tucson. AZ. pp. 183-198.
Goward. S.N. and Huemmrich. K.E 1992. Vegetation canopy PAR absorptance and the
normalized difference vegetation index: An assessment using the SAIL mode\. - Rem.
Goward. S.N .. Cruickshanks. G.e. and Hope. A.S. 1985a. Observed relation between thermal
emissions and reflected spectral rellectance from a complex vegetated landscape. - Rem.
Goward. S.N .. Tucker. e.J. and Dye. D.G. 1985b. North American vegetation patterns
observed with the NOAA-7 advanced very high resolution radiometer. - Vegetatio 64:
3-14.
Goward. S.N.. Kerber. A. Dye. D.G. and Kalb. V. 1987. Comparison of North and South
American biomes from AVHRR observations. - Geocarto 2( I): 27-40.
Goward. S.N. Markham. B.. Dye. D.G.. Dulaney. W. and Yang, J. 1991. Normalized difference
vegetation index measurements from advanced very high resolution radiometer. - Rem.
Sens. Environ. 35(2,3): 259-279.
Goward. S.N.. Huemmrich, K.E and Waring. R.H. 1994a. Visible-near infrared spectral
reflectance properties of landscape components in western Oregon. - Rem. Sens. Environ.
4(2): 322-343.
Goward. S.N.. Waring. R.H .. Dye. D.G. and Yang. J. I994b. Ecological remote sensing at
OTTER: Macroscalc satellite observations. - Eco\. App\. 4(2): 322-343.
Graetl. R.D. and Gentle. R.D. 1982. The relationships between reflectance in the Landsat
wavebands and the composition of an Australian semi-shrub rangeland. - Photogramm.
Engin. Rem. Sens. 48: 1721-1730.
Grigg. D. 1987. The industrial revolution and land transformation. - In: Wolman. M.G. and
Fournier. EG.A. (cds). Land Transformation in Agriculture. J. Wiley and Sons. New York.
pp.79-109.
Gurney, R.J., Foster, J.L. and Parkinson, c.L. (eds). 1993. Atlas of Satellite Observations
Related to Global Change. - Cambridge University Press, Cambridge, UK. 470 pp.
Gutman, G., Tarpley, D. and Ohring, G. 1987. Cloud screening for determination of land
surface characteristics in reduced resolution satellite data set. - Int. J. Rem. Sens. 8(6):
859-870.
Hansen, J., Johnson, D., Lacis, A., Lebedeff, S., Lee, P., Rind, D. and Russell, G. 1981.
Climatic impact of increasing atmospheric carbon dioxide. - Science 213: 957-966.
Harris, A.R. and Mason, I.M. 1992. An extension of the split-window technique giving
improved atmospheric correction and total water vapor. - Int. J. Rem. Sens. 13(5):
881-892.
Hatfield, J.L., Asrar, G. and Kanemasu, E.T 1984. Intercepted photosynthetically active
radiation estimated by spectral reflectance. - Rem. Sens. Environ. 14: 65-75.
Heilman, J.L., Kanemasu, E.T, Rosenberg, N.J. and Blad, B.L. 1976. Thermal scanner
measurement of canopy temperatures to estimate evapotranspiration. - Rem. Sens.
Environ. 5: 127-145.
Heimann, M. and Keeling, C.D. 1989. A three dimensional model of atmospheric CO 2
transport based on observed winds: 2. Model description and simulated tracer experiments.
- Geophy. Monogr. 55: 237-275.
Hellden, U. 1987. An assessment of woody biomass, community forests, landuse and soil
erosion in Ethiopia. A feasibility study on the use of remote sensing and GIS-analysis for
planning purposes in developing countries. - University of Lund, Department of
Geography, Lund, Sweden. 75 pp.
Holben, B.N. 1986. Characteristics of maximum-value composite images from temporal
AYHRR data. - Int. J. Rem. Sens. 7: 1417-1434.
Holben, B.N., Tucker, c.J. and Fan, C.I. 1980. Assessing soybean leaf area and leaf biomass
with spectral data. - Photogramm. Engin. 46: 651-656.
Hope, A.S. and McDowell, TP. 1992. The relationship between surface temperature and a
spectral vegetation index of a tallgrass prairie: Effects of burning and other landscape
controls. - Int. J. Rem. Sens. 13(15): 2849-2863.
Hope, A.S., Petzold, D.E., Goward, S.N. and Ragan, R.M. 1987. Simulating canopy
reflectance and thermal infrared emissions for estimating evapotranspiration. - Water Res.
Bull. 22(6): 1011-1019.
Houghton, J.T, Jenkins, G.I. and Ephraums, J.J. 1990. Climate Change: The IPCC Scientific
Assessment. - Cambridge University Press, Cambridge, UK.
Huete, A.R. 1988. A soil-adjusted vegetation index (SAYI). - Rem. Sens. Environ. 25:
295-309.
Huete, A.R. and Tucker, C.I. 1991. Investigation of soil influences in AYHRR imagery. - Int.
J. Rem. Sens. 12(6): 1223-1242.
Huete, A.R., Jackson, R.D. and Post, D.E 1986. Spectral response of a plant canopy with
different soil backgrounds. - Rem. Sens. Environ. 17: 37-53.
Hunt, E.R., Jr. 1994. Relationship between woody biomass and PAR conversion efficiency for
estimating net primary production from NDYI. - Int. J. Rem. Sens. 15: 1725-1730.
Hussin, YA., Reich, R.M. and Hoffer, R.M. 1991. Estimating slash pine biomass using
RADAR backscatter. - IEEE Trans. Geosci. Rem. Sens. 29: 427-438.
Ives, R.L. 1939. Infra-red photography as an aid in ecological surveys. - Ecology 20(3):
433-439.
Jackson, R.D. 1983. Spectral indices in n-space. - Rem. Sens. Environ. 13: 1401-1429.
Jackson, R.D., Idso, S.B., Reginato, R.I. and Pinter, P.I. 1981. Canopy temperature as a crop
stress indicator. - Water Res. Res. 4: 1133-1138.
267
James. MJ. and Kullari, S. 1994. The Pathfinder AVHRR land data set: An improved coar~e
resolution data set for terrestrial monitoring. - Int. J. Rem. Sens. 15( 17): 3347-.'364.
Jarvis. P.G. and Leverenz, J.W. 1983. Productivity of temperate. deciduous and evergreen
forests. - In: Lange, O.L., Nobel, P.S., Osmond, CB. and Ziegler, H. (cds). Physiological
Plant Ecology IV. Springer-Verlag, New York, pp. 223-280.
Jasinski, M.F. and Eagleson, P.S. 1990. Estimation of subpixel vegetation cover using
red-infrared scattergrams. - IEEE Trans. GeoseL Rem. Sens. 28(2): 253-267.
Jayroe, R.R., Jr. 1978. Some observations about Landsat digital analysis. - NASA. Marshall
Space Flight Center, NASA Technical Memo 78184, Huntsville, AL. 35 pp.
Jedlovec, GJ. 1990. Precipitable water estimation from high-resolution split window radiance
measurements. - J. App!. Meteoro!. 29: 863-877.
Jordan, CF. 1969. Derivation of leaf area index from quality of light on the forest 1100r. -
Ecology 50(4): 663-666.
Justice, CO., Townshend, J.R.G., Holben, B.N. and Tucker, CJ. 1985. Analy~is of the
phenology of global vegetation using meteorological satellite data. - Int. J. Rem. Sens.
6(8): 1271-1381.
Justice, CO., Eck, T., Tanre, D. and Holben, B.N. 1991. The effect of water vapor on the
normalized difference vegetation index derived for the Sahelian Region from NOAA
AVHRR data. - Int. J. Rem. Sens. 12(6): 1165-1188.
Kaufman, Y. and Tanre, D. 1992. Atmospherically Resistant Vegetation Index (ARVI). -IEEE
Trans. GeoscL Rem. Sens. 30(2): 261-270.
Kauth, R.J. and Thomas, G.S. 1976. The tasseled cap - a graphic description of the
spectral-temporal development of agricultural crops as seen by Landsat. - In: Proceedings
of the Tenth Symposium on Machine Processing of Remotely Sensed Data, Purdue
University, We~t Lafayette. IN, pp. 41-51.
Kettig, R.L. and Landgrebe, D.A. 1976. Classification of spectral image data by extraction and
classification of homogeneous ohjects. - IEEE Trans. GeoscL Electr. 14: 19-25.
Kimes, D.S. and Kirchner, J.A. 1982. Radiative transfer model for heterogeneous 3-D scenes.
-App!. Opt. 21: 4119-1438.
Klcespies, TJ. and McMillin, L.M. 1990. Retrieval of precipitahle water from ohservations on
split window over varying surface temperatures. - J. App!. Meteoro!. 29: 851-862.
Koomanoff, V.A. 1989. Analysis of glohal vegetation patterns: A comparison between
remotely sensed data and a conventional map. - University of Maryland, Department of
Geography, Biogeography Research Series Report 890201. College Park, MD. III pp.
Krinov, E.L. 1947. Spectral Renectance of Natural Formations (trans!. by NEC of Canada,
T1439, G. Belkov). - Akad. Nank, USSR. Laboratorica Aerometodov, Moscow. 271 pp.
Kumar, M. and Monteith, J.L. 1981. Remote sensing of crop growth. - In: Smith, H. (cd).
Plants in the Daylight Spectrum. Academic Press, New York, pp. 134-144.
Landsberg, J.J. 1986. Physiological Ecology of Forest Production. - Academic Press, London.
198 pp.
Landsberg, J.J., Prince, S.D., Jarvis, P.G., McMurtrie. R.E., Luxmoore, R. and Medlyn, B.E.
1996. Energy conver~ion and usc in forests: An analysis of forest production in terms of
radiation utilisation erticiency (f). - In: Gholz. H.L., Nakane. K. and Shimoda, H. (ed~).
The Use of Remote Sensing in the Modeling of Forest Productivity. Kluwer Academic
Puhlishers. Dordrecht. The Netherlands. pp. 273-298.
Li. X. and Strahler, A.H. 1985. Geometric-optical modeling of a conifer fore~t canopy. - IEEE
Tran~. Geosci. Rem. Sens. GE-23(5): 705-721.
LOf, G.O.G., Duffie, J.A. and Smith, e.O. 1966. World distribution of solar radiation. -
University of Wisconsin, Solar Energy Lab, Engineering Experiment Stations, Report 21,
Madison, WI. 70 pp.
Long, S.P. 1983. C4 photosynthesis at low temperatures. - Plant Cell Environ. 6: 345-363.
Major, DJ., Baret, F. and Guyot, G. 1990. A ratio vegetation index adjusted for soil brightness.
- Int. J. Rem. Sens. 11: 727-740.
McClain, E.P., Pichel, WG. and Walton, e.e. 1985. Comparative performance of
AVHRR-based multichannel sea surface temperatures. - 1. Geophys. Res. 90:
11587-11601.
McDonald, RB. and Hall, F.G. 1980. Global crop forecasting. - Science 208: 670-679.
Monteith, J.L. 1977. Climate and efficiency of crop production in Britain. - Phil. Trans. Roy.
Soc. London (Series B) 281: 277-294.
Monteith, J.L. and Unsworth, M.H. 1990. Principles of Environmental Physics. 2d ed. -
Edward Arnold, London. 291 pp.
Mooney, H.A. and Field, e.B. 1988. Photosynthesis and plant productivity - scaling to the
biosphere. - In: Proceedings of Photosynthesis: e.S. French Symposium, Stanford, CA.
Alan R Liss, Inc., New York, pp. 19-44.
Mooney, H.A. and Hobbs" R (eds). 1990. Remote Sensing of Biospheric Functioning. -
Springer-Verlag, New York. 312 pp.
National Research Council. 1986. Remote Sensing of the Biosphere. - Committee on
Planetary Biology, Space Science Board National Academy Press, Washington, De. 91 pp.
Nemani, R.R. and Running, S.W. 1989. Estimation of surface resistance to evapotranspiration
from NDVI and thermal-IR AVHRR data. - J. Clim. Appl. Meteorol. 28: 276-294.
Nemani, R.R., Pierce, L.L., Running, S.W. and Goward, S.N. 1993. Developing
satellite-derived estimates of surface moisture status. - J. Appl. Meteorol. 32(3): 548-557.
driving factors. - Rem. Sens. Environ. 48: 319-331.
Njoku, E.G. 1982. Passive microwave remote sensing of the earth from space - a review. -
Proc. IEEE 70(7): 728-750.
Novoseltsev, YP. 1964. Spectral Reflectivity of Clouds (in Russian). Spectrol. Otrazhatel.
Oblakov, Tr. Gla: Geofiz. Obs. A.D. Voyykova: 186-197, NASA TT F-328(152).
Otterman, J. 1984. Albedo of a forest modeled as a plane with dense protrusions. - J. Clim.
Appl. Meteorol. 23: 297-307.
Peterson, D.L. 1996. Forest structure and productivity along the Oregon transect. - In: Gholz,
173-218.
Peterson, D.L. and Waring, R.H. 1994. Overview of Oregon transect ecosystem research
project. - Ecol. Appl. 4(2): 211-225.
Peterson, D.L., Spanner, M.A., Running, S.W and Teuber, K.B. 1987. Relationship of
Thematic Mapper Simulator data to leaf area index of temperate coniferous forest. - Rem.
Petzold, D.E. and Goward, S.N. 1988. Reflectance spectra of subarctic lichens. - Rem. Sens.
Environ. 24: 481-492.
Pinker, R.T. 1990. Satellites and our understanding of the surface energy balance. -
Palaeogeog. Palaeoclimatol. Palaeoecol. 82: 321-342.
269
Pinker, R.T. and Laszlo, I. 1992a. Global distribution of photosynthetically active radiation as
observed from satellites. - J. Clim. 5: 56-65.
Pinker, R.T. and Laszlo, I. 1992b. Modeling surface solar irradiance for satellite applications
on a global scale. - J. Appl. Meteorol. 31: 194-211.
Post, WM., Peng, T.-H., Emanuel, WR., King, A.W and Dale, V.H. 1990. The global carbon
cycle. -Am. Sci. 78: 310-326.
Potter, e.S., Randerson, J.T., Field, e.B., Matson, P.A., Vitousek, P.M., Mooney, H.A. and
Klooster, S.A. 1993. Terrestrial ecosystem production: A process model based on global
satellite and surface data. - Glob. Biogeochem. Cycl. 7(4): 811-841.
Prabhakara, e., Dalu, VG. and Kunde, VG. 1974. Estimation of sea surface temperature from
remote sensing in the 11-13 J.Im window range. - J. Geophys. Res. 79(33): 5039-5044.
Price, J.e. 1980. The potential of remotely sensed data to infer surface soil moisture and
evaporation. - Water Res. Res. 16: 787-795.
Price, J.e. 1989. Quantitative aspects of remote sensing in the thermal infrared. - In: Asrar, G.
(ed). Theory and Applications of Optical Remote Sensing. J. Wiley and Sons, New York,
pp. 578-603.
Price, J.e. 1990. Using spatial context in satellite data to infer regional scale evapotranspiration.
- IEEE Trans. Geosci. Rem. Sens.' 28(5): 940-948.
Prihodko, L. 1992. "Estimation of air temperature from remotely sensed observations." -
Master's thesis, University of Maryland, College Park, MD. 178 pp.
Prince, S.D. 1991. A model of regional primary production for use with coarse resolution
satellite data. - Int. J. Rem. Sens. 12(6): 1313-1330.
Prince, S.D. and Goward, S.N. 1996. Global net primary production: The remote sensing
approach. - J. Biogeog. Lett. (in press).
Prince, S.D., Justice, e.O. and Moore, III, B. (eds). 1994. - Monitoring and modelling of
terrestrial net and gross primary production. Joint IGBP-DIS-GAIM working paper I.
International Geosphere Biosphere Program Data & Information System, Paris. 56 pp.
Prince, S.D., Kerr, YH., Goutorbe, J.-P., Lebel, T., Tinga, A, Bessemoulin, P., Brouwer, 1.,
Dolman, AJ., Engman, E.T., Gash, J.H.e., Hoepffner, M., Kabat, P., Monteny, B., Said, F.,
Sellers, P.J. and Wallace, J. 1995. Geographical, biological, and remote sensing aspects of
the Hydrologic Atmospheric Pilot Experiment in the Sahel (HAPEX-Sahel). - Rem. Sens.
Environ. 51: 215-234.
Rasool, S.1. (ed). 1987. Potential of remote sensing for the study of global change. - Adv.
Space Res. 7(1): 1-90.
Rosenzweig, e. and Dickerson, R. 1986. Climate-vegetation interactions. - In: Proceedings of
a Workshop at NASA Goddard Space Flight Center OIES-2, Office of Interdisciplinary
Earth Studies, UCAR, Boulder, CO. 156 pp.
Rosswall, T., Woodmansee, R.G. and Risser, P.G. (eds). 1988. Scales and Global Change.
SCOPE 35. - J. Wiley and Sons, New York. 355 pp.
Ruimy, A, Dedieu, G. and Saugier, B. 1994. Methodology for the estimation of terrestrial net
primary productivity from remotely sensed data. - J. Geophys. Res. 99(D3): 5263-5283.
Running, S.W., Justice, e.O., Salomonson, V, Hall, D., Barker, J., Kaufman, J., Strahler, A.H.,
Huete, AR., Muller, J.-P., Vanderbilt, V, Wan, Z.M., Teillet, P. and Carneggie, D. 1994.
Terrestrial remote sensing science and algorithms planned for EOS MODIS. - Int. J. Rem.
Sens. 15(17): 3587-3620.
Runyon, J., Waring, R.H., Goward, S.N. and Welles, J.M. 1994. Environmental limits on
above-ground production: Observations from the Oregon transect. - Ecol. Appl. 4(2):
226-238.
Saint, G. 1994. VEGETATION Preparatory Programme: Call for Proposals. - European Joint
Research Center, Institute for Remote Sensing Applications, VGT/PS/9407211l. 75 pp.
Saunders, R.U. and Kriebel, K.T. 1988. An improved method for detecting clear sky and
cloudy radiances from AVHRR data. - Int. J. Rem. Sens. 9: 123-150.
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.
MeteoroI. Soc. 64: 779-784.
Schmetz, J. 1991. On the retrieval of surface radiation budget components from satellites. -
Palaeogeog. Palaeoclimatol. Palaeoecol. (Global and Planetary Change Section) 90: 17-24.
Schneider, S.R. and McGinnis, D.E, Jr. 1982. The NOAA AVHRR: A new sensor for
monitoring crop growth. - In: Proceedings of the Tenth Symposium on Machine Processing
of Remotely Sensed Data, Purdue University, West Lafayette, IN, pp. 211-223.
Scientific Committee on Problems of the Environment (ed). 1984. The Role of Terrestrial
Vegetation in the Global Carbon Cycle: Measurement with Remote Sensing. SCOPE
Report. - J. Wiley and Sons, New York. 247 pp.
Seguin, B. and Itier, B. 1983. Using midday surface temperature to estimate daily evaporation
from satellite thermal IR data. - Int. J. Rem. Sens. 4: 371-383.
Sellers, P.J. 1985. Canopy reflectance, photosynthesis, and transpiration. - Int. J. Rem. Sens.
6: 1335-1371.
Sellers, P.J. 1991. Modeling and observing land-surface-atmosphere interactions on large
scales. - Surv. Geophy. 12: 85-114.
Sellers, P.J., Hall, EG., Asrar, G., Strebel, D.E. and Murphy, R.E. 1988. The first ISLSCP field
experiment. - Bull. Am. MeteoroI. Soc. 69: 22-27.
Sellers, P.J., Hall, EG., Margolis, H., Kelly, B., Baldocchi, D., den Hartog, G., Chilar, J., Ryan,
M.G., Goodison, B., Crill, P, Ranson, J.K., Lettenmaier, D. and Wickland, D. 1995. The
Boreal Ecosystem-Atmosphere Study (BOREAS): An overview and early results from the
1994 field year. - Bull. Am. MeteoroI. Soc. 76(9): 1549-1577.
Sellers, W.O. 1965. Physical Climatology. - University of Chicago Press, Chicago. 272 pp.
Smith, J.A., Ranson, K.J., Nguyen, D. and Balick, L. 1985. Thermal vegetation canopy model
studies. - Rem. Sens. Environ. II: 31-26.
Smith, R.C.G. and Choudhury, B.J. 1991. Analysis of normalized difference and surface
temperature observations over southeastern Australia. - Int. J. Rem. Sens. 12(10):
2021-2044.
Spanner, M., Johnson, L., Miller, J., McCreight, R., Freemantle, J., Runyon, J. and Gong, P.
1994. Remote sensing of seasonal leaf area index across the Oregon transect. - Ecol. AppI.
4(2): 258-271.
Stanhill, G. 1970. Some results of helicopter measurements of the albedo of different land
surfaces. - Sol. Ener. 13: 59-66.
Stoner, E.R. and Baumgardner, M.E 1981. Characteristic variations in reflectance of surface
soils. - Soil Sci. Soc. Am. J. 45(6): 1161-1165.
Stowe, L.L., McClain, E.P, Carey, R., Pellegrino, P., Gutman, G., Davis, P., Long, C. and Hart,
S. 1991. Global distribution of cloud cover derived from NOAAIAVHRR operational
satellite data. - Adv. Space Res. 3: 51-54.
Suits, G.H. 1972. The calculation of the directional reflectance of a vegetative canopy. - Rem.
Susskind, 1., Rosenfield, J., Reuter, D. and Chahine, M. 1984. Remote sensing of weather and
climate parameters from HIRS2-MSU on TIROS-N. - 1. Geophys. Res. 89: 4677-4697.
Swain, P.H. and Davis, S.M. (eds). 1978. Remote Sensing: The Quantitative Approach. -
McGraw-Hill Book Company, New York. 396 pp.
271
Tame, D., Holben, B.N. and Kaufman, Y.J. 1992. Atmospheric correction algorithm for
NOAA-AVHRR products: Theory and application. - IEEE Trans. Geosci. Rem. Sens.
30(2): 231-248.
Tans, P.P., Fung, I.Y. and Takahashi, T. 1990. Observational constraints on the global
atmospheric CO2 budget. - Science 247: 1431-1438.
Tarpley, J.D. 1979. Estimating incident solar radiation at the surface from geostationary
satellite data. - J. Appl. Meteorol. 18: 1172-1181.
Terri, J.A. and Stone, L.G. 1976. Climatic patterns and the distribution of C4 grasses in North
America. - Oecologia 23: 1-12.
Thomas, WL, Jr. (ed). 1966. Man's Role in Changing the Face of the Earth. - University of
Chicago Press, Chicago. 1193 pp.
Tieszen, L.L. and Johnson, P.L. 1968. Pigment structure of some arctic tundra communities. -
Ecology 49(2): 370-373.
Townshend, J.R.G. and Tucker, e.J. 1981. Utility of AVHRR NOAA-6 and NOAA-7 for
vegetation mapping. - In: Matching Remote Sensing Technologies and Applications.
Remote Sensing Society, London, pp. 97-107.
Townshend, J.R.G. and Justice, e.O. 1986. Analysis of the dynamics of African vegetation
using the normalized difference v,egetation index. - Int. J. Rem. Sens. 7: 1189-1207.
Townshend, J.R.G. and Justice, e.O. 1990. The spatial variation of vegetation changes at very
coarse scales. - Int. J. Rem. Sens. 11: 149-157.
Townshend, J.R.G., Justice, e.O., Li, W, Gurney, e. and McManus, J. 1991. Global land cover
classification by remote sensing: Present capabilities and future possibilities. - Rem. Sens.
Environ. 35: 243-255.
Tucker, e.J. 1979. Red and photographic infrared linear combinations for monitoring
Tucker, C.J., Holben, B.N., Elgin, J.H., Jr. and McMurtrey, J.E. 1981. Remote sensing of total
dry-matter accumulation in winter wheat. - Rem. Sens. Environ. 11: 171-189.
Tucker, e.J., Townshend, J.R.G. and Goff, T.E. 1985. African land-cover classification using
satellite data. - Science 227: 369-375.
Turner, B.L., II. 1989. The human causes of global environmental change, - In: DeFries, R.S.
and Malone, T. (eds). Global Change and Our Common Future: Papers from a Forum.
National Academy Press, Washington, DC, pp. 90-99.
Turner, BL, II, Clark, We., Kates, R.W, Richards, J.F., Matthews, J.T. and Meyer, WB.
(eds). 1990. The Earth Transformed by Human Action: Global and Regional Changes in the
Biosphere of the Last 300 Years. - Cambridge University Press, Cambridge, UK. 713 pp.
Van Wijk, WR. and De Vries, D.A. 1963. Periodic temperature variations. - In: Van Wijk, WR.
(ed). Physics of Plant Environment. North-Holland Publishing Company, Amsterdam.
Verhoeff, W 1984. Light scattering by leaf layers with application to canopy reflectance
modeling: The SAIL model. - Rem. Sens. Environ. 16: 125-141.
Vitousek, P.M., Ehrlich, P.R., Ehrlich, A.H. and Matson, P.A. 1986. Human appropriation of
the products of photosynthesis. - Bioscience 36: 368-373.
Waring, R.H. and Peterson, D.L. 1994. The Oregon Transect Ecosystems Research (OTTER):
Summary and overview. - Ecol. Appl. 4(2): 211-225.
Watson, K. 1973. Periodic heating of a layer over a semi-infinite solid. - J. Geophys. Res. 78:
5904-5910.
Whittaker, R.H. and Likens, G.E. 1975. The biosphere and man. - In: Lieth, H. and Whittaker,
R. (eds). Primary Productivity of the Biosphere. Springer-Verlag, New York, pp. 306-328.
Williams, D.L. 1991. A comparison of spectral reflectance properties at the needle, branch and
canopy level for selected conifer species. - Rem. Sens. Environ. 35: 79-91.
Wolfe, w.L. (ed). 1965. Handbook of Military Infrared Technology. - Office of Naval
Research, Department of the Navy, Washington, DC. 906 pp.
Wooley, J.T. 1971. Reflectance and transmittance of light by leaves. - Plant Physiol. 47:
656-662.
Yates, H., Strong, A., McGinnis, D., Jf. and Tarpley, D. 1986. Terrestrial observations from
NOAA operational satellites. - Science 23 I: 463-470.
Yefimova, N.A. 197 I. Geographical distribution of the sums of photosynthetically active
radiation. - Soviet Geog. 17-66-74.
ELEVEN
Energy Conversion and Use in Forests: An Analysis of Forest

production in Terms of Radiation Utilisation Efficiency (E)
J.J. Landsberg, S.D. Prince, P.G. Jarvis, R.E. McMurtrie, R. Luxmoore and B.E.
Medlyn
Landsberg, J.J., Prince, S.D., Jarvis, P.G., McMurtrie, R.E., Luxmoore, R. and
Medlyn, B.E. 1996. Energy conversion and use in forests: An analysis of forest
production in terms of radiation utilisation efficiency (E). - In: Gholz, H.L., Nakane,
Productivity. Kluwer Acad. Pub!., Dordrecht, The Netherlands, pp. 273-298.
The linear relationship between the photosynthetically active solar radiation (PAR)
absorbed by forest canopies (APAR) and the production of dry mass by forests
provides a simple, robust model with only one parameter for the estimation of forest
production. The slope of the relationship is normally denoted E. The E model has been
developed from plant production studies and is soundly based physiologically. It has
also evolved from remote sensing studies. Although the relationship between APAR
and canopy photosynthesis may be highly variable over short periods, it remains
constant over longer periods, such as months or seasons.
The effectiveness with which canopies convert energy into dry mass depends on
environmental conditions such as temperature, water status and nutrition. Values of E
derived experimentally, in terms of net primary production (NPP) and APAR, range
2
from 2.8 g Mr for agricultural crops to a low value of 0.2 g Mr2 for tropical
rainforests. Values for other forests tend to lie in the range of 1-2 g Mr2 . Modifying
factors with values between 0 and 1, dependent upon the availability of water in the
soil, atmospheric vapour pressure deficits or nutrition, are used to account for the
effects of these variables on E. If the modifiers are applied to E, the result is an estimate
of potential E. Analysis of experimental results, therefore, should lead to the
identification of optimum, unconstrained values of E, which may vary considerably
less than the current experimental values.
Since the proportion of incident energy absorbed by a plant canopy depends on the
leaf area index (L*) of the canopy, good estimates of L* provide a sound basis for
estimating APAR. Strong statistical relationships exist between reflectance ratios in
the infrared, near-infrared and visible spectra, monitored from space, and L*, but these
vary with vegetation type and surface conditions. More important are recent
developments showing that the fraction of incoming energy absorbed by a canopy can
be calculated directly from canopy reflectance properties measured from space. These
allow direct application of the E model. As our knowledge of the factors causing
variation in radiation utilisation by canopies improves, along with our capacity to
273
estimate environmental conditions and their modifying effects on radiation utilisation,

it should be possible to use the model with remotely sensed data to estimate forest
productivity over large areas.
J.J. Landsberg, CSIRO Centre for EnvironmentaL Mechanics, GPO Box 821,
Canberra ACT 26/0, Australia. S.D. Prince, Department of Geography, University of
MaryLand, College Park, MD 20742-8225, USA. PG. Jarvis, Institute ofEcoLogy and
Resource Management, University of Edinburgh, Edinburgh EH9 JU, UK. R.E.
McMurtrie and B.E. MedLyn, SchooL of BioLogicaL Sciences, University of New South
WaLes, Sydney NSW 2052, Australia. R. Luxmoore, EnvironmentaL Sciences Division,
Oak Ridge NationaL Laboratory, Oak Ridge, TN 37831-6038, USA.
275
Introduction
The carbon (C) balance of forest stands is an important factor in the global C budget:
it is estimated (Melillo et al. 1990) that more than 70% of the C stored in terrestrial
ecosystems is stored in forests. It is, therefore, important to be able to accurately
calculate rates of tree growth and C storage as a basis for evaluating the global
consequences of maintaining mature forests, and of forest clearance and land use
change, regrowth and the growth of plantations. The ability to estimate forest growth
rates is also important at local and regional levels, where forest managers are
continually concerned about growth rates and production rates for commercial timber.
For these reasons, the C balance of forest stands has been the subject of a great
many studies, and a number of models have been developed to describe forest growth
in terms of the physiological and physical processes that govern CO 2 uptake by
photosynthesis and losses by respiration (Running and Coughlan 1988, McMurtrie et
al. 1990, Wang and Jarvis 1990, Running and Gower 1991). These models are, for the
most part, research tools based on detailed knowledge and descriptions of canopy
structures and of the physiologic~l processes concerned; they usually involve many
parameters and require the input of detailed weather data. It is not feasible to use such
models to calculate the C balance of large areas encompassing significant spatial
variation in stand structure and canopy characteristics because it is not possible to
specify, for each (quasi-) homogeneous area of forest, the parameter values needed for
the models, nor is it usually possible to provide the detailed weather and soil data
needed. There is, therefore, a need for a simple model with a small number of
parameters, preferably with robust, conservative values, that can be applied to large
areas. This chapter is concerned with development of such a model - essentially, a
linear relationship between the photosynthetically active solar radiation absorbed by
forest canopies (APAR, denoted Q,) and the net primary dry mass production (NPP,
denoted Pp) by forests,
Pp=EQa (I)
where P p and Qa are integrated over a period usually comprising about a year.
The coefficient describing the slope of the relationship (E) is often called the
radiation utilisation efficiency. Although efficiency is technically dimensionless, E
usually has associated units such as grams dry matter produced per megajoule
absorbed radiation (g MT\ This model is now being evaluated as a tool in global
ecological studies (e.g., Potter et al. 1993, Ruimy et al. 1994, Prince and Goward
1996) as well as for the estimation of stand-scale forest productivity. While much of
the discussion will be general in its implications, we are primarily concerned here with
the model's application to forests.
The E model has been developed from two separate approaches -
ecophysiological studies of forest productivity (Jarvis and Leverenz 1983, Landsberg
1986) and remote sensing studies (Kumar and Monteith 1982, Goward et al. 1985,
Prince 1991b). The evolution ofthese approaches is described later.
Although simplicity is essential for models used to estimate C budgets over large
areas and relatively long periods, it is also essential that models be based on sound
physical and physiological principles. We therefore examine the question of whether
the model is soundly based and, if so, whether the parameter is robust and
conservative and can be applied to large areas of diverse forest vegetation without
leading to errors large enough to render the results useless. There is a wide range of
values of in the literature, and some guiding principle is required to provide a basis
for generalisation and convergence; we have attempted to distil such a principle from
an evaluation of some of the published data.
Since the model is based on the relationship between the solar radiation absorbed
by canopies and dry matter production, the driving variable must be some measure of
energy input and absorption. Incident short-wave radiation values (Q) can be obtained
from ground measurements or satellites (Goward et al. 1994), while radiation
absorption depends on the structure and leaf area density of the plant canopy. For
many purposes, these can be accurately specified by the leaf area index (L *). There are
a number of techniques for measuring L * from the ground, including destructive
sampling and nondestructive methods based on the measurement of
photosynthetically active radiation (PAR, or radiation in wavebands between 0.4 and
0.7 J.lm) beneath canopies and inversion of detailed PAR interception models. L * can
also be estimated by remote sensing, which provides the only method available for
describing land surface and vegetation properties over large areas (see later discussion,
and Goward and Dye 1996).
In the last section of the chapter, we consider the application of the model in
calculating forest productivity and estimating the terrestrial component of the global C
cycle. Much of the discussion is general rather than specific to forests, but all of it is
applicable to forests.
Definitions
The literature on the use of to describe plant productivity in terms of absorbed solar
radiation contains many different measures of growth and radiation. This can lead to
confusion if only observed values of are given without a full explanation of the basis
of the calculations and the units employed. Variables used to define radiation
conversion efficiency include incident, intercepted or absorbed radiation; total solar or
photosynthetically active radiation; net or gross dry mass production; and above-
ground or total biomass or CO 2 exchange.
We will deal in terms of NPP, defined as
Pp = G p - Ra (2)
where gross primary production (GPP, defined as Gp) is the net photosynthesis of the
leaves of the trees and R a is the autotrophic respiration of the stems, branches and roots
integrated over an appropriate period of time. NPP is the increase in dry mass of the
living trees over the period of integration. If NPP is estimated from sequential
measurements, or sampling, over an interval 8t = t2-tl' then NPP is the change in
standing biomass plus the material produced after t l that is lost as litterfall before ~.
Ideally, Eq. (2) describes total (above- + below-ground) NPP, but in many
experiments values of below-ground production (P b) are not available, and has been
determined on the basis of above-ground dry mass production (P) only.
277
Defining in terms of dry mass production has advantages in terms of forest

productivity; however, the disadvantage, in relation to global C balance, is that values
of dry mass must be converted to C or CO 2 , depending on the basis of the calculations.
This is a relatively unimportant problem; despite the variations in dry matter
composition (Vogt 1991), the errors arising from the assumption that C accounts for
50% of plant dry mass are negligible compared with other uncertainties in this type of
calculation.
Although the above-canopy ratio of incident PAR to total incident solar radiation is
generally between 0.45 and 0.50 (PAR "" 0.5 Q) and CO2 exchange can be expressed
in grams of C or converted into an equivalent amount of carbohydrate, given in any
of the other forms cannot be interconverted without additional information that is
generally specific to the measured vegetation. In relation to the PARlQi ratio, Stigter
and Musabilha (1982) determined that PAR was approximately 51 % of Q i in the
tropics on clear days, but that on cloudy days the proportion increased to 63%.
In this chapter, (g MJI) will denote the ratio of total NPP to PAR absorbed by
foliage (APAR), .
= (Pb + Pa)/Qa (3)
where Q a = 0.5(Qi - Qr - Q, + Q,,). The subscripts r, t and sr denote reflected,

transmitted and soil-reflected radiation. The relationship is not strictly accurate
because conversions between PAR and solar radiation apply only to incident radiation;
below canopies the conversions are quite different because of the preferential
absorption of red radiation by chlorophyll. In plant physiological literature, PAR is
often defined in terms of photon flux density (PFD, here denoted </lp' 2.1mol quanta,
are equivalent to IMJ). PAR, with energy units, is more commonly used in relation to
remote sensing and will be used throughout this chapter, except in the discussion of
canopy photosynthesis.
A distinction must be drawn among experimental results obtained from studies of
individual trees, canopies and entire communities. The radiation measurements must
correspond to the appropriate scale at each level. Many forest production studies are
concerned only with the canopy-forming trees, whereas it is not generally possible to
isolate individual components of the community from remote sensing measurements.
The dry mass production relevant to remote sensing studies is that of the entire forest
community, including shrubs and herbs, if present, as well as the tree layer.
Evolution of the E model

Since photosynthesis is a quantum-dependent process, it is intuitively obvious
(although disputed by Demetriades-Shah et al. 1992, 1994; see the replies by
Arkenbauer et al. 1994, Kiniry 1994, Monteith 1994) that there is likely to be some
relationship between the solar energy incident on the canopy of a plant community,
and dry matter production. This has been explicitly recognised, and extensively
exploited, in numerous models of crop and forest growth (see next section), based on
PAR absorption by canopies and leaf photosynthetic characteristics. The possibility of
a simple relationship between PAR and crop dry matter production was recognised by
Landsberg (1967), when he derived radiation utilisation efficiencies for lucerne

(alfalfa) grown under irrigation in Africa, and found that the values obtained increased
with decreasing water stress (integrated over growth periods), up to the point where
water was not limiting the growth of the crop. However, it was Monteith (1977) who
showed that the relationship between APAR and the annual p. of dry mass by four very
different crops (sugar beets, potatoes, barley and apples) in Britain was linear, and
developed this into a model of crop growth. Monteith's work was followed by other
research on crops (Gallagher and Biscoe 1978, Legg et al. 1979). Rauner (1976)
provided some "photosynthetic efficiency" values ( as a percentage of absorbed PAR)
for deciduous forest, although it was not clear how the data were obtained, and it was
not until later that a thorough treatment of the application of the model to forests was
provided by Jarvis and Leverenz (1983). They made estimates of the efficiency with
which forests converted absorbed radiation into dry matter on the basis of
physiological information and empirical relationships. The first convincing empirical
demonstration that the approach could be applied to forests was provided by Linder
(1985), who derived above-ground values of for plantations of Eucalyptus and Pinus
radiata.
The possibility of applying the model over large areas, using remotely sensed
measurements, began to emerge from the recognition (e.g., Tucker 1979) that ratios of
the radiation reflectance by plant canopies in different wavebands could provide
information about green, photosynthetically active foliage, and that time integrals of
vegetation indices were strongly related to net production (Tucker et al. 1985, Prince
1991a). Kumar and Monteith (1982) showed that APAR relates to the ratio of red to
near-infrared (NIR) reflectance and combined the resulting estimates of APAR with
the simple model to calculate dry mass production by sugar beets and wheat. Goward
et al. (1985), working with data from the Advanced Very High Resolution Radiometer
(AVHRR) on the NOAA TIROS satellite, integrated the Normalised Difference
Vegetation Index (NDVI=(NIR-VIS)/(NIR+VIS), where VIS denotes spectral
reflectances in the visible wavebands) over a growing season for a number of locations
in the United States for which estimates ofNPP were available. They showed a strong
linear relationship between NDVI and NPP. Goward and Dye (1987) developed a
crude canopy photosynthesis model not based on but driven by APAR estimates
based on NDVI. The model included water balance calculations and, using climatic
data from a number of locations, was run for 12 different biome types. Goward and
Dye obtained a relationship that was "positive and encouraging." Goward (1989)
noted that "although the concept (of a direct link between remotely sensed
visible/near-infrared vegetation indices and climate/vegetation interactions) needs
further investigation, the empirical evidence strongly supports the link."
The questions that arise in relation to the use of the simple model, at whatever
level, are: Is it a soundly based approach? What are the appropriate values of for
different vegetation types? Are these values stable over time? What are the factors that
cause to vary, and are we in a position to predict the variation? These questions are
addressed in the following sections of this chapter.
279
Physical and physiological basis of

That the model is soundly based can be established by examining the physical and
physiological bases for it, and determining whether the assumption of a linear
relationship between radiation absorption and the rate of C fixation by plant canopies
is tenable.
The rate of C fixation by a plant canopy depends on the radiation absorbed by the
canopy and the photosynthetic characteristics of the leaves. Radiation absorption by
plant canopies has been exhaustively analysed and modelled in great detail (e.g., Wang
and Jarvis 1990, Wang et al. 1992 and references cited therein), but for most purposes
simple models provide adequate descriptions of photon flux (PF) interception (Eq. 8).
The change in the rate of photosynthesis by a single leaf with increasing PFD is
nonlinear; however, if a canopy is closed, so that the foliage intercepts all or most of
the incident radiation and much of the foliage is not PFD-saturated, photosynthesis by
the canopy as a whole is likely to be PF-limited and the relationship between
intercepted energy and net photosynthesis by the canopy tends towards linearity. This
tendency is strengthened if a high proportion of incoming radiation is diffuse. We can,
therefore, assume that the CO2 assimilation rate (A) of a PF-limited canopy will be
proportional to the absorbed photon flux density <Ppa
A= lO'<Ppa (4)
where lO' has units of moles CO 2 per mole quanta.
Nonlinearities are likely to be less apparent over long periods (e.g., weeks or
seasons). By integrating over time, converting assimilated CO 2 to plant dry mass
(Russell et al. 1989) and absorbed photon flux to absorbed energy flux, Eq. (4)
provides the G p term in Eq. (2),
G p =109 Qa (5)
where 109 is maximum energy conversion efficiency. The relationship between APAR
and photosynthetic productivity has been analysed formally by Charles-Edwards
(1982), who provided an expression for the daily net photosynthetic integral of a
canopy in terms of the quantum efficiency of leaf photosynthesis (a), the rate of PFD-
saturated photosynthesis (A ma,) and the canopy extinction coefficient (k). He showed
that the efficiency (lO) with which a closed crop canopy uses intercepted radiation in the
production of new dry mass is determined by a, A max and k, with some conversion
factors, but that fivefold differences in mean daily PAR had relatively little effect on
values (i.e., the relationship tended to be linear).
Wang et al. (1992) carried out a useful analysis on the relationship between daily
photosynthesis values and absorbed PAR. Using the model BIOMASS (McMurtrie et
al. 1990), which incorporates a detailed mechanistic model of leaf photosynthesis,
they showed that on a daily basis there was considerable scatter between simulated
daily photosynthesis and absorbed PAR. However, the variability collapsed to a single,
approximately linear, relationship between canopy photosynthesis and absorbed PAR
when expressed in annual terms, providing a clear demonstration that the simplified
relationship encapsulates the complexities of the photosynthetic process and the

interactions of directly illuminated and shaded foliage.
Jarvis and Leverenz (1983) demonstrated that the rate of net photosynthesis by a
spruce canopy tended to be linear with PFD, largely as a consequence of shoot and
canopy structure. They went on to analyse canopy productivity (dry mass production)
in terms of leaf photosynthetic properties, respiratory losses and mortality losses, all
of which can have substantial effects on forest growth. The most important factors
appear to be PAR interception by the canopy and losses resulting from root, leaf and
tree mortality.
Prince (1991a), following Jarvis and Leverenz (1983), wrote the 10 model in the
form
(6)
wherefis a factor that measures physiological constraints such as stomatal closure, Yg
and Y m are respiratory coefficients defining the assimilate lost in growth and
maintenance respiration, respectively, and d is the proportion of biomass not lost by
litterfall, decay or grazing. Prince argued that, since the most direct and immediate
outcome of photon flux absorption by canopies is C fixation by the process of (gross)
photosynthesis, the 10 values relating to G p (Eqs. 2 and 5) are more likely to be
fundamental and conservative than values derived from measurements of NPP. Gross
photosynthesis includes photorespiration but excludes C losses associated with growth
and maintenance respiration.
This approach highlights a potential practical limitation of the 10 model, particularly
in the case of forests, since respiration depends on factors such as biomass and its
components (leaves, bark and living stemwood, which vary between species and
between young and old forests) and temperature, so that we are forced back to
empirical data and calculations based on generally inadequate descriptions of forest
biomass and characteristics. The concept of an unstressed, or potential, value of 10,
embodied in Eq. (6), may well prove to be of considerable value.
The potential, or optimal, value for 10 could, in principle, be estimated from the
quantum efficiency of photosynthesis at the chloroplast level. This tends towards a
theoretical maximum value (about 9 mole quanta per mole CO 2), when all variables
are optimum. This theoretical value at the chloroplast level is of little use for
estimating 10 values for forest canopies, but values of the quantum yield for single
leaves, obtained from measurements of photosynthesis at very low PFD, could be used
as the upper, unstressed limit value for the canopy in question. That value would be
modified by the effects of environmental factors on the physiological processes that
influence the final values of E. This approach has already been applied to global NPP
modelling in the GLO-PEM model (Prince and Goward 1995) and to net CO 2 flux data
from a mixed deciduous forest in New England (Waring et al. 1995).
Effects of environmental variables. Environmental variables may have both

immediate and long-term effects on the effectiveness of radiation utilisation by
canopies. Over short time intervals (hours), the relationship between photosynthesis
and APAR can be highly variable, partly because of the feedback regulation of
281
photosynthesis by the internal source-sink relationships of plants. Maier and Teskey

(1992), for example, showed that net photosynthesis of Pinus strobus shoots was
higher in the spring, during active shoot growth, than during the summer, even though
temperatures were lower in the spring. Whether this effect is significant over the time
scales of concern for the use of the simple e model is yet to be demonstrated. However,
particularly in the case of trees, there is clearly a very large source-sink effect in
relation to carbohydrate allocation to component parts of the plants. One of these parts
is the root system, where considerable amounts of carbohydrate are used in the
production of fine roots, which may have very short lifetimes (high turnover).
Empirical data indicate that forest stands allocate highly variable proportions of their
assimilate to roots (P b, Vogt 1991), but there are grounds for assuming that the
proportion of NPP allocated to roots may be relatively constant (i.e., P/(Pa + Pb) ""
constant). A study by Aber et al. (1985) showed that the rate of fine root turnover
increased with N availability in the soil, while a companion study by Nadelhoffer et al.
(1985) showed that above-ground NPP (Pa) is strongly related to N uptake. Raich and
Nadelhoffer (1989) have since de~onstrated that annual rates of net fine root and Pain
forest ecosystems are strongly related.
In the short term, water stress caused by high evaporative demand would be
expected to have an immediate effect on a canopy's energy conversion efficiency by
causing stomatal closure. There are also longer-term effects on growth and
development processes such as cell division and expansion, which lead to reductions
in dry mass production by trees. These are proportional to a water stress integral
(Myers 1988), which is the summation of predawn leaf water potential values over
time. Long periods of water shortage are likely to affect the capacity of a canopy to
utilise radiant energy by reducing L *- and hence APAR - through accelerated rates
of leaf fall (Pook 1986, Linder et al. 1987), or by reducing leaf growth and canopy
development. Effects on canopy development are different for species that have
determinate and indeterminate shoot growth. Where growth is indeterminate (e.g.,
Pinus taeda, P. radiata), water stress affects leaf production and development; in
determinate species (e.g., Picea sitchensis, Pinus sylvestris), where the number of
apices is set in the autumn and cannot be changed during the subsequent growing
season, water stress affects leaf expansion but not the number of leaves.
The effects of temperature on e are likely to act, for the most part, in the short term.
Both high and low temperatures may reduce photosynthesis (e.g., Harley et al. 1992),
and hence e. Temperatures below freezing inactivate the photosynthetic apparatus,
which may then take several days to recover fully (i.e., there is a "carry-over" effect).
High temperatures inactivate enzyme systems and reduce photosynthesis; for most
plants, photosynthetic rates fall to zero above about 40C (Larcher 1983). Recovery is
fairly rapid, unless the temperatures are high enough and the period of exposure
sufficiently prolonged to damage biochemical systems. There are also longer-term
temperature responses associated with seasonal growth patterns: in deciduous trees,
leaves develop in the spring - when both L * and e increase - and reach their
maximum photosynthetic capacity (maximum e) during the summer. Photosynthetic
capacity declines rapidly in the autumn (fall), when both L * and e fall rapidly. This
decline is usually represented as a daylength response but can be described adequately
in terms of temperature. Temperature also influences decomposition and hence

nutrient mineralization - processes with time scales ranging from months to
centuries.
Nutritional effects on E operate through effects on leaf enzyme systems -
primarily ribulose bisphospate carboxylase (Rubisco) - and hence directly on
photosynthetic rates. They are likely to be medium- to long-term (i.e., to operate over
periods of weeks to seasons). There is evidence from the agricultural literature that
high leaf N concentrations increase the radiation use efficiency of crops (e.g., Sinclair
and Horie 1989, Sinclair and Shiraiwa 1993). Nitrogen also has direct effects on leaf
expansion. Data from forestry are more limited than from agriculture, but Wang et al.
(1991) found that N fertilisation increased E (for above-ground NPP) in a stand of
Sitka spruce by about 30%, relative to unfertilised stands. McMurtrie et al. (1994)
carried out a detailed, model-based analysis of E for a number of pine stands round the
world (see discussion in "Values of E in woody plants"), some of which were subject
to widely varying fertiliser regimes, but they did not identify an effect of N on E.
However, McMurtrie et al. (1992) and Kirschbaum et al. (1994) have run the model
MAESTRO (Wang and'Jarvis 1990) with a range of leaf N concentrations to explore
the dependence of E on leaf N, and found a strong, positive relationship. On the basis
of net Pa only, Raison and Myers (1992) found that E for irrigated and fertilised stands
of P. radiata was 36% higher than for irrigated stands without fertilisation. However,
when data on soil respiration were used to estimate root production and turnover, the
values of E for total NPP (Pa + Pb) were slightly lower for the fertilised than for the
irrigated, nonfertilised stands (S. Pongracic, personal communication 1994):
Fertilised and irrigated Irrigated only

E (P) 0.81 0.63
E (Pa + Pb) 1.11 1.23
In summary, plant canopies exhibit a hierarchy of responses to environmental

variables in terms of radiation utilisation efficiency. The most rapid responses include
activation of photosynthetic enzymes such as Rubisco, followed by stomatal
responses, changes in leaf attitude and light capture, variation in the quantity of
photosynthetic machinery, leaf area adjustments through growth, changes in specific
leaf area and senescence and, finally, changes in community species composition.
Water and nutrition may elicit plant canopy responses through their influence on the
proportion of C fixed by a stand allocated to roots, through effects on stomatal
conductance and hence CO 2 uptake and through effects on leaf enzyme systems and
photosynthetic rates. Many of these environmental variables have an immediate effect
on E, which is likely to be followed by the gradual adjustment and return of E to its
normal range.
Values of in woody plants

We noted earlier that it may be possible to identify some intrinsic, unstressed,
maximum value of E. A number of approaches can be adopted to identify such a value,
or values: first, undertake additional experimental work, with plants grown under
283
optimum, unstressed conditions; second, search the literature for upper-boundary

values of E (Hunt 1994); third, attempt to calculate values, using models that allow
estimation of the effects of variables likely to reduce E; and finally, as mentioned
earlier, estimate values of E from the quantum yields of individual leaves, determined
at very low PFD. In the next section, we consider some of the values reported in the
literature and values obtained using models.
Values derived from experiments. In Monteith's (1977) study, the slope of the line
relating annual dry matter production by sugar beets, barley, apples and potatoes to
intercepted total solar radiation was 1.4 g dry matter MT 1 Similar values were
obtained for woody crops by Cannell et al. (1987) who, in a two-yr study on irrigated
and fertilised willow, obtained E values for total Pa' for stems only and for total
1
production, including roots. Their overall average value for Pa was 1.4 g Mr total
short-wave radiation. In a later study, Cannell et al. (1988) compared pot-grown Salix
and Populus stands and found that "two very different clones, one of Salix and one of
Populus, used the light energy ~hat they absorbed with approximately the same
efficiency. For both clones, the seasonal mean light use efficiency value (E) was about
1.5-1.6 g Mr 1 total short-wave radiation." Cannell et ai. (1988) concluded that "the
implication is, as Monteith (198 I) suggested, that E is a conservative value for C3 crops
growing in similar radiation environments, differing relatively little among species,
especially when differences in root/shoot partitioning are taken into account..." There
is considerable evidence (e.g., references cited by Russell et ai. 1989 and Russell and
Prince 1991b) that E values of about 2.8 g Mr 1 APAR are the norm for P a by
intensively managed field crops. This has, apparently, led to the expectation that E
values for forests might turn out to be "conservative, differing relatively little among
species..."
However, values of E obtained for tree crops in the field are not nearly so high. A
study by Landsberg and Wright (1989) on Populus clones grown under intensive
culture in the field at two locations in the United States yielded above-ground values
for E, at one site, of 0.32 g Mr 1 total solar radiation in the first year of growth and 1.4
g Mr 1 in the second year. It was assumed that the difference was due to the fact that a
large proportion of the dry matter produced went to root development in the first year.
At the second site, where the trees were older, an average value for E of 1.28 g Mr 1
total solar radiation was obtained. The first empirical E values for forests were
provided by Linder (1985), who used Beers' Law (Eq. 8) to calculate intercepted
1
radiation and obtained E = 0.9 g MT APAR for young Eucalyptus globulus stands in
1
Australia and 1.7 g Mr APAR for Pa by a number of forest stands in Australia, New
Zealand and Europe. He made no estimates ofPb .
Jarvis and Leverenz (1983) derived estimates of the likely values of E for forests of
various types from information about radiation interception by canopies, leaf
photosynthetic properties, respiration and other C losses. They arrived at estimates of
E for NPP in relation to total solar radiation ranging from 0.15 for warm-area
deciduous forests to 0.78 for cool-area evergreens. Wang et al. (199 I) measured APAR
over four years in six plots of Picea sitchensis in central Scotland, with different
thinning and fertiliser treatments. Estimating Pa from allometric relationships, they
I
obtained values for c of 0.95 g Mr total solar radiation for stands fertilised with N,
1
and 0.75 g Mr for unfertilised stands (a difference of about 30% attributable to N).
However, using the model MAESTRO, Wang et al. calculated Gp in terms of CO 2
uptake by the stands and obtained a value of 1.50 g Mr 1, the difference being
attributed to losses by autotrophic respiration.
The OTTER Project, a study of NPP in a range of forest types across the Oregon
transect (an area extending from the Oregon coast to the inland rain shadow of the
Cascade Mountain Range and encompassing a wide range of climates and vegetation
types) (Runyon et al. 1994), provided information on APAR, Pa , Pb and L*, as well as
on a range of environmental variables (temperature, atmospheric vapour pressure
deficit, a measure of the water stress to which the trees were subject and estimates of
soil water status). Runyon et al. found no general relationship between NPP and APAR
until they corrected for the effects of the environmental variables, using the equation
presented here (Eq. 7). The modifiers were based on the occurrence of subzero
temperatures, "drought" - as defined by predawn foliage water potentials - and
large vapour pressure deficits. Their analysis led to c values of 0.8 g Mr' APAR for
above-ground NPP and 1.3 g Mr 1 for total NPP. All vegetation types studied fell on the
2
regression lines (Pavs Qa; Pa+ Pb vs Q) that had R values of 0.99.
The lowest reported c values appear to be those derived by Saldarriaga and
Luxmoore (1991), who investigated the productivity of tropical rain forest stands at 23
sites in Colombia and Venezuela. The stands at these sites represented a
chronosequence from areas recently abandoned after the use of slash-and-burn
agriculture to those characterised by mature forest. Biomass and leaf area calculations
were based on careful sampling procedures, solar radiation was measured in the area
for many years and a series of studies on canopy absorption of PAR was carried out.
1
The values of c obtained for NPP declined from about 0.3 to 0.4 g Mr in young stands
1
to an almost constant 0.2 g Mr for stands 10-50 years old. These values reflect the
very low mineral nutrition of the study sites (Saldarriaga et al. 1988).
It is unlikely that environmental variables alone can account for all the
experimental variation in c. There is a decline in the reported values for forests that
corresponds roughly to the age of the stands (Fig. 1), and it can be seen in the
experimental data discussed here that the highest c values come from very young
plants grown under nonlimiting conditions (Cannell et al. 1987, 1988); values from
plantations (Linder 1985, Landsberg and Wright 1989, Wang et al. 1991) are lower,
while those from old, mature forests (Rauner 1976, Saldarriaga and Luxmoore 1991,
Runyon et al. 1994) are at the bottom of the range. Despite the uncertainties in the
data, this age effect appears to be genuine, although the reasons for it are not yet clear.
Ryan and Waring (1992) found that respiration alone was apparently not sufficient to
account for the low NPP of mature forests. Waring and Silvester (1994) have proposed
a partial explanation: reduced rates of leaf photosynthesis caused by low water
potentials in foliage with very long water-conducting pathways from the roots.
Another (or additional) explanation may lie in the large proportions of old shaded
foliage with low photosynthetic rates (Jarvis et al. 1976) that may occur in old forests.
The total amount of C fixed by this foliage is low compared with the amount that
would be fixed by the same proportion of young foliage, even in the same PF regimes.
285
1.4 I I I I
\A Chestnut coppice
" , -
(SabatierTarrago 1989)
1.2 .\ o Oak (Rauner 1976)
..,,
\~ A Trorcical rainforest ._
f- (Sa darriaga & Luxmoore 1991)
1.0
~
A Fir (Kira 1975)
-
"
0
0.8 f- 0
""'-'-'
~
~
0.6 f- -
~
W A
" ' o............
-
0
.
0.4 fj-
....
........ 0
-. ..... -
0.2 f-
I I I I
0
0 40 80 120 160 200
Forest Age (years)
Figure 1. Values of E and its variation with age, derived from studies reported by
Ruimy et aI. (1994). The line was drawn by eye. The original papers (Sabatier-
Tarrago 1989, Rauner 1976, Saldarriaga and Luxmoore 1991, Kira 1975) should
be consulted for details about how E was determined.
Values derived using models. McMurtrie et al. (1994) analysed several years of data
from five pine stands, including four species located in Australia, New Zealand,
Florida, Sweden and Wisconsin. Measured annual above-ground NPP (P.) ranged
from 0.2 to 1.6 kg C m-2 They established that the model BIOMASS (McMurtrie et al.
1990) provided accurate simulation of the measured values, then used the model to
obtain "effective," or utilisable, values of APAR. These were derived on the
assumption that the energy incident on and absorbed by a canopy on a particular day
is not necessarily utilised for photosynthesis if conditions exist that prevent or restrict
CO2 uptake. Potential photosynthetic gain is therefore modified on the basis of adverse
environmental factors. (This idea was proposed in outline by Landsberg (1986) and
applied by Runyon et al. (1994)). McMurtrie et al. (1994) used the form of equation
for annual G p suggested by Landsberg,
G p = L Q a fw fD fT (7)
where the modifying factors (f) describe, respectively, reductions on day i resulting
from soil water deficit (W), low temperature effects (T) and saturation vapour deficit
of the air (D), which affects stomatal conductance. The value fw was taken as unity at
high soil water levels and as a linear function of root-zone soil water content below a
critical point; fD was calculated as a linear function of the daytime values of D. The
temperature modifier integrates reductions caused by the occurrence of frost on day i,
incomplete recovery of photosynthetic capacity after winter and cessation of
photosynthesis late in the growing season. When McMurtrie et al. plotted simulated
AUllr.U.
Flor1da .
0
A
A
-
WIfC:Oft.ht
J
'-
......,.
S dtn X
-
A
A

A
E
:t 0
0
0
. A
0
~
.
0
:;
E
:1.
x
x
x
~ .,-- .,---.!!1I(')

Figure 2. (a) Simulated gross annual photosynthetic C gain (GPP, or G p )

calculated for pine stands near Canberra, Australia, in Florida and Wisconsin,
USA, at Jadraas, Sweden and at Puruki, New Zealand. GPP is plotted against
simulated APAR, calculated from diffuse and direct radiation and a radiation
absorption model; (b) simulated GPP (G p ) plotted against simulated utilisable
APAR (corrected for environmental variables, Eq. 7). The line has a slope of 1.86
g C MJ- 1 (McMurtrie et al. 1994).
opp (Op) against uncorrected APAR, there was considerable scatter; the plot of
simulated OPP against utilisable APAR was very much tighter, justifying the use of
utilisable APAR. Figure 2a shows OPP plotted against unmodified APAR for the five
sites during the seasons studied. Figure 2b shows the plot of simulated Op against
simulated utilisable APAR. The line has a slope of 1.86 g C Myl.
No effects of nutrition are shown in Figure 2a because leaf N content expressed on
a unit area basis differed little - from 2.75 to 3.07 g m-2 - across the five sites studied
by McMurtrie et aI., although on a dry mass basis the concentrations ranged from 8 to
15 mg gol. In the BIOMASS model, leaf N contents are substituted into equations from
Field (1983) to calculate photosynthetic rates. The question of whether photosynthesis
should be related to leaf N content or concentrations on a dry mass basis is yet to be
decided (Field 1991, Schulze et al. 1994).
MAESTRO (Wang and Jarvis 1990) has much more detailed radiation absorption
routines than does BIOMASS, so the MAESTRO and BIOMASS models have been
compared using canopy structure and radiation data from three of the sites included in
the McMurtrie et al. study. The results are presented in Figure 3, which shows that the
performance of the two models is similar for homogeneous canopies. Consequently,
for most purposes there is no need to use the more complex and highly parameterized
model. MAESTRO includes routines that deal with spaced trees, so it is particularly
useful for analysis of plant communities that can be treated as assemblages of
287

,-...
...>-
A
';" 4 AA
~Yoo A
A
"! D 0.0 A
E 0
0>
:::.. 2
DO
a.. AA
a..
<.9 ,fA
0
0 2 3
Absorbed PAR (GJ m-2 yr- 1 )
Figure 3. Comparative values of GPP obtained from two models, MAESTRO

(solid symbols) and BIOMASS (open symbols), applied to three ofthe pine stands
considered in Figure 2. The v,alues obtained from the models are similar,
indicating higher values of E at Furuld, New Zealand (squares), followed by the
Australian (circles) and Florida (triangles) sites.
individual trees. Neither model includes a decomposition module, although a later

version of BIOMASS (G'DAY, Comins and McMurtrie 1993) includes decomposition
and N dynamics.
The significance of this analysis of utilisable radiation is that the values of
obtained can be interpreted as indicating the PAR utilisation efficiency of pine stands
growing under nonlimiting conditions, when approaches the values obtained for
field crops. Furthermore, since data from all the stands fall on the same line (Fig. 2b),
it is clear that there are no differences in the energy utilisation efficiency of the
different pine species. Therefore, a single value of maximum can be used for all the
species, and it appears that the model, with its simple modifying functions, can be used
for large-scale calculations.
Hunt and Running (1992) used the model BIOME-BGC, a generic version of
FOREST-BGC (Running and Coughlan 1988, Running and Gower 1991)
parameterized for spruce (Picea species) and aspen (Populus species) at two sites in
Canada, to calculate C budgets for the two species. They observed that "differences in
annual net photosynthesis (and hence, growth respiration) for the same genus between
the two sites were almost completely due to differences in L * and (length of) growing
season..." The average annual values obtained were 1.59 g Myl APAR for spruce and
1.54 g Myl for aspen at the southern site (Prince Albert, Saskatchewan) and 1.03 and
0.68 g Myl for the two species, respectively, at the northern site (Thompson,
Manitoba).
FOREST-BGC and BIOME-BGC have routines that account for the effects of soil
water and D on stomatal conductance; they also have a soil N routine, which leads to
estimates of leaf N concentrations. The relatively high values obtained by Hunt and
Running suggest that these procedures are largely effective in modifying the
interactions between APAR and CO 2 uptake. This is supported by comparisons
between the output of FOREST-BGC and measurements made in the OTTER Project
(Running 1994). FOREST-BGC was carefully parameterized for the vegetation on the
sites across the transect to establish initial conditions and plant functional parameters;
then a series of simulation runs was carried out and the results compared with
measurements of predawn plant water potential, transpiration, Pa and L*. Although
FOREST-BGC performed reasonably well, the results were not as impressive in terms
of correspondence between measured and calculated variables as those obtained by
Runyon et al. (1994). However, Runyon et al. focused only on Pa' and there was an
element of fitting to data in their analysis. Running points out that "it is difficult and
costly to obtain rigorous parameterization and complete validation data for complex
ecosystem models..." and that the differences in the time frames of measurement and
calculation often result in comparisons for which there is ample explanation for lack
of correspondence.
Running and Hunt (1993) have sufficient confidence in their models to use
BIOME-BGC for landscape-scale simulations of the C, water and N cycles of
different ecosystems, and have used the results to calculate E for a range of plant types
and life forms. The values obtained indicate that differences in E between coniferous
and deciduous trees in the same climate are small; values for trees were much smaller
than those for grass. However, they did not calculate utilisable radiation. To obtain this
value, APAR must be reduced on a pro rata basis by the amount of energy intercepted
by the canopy when photosynthesis is reduced relative to photosynthetic rates when all
variables are optimum. We suggest that this normalisation approach will facilitate
identification of optimal, or unconstrained, values of E. Where normalisation is not
used, the variation resulting from environmental and physiological constraints will be
incorporated into estimated E.
It seems clear that we must expect E to vary as a result of limiting environmental
factors, physiological feedbacks (e.g., C accumulation) and stand age. For this reason,
attempts to determine E experimentally, although they should continue, will always
provide highly variable results. Normalisation of those results in terms of utilisable
radiation appears to provide a means of removing considerable variation, so that
second-order effects such as nutrition and stand age may be expected to emerge.
Experimental work and model development should, therefore, continue to focus on
identification and definition of the modifiers and their operation.
Estimation of L* and APAR

The values for APAR required to apply the E model to forests can be obtained either
directly, through measurements made on the ground or through satellite
measurements, or indirectly, by calculation. The most straightforward approach is to
measure APAR using arrays of radiation or quantum sensors, or mobile, trolley-
mounted instruments. Such measurements are not feasible where the model is to be
applied over large areas, but they provide the means of testing radiation interception
models, which allow extrapolation of APAR estimates over large areas, if values for
L * and Qi are available. The simplest of these models, which is adequate for most
purposes, is the widely used expression for exponential absorption of PAR by the
foliage of a canopy:
289
Qa = 0.5 Q; (1 - e okLo) (8)

where k is the light extinction coefficient. Pierce and Running (1988) used an
integrating radiometer to derive values of k for large areas of forest as a means of
comparing the L * values they obtained by this method with those obtained by remote
sensing.
For calculations of NPP over large areas, where remotely sensed canopy
reflectance data are available, the use of L* is likely to be superseded by better
measures of the photosynthetic capacity of canopies such as chlorophyll concentration
per unit area, which influences the red-to-near-infrared reflectance ratios of forest
canopies (Yoder and Waring 1994) or the measurement of APAR itself. However, L*
is also valuable for other purposes - most notably, for calculating evapotranspiration
- and is a key variable in many models, such as BIOMASS, FOREST, BIOME-BGC
and MAESTRO.
The importance of obtaining reliable values of L * to calculate forest productivity
was well illustrated by Bonan (1993), who applied an ecosystem production model to
21 forest stands located on river floodplain and upland sites near Fairbanks, Alaska, to
determine the sensitivity of simulated canopy photosynthesis to L* and forest type.
The five tree species in the 21 stands differed both in structure and in physiological
characteristics such as photosynthesis per unit foliage mass, N requirements for
growth, respiration and C allocation. They did not differ in canopy geometry. Bonan's
model successfully simulated a range of physiological responses, as well as observed
Pa' Sensitivity and statistical analysis showed that net canopy CO 2 assimilation was
most sensitive to L* and less sensitive to physiological differences among species,
although the species composition of stands also accounted for significant variation.
Bonan concluded that "uncertainty in L* can cause net canopy assimilation to be in
error by up to 42-70%, depending on forest type..." The analysis confirms the
importance of good estimates ofL* (for which remote sensing may be necessary), but
indicates that physiological differences between species must not be ignored.
The correlation between NDVI and L* has been extensively investigated. Spanner
et al. (1990a) used data from the Advanced Very High Resolution Radiometer
(AVHRR) on a NOAA satellite to calculate NDVI, which was then used to examine
the temporal variation in L* in forest stands in Oregon, Montana, Washington and
California. They found a strong relationship between summer maximum values of
NDVI and measured L* and identified clear seasonal trends in L*. In a detailed study
as part of the OTTER Project, Spanner et al. (1994) used a number of broad- and
narrow-band instruments to study spatial and seasonal variation in L* across the
Oregon transect. They concluded that "the strength of the relationships observed
between minimum and maximum L* and the Simple Ratio (near-infrared-to-red
reflectance) calculated from four remote sensing instruments on three different
platforms and at several altitudes indicates that remote sensing offers utility for
assessing and monitoring L*." However, in another paper (Spanner et al. 1990b),
Spanner and his colleagues noted that "the relationships between the L* of coniferous
forests and (Landsat Thematic Mapper) spectral bands... were affected by canopy
closure, understory vegetation and background reflectance...Solar radiation reflected
from the forest canopy was related to coniferous forest L*...(but) differences in forest
stand canopy closure permitted understory vegetation and the background to
contribute to the integrated spectral radiance of the forest stands..." Despite this
finding, Curran et al. (1992) used NDVI calculated from Landsat Thematic Mapper
(TM) to evaluate seasonal L* variation in 16 slash pine plots in Florida that had been
given different fertiliser treatments. The results of this rigorous test of the method were
positive enough to warrant the conclusion that the use of remotely sensed data has
potential for the study of seasonal dynamics in forest canopies.
Goward et al. (1993) made a thorough study of the spectral reflectance properties
of many important landscape constituents across western Oregon and evaluated the
implications of their results for remote sensing. They concluded that the simple "green
foliage" description provided by remotely sensed NDVI is inadequate and wrote: "In
the extreme it appears possible to record an NDVI of less than 0.55 for a complete
canopy of juniper and sagebrush and an NDVI greater than 0.4 from litter and lichen
covered landscapes without green foliage ..."
The fraction of incorping PAR absorbed by a canopy (FPAR) can be calculated
directly from canopy reflectance properties. Based on an analysis of the SAIL model,
Hall et al. (1990) proposed the use of the second derivative of the reflectance/
wavelength function obtained from narrow-band spectra to estimate APAR (Q);
however, this application does not appear to have been tested in the field. SAIL
assumes random foliage distribution, which is a poor assumption for many forest
stands. From a later analysis using the same model, Goward and Huemmrich (1992)
concluded that "Under typical observing conditions NDVI is near linearly related to
both instantaneous and daily total APAR..." They noted that the potential influence of
such landscape variables as leaf angle distribution, leaf spectral optics and background
spectral reflectance on the APARINDVI relationship is significant and suggested
alternative remote sensing approaches to resolve the structural and spectral variations
within vegetated landscapes. However, they also recommended that "before either
approach is seriously considered, an aggressive effort should be undertaken to
demonstrate a need for such added complexity." Sellers (1985), Sellers et al. (1992)
and Goward and Dye (1996) provide more detailed treatments of the estimation of
APAR and its relationship to canopy photosynthesis.
Estimation of environmental variables

Since NPP depends not only on L* and APAR but also on the environmental variables
affecting the efficiency of radiant energy conversion by plant canopies, it is essential
that information about environmental constraints that limit the utilisation of APAR be
available. Since these constraints are likely to operate over different periods, the
methods used to estimate them and the intervals over which the variable values are
averaged will vary with different applications. When long-term averages are required
for particular sites and regions, climatological data will often provide the most useful
values (e.g., as used on a global scale by Potter et al. 1993, Melillo et al. 1993).
Running et al. (1987) and Running and Coughlan (1988) have developed procedures
for extrapolating meteorological measurements made at a single site to other areas on
the basis of altitude and azimuth and slope. This approach can be applied to regions
291
where the relationships between variables at individual sites and with reference
stations are known or can be reasonably well estimated. An approach that perhaps has
greater promise for large regions involves the use of satellite-derived estimates of
variables such as surface temperature, which can be combined with information about
vegetation type - also derived from satellite measurements - to obtain estimates of
soil moisture status (e.g., Nemani et al. 1993). Goward et al. (1994) outlined
procedures for estimating Q, from the Total Ozone Mapping Spectrometer (TOMS)
developed by Eck and Dye ( 1991), APAR from NDVI and atmospheric humidity and
air temperature from split-window observations using AVHRR. When these
techniques were tested using data from the OTTER Project, the results were
favourable enough to encourage further development and evaluation.
Applications of the E model

Applications to forestry. To date, there have been few attempts to use the model as a
predictive tool in forestry studies. Byrne et al. (1986) adopted the value of obtained
by Linder (1985) for P. radiata, and tested the ability of the model to calculate
above-ground dry matter accumulation by 10 stands of different ages growing in
Australia and New Zealand. To estimate APAR, they assumed that the fraction of
incident radiation intercepted by the canopies was proportional to the total crown area
of young trees projected normal to the horizontal. This projected crown area, in turn.
was assumed to be directly proportional to canopy biomass until canopy closure, after
which time Eg. (8) was used. A water balance modifier, based on the proportion of
available water in the root zone, was applied. There was little deviation from a I: I
relationship between estimated and simulated production for the 10 stands. Linder et
al. (1985) used the model to analyse the factors affecting productivity of Eucalyptus
stands, and West and Osler (1995) have used it to estimate biomass accumulation and
to analyse stand thinning experiments. They concluded: .... .the results suggested that a
common value of light-use efficiency can be applied to individual trees in the stand, so
that model 2 [essentially Eq. 2 of this paper] can be applied to describe either stand or
individual tree growth...The results also suggest a method by which growth response
to thinning may be predicted in such model systems for sites where water availability
is a major factor determining forest growth..."
Applications to terrestrial ecosystems. The most globally comprehensive efforts to

apply the model have been the studies on terrestrial ecosystem production by Potter
et al. (1993), Ruimy et al. (1994) and Prince and Goward (1996). Potter et al. (1993)
derived a single basic value for total global vegetation. They obtained a "maximum
efficiency " using NPP data collated for 17 sites by Raich et al. (1991) and calibrated
this using a modified least-squares technique relative to preliminary estimates of for
the sites concerned. The maximum value they obtained was 0.389 g C Mr 1
(approximately 0.8 g dry mass Mr' APAR; however, note the reservations expressed
earlier). This value was applied on a monthly basis, subject to constraints imposed by
water and temperature factors. Potter et al. were primarily concerned with seasonal
patterns of net ecosystem productivity across the globe; their model was driven by
NDVI and climatic data and included plant and soil respiration routines and the
turnover of soil C pools. Carbon fixation calculated by the E model was, therefore,
only one component of the exercise - albeit a very important one. The value of E used
by Potter et al. is low relative to some of the values discussed here, but it may well be
appropriate as an average ecosystem value. Ruimy et al. (1994) consider that better
results would most likely be obtained using different values appropriate to specific
ecosystems, but see Prince and Goward (1996). Lack of information about ecosystem
productivity is itself a constraint, and further development of this model will be
necessary before we are able to propose soundly based E values for different
ecosytems.
Ruimy et al. (1994) used a biome map of the world and, to the extent possible,
obtained average empirical values of E from the literature for each biome type. This
proved impossible for some biomes and led to very biased NPP values for biomes for
which there are few measurements. Their survey of the literature also provided
estimates of Pa and Pb and of relationships between APAR and NDVI. A series of
model runs was carried out using different vegetation maps, values of E, sources of
solar radiation data and APARlNDVI relationships. The values of E used for forested
biomes ranged from 0.3';]. to 1.57 g total (Pa and P b) MJ I APAR; the E value for tropical
I
managed forests was 2.72 g MJ .
Another global application of the E model (GLO-PEM) is under development at the
University of Maryland (Prince and Goward 1995). Not only solar radiation and
APAR but also environmental variables that affect E are derived from satellite data.
The potential Eg, in terms of gross production (Eq. 6), is calculated from first principles
so that no element of fitting or tuning is involved; this enables field observations of E
to be used for validation. First runs of GLO-PEM gave values of E ranging from 0-11
(tundra) to 0.71 (broadleaf evergreen forest). Most forested biomes had values
between 0.6 and 0.7.
An ambitious application of the E model is currently under consideration for use
with the MODIS instrument to be launched by NASA as part of the EOS (Earth
Observing System) program. It has been proposed that BIOME-BGC be used to
calculate NPP and derive "biome-specific" values of E. These, in conjunction with
global vegetation databases, MODIS-derived vegetation classes (evergreen needleleaf,
evergreen broadleaf, deciduous needleleaf, deciduous broadleaf, annual broadleaf and
grasses) and land cover algorithms, as well as a global climate database and estimates
of water balance and temperature obtained from appropriate meteorological data, are
to be used to produce weekly estimates of net photosynthesis for the global terrestrial
biosphere at l-km resolution (Running et al. 1994).
Conclusions
We conclude from this review and assessment that the radiation utilisation efficiency
factor, E, is more than the coefficient of an empirical relationship: it is a legitimate and
soundly based approach to forest and general ecosystem modelling. The model has the
great benefit of being extremely simple and offers the possibility of deriving the main
driving variable (APAR) from remote sensing.
293
Progress will lie in determining unconstrained values of and incorporating

environmental constraints, or modifiers, based on sound physical and physiological
processes. Incorporation of respiration estimates into the calculations is also likely to
improve the estimates of NPP. The equations used to define modifiers may have to be
designed to permit analyses on different time scales but should not become so
complex that they suffer from the limitations of detailed models for which the data
needs are so great that they usually cannot be met. Determination of unconstrained
values of should be done theoretically, empirically and using complex models to
promote the derivation of simple relationships with environmental modifiers (e.g.,
Running and Hunt 1993, McMurtrie et al. 1994 and others).
The use of eddy covariance measurements of CO 2 flux will undoubtedly provide
invaluable data on net ecosystem productivity (NEP), but quantitative estimation of
the contribution of heterotrophic respiration by soil microorganisms is difficult.
Similarly, measurement and modelling of canopy photosynthesis facilitates progress,
but the quantitative determination of autotrophic respiration by branches, stems and
roots of trees is difficult. Consequently, mundane but essential biomass measurerilents
and various methods of measuring and estimating biomass production, particularly
below ground, are essential for identification of values that can be used in forest
productivity models. In this respect, above-ground values of NPP remain very
valuable.
Foresters are interested in wood yield: a practical approach to forest yield
estimation is the use of the model, with modifiers calculated for particular sites, and
regression of the resulting biomass predictions against forestry measures of wood
production. The resulting semiempirical estimate is likely to be considerably more
accurate than conventional yield estimates based on concepts such as site quality and
complex statistical analyses that attempt to incorporate environmental variables.
Acknowledgments - McMurtrie and Medlyn acknowledge the support of the

Australian Research Council and the dedicated grants scheme of the Australian
National Greenhouse Advisory Committee.
References
Aber, J.D., Melillo, J.M., Nadelhoffer, KJ., McClaugherty, c.A. and Pastor, J.A. 1985. Fine
root turnover in forest ecosystems in relation to quantity and form of nitrogen availability:
A comparison of two methods. - Oecologia 66: 317-321.
Arkenbauer, T.J., Weiss, A., Sinclair, T.R. and Blum, A. 1994. In defense of radiation use
efficiency: A response to Demetriades-Shah et a1. (1992). - Agric. For. Meteoro1. 68:
221-227.
Bonan, G.B. 1993. Importance of leaf area index and forest type when estimating
photosynthesis in boreal forests. - Rem. Sens. Environ. 43: 303-314.
Byrne, G.P., Landsberg, U. and Benson, M.L. 1986. The relationship of above-ground dry
matter accumulation by Pinus radiata to intercepted solar radiation and soil water status. -
Agric. For. Meteoro1. 37: 63-73.
Cannell, M.G.R., Milne, R., Sheppard, LJ. and Unsworth, M.H. 1987. Radiation interception
and productivity of willow. - J. App1. Eco1. 24: 261-268.
Cannell, M.G.R., Sheppard, LJ. and Milne, R. 1988. Light use efficiency and woody biomass
production of poplar and willow. - Forestry 61: 123-136.
Charles-Edwards, D.A. 1982. Physiological Determinants of Crop Growth. - Academic Press,
Sydney, Australia. 161 pp.
Comins, H.N. and McMurtrie, R.E. 1993. Long-term response of nutrient-limited forests to
CO 2 enrichment: Equilibrium behavior of plant-soil models. - Ecol. Appl. 3: 666-681.
Curran, P.J., Dungan, J.L. and Gholz, HL 1992. Seasonal LAI of slash pine estimated by
Landsat TM. - Rem. Sens. Environ. 39: 3-13.
Demetriades-Shah, T.H., Fuchs, M., Kanemasu, E.T. and Flitcroft, I. 1992. A note of caution
concerning the relationship between cumulated intercepted solar radiation and crop growth.
- Agric. For. Meteorol. 58: 193-207.
Demetriades-Shah, T.H., Fuchs, M., Kanemasu, E.T. and Flitcroft, I.D. 1994. Further
discussions on the relationship between cumulated intercepted solar radiation and crop
growth. -Agric. For. Meteorol. 68: 231-242.
Eck, T.E and Dye, D.G. 1991. Satellite estimation of incident photosynthetically active
radiation using ultra-violet reflectance. - Rem. Sens. Environ. 38: 135-146.
Field, C.B. 1983. Allocating leaf nitrogen for the maximization of carbon gain: Leaf age as a
control on the allocation program. - Oecologia 56: 341-347.
Field, C.B. 1991. Ecological scaling of carbon gain to stress and resource availability. - In:
Mooney, H.A., Winner, W.E. and Pell, E.J. (eds.) Responses of Plants to Multiple Stresses.
Academic Press, New York, pp. 35-65.
Gallagher, J.N. and Biscoe, P.Y. 1978. Radiation absorption, growth and yield of cereals. -
Agric. Sci. (Cambridge) 91: 47-60.
Goward, S.N. 1989. Satellite bioclimatology. - J. Clim. 2: 710-720.
Goward, S.N. and Dye, D.G. 1987. Evaluating North American net primary productivity with
satellite observations. - Adv. Space Res. 7: 165-174.
Goward, S.N. and Huemmrich, K.E 1992. Vegetation canopy absorptance and the normalized
difference vegetation index: An assessment using the SAIL model. - Rem. Sens. Environ.
39: 119-140.
Gholz, HL, Nakane, K. and Shimoda, H. (eds). The Use of Remote Sensing in Modeling
241-272.
Goward, S.N., Huemmrich, K.E and Waring, R.H. 1993. Visible-near infrared spectral
reflectance of landscape components in Western Oregon. - Rem. Sens. Environ. 47:
190-203.
Goward, S.N., Tucker, CJ. and Dye, D.G. 1985. North American vegetation patterns observed
with the NOAA-7 advanced very high resolution radiometer. - Vegetatio 64: 3-14.
Goward, S.N., Waring, RH., Dye, D.G. and Yang, J. 1994. Ecological remote sensing at
OTTER: Satellite macroscale observations. - Ecol. Appl. 4: 322-342.
Hall, EG., Huemmrich, K.E and Goward, S.N. 1990. Use of narrow-band spectra to estimate
the fraction of absorbed photosynthetically active radiation. - Rem. Sens. Environ. 32:
47-54.
Harley, P.c., Thomas, R.B., Reynolds, J.E and Strain, B.R 1992. Modelling photosynthesis of
cotton grown in elevated CO 2 , - Plant Cell Environ. 15: 271-282.
Hunt, E.R 1994. Relationship between woody biomass and PAR conversion efficiency for
estimating net primary production from NDVI. - Int. J. Rem. Sens. 15: 1725-1730.
Hunt, E.R and Running, SJ. 1992. Simulated dry matter yields for aspen and spruce stands in
the North American boreal forest. - Can. J. Rem. Sens. 18: 126-133.
295
Jarvis, P.G. and Leverenz, J.W 1983. Productivity of temperate, deciduous and evergreen
forests. - In: Lange, O.L., Nobel, P.S., Osmond, C.B. and Ziegler, H. (eds). Encyclopedia
of Plant Physiology. Vol. 12D, Physiological Plant Ecology 4. Springer-Verlag, New York,
pp. 234-280.
Jarvis, P.G., James, G.B. and Landsberg, 1.1. 1976. Coniferous forest. - In: Monteith, J.L. (ed).
Vegetation and the Atmosphere. Vol. 2. Academic Press, London, pp. 171-240.
Kiniry, J.R. 1994. A note of caution concerning the paper by Demetriades-Shah et al. (1992).
- Agric. For. Meteorol. 68: 229-230.
Kira, T. 1975. Primary production of forests. - In: Photosynthesis and Productivity in Different
Environments. J.P. Cooper (ed). Cambridge University Press, Cambridge, UK, pp. 5-40.
Kira, T., Sinozaki, K. and Hozumi, K. 1969. Structure of forest canopies as related to their
productivity. - Plant Cell Physiol. 10: 129-142.
Kirschbaum, M.U.E, King, D.A., Comins, H.N., McMurtrie, R.E., Medlyn, B.E., Pongracic,
S., Murty, D., Keith, H., Raison, R.J., Khanna, P.K. and Sheriff, D.W 1994. Modelling
forest response to increasing CO2 concentration under nutrient-limited conditions. - Plant
Cell Environ. 17: 1081-1099.
Kumar, M. and Monteith, J.L. 1982. Remote sensing of plant growth. - In: Smith, H. (ed).
Plants and the Daylight Spectrum: Academic Press, London, pp. 133-144.
Landsberg, J.J. 1967. Responses of lucerne to different soil moisture regimes. - Exp. Agric. 3:
21-28.
Landsberg, J.J. 1986. Physiological Ecology of Forest Production. - Academic Press, London.
198 pp.
Landsberg, J.J. and Wright, L.L. 1989. Comparisons among Populus clones and intensive
culture conditions, using an energy conversion model. - For. Ecol. Manage. 27: 129-147.
Larcher, W 1983. Physiological Plant Ecology. - Springer-Verlag, Berlin. 303 pp.
Legg, B.J., Day, W., Lawlor, D.W. and Parkinson, K.J. 1979. The effects of drought on barley
growth: Models and measurements showing the relative importance of leaf area and
photosynthetic rate. - J. Agric. Sci. 92: 703-716.
Linder, S. 1985. Potential and actual production in Australian forest stands. - In: Landsberg,
J.J. and Parsons, W (eds). Research for Forest Management. CSIRO, Canberra, Australia,
pp. 11-35.
Linder, S., McMurtrie, R.E. and Landsberg, 1.1. 1985. Growth of eucalyptus: A mathematical
model applied to Eucalyptus globulus. - In: Tigerstedt, P.M.A., Puttonen, P. and Koski, V.
(eds). Crop Physiology of Forest Trees. Helsinki University Press, Helsinki, pp. 117-126.
Linder, S., Benson, M.L., Myers, B.J. and Raison, R.J. 1987. Canopy dynamics and growth of
Pinus radiata. Effects of irrigation and fertilization during a drought. - Can. J. For. Res. 17:
1157-1165.
Maier, C.A. and Teskey, R.O. 1992. Internal and external control of net photosynthesis and
stomatal conductance of mature eastern white pine (Pinus strobus). - Can. J. For. Res. 22:
1387-1394.
McMurtrie, R.E., Rook, D.A. and Kelliher, EM. 1990. Modelling the yield of Pinus radiata on
a site limited by water and nitrogen. - For. Ecol. Manage. 30: 381-413.
McMurtrie, R.E., Comins, H.N., Kirschbaum, M.U.E and Wang, Y-P. 1992. Modifying
existing forest growth models to take account of effects of elevated CO2 , - Aust. J. Bot. 40:
675-677.
McMurtrie, R.E., Gholz, HL, Linder, S. and Gower, S.T. 1994. Factors controlling the
productivity of pine stands: A model-based analysis. - Ecol. Bull. 43 (Copenhagen):
173-188.
Melillo, J.M., McGuire, A.D., Kicklighter, D.W., Moore, B., Vorosmarty, C.I. and Schloss, A
1993. Global climate change and terrestrial net primary production. - Nature 363: 34-240.
Melillo, J.M., Woodward, I.E, Salati, E. and Sinha, S.K. 1990. Effects on ecosystems. - In:
Houghton, J.T., Jenkins, G.J. and Ephraums, J.J. (eds). Climate Change. The IPCC
Assessment. Cambridge University Press, Cambridge, UK, pp. 283-310.
Monteith, J.L. 1977. Climate and the efficiency of crop production in Britain. - Phil. Trans.
Roy. Soc. London (Ser. B) 281: 277-294.
Monteith, J.L. 1981. Does light limit crop production? - In: Johnson, C.B. (ed). Physiological
Processes Limiting Plant Productivity. Butterworths, Toronto, pp. 23-38.
Monteith, J.L. 1994. Validity of the correlation between intercepted radiation and biomass. -
Agric. For. Meteorol. 68: 213-220.
Myers, B.I. 1988. Water stress integral - A link between short-term stress and long-term
growth. - Tree Physiol. 4: 315-232.
Nadelhoffer, K.I., Aber, J.D. and Melillo, J.M. 1985. Fine roots, net primary production, and
soil nitrogen availability: A new hypothesis. - Ecology 66: 1377-1390.
Nemani, R., Pierce, L. and Running, S.W. 1993. Developing satellite-derived estimates of
surface moisture status. - J. App!. Meteorol. 32: 548-557.
Pierce, L.L. and Running! S.W. 1988. Rapid estimation of coniferous forest leaf area index
using a portable integrating radiometer. - Ecology 69: 1762-1769.
Pook, E.W. 1986. Canopy dynamics of Eucalyptus maculata Hook. 4. Contrasting responses
to two severe droughts. - Aust. J. Bot. 34: 1-14.
Potter, e.S., Randerson, J.T., Field, C.B., Matson, P.A, Vitousek, P.M., Mooney, H.A. and
Klooster, S.A. 1993. Terrestrial ecosystem production: A process model based on global
satellite and surface data. - Glob. Biogeochem. Cycl. 7: 811-841.
Prince, S.D. 1991a. Satellite remote sensing of primary production: Comparison of results for
Sahelian grasslands 1981-1988. - Int. J. Rem. Sens. 12: 1301-1311.
Prince, S.D. 1991 b. A model of regional primary production for use with coarse-resolution
satellite data. - Int. J. Rem. Sens. 12: 1313-1330.
Prince, S.D. and Goward, S.N. 1996. Global net primary production: The remote sensing
approach. - J. Biogeog. (in press).
Raich, J.W. and Nadelhoffer, K.J. 1989. Below ground carbon allocation in forest ecosystems:
Global trends. - Ecology 70: 1346-1354.
Raich, J.w., Rastetter, E.B., Melillo, J.M., Kicklighter, D.W., Steudler, P.A., Peterson, B.I.,
Grace, AL., Moore, B. and Vorosmarty, e.J. 1991. Potential net primary production in
South America: Application of a global model. - Ecol. Appl. 1: 399-429.
Raison, R.I. and Myers, B.J. 1992. The Biology of Forest Growth experiment: Linking water
and nitrogen availability to the growth of Pinus radiata. - For. Ecol. Manage. 52: 279-308.
Rauner, J.L. 1976. Deciduous Forests. - In: Monteith, J.L. (ed). Vegetation and the
Atmosphere. Vol. 2. Academic Press, London, pp. 241-264.
Ruimy, A., Saugier, B. and Dedieu, G. 1994. Methodology for the estimation of net primary
production from remotely sensed data. - J. Geophys. Res. 99: 5263-5283.
Running, S.W. 1994. Testing FOREST-BGC ecosystem process simulations across a climatic
gradient in Oregon. - Ecol. Appl. 4: 238-247.
Running, S.w. and Coughlan, J.e. 1988. A general model of forest ecosystem processes for
regional applications. 1. Hydrologic balance, canopy gas exchange and primary production
processes. - Ecol. Model. 42: 125-154.
Running, S.w. and Gower, S.T. 1991. FOREST-BGC, a general model of forest ecosystem
processes for regional applications. 2. Dynamic carbon allocation and nitrogen budgets. -
Tree Physiol. 9: 147-160.
297
Running, S.W. and Hunt, E.R 1993. Generalization of a forest ecosystem process model for
other biomes, BIOME-BGC, and an application for global scale models. - In: Ehleringer,
J.R. and Field, C.B. (eds). Scaling Physiological Processes: Leafto Globe. Academic Press,
San Diego, CA, pp. 141-158.
Running S.w., Justice, C.O., Salomonson, v., Hall, D., Barker, J., Kaufman, Y.J., Strahler,
A.H., Huete, A.R., Muller, J-P., Vanderbuilt, v., Wan, Z.M., Teillet, P. and Carnegie, D.
1994. Terrestrial remote sensing science and algorithms planned for EOSIMODIS. - Int. J.
Rem. Sens. 15: 3587-3620.
Running, S.w., Nemani, R.R. and Hungerford, R.D. 1987. Extrapolation of synoptic
meteorological data in mountainous terrain and its use for simulating forest evapotranspiration
and photosynthesis. - Can. J. For. Res. 17: 472-483.
Runyon, J., Waring, R.H., Goward, S.N. and Welles, J.M. 1994. Environmental limits on net
primary production and light-use efficiency across the Oregon transect. - Eco!. App!. 4:
226-237.
Russell, G., Jarvis, P.G. and Monteith, J.L. 1989. Absorption of radiation by canopies and stand
growth. - In: Russell, G., Marshall, B. and Jarvis, P.G. (eds). Plant Canopies: Their Growth,
Form and Function. Cambridge University Press, Cambridge, UK, pp. 21-39.
Ryan, M.G. and Waring, R.H. 1994. Maintenance respiration and stand development in a
subalpine lodgepole pine forest...:. Ecology 73: 2100-2108.
Sabatier-Tarrago, C. 1989. "Production de taillis de Chataignier (Catanea sativa, Mill.) en
relation avec les characteristiques stationelles." - Ph.D. thesis, Univ. Paris-Sud, Orsay. 250
pp.
Saldarriaga, J.G. and Luxmoore, RJ. 1991. Solar energy conversion efficiencies during
succession of a tropical rainforest in Amazonia. - J. Trop. Eco1. 7: 233-242.
Saldarriaga, J.D., West, D.C., Tharp, M.L. and Uhl, C. 1988. Long-term chronosequences of
forest succession in the upper Rio Negro of Colombia and Venezuela. - J. Eco!. 76:
938-958.
Schulze, E-D., Kelliher, EM., Korner, c., Lloyd, J. and Leuning, R 1994. Relationship among
maximum stomatal conductance, ecosystem surface conductance, carbon assimilation rate,
and plant nitrogen nutrition: A global ecology scaling exercise. - Ann. Rev. Eco!. Syst. 25:
629-660.
Sellers, P:J. 1985. Canopy reflectance, photosynthesis and transpiration. -lnt. J. Rem. Sens. 6:
1335-1372.
Sellers, PJ., Berry, J.A., Collatz, G.J., Field, C.B. and Hall, EG. 1992. Canopy reflectance,
photosynthesis, and transpiration. 3. A reanalysis using improved leaf models and a new
canopy integration scheme. - Rem. Sens. Environ. 42: 187-216.
Sinclair, T.R. and Horie, T. 1989. Leaf nitrogen, photosynthesis and crop radiation efficiency:
A review. - Crop Sci. 29: 90-98.
Sinclair, T.R and Shiraiwa, T. 1993. Soybean radiation use efficiency as influenced by non-
uniform specific leaf nitrogen distribution and diffuse radiation. - Crop Sci. 33: 808-812.
Spanner, M.A., Pierce, L.L., Running, S.w. and Peterson, D.L. 1990a. The seasonality of
Spanner, M.A., Pierce, L.L., Running, S.w. and Peterson, D.L. 1990b. Remote sensing of
temperate coniferous forest leaf area index. The influence of canopy closure, understory
vegetation and background reflectance. - Int. J. Rem. Sens. II: 95-111.
Spanner, M.A, Johnson, L., Miller, J., McCreight, R., Freemantle, J., Runyon, J. and Gong, P.
1994. Remote sensing of seasonal leaf area index across the Oregon transect. - Eco1. App!.
4: 258-271.
Stigter, CJ. and Musabilha, Y.M.M. 1982. The conservative ratio of photosynthetically active
radiation to total radiation in the tropics. - J. Appl. Ecol. 19: 853-858.
Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring
Tucker, c.J., Vanpraet, C.L., Sharman, M.J. and Van Ittersum, G. 1985. Satellite remote
sensing of total herbaceous biomass production in the Senegalese Sahel: 1980-1984.-
Vogt, K. 1991. Carbon budgets of temperate forest ecosystems. - Tree Physiol. 9: 69-86.
Wang, YP. and Jarvis, P.G. 1990. Description and validation of an array model- MAESTRO.
-Agric. For. Meteorol. 51: 257-280.
Wang, YP., Jarvis, P.G. and Taylor, C.M.A. 1991. PAR absorption and its relation to above-
ground dry matter production of Sitka spruce. - J. Appl. Ecol. 28: 547-560.
Wang, Y-P., McMurtrie, RE. and Landsberg, U. 1992. Modelling canopy photosynthetic
productivity. - In: Baker, N.R. and Thomas, H. (eds). Crop Photosynthesis: Spatial and
Temporal Determinants. 5. Elsevier Science Publishers, Amsterdam, pp. 43-67.
13
Waring, RH. and Silvester, W.B. 1994. Variation in foliar 8 C values within the tree crowns
of Pinus radiata. - Tree Physiol. 14: 1203-1213.
Waring, RH., Law, B.E.,. Goulden, M.L., Bassow, S.L., McCreight, RW., Wofsy, S.c. and
Bazzaz, EA. 1995. Scaling gross ecosystem production at Harvard Forest with remote
sensing: A comparison of estimates from a constrained quantum-use efficiency model and
eddy correlation. - Plant Cell Environ. 18: 1201-1213.
West, P.w. and Osler, G.H.R. 1995. Growth responses to thinning and its relation to site
resources in Eucalyptus regnans E Muell. (in press).
Yoder, BJ. and Waring, RH. 1994. The normalized difference vegetation index of small
Douglas-fir canopies with varying chlorophyll concentrations. - Rem. Sens. Environ. 49:
81-91.
Color Plates
301
A.
open spruce bog

hemlock/hardwood hardwood bog
mixed softwood mixed hardwood
Norway spruce hardwood/conifer mix 1
white pine hardwood/conifer mix 2
D red pine
Chapt. 3, Plate J. PnET input parameters derived from AVIRIS data: (a) species
derived from AVIRIS classification, (b) N concentration (%) and (c) foliar
biomass (kg ha- l ).
B.
1.0-1.29%
1.3-1.59%
1.6-1.89%
1.9-2.19%
2.2-2.49%
303
c.
D 0-1,999 8,000-9,999
2,000-3,999 10,000-11,999
4,000-5,999 12,000-13,999
6,000-7,999 14,000+
A.
1 - 200
201 - 400
401 - 600
601 - 800
Chapt. 3, Plate 2. Net photosynthesis (g C m,2 yr,l) predicted from PnET using (a)
GIS species data and average field data for foliar N concentration and foliar
biomass and (b) species, foliar biomass and foliar N concentration derived from
AVIRIS data.
305
B.
1 - 200
201 - 400
401 600
601800
801 1,000
1,001 - 1,200
1,200+
Green: Deciduous broadleaf forest.

Yellow: Pine forest.
Red: Japanese cedar plantation.
Blue: Clearcut area.
Magenta: Cultivated land area.
Chapt. 7, Plate 1. Vegetation map based on corrected Landsat TM data obtained

on 8 May 1987.
"
, j'
}
.' :t ' '"
.... yo ... "
,' ...... '~":: . ",t'o!. "

.
'. iJ '
. 4,.' 't',,1-'
,..":' # . .,..
.'''- . ,
. . , .. ,,,~_ ..
(.,,,,: '~1,'('i
\.~' ~/~i/'" \~~~ \.~/.'
~ C"

~ . '.:....~
~'..
.' .. , .'t.
.
."~ ;', 1'--'"
'f'
\ ....
~:t,.. ~
' . . 't~" ' ,.,.
't. W. ,)
.. '
;.it, ~;"'S",,-I.'~''''''
. J
, ' p . ' .<
'-.'~ '" ) ' ;~,. \ ,.,,~
" ,
... .,' ,.... :.

. '
; J::'J
S., '
.' tJII ,i;-:
(8)
( )
Chapt, 7, Plate 2. Biomass maps based on relationships between vegetation

indices and above-ground biomass in the deciduous broadleaf forest (A), pine
forest (B) and Japanese cedar plantation (C); total biomass within
the forest area is shown in (D).
Blue: 0-50; Magenta: 50-100; Red: 100-150; Yellow: 150-200;
Green: 200-250; Light blue: 250-300 t ha).
307
I~l

Chapt. 9, Plate I. Central Oregon Cascade Mountain Range study area showing
land ownership and site-class distributions.
Carbon Rux
19721991
Mgha'yr'
Sinks
-4.0010-5.00
-2.0010-400
-0.2510 -2.00
Stable
-02510 025
SourCjt
025 0 2.00
2.00 10 4.00
400 0 8.00
o 20 8001018.00
I I Other
Non oreal
KIlometers
Chapt. 9, Plate 2. Total C flux for the area in Figure 3. Positive values represent a
net release of C to the atmosphere (source) and negative values represent a
removal of C from the atmosphere (sink).
309
-180 - t 0 -120 - 0 -60 - 0 0 0 0 0 120 I 0 180
JO
10
-10
-JO
-50
MJ m- 2
0 100 200 300 100
Chapt. 10, Plate 1. Monthly incident PAR (PARi) at the earth's snrface, July
1992, derived from Total Ozone Mapping Spectrometer (TOMS) ultraviolet
observations. Measurements have been validated for a few sites around the
globe; comparison to date indicates excellent agreement between estimates
based on these satellite observations and estimates resulting from ground
measurements.
Chapt. 10, Plate 2. Visible wavelength minimum brightness image for six
months (March to October 1987), based on AVHRR 8-km NOAA/NASA
Pathfinder data. Differences may be evaluated through comparison with NDVI
(Fig. 1). This minimum visible reflectance reveals an additional aspect of
vegetation related to within-canopy shadows, cast predominately by the woody
portion. The geography of the observed brightness variations is in relative
accord with known patterns of forest occurrence and appears to be reasonably
well related to known patterns of above-ground biomass (Fig. 7).
Index
Abies balsamea, 115
absorbed photosynthetically active radiation, APAR, 173, 178,249,262,273,275,
277-281,283-292,294
Acer rubrum, red maple, 62
Advanced Solid-state Array Spectrometer, ASAS, 182, 184, 198,200,204,211
Advanced Very High Resolution Radiometer, AVHRR, 21, 62,105,130,137,
157-158,178-180, 183-184, 194-197,200,211-213,215-216,241,243-245,
247-249,252-268,270-271,278,289,291,294,297,310
aerodynamic resistance, 143
aerosol, 198,246,248,256,260
Affine transformation, 18, 163
Africa, 278
agriculture, 77, 79, 115, 155,196, 198,222-223,247,251,259,263,265,267,273,
282,284
aucraft,6-7, 77,100,102-104,174-176,183-185,194,196,198,200
AIRSAR, 115-116, 118, 127, 184
Auborne Visible Infrared Imaging Spectrometer, AVIRIS, 3, 6-7, 9, 11, 15-19,21,
61,63-67,70-71,184,198-199,202-203,207,213,301,304
Alaska, 128, 132,217,289
albedo, 48-49, 95-97, 103, 106, 170, 182,268,270
allocation, 4, 19, 143, 145, 151, 190-193,204,215,281-282,289,294,296
allometry, 19, 163, 171, 181-183,203-205,283
Alnus rubra, red alder, 179, 186
altitude, 161, 163, 178, 184-185, 196, 198,248,289-290
Amazon, 113, 127, 130-131,256,297
amino acid, 177, 182, 186,213
angular distribution, 26-27, 29, 34,42,46,48-49,51
architecture, 53-54, 58-59, 80, 200, 217
Arctostaphylous patula, greenleaf manzanita, 200
atmospheric correction, 63, 76, 95, 102-103, 198-200,211,217,266,271
Australia, 75-78,83,87-88,93, 104-108, 183,264,270,274,283,285-286,291,
294-295
azimuth, 8, 28-30, 33-37, 42, 47,108, 162-163, 182,290
backscatter, 38-40, 57, 103, 108, 115-116, 118, 133, 178,200,206,212,216,246,
264,266
bark, 48, 246, 280
basal area, 5, 51,57, 81,86,98, 103, 112, 121, 124, 183,206
Beer's Law, 121, 152, 162
Betula allegeniensis, 115
Betula papyrifera, 115
Bidirectional Reflectance Distribution Function, BRDF, 75-76,82,86-87,93,
99-104, 106, 204
biochemistry, 3-5,10-19,21,72,137,156,171,173,207-209,212-213,215,281
311
biomass, 8, 11,19,21,61-67,69-71,75,77,79,92,99, 10 3, 108, 112, 115, 118,

122-127,130-131,133,137,141,151-152,154,159-160, 163, 165-171, 173,
176-178,180,182,186, 188, 191-193,202-208,210,212-213,217,235-236,
241-242,245,247,256,258-260,263-264,266,276,279-280,284-287,289,
291,293-294,296,298,301,304,306,310
BIO~SS,279,285-287,289
BIOME-BGC, 72, 287-289, 292, 297
biosphere, 129, 156-158,211,220,241-242,244-245,261,263,268-269,271,292
bog, 63-64,70, 156
boreal, 71,114,118,128,131,155,251,270,293-294
BOREAS, 261, 270
boundary layer, 57, 179
broadleaf,25-26, 35,43,57-58,62, 64,67-70, 121, 124, 141, 159, 161, 163-164,
166-171,194,247,292,306
budbreak, budburst, 183, 188
burn, 5, Ill, 198,223,~26,233,266
calibration, 21, 63-64, 67, 70,114,137,141,170,194,196,199,204,256
California, 59, 129,212,214,217,289
Calluna,44
Canada, 104, 183, 185,210,236,251,259,267,287
canopy, 3-7, 10-12, 16-17, 19-21,23-26,33,35,41-43,46-49,51-52,56-65,
67-68,70-72,78-79,81-83,86,89,95,99-100,105-108,111-115, 119, 121,
128-130,133,137,141,143,145,156-158,160-161, 163-164, 166-167, 169,
171,173,178-179,181-183,188-189,191-192,194-195, 199-202,204,
207-210,212-217,241,243,245-249,252,254-256,259,262-268,270-271,
273,275-286,288-291,293-298
carbohydrate, 177,214,277,281
carbon, 3-4, 20, 61-62, 72, 130, 132, 136, 143, 155, 157, 171, 173,211,214-217,
219-220,224-225,228,230-231,233-237,242,263-266,269-270,275,292,
294,296-298
carbon dioxide, CO 2, 52, 71-72, Ill, 145, 152, 155, 157,215,220,224,234,
236-237,242,263-264,266,271,275-277,279-280,282,284-285,287,289,
293-295
Carpinus laxiflora, 161
Castanea crenata, 161
cellulose, 3, 6, 10-14, 16-18,21, 182, 188,209,217
Chamaecyparis obtusa, cypress, 161
chemistry, 20-21, 62-63, 72,112,133,158,173,177-178,182,187-188,207-211,
214,217,220,234,242
chlorophyll, 3, 6,10-14,18-21,57,62,160,177,182,186-188,198-199,202-204,
208-209,212,214,247,263,277,289,298
classification, 61, 64, 67, 80,103,110,112,130,132,135,137,158-160,163,
165-166,169-171,178,212,216,229,262,264,267,271,301
clearcut,52, 111-112, 126, 128, 159, 163, 166, 169,225-226,228,306
313
climate, 19,21,59,65-66,75,77,79,92, 110, 112, 114, 132, 135, 137-139, 141,

150-151,154,156,158,173-179,186,188-191,193-194, 210, 212, 214-215,
220-222,224,228,236-237,242,246,253,262-263,265-266,268-271,278,
284,288,290-292,296
cloud, 119, 175, 178, 180, 183, 186, 194, 196,206,245-246,252,256,263-264,
266,268,270,277
clump, 25, 56, 59, 80, 91, 106, 113, 194
cluster, 25, 43, 46-47,84, 178
combustion, 220
Compact Airborne Spectrographic Imager, CASI, 185, 198-199,202,207,212
competition, 78-81, 92,105-107, 130, 166,223
conductance, 143, 145, 180, 189,254,282,285,287,295,297
conifer, 14,20-21,23,25-28,32,35-36,42-43,46,48,52,56, 58-59, 62-64,
66-67,72,106-107,128,131-132,141,143,154,158,165,171, 173, 175-176,
178-179,186,188,191, 193-194, 196,200,202-204,211-217,223,229,
236-237,264,267-268,271,288-290,295-297
crop, 4-5, 8, 19-21,77,79, 105; 170,214,216,249,259,263,266-268,270,273,
277-279,282-283,287,294-298
crown, 8, 10-11,23,25-26,41-48,50,56,59,75,78,80-86,88-91,93-94,
96-100, 106, 108, 114, 120, 133, 167, 169, 178, 182,202-205,214,217,291,
298
Cryptomeriajaponica, Japanese cedar, 159, 161, 163-170,306
cuticle, 27, 32-33
cuvette, 54
database, 66,104,138-139,148,150-152,157,177,224-225,292
debris, 223-225, 234-235
deciduous, 51-52, 63, 70, 72,112,159,161,163-171,179,183,191,211,223,251,
267,278,280-281,283,288,292,295-296,306
decomposition, 16, 19,21,143,177,181,191,193,211,214,221,226,228,231,
235,282,287
Defense Mapping Agency, DMA, 139
deforestation, 113, 131, 133
deposition, 72,80, 143,215
desert, 110, 131, 175, 183,203,242,248,251,254,256-257,264-265
detritus, 219, 221, 224, 228, 230-231, 234
Differential Vegetation Index, DVI, 159-160, 163, 167-170
digital number, DN, 28-30, 38, 159, 163-165
Digital Elevation Model, DEM, 139, 149, 152, 155, 158, 162-163
Digital Terrain Model, DTM, 160, 163, 169
disturbance, 77-80, 91, 99,110-112,114,141,176,221,224-226,229,233
diversity, 67,105,110,112,129,218,244
drought, 5, 19, 152, 154, 176, 179, 183, 188, 192, 198,210,245,253,256,284,
295-296
DSVM,149
dwell time, 16
Earth Observing System, EOS, 102, 104, 107, 132,260-262,269,292,297

ecological field theory, 81, 108
ecophysiology, 128, 180,265,275
ecotone, 111, 129, 131
EDiS,149
eddy correlation, eddy covariance, 65, 71, 293, 298
elevation, 5, 47, 56,115,139-140,145,149,152,155,158,162-163,198,222-223
end-member analysis, 88-89
energy balance, 245, 268
enzyme, 281-282
epsilon, , 273-285, 287-288, 291-293
erosion, 77, 102, 266
Estonia, 23, 37, 58, 106
Eucalyptus, eucalyptus, 83, 92, 278, 283, 291, 295-296, 298
evaporation, 121-122, 143, 154, 156,255,263,269-270,281
evapotranspiration, 71,107,131,137,150,157-158,173,177,181,212,216,236,
254,265-266,268-269,289,297
evergreen, 71-72,112,158,186,211,217,237,251,267,283,292,295
extinction coefficient, 35, 105, 145, 183,279,289
fertility, 44, 119,211
fertilizer, 3-5, 8, 10-11, 13-15, 18-20,44,111, 119, 173, 177, 179, 186, 188, 191,
193,211,216,224,282-284,290,295
Festuca idahoensis, Idaho fescue, 200
field capacity, 119, 139
FTIFE, 212, 253,255,261-262, 265
Finland, 54
fire, 79,92,99,102,111,226,233-234
flood, 77, 213, 289
Florida, 3-5,9,18,20-21,72,109,156,212,285-287,290
foliage area index, 94
FOREST-BGC, 141, 143-146, 148-150, 152, 157, 177, 180-182, 186, 188-193,
210,215,287-288,296
forest floor, 25,46,48,206,224-225,251, 267
forest management, 234, 236, 295
forestry, 56-57, 59,77, 79,111,212,216,219,236,282,291,293-294
fossil fuel, 220, 236
fractal, 24, 113, 131
freeze, 11, 14, 112,143,152,179,183,188,251,281
frost, 176, 285
G-function, 25-26, 30
/lex glabra, gallberry, 5
gap, 35, 39,41-43,46, 75,79-80,83-84,90,95,98-99,109-115,119,121,
128-129,131-133,183
geographic information system, GIS, 57, 64, 66,129,138,155-157,235,304
geometrical cross section, 28-30, 35-36
315
geometric-optical model, 58, 75, 94,105-107,137,156,178,181-182,202,

204-205,213,267
GLO-PEM, 280, 292
GO model, 75-76, 79,86,89-96, 103
Goddard Space Flight Center, 109, 114, 131, 262, 269
GOES, 183-184, 234
grass,75-78,80,88-89,92-95, 102, 130, 157, 194,235,242,247-248,264,271,
288,292,296
grazing, 77, 79, 92-93, 102,264,280
greenness, 77, 103
gross primary production, 269, 276
growing season, 7, 71, 121, 188,256,278,281,285,287
growing-degree day, 65
growth form, 62, 80,211,243,297
H.J. Andrews Experimental Forest, 114, 176,234
HAPEX,261
hardwood,62-67,70, 72,114,214,263
Harvard Forest, 61-68, 70, 298
harvest, 5, 102, 111, 114, 121-122, 124-125, 127, 194,221-223,225-228,
230-231,233-234
hemlock, 63-64,114-115,128,179,183,189,214,222,235
herbicide, 223
heterogeneity, 16,89,109-114,118,121,126-130,141,150,219-221,224,228,
233,267
hotspot, 32,39-43,51,58, 82, 86,99-100, 104-106,213
hydrology, 154-156
ice, 14,260
illumination, 14-16,27,31-35,37-43,53-54,87,90,145,178,200, 202, 204-205,
249,280
inclination, 28-30, 33-37, 42, 46-48
Infrared Intelligent Spectroradiometer, IRIS, 6-7,11, 14, 19,21,243
insect, 79
invasion, 76-77, 92, 97, 108
IPAR, 176, 178, 186, 188-189, 194-195,201
irradiance, 8,27,32,52-54,59,245-246,269
Jet Propulsion Lab, JPL, 9, 17-18,21,71, 115, 129
Juniperus occidentalis, western juniper, 183, 203
Kansas, 255
Japan, 59, 130, 159-161, 171,241,259,262
LAI-2000, 46, 58-60
land use, 77, 79, 113, 127, 130, 132, 161,217,219-220,222,224-225,236,266,
275
LANDCARB, 221, 224-228, 230
Landsat, 5-6, 8, 10,20,88,93, 105, 108, 128, 130,133,141,152,159-160,
162-165,169-171, 177-178, 185,212,216,230,235,243,247,259,261,
264-265,267,289-290,294,306
landscape, 71, 73, 77,107,109-114,122,127,129-133,141,149-150,155-156,

173-174, 176, 196,204,210,212,223-224,228,235-236,243,247,251,253,
264-266,288,290,294
leaf angle distribution, 79, 247-249, 290
leaf area index, LAI, 3-5, 7-8, 10, 19-21,24,46,48-49,57-60,62-63,67,70-72,
75,77,79,103, 112,132,137-141,143,145,150-152,154,156-157,171,173,
176-178,181,183,186-187,191-194,198-204,208-210, 212-213, 215-216,
226,243,247-248,262-263,267-268,270,273,276,293-294,296-297
life form, 242, 288
light, 14,21-22,25,38,47,52-55,57-58,60,65,75,80-81,83,91-92, 113, 119,
123,126,145,152,157,160, 183,185-186,207,214,217,248,254,256-258,
262,267,271-272,282-283,289,294,296,306
light-use efficiency, 173, 186, 188-189, 196, 198,210,216,291,297
lignin, 3, 6,10-14,16-18,21,63-64,72,178,182,187-188,208-209,214,217
litter, litterfall, 8,19,21,63,80,92-94,137,143,177,180-181,186,193,211,214,
226-227,246,248-249,251,276,280,290
livestock, 77
logging, 81, 226-227
macronutrient, 5
MAESTRO, 282, 284, 286-287, 289, 298
Maine, 109, 114, 119, 130
management, 21, 51-52, 56, 77,105,129,156,183,215-217,222,226,229,
233-237,274,295
mapping,3,5-6, 10,21,47,62,64,72,80,97-98,102-108,110,131,138-139,
141,152,156,158-163,166,169-171,175,177,184,211, 215, 221, 224,
228-230,233,236,242,244-247,267-268,271,289-292,306,309
Massachusetts, 57, 61-62
mesophyll, 32, 145
meteorology, 102, 158, 173, 180, 188, 190, 196-197,216,267,290,292,297
microclimate, 128,132-133,141,145,151,156-157,215,222
micronutrient, 5
mineralization, 143, 177, 188,212,282
Minnaert's Law, 162
Minnesota, 114
Moderate Resolution Imaging Spectrometer, MODIS, 102-104, 107,259-261,269,
292,297
Montana, 135, 139, 141, 150-1.53, 155-157, 196,289
mortality, 112-113, 121-122, 128, 192-193,280
MTCLIM, 141, 145, 149-150, 156, 180, 182, 189,212
Multispectral Scanner, MSS, 88,93,105,108,130,170-171,185,230,247,264
NASA, 9, 17, 19,21,61-63,71,109,114,128-129,131,135,155,173,183-184,
210,214,216, 234,244,248,253-255,257,262,267-269,292,310
needleleaf, 62, 67-68, 70,121,247,292
net primary production, NPP, 4, 71, Ill, 138, 173, 176, 186, 188-190,212,216,
245,247,249, 251,265-266,269,273,275-278,280-285,289-294,296-297
317
New Zelliand, 133,217,283,285-287,291

nitrogen, 3, 6, 11-12, 16-17,21,61-63,72, 133, 138, 157-158, 173, 182, 187, 193,
208,211,214-217,282,293-297
NOAJ\,5,21, 137, 157, 178,180,184,243-244,248,251,253-254,257,260,262,
267,270,272,278,289,310
Normalized Difference Vegetation Index, NDVI, 3, 8, 10, 19,62,64,67,69,71,
95-97,103,157,159-160,163,166-170,178-180,188,194-196, 198,200--201,
203,208,215,248, 251,254-255,257,263,265-268,271,278,289-292,294,
298,310
North America, 251, 271
nutrient, 4-5,10--12,19-21,72,75,80--81,83,92,110,113,173,176-178,180,
214,216-217,282,294-295
nutrition, 4, 19,21,211,273,282,284,286,288,297
ocean, 21, 61, 175, 196,220,236,242-243,253
old-growth,98, 112, 114, 128, 179, 186, 189,204,221,225-231,234-236
optic, 14,21,23-24,26-28,32,35-36,43,46-48,53,56-60,67,71,79, 100,
104-107,115,136,156,164;180--182, 196, 198,204,206,215,247-249,256,
259,262,269,290
Oregon, 21, 107-108, 114, 132-133, 173-179, 183, 189, 194, 199,203-204,206,
210--219,221-222,224,234-236,253,258,265,268-271,284,289-290,294,
296-297,307
011l.ER, 173-174, 176-179, 183, 186, 188, 190--191, 194, 196, 198,202,204,207,
210,213-214,216,253,255,258,261,265,271,284,288-289,291,294
organic matter, 5, 20, 211
orientation, 14,24-25,42,90--91,109,115-116,118,121-123,126-128,169,178,
206,248,265
Pacific Northwest, 21, 112-114, 127, 132, 139, 155, 158, 176,212,215,217,219,
221,234-235,237
partitioning, 20, 24,131,136,264,283
pasture, 161, 222-223
peatiand,229
Penman-Monteith Equation, 143
penumbra, 27,56-57
phase function, 23-24, 26, 31-34, 36, 38-39,41--42,49
phenology,7,56,96, 112, 166,264,267
phosphorus, 11,21,176,211,216
photon flux density, PFD, 277, 279-280, 283
photosynthesis, 11, 19,21,23,25,35,48,52-59,61-62,65-66,69-72,83,
106-107,131,137,141,143,145,150--153,157-158,171,173,176-180,
188-189,191-192, 194,212,214-217,236,241-242,244-248,251-252,259,
262-264,266,268-273,275-290,292-295,297-298,304
photosynthetically active radiation, PAR, 42, 48--49, 52-53, 62, 173, 176, 178, 181,
183,186,188-190, 194-195, 198,212,241,244-249,251-252,259,264-266,
269,272-273,276-280,284,287-288,290,294,298,309
phytomass,79, 137,217,264
Picea abies, Norway spruce, 35, 49, 63-64, 67

Picea mariana, black spruce, 63-64, 115
Picea rubens, red spruce, 115
Picea sitchensis, Sitka spruce, 59, 186,281-283,298
pigment, 7, 160,202,247-248,260,271
pine, 3-5, 10-12, 14-21,33,35-39,44,46,48-52,55,59-60,63-67,72, 156-157,
159,161,163-171,179, 183, 194,200-202,211-214,266,282,285-287,290,
294-295,297,306
Pinus densiflora, 161
Pinus elliottii, slash pine, 3-5, 10-12, 14-21,209,212,266,290,294
Pinus ponderosa, ponderosa pine, 157, 179, 183, 194,200-202,212
Pinus radiata, Monterey pine, 278, 281-282, 291, 293, 295-296, 298
Pinus resinosa, red pine, 63-64, 66, 171
Pinus strobus, white pine, 63-64, 281, 295
Pinus sylvestris, Scots pine, 33, 35-39, 44, 46, 49, 51, 55, 59-60,157,214,281
Pinus taeda, loblolly pine, 21, 211, 281
pixel, 16, 18,64-67,69--70,76,83-88,96,98-100, 102-103, 109, 112, 115-118,
126-127,136-137,162-163,169,198,202,204,252,260
planimeter, 28
plantation, 3-5, 20-21, 72,156,159,161,163-170,212,221,275,278,284,306
PnET-Day, 65-66, 70
polarization, 57-58, 60, 115,206,212,263
Populus, 115,283,287,295
Populus grandidentata, 115
Populus tremuloides, 115
precipitation, 5,119,139,143,145,148,152,154,175,180
primary production, 3-4, 71-72, Ill, 138, 157, 173, 176, 189,212,215-216,241,
245,261,263,265-266,269,271,273,276,294-297
protein, 11, 14, 16,209,211
Pseudotsuga menzeisii, Douglas-fir, 46, 114, 128, 179, 183, 186,204,216,221-224,
234-236,298
psychrometric constant, 143
Purshia tridentata, bitterbrush, 200-201
quantum yield, 264, 280
Quercus rubra, red oak, 62
Quercus serrata, 161
radar, 76,86,103,109,114-115,132, 178,184,206-207,212-214,256,264,266
radiation
absorbed, 7,14,16-18,21,24-25,27,47-48,51-53,64,67,70,136, 152, 160,
173,177-178, 194, 198-199,202-204,207-209,242,245-249,255,259,
262,264-265,273,275-280,283-286,288,290,294,297-298
blue, 14,37-39,41, 183,260
diffuse, 32,41,47,49, 57, 119, 121-122, 162, 178, 183,209,279,286,297
direct, 37,41,47,49,52-53, 156,286
green, 14, 37-41, 209, 246
319
infrared, 3, 6-7, 22,39,58,61-63, 71-72,89,141,160,171,184,202,207,209,

252-254,259-260,263,265-266,269,271-273,294,298
intercepted, 21, 25, 35,43,46-50,58,62,164,176,178,181,183,186,188-190,
194-195,213,264,266,276,279-280,283,288,291,293-294,296
rrllcrowave, 108, 115, 133,216-217,252,256,259,268
near-infrared, 7,14,27,32,37-38,40-43,50-51,76,91,95,137,198,202-203,
211,243-244,246-247,251,256,260,273,278
penetration, 24-25, 46-47, 49, 59, 80, 106, 145
red, 3, 7,14,18,20,37-39,41,50-52,62-64,66,84,94-96,101,137, 160, 181,
199-200,202-203,208-209,212-214,260,271,277-278,298,306
reflected, 3, 7,11,14-18,20-22,24-25,27,31-34,38,40-43,47-52,56-59,
62-64,71-72,75-76,79,84,88-96,100-102,105-108,111,118,137,141,
156,160,162,171,173,176-178,180-181,186, 194, 198-204,207-209,
211-214,216,244,246-247,249,251,256,258-260,262-268,270-273,
277-278,284,289-290,294,297,310
refracted, 32-33
scattered, 7,24-27,31-32,34,36-43,46,49-50,57,59-60, 79,83,88-89,91,
103,105,107-108,167,170,177,194, 196, 199-200,202-203,205-206,
209,247,258,267,271,279,286
short-wave, 132, 139, 143, 145, 156, 180,202,245,259,265,276,283
thermal, TIR, 6, 58, 76, 86, 121-122, 127-128, 132, 137, 162, 179-180, 184-185,
196,212,252-255,260,263,265-266,269-270
transmitted, 22, 25, 27, 31-32, 34-36, 38, 41, 46-48, 57, 79, 115, 145, 148, 162,
177,183,195,204,209,213,272,277
ultraviolet, 194,246,264,294,309
utilizable, 189,285-288
visible, 3, 6-7, 31-33,41-42,61,63,71-72,76,81,86-87,95,99,103,141,
184-185, 194,202-203,211,243-244,246-247,251,256,258,260,263,
273, 278, 310
radiative transfer, 23-26, 28, 31,41,48-49,56-59,63,106,173,176,178,203,
209-210,214,247,267-268
radiation utilization efficiency, 213, 267, 273, 275, 282, 292
rainfall, 5, 76, 78-79, 93,103,161,175,183,284
red edge, 3, 18,20,202-203,208-209,212-214
regeneration, 49-50, 79, 113, 122, 126, 167, 193,223,229,234
rehabilitation, 78, 102
relative humidity, Ill, 145, 173, 175, 178-179, 189, 196
respiration, 55, 65, 71,143,145,153,158,177,180-181,191-192,245,259,
275-276,280,282-284,287,289,291,293,297
restoration, 129
retention time, 63
Regional HydroEcological Simulation System, RHESSys, 135, 138-142, 148-152,
154, 157,211
RESCOMP, 75, 79,82,91-93,95, 104, 107
root, 8,19,81,92,141,145,163,180-181,186,191,193,198,210, 214, 221,

224-226,246,276,280-285,291,293,296
rotation, 19,37,44, 117,223,234
runoff, 143
SAIL, 265, 271, 290, 294
salinization, 77-78,92, 102
Salix, 283
SAR, 109, 115-116, 126-132,206,213,217
satellite, 5-6, 72, 77, 99,102-103,107-108,129-132,135-136,155,157-158,162,
170-171,174,176,183-184,194-198,210,212-213,219,221-222,224,
229-230,236,241-247,253,259,261-272,276,278,288-289,291-292,294,
296,298,309
saturation vapor deficit, 252, 254, 285
scale, 20, 62, 71-72,76-79,82,88-89,91,99-100,103,108-114, 121, 126-129,
131,133,136,138,145,155-158,174-175,210,215,219-220, 224, 228, 233,
237,242-244,247,249,259-262,264,268-271,277,281-282,290,293-294,
297-298
Scotland, 283
SE-590,7, 198,201
senescence, 12,65,200,282
sensor, 5-7,11,14-15,21,52,88-89,92,99,110,112-113,115,136-137,160,
162,174,176-177,182-185, 196, 198-200,204,207,209-210,230,241,
243-247,252-254,256,259-261,270,288
Serenoa repens, saw palmetto, 5
shade, 27, 34-35,38,41-43,46,48-49, 53, 80, 86-87, 89-90,93-95,99,107,114,
140,148,150,202,204,280,284
shadow, 14,25,27,43,75,78,84,87,89-91,93,95-96,99,104, 106, 120-121,
167,175,200,204,214,256,284,310
SHE, 149, 155
shrub, 49, 75-78,80,88-89,92-102,106,183,200,223,264,277
signal-to-noise ratio, SNR, 3, 16-18,207
silhouette, 25, 28, 53, 59, 88
silhouette-to-total-area ratio, STAR, 25, 35, 53
silviculture, 58, 222, 234
simple ratio, 62, 177, 199,201,289
simple vegetation index, SVI, 246-249, 251, 253, 255
site index, 226, 232
site preparation, 5, 223, 226
snow, 143, 154, 156, 196,246,248-249,251
snowmelt, 139, 143
soil, 4-5, 19-20,48,50,65,75-77,79-80,92-95, 100, 105, 108, 110-111,
113-115,119,126-128,130,135,137-141,143,145,148, 150-152,154-155,
157-158,171,177-181,183,186,188-191,193-194,198, 202, 204, 206,
210-212,214,221,225,228-229,236,241,245-246,248-249,251-255,257,
259,262-263,266,268-270,273,275,277,281-282,284-285,287,291-293,
295-296
321
soil water-holding capacity, SWHC, 139, 145, 148, 154, 190-191,210

solar angle, 7, 121, 159, 162-163
soybean, 14,262,266,297
spatial heterogeneity, 16,20,56,61,70, 107, 109-112, 127,219,224
species composition, 61-Q2, 64, 66-Q7, 1I2, 1I5, 121, 128,282,289
specific heat, 143
specific leaf area, specific needle area, 8, 173, 181, 187,203,215,282
specific leaf weight, specific needle weight, 65
spectrum, 3, 5-7,9, 11, 14-21, 37-41,48-50,56, 58-59, 61-Q3, 72, 76, 78, 87-89,
93-94,96,98-99,102,104,107,130,136-137,141,157, 160, 166, 170-171,
173-174,176,178,181, 184-185, 198,201-203,207-209,211-212,214,217,
243-244,246-247,249,251-252,260-262,264-268,271,273,278,289-290,
294
spectrophotometer, 11
spectroradiometer, 3, 6-7, 19,21, 185, 194,201
spectroscopy, spectrometer, 3, 6-7, 14, 16,22,61,63,71, 104-105, 178, 182, 184,
200,207, 21I, 244, 246, 260;'291, 309
specular reflectance, 33, 41-42, 57
spherical distribution, 30, 36, 47
SPOT, 39, 51, 58, 106, 128,235,243,259,261
starch, 176, 182, 187-188,209
stomata, 152, 179, 189,280-282,285,287,295,297
succession, 51, 56, 78,107, 109-1I2, 1I4, 121, 125, 129-130, 171,210,221-222,
228-229,231,297
surface area, 25-26, 54, 141, 187,204
Sweden, 266, 285-286
Synthetic Aperature Radar, SAR, 109, 115-116, 126-132,206,213,217
temperate, 21, 65, 71-72,133,158,169-171,215-217,220,234-235,267-268,
295,297-298
temperature, 5, 20, 65, 89, 92, Ill, 113, 119, 128, 137, 139, 143, 145, 152, 154-155,
161,173,175,178-180,182-183,186,188-189,193,196-198, 206, 21I,
214-215,217,241,245,251-257,260,262-263,266-271,273,280-282,
284-285,291-292
terrain, 1I5, 138, 141, 156, 158-160, 162-163, 170-171, 175, 178-179, 196,210,
216,222,297
Thematic Mapper, TM, 3, 5-8,10,19-21,64,72,106-108, 130, 141, 149, 152,
159-160,162-166,169-171, 177, 184,215,229-230,247,259,268,289-290,
294,306
thinning, 51-52, 116-118, 156,200,283,291,298
Thuja plicara, western red cedar, 222
tirnber, 81, 84, 178,221-223,228,233-235,275
Tiros Operational Yertical Sounder, TOYS, 252
TMS, 184, 198-200
tolerance, 80
Total Ozone Mapping Spectrometer, TOMS, 180, 184, 194-195,244-246,261,291,
309
topography, 111, 114-115, 135, 138-141, 148, 150, 155, 159-162, 164-166,
169-171, 174,237
TOPMODEL, 139-141, 148, 150, 152
trace gas, 220
transfer function, 135-137, 141, 154
translocation, 11
transmissivity, 145, 148
transparency, 46
transpiration, 72, 141, 143, 157, 165, 171,213,215-217,270,288,297
tropic, 65, 105, Ill, 130-131, 171,256,273,277,284,292,297-298
Tsuga canadensis, 63, 115
Tsuga heterophylla, western hemlock, 128, 179, 183, 189,222,235
understory, 5, 7-8,21,49,141,179,181,183,194,200-202,204,206,213,216,
289-290,297
United States, 3-4, 9,19,61,109,122,129-133,135,156-158,173-174,180,183,
185,202,210,
217,219-221,228-229~235-237,241,251,255,259,274,278,283,286
uptake, 110, 143, 152,217,221,236,242,264,275,281-282,284-285,287
USSR, 44, 267
Vaccinium, 49, 51
validation, 49,92, 135, 155, 193,233,246,253,256,288,292,298
vapor pressure deficit, VPD, 65, 143, 152, 173, 179, 188, 196-198,241,245,273,
284
vegetation index, VI, 62, 67, 72,106,135,157,159-160,163,166-167,169-170,
199,214,243,251,262,278,306
volume density, 23-24, 26, 29, 35,41-42,45-46
VVashington, 129, 132-133, 176,212,219,234-236,268,271-272,289
water, 3-6, 1 0-14, 17-20,57,71,75-77,80-81,91-92, 104, 106-108, 119,
137-141,143-146,148-150,152-158,173,176-177,180,186, 189, 191-192,
196, 198,202,208,210,213,215-217,236-237,248,252-254,256,260,
262-263,265-267,269,273,278,281-282,284-285,287-288,291-293,
295-296
water stress, 137, 139, 152, 154, 156, 191,263,278,281,284,296
water table, 5, 11,77, 148
waterlogging, 77
watershed, 135, 139-140, 148-157, 174,235
wetness, 179,255
wilderness, 79, 222, 232-233
wildfire, 77, 198,222-226,228,230
wilting point, 119, 139
wind, windthrow, Ill, 143,233
VVisconsin, 63, 121,268,285-286
wood, 26,47, 58, 65, 71,75-78,80-83,86-87,90-93,95-99,101-108,113,129,
133,156, 178, 181,203-204,207,213,216,220-221,223-225,227,234-236,
246-247,256,259,266,269,282-283,293-294,296,310
323
woodland, 58, 71, 75-78,80-81,83,86-87,90-93,95-98,101-102,104-105,

107-108,156,213
yield, 31-32, 49, 51-52, 65, 70, 79,112,115,198,224,230-231,235,259,264,
280,283,293-295
ZELIG, 114, 118-123
zenith, 30, 36, 50-51, 87, 90,100-101,183,194,255
FORESTRY SCIENCES
I. P. Baas (ed.): New Perspectives in Wood Anatomy. Published on the Occasion of the
50th Anniversary of the International Association of Wood Anatomists. 1982
ISBN 90-247-2526-7
2. C.F.L. Prins (ed.): Production, Marketing and Use of Finger-Jointed Sawnwood.
Proceedings of an International Seminar Organized by the Timber Committee of the
UNECE (Ha1mar, Norway, 1980). 1982 ISBN 90-247-2569-0
3. R.A.A. 01deman (ed.): Tropical Hardwood Utilization. Practice and Prospects. 1982
ISBN 90-247-2581-X
4. P. den Ouden (in collaboration with B.K. Boom): Manual of Cultivated Conifers.
Hardy in the Co1d- and Warm-Temperate Zone. 3rd ed., 1982
ISBN Hb 90-247-2148-2; Pb 90-247-2644-1
5. lM. Bonga and DJ. Durzan (eds.): Tissue Culture in Forestry. 1982
ISBN 90-247-2660-3
6. T. Satoo: Forest Biomass. Rev. ed. by HAL Madgwick. 1982 ISBN 90-247-2710-3
7. Tran Van Nao (ed.): Forest Fire Prevention and Control. Proceedings of an Interna-
tional Seminar Organized by the Timber Committee of the UNECE (Warsaw, Poland,
1981). 1982 ISBN 90-247-3050-3
8. J.J. Douglas: ARe-Appraisal ofForestry Development in Developing Countries. 1983
ISBN 90-247-2830-4
9. J.C. Gordon and C.T. Wheeler (eds.): Biological Nitrogen Fixation in Forest
Ecosystems. Foundations and Applications. 1983 ISBN 90-247-2849-5
10. M. Nemeth: Virus, Mycoplasma and Rickettsia Diseases of Fruit Trees. Rev.
(English) ed., 1986 ISBN 90-247-2868-1
11. M.L. Duryea and T.D. Landis (eds.): Forest Nursery Manual. Production of Bareroot
Seedlings. 1984; 2nd printing 1987 ISBN Hb 90-247-2913-0; Pb 90-247-2914-9
12. EC. Hummel: Forest Policy. A Contribution to Resource Development. 1984
ISBN 90-247-2883-5
13. P.D. Manion (ed.): Scleroderris Canker of Conifers. Proceedings of an International
Symposium on Scleroderris Canker of Conifers (Syracuse, USA, 1983). 1984
ISBN 90-247-2912-2
14. M.L. Duryea and G.N. Brown (eds.): Seedling Physiology and Reforestation Success.
Proceedings of the Physiology Working Group, Technical Session, Society of
American Foresters National Convention (portland, Oregon, USA, 1983). 1984
ISBN 90-247-2949-1
15. K.A.G. Staaf and N.A. Wiksten (eds.): Tree Harvesting Techniques. 1984
ISBN 90-247-2994-7
16. J.D. Boyd: Biophysical Control of Microfibril Orientation in Plant Cell Walls.
Aquatic and Terrestrial Plants Including Trees. 1985 ISBN 90-247-3101-1
17. W.P.K. Findlay (ed.): Preservation ofTimber in the Tropics. 1985
ISBN 90-247-3112-7
18. I. Samset: Winch and Cable Systems. 1985 ISBN 90-247-3205-0
FORESTRY SCIENCES
19. R.A. Leary: Interaction Theory in Forest Ecology and Management. 1985
ISBN 90-247-3220-4
20. S.P. Gessel (ed.): Forest Site and Productivity. 1986 ISBN 90-247-3284-0
21. T.C. Hennessey, P.M. Dougherty, S.V. Kossuth and J.D. Johnson (eds.): Stress
Physiology and Forest Productivity. Proceedings of the Physiology Working Group,
Technical Session, Society of American Foresters National Convention (Fort Collins,
Colorado, USA, 1985). 1986 ISBN 90-247-3359-6
22. K.R. Shepherd: Plantation Silviculture. 1986 ISBN 90-247-3379-0
23. S. Sohlberg and V.E. Sokolov (eds.): Practical Application of Remote Sensing in
Forestry. Proceedings of a Seminar on the Practical Application of Remote Sensing in
Forestry (Jonkoping, Sweden, 1985). 1986 ISBN 90-247-3392-8
24. J.M. Bonga and D.J. Durzan (eds.): Cell and Tissue Culure in Forestry. Volume 1:
General Principles and Biotechnology. 1987 ISBN 90-247-3430-4
25. J.M. Bonga and D.J. Durzan (eds.): Cell and Tissue Culure in Forestry. Volume 2:
Specific Principles and Methods: Growth and Development. 1987
ISBN 90-247-3431-2
26. J.M. Bonga and D.I. Durzan (eds.): Cell and Tissue Culure in Forestry. Volume 3:
Case Histories: Gymnosperms, Angiosperms and Palms. 1987 ISBN 90-247-3432-0
Set ISBN (Volumes 24-26) 90-247-3433-9
27. E.G. Richards (ed.): Forestry and the Forest Industries: Past and Future. Major
Developments in the Forest and Forest Industries Sector Since 1947 in Europe, the
USSR and North America. In Commemoration of the 40th Anniversary of the Timber
Committee of the UNECE. 1987 ISBN 90-247-3592-0
28. S.V. Kossuth and S.D. Ross (eds.): Hormonal Control ofTree Growth. Proceedings of
the Physiology Working Group, Technical Session, Society of American Foresters
National Convention (Birmingham, Alabama, USA, 1986). 1987 ISBN 90-247-3621-8
29. U. Sundberg and C.R. Silversides: Operational Efficiency in Forestry.
Vol. 1: Analysis. 1988 ISBN 90-247-3683-8
30. M.R. Ahuja (00.): Somatic Cell Genetics of Woody Plants. Proceedings of the IUFRO
Working Party S2.04-07 Somatic Cell Genetics (Grosshansdorf, Germany, 1987).
1988. ISBN 90-247-3728-1
31. P.K.R. Nair (00.): Agroforestry Systems in the Tropics. 1989 ISBN 90-247-3790-7
32. C.R. Silversides and U. Sundberg: Operational Efficiency in Forestry.
Vol. 2: Practice. 1989 ISBN 0-7923-0063-7
Set ISBN (Volumes 29 and 32) 90-247-3684-6
33. T.L. White and G.R. Hodge (eds.): Predicting Breeding Values with Applications in
Forest Tree Improvement. 1989 ISBN 0-7923-0460-8
34. H.I. Welch: The Conifer Manual. Volume 1. 1991 ISBN 0-7923-0616-3
35. P.K.R. Nair, H.L. Gholz, M.L. Duryea (eds.): Agroforestry Education and Training.
Present and Future. 1990 ISBN 0-7923-0864-6
36. M.L. Duryea and P.M. Dougherty (eds.): Forest Regeneration Manual. 1991
ISBN 0-7923-0960-X

1shimoda H Gholz H L Nakane K Eds The Use of Remote Sensing I

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

1shimoda H Gholz H L Nakane K Eds The Use of Remote Sensing I

Hochgeladen von

Copyright:

Verfügbare Formate

THE USE OF REMOTE SENSING IN THE MODELING

SPRINGER SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-94-010-6290-9 ISBN 978-94-011-5446-8 (eBook)

Printed on acid-free paper

All Rights Reserved

Section One: Stand-Level Analyses

Section Two: Landscape/Regional-Level Analyses

Section Three: Global-Level Analyses

Acknowledgments - Financial support for the Congress symposium and subsequent

Henry L. Gholz, Gainesville, Florida, USA

Assessing Leaf Area and Canopy Biochemistry of Florida Pine

H.L. Gholz, School of Forest Resources and Conservation, University of Florida,

1986 - 1987 1987 - 1988 1988 - 1989

(specifically LAI) and biochemical attributes of slash pine canopies could be

Remotely sensed data

Sensor Platform Swath Instantaneous Spectral Bandwidth Number Vegetation ::l

Airborne Visible ER-2 11 20 410-2450 9.4-9.7 224 Chlorophyll,

the narrow-band Airborne Visible Infrared Imaging Spectrometer was carried by a

Date Number of samples TM AVIRIS IRIS

February 1988 124, Y N N

Relating LAI to forest canopy reflectance

... -...".......i.. I ...

Figure 3. The relationship between NDVI recorded by the Landsat Thematic

Patterns in the canopy biochemistry of slash pine stands

biochemical concentrations to reflectance spectra measured in the laboratory (IRIS)

tissue prior to senescence. Significantly lower concentrations of lignin and cellulose

Table 3. Nutrient concentrations (% mean 1 std. dev.) for each of three

New needles U 0.1130.036

Old needles U 0.056O.OO8 O.l350.019 0.2160.079 0.1030.028

Old twigs U 0.1900.059

Unfoliated U 0.2440.040 0.022O.OO7 O.O85O.030 O.I920.079 O.084O.O17

Table 4. Biochemical concentrations for 80 slash pine foliage samples (August

Biochemical Mean Standard Minimum Maximum

Water (%) 58.3 4.4 37.6 78.0

Table 5. Simple correlations (r) among biochemical concentrations from

Water Chlorophyll N Lignin Cellulose

* Indicates significance at the p = 0.01 level.

Table 6. Differences in mean biochemical concentrations between control and

Biochemical Control Fertilized

Water (%) 58.6 (2.7) 58.0 (3.1)

Table 7. Differences in mean concentrations between needle ages for needles

Needle age class

Biochemical New Old

Water (%) 60.8 (1.7) 55.9 (1.9)

Laboratory reflectance spectra of slash pine foliage

Figure 4. Mean laboratory reflectance spectra of 40 control plot samples and 40

Coupling the AVIRIS measurements and foliar biochemistry: The

is typically a variable, uncontrollable and relatively weak radiation source

Table 8. The wavebands of maximum correlation between foliar biochemical

First derivative Second derivative

Foliar Waveband Waveband

Nitrogen 1491 Protein 0.5 1195 Protein 0.4

Table 9. The SNR threshold (MSNR) at which the strongly correlating

First derivative Second derivative

Foliar Waveband Waveband

Nitrogen 1491 37:1 1195 61:1

JPLSNR@ Vegetation Approx. JPL

1491 175:1 7.1 25:1 37:1

Coupling the AVIRIS measurements and foliar biochemistry: An

Acknowledgments - Gholz acknowledges support from USDA Cooperative

Modeling Radiative Transfer through Forest Canopies:

In order to theoretically describe the radiative transfer inside a forest canopy,

T Nilson and J. Ross, Institute of Astrophysics and Atmospheric Physics, Estonian