(Geology) Remote Sensing of Environment - IKONOS Special Issue (Geologos)

Remote Sensing of Environment 88 (2003) 1 2
www.elsevier.com/locate/rse
Preface
A new direction in Earth observations from space: IKONOS

As guest editors, we are pleased to present this special
issue of Remote Sensing of Environment, dedicated to the
characterization and scientific applications of Space Imaging IKONOS data. In recent years, U.S. federal agencies
have more aggressively explored using commercial sources
of remote sensing data to meet their science and applications
needs. This special issue documents progress achieved
during the last 5 years in working with one commercial
vendor, Space Imaging, and its IKONOS land imaging
satellite.
The Space Imaging IKONOS system represents a significant technical advancement in space-acquired land observation. Upon its successful launch in 1999, the system
provided the finest spatial resolution publicly available from
space1-m panchromatic and 4-m multispectral (blue,
green, red, and near-infrared)with high radiometric fidelity and impressive geometric accuracy (Dial, Bowen, Gerlach, Grodecki, & Oleszczuk, 2003). This system provided a
major new complement to the multiscale observations
provided by systems, such as Landsat, ASTER, SPOT,
AVHRR, and MODIS.
The governments purchase of IKONOS imagery presented a new management and fiscal model for supplying
satellite remote sensing imagery to the user community.
Until quite recently, most satellite-based Earth observations
have been acquired by government-developed and -funded
systems. Space Imagings IKONOS, and more recently
DigitalGlobes QuickBird, have offered a new model for
acquiring Earth observations, where private companies
raised the financial capital required to develop and operate
a remote sensing satellite. Rather than the U.S. government
building and operating such systems, federal agencies can
simply purchase imagery and products from the commercial
sector as needed (Birk et al., 2003). The successful use of
IKONOS imagery in science- and defense-related applications is a testament to the level of cooperation and collaboration that has developed between the government and
Space Imaging. The experiences gained through this relationship should serve as an example for other current and
future government data purchase programs.
This special issue explores the successes and possible
shortcomings of one source of commercial imagery, and
addresses some of the skepticism regarding the U.S. private
sectors ability to acquire the level of science and applications measurements that have historically been supplied by
0034-4257/$ - see front matter D 2003 Published by Elsevier Inc.
doi:10.1016/j.rse.2003.08.011
the government. The technical performance of IKONOS has

proven to be exceptional, while independent characterizations have provided further confidence in IKONOS product
quality (Goward, Davis, Fleming, Miller, & Townshend,
2003; Helder, Coan, Patrick, & Gaska, 2003; Pagnutti et al.,
2003; Ryan et al., 2003).
This publication is a result of cooperation between
several U.S. government agencies, Space Imaging, and the
international Earth science community. Through a Joint
Agency Commercial Imagery Evaluation (JACIE) team,
the National Aeronautics and Space Administration
(NASA), the U.S. Geological Survey (USGS), and the
National Imagery and Mapping Agency (NIMA) have
performed thorough and independent characterizations of
IKONOS radiometric accuracy, image quality, and geopositional accuracy (Zanoni et al., 2003). The results of
these efforts are published for the first time in this issue.
Several examples of IKONOS imagerys extensive use in
Earth science research are also presented here (Andrefouet
et al., 2003; Goetz, Wright, Smith, Zinecker, & Schaub,
2003; Hurtt et al., 2003; Masuoka et al., 2003; Morisette et
al., 2003; Sawaya, Olmanson, Heinert, Brezonik, & Bauer,
2003; Seelan, Laguette, Casady, & Seielstad, 2003; Small,
2003). These results clearly show that confidence in this
commercial approach requires substantial interaction over
extended periods of time between the users and the data
providers (Goward et al., 2003), not unlike experiences with
the traditional government approach. The role of the government agencies as brokers in this process is vital to the
successful use of commercial remote sensing measurement
sources.
As guest editors, we are pleased to present the following
manuscripts encompassing technical and scientific aspects
of IKONOS for Earth observation. We thank the many
authors and co-authors for their research and documentation, for their willingness to participate in this endeavor, and
for their patience in bringing this issue to fruition. We also
thank the many anonymous peer reviewers who donated
their time to ensuring the technical quality of each manuscript. Special thanks also go to Ms. Laura Pair, Ms. Marcia
Wise, and Ms. Kim Levens of Stennis Space Center for their
hard work and administrative support to this publication.
We hope that you find this special issue informative and
that the articles provide a valuable stimulus and a reference
for future research.
Preface
References
Andrefouet, S., Kramer, P., Torres-Pulliza, D., Joyce, K. E., Hochberg,
E. J., Garza-Perez, R., Mumby, P. J., Riegl, B., Yamano, H., White,
W. H., Zubia, M., Brock, J. C., Phinn, S. R., Naseer, A., Hatcher, B.
G., & Muller-Karger, F. E. (2003). Multi-sites evaluation of IKONOS
data for classification of tropical coral reef environments. Remote
Sensing of Environment, 88, 127 142. (this issue).
Birk, R. J., Stanley, T., Snyder, G. I., Hennig, T. A., Fladeland, M. M., &
Policelli, F. (2003). Government programs for research and operational
uses of commercial remote sensing data. Remote Sensing of Environment, 88, 3 16. (this issue).
Dial, G., Bowen, H., Gerlach, F., Grodecki, J., & Oleszczuk, R. (2003).
IIKONOS satellite imagery, and products. Remote Sensing of Environment, 88, 23 36. (this issue).
Goetz, S. J., Wright, R. K., Smith, A. J., Zinecker, E., & Schaub, E. (2003).
IKONOS imagery for resource management: Tree cover, impervious
surfaces and riparian buffer analyses in the Mid-Atlantic region. Remote
Sensing of Environment, 88, 194 207. (this issue).
Goward, S. N., Davis, P. E., Fleming, D., Miller, L., & Townshend, J. R.
(2003). Empirical comparison of Landsat 7 and IKONOS multispectral
measurements for selected Earth Observation System (EOS) validation
sites. Remote Sensing of Environment, 88, 79 98. (this issue).
Goward, S. N., Townshend, J. R. G., Zanoni, V., Pollicelli, F., Stanley, T.,
Ryan, R., Holekamp, K., Underwood, L., Pagnutti, M., & Fletcher, R.
(2003). Acquisition of Earth science remote sensing observations from
commercial sources: Lessons learned from the Space Imaging IKONOS
example. Remote Sensing of Environment, 88, 200 218. (this issue).
Helder, D., Coan, M., Patrick, K., & Gaska, P. (2003). IKONOS geometric characterization. Remote Sensing of Environment, 88, 68 78.
(this issue).
Hurtt, G., Xiao, X., Keller, M., Palace, M., Asner, G. P., Braswell, R.,
Brondzio, E. S., Cardoso, M., Carvalho, C. J. R., Fearon, M. G., Guild,
L., Hagen, S., Hetrick, S., Moore III, B., Nobre, C., Read, J. M., Sa, T.,
Schloss, A., Vourlitis, G., & Wickel, A. J. (2003). IKONOS imagery for
the large scale biosphere atmosphere experiment in Amazonia (LBA).
Remote Sensing of Environment, 88, 110 126. (this issue).
Masuoka, P. M., Claborn, D. M., Andre, R. G., Nigro, J., Gordon, S. W.,
Klein, T. A., & Kim, H. (2003). Use of IKONOS and Landsat for
malaria control in the Republic of Korea. Remote Sensing of Environment, 88, 186 193. (this issue).
Morisette, J. T., Nickeson, J. E., Davis, P., Wang, Y., Tian, Y., Woodcock,
C. E., Shabanov, N., Hansen, M., Cohen, W. B., Oetter, D. R., &
Kennedy, R. E. (2003). High spatial resolution satellite observations
for validation of MODIS land products: IKONOS observations acquired
under the NASA scientific data purchase. Remote Sensing of Environment, 88, 99 109. (this issue).
Pagnutti, M., Ryan, R. E., Kelly, M., Holekamp, K., Zanoni, V., Thome, K.,
& Schiller, S. (2003). Radiometric characterization of IKONOS multispectral imagery. Remote Sensing of Environment, 88, 52 67. (this
issue).
Ryan, R., Baldridge, B., Schowengerdt, R. A., Choi, T., Helder, D. L., &
Blonski, S. (2003). IKONOS spatial resolution and image interpretability characterization. Remote Sensing of Environment, 88, 37 51. (this
issue).
Sawaya, K., Olmanson, L., Heinert, N., Brezonik, P., & Bauer, M. (2003).
Extending satellite remote sensing to local scales: Land and water resource monitoring using high-resolution imagery. Remote Sensing of
Environment, 88, 143 155. (this issue).
Seelan, S. K., Laguette, S., Casady, G. M., & Seielstad, G. A. (2003).
Remote sensing applications for precision agriculture: A learning community approach. Remote Sensing of Environment, 88, 156 168. (this
issue).
Small, C. (2003). High spatial resolution spectral mixture analysis of urban
reflectance. Remote Sensing of Environment, 88, 169 185. (this issue).
Zanoni, V., Stanley, T., Ryan, R., Pagnutti, M., Baldridge, B., Roylance, S.,
Snyder, G., & Lee, G. (2003). The Joint Agency Commercial Imagery
Evaluation (JACIE) team: Overview and IKONOS joint characterization approach. Remote Sensing of Environment, 88, 17 22. (this issue).
Vicki M. Zanoni
NASA Earth Science Applications Directorate,
Mail Code MA20, Stennis Space Center, MS 39529, USA
E-mail address: Vicki.M.Zanoni@nasa.gov
Samuel N. Goward
Department of Geography, University of Maryland,
2181 LeFrak Hall,
College Park, MD 20742, USA
E-mail address: sgoward@umd.edu

Government programs for research and operational uses

of commercial remote sensing data
Ronald J. Birk a,*, Thomas Stanley b, Gregory I. Snyder c, Thomas A. Hennig d,
Matthew M. Fladeland e, Fritz Policelli b
a
b
NASA Office of Earth Science, 300 E Street SW, Washington, DC 20546, USA
NASA Earth Science Applications Directorate, Stennis Space Center, MS USA
c
USGS National Center, Reston, VA, USA
d
National Imagery and Mapping Agency, Bethesda, MD, USA
e
NASA Ames Research Center, Moffett Field, CA, USA
Received 14 February 2003; received in revised form 6 June 2003; accepted 30 July 2003
Abstract
The private sector is delivering products and services derived from an expanding array of airborne and spaceborne remote sensing
systems. The successful commercial launches of the IKONOS, QuickBird, and OrbView-3 satellites in 1999, 2001, and 2003,
respectively, combined with commercial airborne sensors such as the Positive Systems ADAR 5500 (multispectral), the Intermap
STAR-3i (interferometric synthetic aperture radar), and TerraPoint, LLC, LIDAR System have ushered in an era of expanded capability
and capacity for the field of remote sensing of our Earth. Remote sensing data from commercial sensors offer the public and private
geospatial information communities important new sources of timely and accurate spatial information that can augment data provided
by public-sector remote sensing systems. Several Federal agencies, including the National Aeronautics and Space Administration
(NASA), the U.S. Geological Survey, and the National Imagery and Mapping Agency (NIMA), have established data purchase
programs and related activities to access, evaluate, and assimilate new commercial remote sensing products to serve research and
operational requirements. Plans for future commercial systems and data products indicate an expanding set of data types using
hyperspectral, radar, LIDAR, and microwave technologies. The availability of new data sources has established the basis for Federal
programs to provide for systematic characterization of the products, consistent with the characterization of data products enabled by
traditional sources that include Landsat, SPOT, and the Advanced Very High Resolution Radiometer (AVHRR). An overview of
commercial remote sensing initiatives within the National Aeronautics and Space Administration, the U.S. Department of the Interior,
and the U.S. Department of Defense (DoD), and of their Joint Agency Commercial Imagery Evaluation (JACIE) team, illustrates these
points, highlights lessons learned from these activities, and outlines recommendations for the future.
D 2003 Elsevier Inc. All rights reserved.
Keywords: Earth science; Commercial remote sensing; Multispectral; Imagery; Digital elevation model; Mapping; Monitoring; IKONOS; QuickBird;
OrbView-3; LIDAR; Radar; Hyperspectral
1. Introduction
The United States Government has significant responsibilities in providing mapping and monitoring information to
meet the needs of its citizens. Traditionally, Federal agencies, including the National Aeronautics and Space Admin-
* Corresponding author. Tel.: +1-202-358-1701; fax: +1-202-3583098.

E-mail address: rbirk@hq.nasa.gov (R.J. Birk).
0034-4257/$ - see front matter D 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2003.07.007
istration (NASA), the National Oceanic and Atmospheric

Administration (NOAA), and the Department of Defense
(DoD), have deployed the primary spaceborne sources of
remote sensing data for our Nation. Spaceborne remote
sensing systems, such as the Advanced Very High Resolution Radiometer (AVHRR) on NOAAs Polar Orbiting
Environmental Satellites and the Landsat satellites jointly
managed by NASA and the U.S. Geological Survey
(USGS), continue to provide observational data for scientific research, economic security, and operational missions
to serve weather prediction, navigation, monitoring, and
R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316
mapping. Information capacity from U.S. civil government

space-based assets, typically of low to moderate spatial
resolution (kilometers down to tens of meters), is being
complemented by data from commercial remote sensing
systems with submeter resolution.
The recent growth in commercially deployed multispectral, hyperspectral, radar, LIDAR, and thermal remote
sensing systems on airborne and spaceborne platforms
offers important sources of timely, quality spatial information that can serve the research and operational needs
of our Nation. In response to these new sources of data
for research and applications, Federal agencies are evolving national data policies to incorporate these new
sources of information into their operations (for a review
of Federal policies on commercial remote sensing data,
see National Research Council, 2002).
Federal agencies are keenly interested in evaluating the
potential of commercial remote sensing solutions to meet
their requirements for geospatial information and have
established activities to verify and validate the characteristics of the data products and their utility to address
agency mission needsparticularly those that contribute
to the protection of life and property. This review process
provides an opportunity for the commercial remote sensing community to establish an understanding of the
Nations research and operational requirements that may
be served by commercially provided solutions.
Federal policies have expanded the opportunities for
commercial suppliers to augment systems owned and
operated by the public sector. Privatization policies of

the 1980s evolved to commercialization and licensing
policies in the 1990s, including the Land Remote Sensing
Policy Act of 1992 (U.S. Government, 1992) and Presidential Decision Directive 23 (OPS, 1994) encouraging
private-sector investment in land-related Earth observing
systems. The U.S. Commercial Remote Sensing Policy
(OSTP, 2003) directs government agencies to rely to the
maximum practical extent on U.S. commercial remote
sensing space capabilities for filling imagery and geospatial needs for military, intelligence, foreign policy,
homeland security, and civil users. These policies encourage the U.S. Government to use data provided by the
private sector to meet mission requirements and not to
compete with commercial suppliers.
Executive Order 12906 (Clinton, 1994), calls for a
National Spatial Data Infrastructure to support public- and
private-sector applications of geospatial data in such areas
as transportation, community development, agriculture,
emergency response, natural resource management, and
communications. This Executive Order established the
basis for a national remote sensing strategy that uses
civil, commercial, and military assets (including airborne
and satellite assets) to support U.S. information needs. A
national civil remote sensing strategy could focus on
solutions to meet the fundamental needs of these communities to optimize data acquisition and utilization
approaches and to extend collaboration in data management and end-user applications.
Table 1
Government data purchase program comparisons
Characteristics
Scientific Data
Purchase (Part I)
Scientific Data
Purchase (Part II)
Commercial imagery
data purchase
SeaWiFS
Sponsor
Purpose
NASA
Determine utility of
commercial data for
NASA science and
applications research
NIMA
Acquire high resolution
imagery to support the
warfighter
NASA
Acquire ocean color
imagery for science
research
Funding
Time frame
Centralized tasking/
distribution hub
Payment on data delivery
Up-front cash payment
Centralized verification
and validation effort
Contract mechanism
US$50 million
1997 2003
Yes
NASA
Examine utility of
additional commercial
datasets for NASA
science and
applications research
US$20 million
2000 2003
Yes
US$30 million
2000 2002
Yes
US$43 million
1997 2003
Yes
Yes
No
Yes
Yes
No
Yes
Yes
No
Yes
Yes (20%)
Yes (80%)
Yes
Indefinite delivery,
indefinite quantity/
fixed price
Yes
Small
Indefinite delivery,
indefinite quantity/
fixed price
Yes
Small
Basic ordering
agreement
(fixed price)
Yes
None
Fixed price
No
None
Yes
Yes
Yes
Yes
Yes
Yes
No
Yes
Tasking panel
Deviation from
commercial
specifications
Multiple data sources
Multiple products
Federal government programs and activities that currently acquire commercial data include the USGS The
National Map,1 the Federal Emergency Management Agency (FEMA) Flood Map Modernization Program, the NOAA
Shoreline Mapping program, the U.S. Department of Agriculture Foreign Agriculture Service programs, the National Imagery and Mapping Agency (NIMA) Commercial
Imagery Program (CIP), the NASA Solid Earth and Natural
Hazards program, and the U.S. Global Climate Research
Program (USGCRP). The implementation of U.S. government policies to procure remote sensing data from the
private sector are reflected in specific agency programs
that assess the use of commercial data products to meet
agency mission requirements. NASA, the USGS, and
NIMA have taken different approaches in implementing
commercial data purchase programs and related activities.
Table 1 lists data purchase programs representative of these
agencies and compares several important program features.
Following are descriptions of the different types of data
products available through these programs, recent research
and operational applications, and discussion on current and
future requirements for commercial geospatial data. Lessons learned in the performance of the commercial data
purchase programs to date are also highlighted, along with
remaining challenges and recommended approaches related
to future commercial data purchase opportunities for Federal agencies.
2. NASA commercial data initiatives

NASAs Earth Science Enterprise conducts research
and development of aerospace science and technology
associated with remote sensing systems to seek answers
to fundamental questions about how the Earth system
functions (see Fig. 1). NASA uses Earth observation
systems to provide detection, monitoring, and mapping
solutions for research on the water and energy cycle, the
carbon cycle, the chemistry-climate connection, weather
and climate predictions, and solid Earth and natural
hazards. NASA Earth science results contribute to global
change research, advanced weather prediction, and natural
hazards research (NASA Office of Earth Science, 2000,
2002).
NASA works with the private sector in partnerships
and through procurements to provide the Earth science
community with remote sensing data provided by commercially owned and operated aerospace systems. Data
purchase projects include the Sea-viewing Wide Field-ofview Sensor (SeaWiFS) Project for ocean color data, the
Scientific Data Purchase (SDP) project for high-resolution
terrestrial and specialized atmospheric data products.
http://www.nationalmap.usgs.gov.
In support of the NASA mission to understand and

protect our home planet, NASA also partners with federal
agencies and national organizations to benchmark the
benefits of assimilating results of NASA Earth science
research and development to enhance decision support for
applications of national priority. The focus is on Earth
science and remote sensings capacity to contribute to
solutions for community growth, energy management, risk
assessments for public health, detection of environmental
indicators for homeland security and biological invasive
species, aviation safety, agricultural efficiency, and management of global to regional issues associated with carbon
sequestration, disasters, coastal ecosystems, water, and air
quality.
2.1. Sea-viewing Wide Field-of-view Sensor project
The Sea-viewing Wide Field-of-view Sensor Project2 is
a public/private partnership to deliver ocean color data to
meet NASA science requirements from a space-based
mission developed and operated by a commercial remote
sensing company. The SeaWiFS system, deployed by
Orbital Sciences Corporation (OSC) in 1995, has provided
measurements of global ocean bio-optical properties
through NASA to the Earth science community and to
commercial applications, such as fisheries. NASA contracted with OSC to pre-purchase SeaWiFS data with
payments made to OSC prior to the construction and
launch of the instrument. NASA shared in the financial
risk of the project and helped enable private-sector deployment of an ocean imaging system that could provide
specific data products to serve the ocean science research
community with systems engineering, instrument calibration, algorithm validation, and ocean science. This approach contributed to OSCs success in building and
deploying the satellite ocean color observatory. The Government benefited from the delivery of ocean color datasets at costs and risks that were shared with the private
sector, as compared to traditional government satellite
development programs where all of the costs and risks
are borne directly by the taxpayer.
2.2. NASA Scientific Data Purchase project Part I
In 1996, Congress funded NASA to support a project
for the systematic acquisition of commercial data and
products for Earth science research. NASAs Earth Science
Enterprise established the Scientific Data Purchase project
in 1997 to explore the viability of using commercial
remote sensing data to meet a subset of Earth science
research objectives. The program was initiated with a
US$50 million budget to serve as a pathfinder to evaluate
the capacity of the private sector to supply remote sensing
http://seawifs.gsfc.nasa.gov/SEAWIFS.html.
Fig. 1. ESE science questions about how the Earth system functions.
data for Earth science research programs. Through this

program, NASA investigators had the opportunity to
submit requests for the acquisition of select commercial
remote sensing data products that augment public data
sources.
With over 600 registered users for the data purchase
project, NASA has distributed over 2700 individual datasets representing over 26 terabytes of data storage capacity.
Fig. 2 shows statistics on the affiliation of science community researchers using commercial data provided by the
NASA Scientific Data Purchase project. Fig. 3 indicates

the types of science research benefiting from the evaluation of the commercial data products. The NASA data
purchase project has demonstrated that commercial data
can be useful for terrestrial and aquatic ecosystem research. By assimilating data from different commercial
sources, a range of remote sensing data products can be
used to contribute to complex models of coral reefs
(Andrefouet et al., 2001; Capolsini, Andrefouet, Rion, &
Payri, 2003; Palandro, Andrefouet, Dustan, & Muller-
Fig. 2. NASAs Scientific Data Purchase project user affiliation.
Karger, 2003), forestry (Ashton et al., 2002; Hansen,

DeFries, Townshend, Marufu, & Sohlberg, 2002; Thenkabail, Lin, Hall, Ashton, & Harris, submitted for publication), and agricultural systems (Mercuri, Engel, &
Johannsen, 2000; Hipple, 2001). In addition, high-resolution imagery has been important for such diverse applications as public health (Masuoka, Claborn, & Andre, 2001),
natural resource conservation (Jenkins & Anderson, 2002),
and archeology (Sever & Irwin, 2003). The National
Imagery and Mapping Agency used over 7000 orthorectified Landsat scenes provided by the Scientific Data Purchase project, covering most of the land surface of the
Earth, to build a global landcover dataset (Koeln, Jones, &
Melican, 2000).
NASA conducted Part I of the data purchase project in
two phases. The first phase started in May 1997 with a
Request for Offers (RFO) inviting interest from vendors
offering commercial remote sensing data products with the
potential to serve the (then) four primary ESE science theme
areas (Table 2). In response to the RFO, NASA received a
total of 18 proposals offering 65 Phase I products. Through
a review and selection process, 10 contracts for 22 Phase I
products were awarded in December 1997. Working with
the vendors and with panels of scientists having expertise in
remote sensing, NASA evaluated, verified, and validated the
products delivered under the Phase I contracts.
Five companies were selected to negotiate contracts for
Phase II data products in 1998: Earth Satellite (EarthSat),
Space Imaging, DigitalGlobe (doing business as EarthWatch
at the time of the contract award), Positive Systems, and
AstroVision International. As of July 2003, four of the five
companies successfully delivered data products to NASA
for use by the Earth science research community. AstroVision International is contracted to provide high-temporal
resolution, full-disk imagery (7-km spatial resolution), and
higher (600-m) spatial resolution imagery of regions of

interest upon successful launch and operational checkout
of its planned geosynchronous satellite system. Table 3
provides an overview of the selected products3 that have
been delivered and profiles a subset of the types of data
available from the private sector. A synopsis of the products
follows.
EarthSat provides orthorectified Landsat imagery covering global land areas for two time periods. The first
dataset is Multispectral Scanner (MSS) imagery collected
during the mid-1970s (representing the earliest multispectral images of Earth taken from space). The second
dataset is Thematic Mapper (TM) imagery collected
around 1990 that serves as a baseline for studies of
change detection. Control points from available government sources were made available to EarthSat to orthorectify both datasets. The contract provides for mosaic
scenes of the TM coverage. The USGS Earth Resources
Observation Systems (EROS) Data Center provided the
systematically corrected data for scenes available in the
U.S. archive. EarthSat was contracted to provide the
required data from foreign ground stations to complete
the dataset. The orthorectified TM data is and the MSS
data will be archived and distributed to the public on a
cost recovery basis by the USGS EROS Data Center
through the Earth Observing System Data Gateway
(http://edcimswww.cr.usgs.gov/pub/imswelcome/). NASA
makes the TM mosaics available to the public at no cost
at http://zulu.ssc.nasa.gov/mrsid.
The Intermap Technologies STAR-3i airborne radar
data, provided through the DigitalGlobe contract, enables
3
Detailed descriptions of each of the products are accessible on the
ESA Directorate Web page at http://www.esad.ssc.nasa.gov/datapurchase/.
Fig. 3. NASAs Scientific Data Purchase project data use statistics on new acquisitions.
scientists to study phenomena related to elevation and

hydrography at high spatial resolutions. Researchers are
using STAR-3i digital elevation model (DEM) data products in monitoring glacial changes (Muskett, Lingle,
Tangborn, & Rabus, 2003), in validating the elevation
models produced from interferometric synthetic aperture
radar (IFSAR) data for mapping Arctic drainage patterns
(Nolan & Prokein, in press), in studying archeological
sites in Central America through the NASA Global
Hydrology and Climate Center (Sever & Irwin, 2003),
Table 2
Four primary ESE science theme areas in 1997
Theme area and purpose
Land-cover and land-use change research
Qualify the past, current and future land cover and land use patterns at
regional and global scales
Understand natural and human-induced influences that lead to changes in
land cover, land use, shorelines, or terrestrial and marine ecosystems
Support the scientific information requirements for national environmental
monitoring and research
Support design of a prototype environmental report on trends in, and the
status of, the U.S. environment
Improve methods for the sustainable management of farmlands, forests,
rangelands, and coastal marine environments
Seasonal-to-interannual climate variability and prediction
Measure globally distributed atmospheric wind profiles
Provide atmospheric soundings from Global Positioning System (GPS)
satellites for weather and climate prediction
Natural hazards research and applications
Improve methods and understanding of how best to characterize and
mitigate the consequences of natural hazards for both managed and
natural ecosystems
Long-term climate: natural variability and change research
Test the utility of new measurements to meet the continuity requirements of
the EOS science program
and in studying hydrological processes in support of

research being conducted by the University of AlaskaFairbanks (Gieck, Kane, Hinzman, Overduin, & McNamara, 2002).
Researchers conducting studies focused on agriculture
(Chang, Clay, Dalsted, & ONeill, 2002), urban sprawl
(Hipple & Daugherty, 2000), and habitat management
(Driscoll & Lawrence, 2002) have validated the use of data
acquired by the Positive Systems ADAR 5500 airborne
system. The 0.7-m multispectral data from this system
emulates the now-current-generation commercial spacebased remote sensing systems, giving Earth system scientists a head start in developing techniques for satelliteacquired, high-spatial resolution datasets.
Scientists use Space Imagings 1-m panchromatic and 4m multispectral IKONOS datasets synergistically, with
techniques such as band sharpening, to observe many
phenomena that previously could only be inferred. Highresolution imagery enables improved solutions for ground
truth validation for accuracy assessments of land use and
land cover classifications derived from coarser spatial resolution systems, such as Landsat and SPOT (Homer, Huang,
Yang, & Wylie, 2002).
NASA participates in the Joint Agency Commercial
Imagery Evaluation (JACIE) team, which is a NASA,
USGS, and NIMA partnership formed to validate data from
commercial satellite vendors (Zanoni et al., 2003). The work
is performed cooperatively with industry and has demonstrated benefits to the government, to industry, and to the
broader user community in understanding the performance
of spatial information products derived from commercially
provided remote sensing instruments.
2.3. NASA Scientific Data Purchase project Part II
Part II of the Scientific Data Purchase, established in
2000 with US$20 million in funding, includes the pro-
Table 3
Overview of NASA Scientific Data Purchase Part I products
Data provider and
image product
Data provider: EarthSat
Circa 1975
orthorectified scene
Circa 1990
orthorectified scene
Circa 1990
orthorectified mosaic
Platform and sensor
Data type
Pixel size
Positional accuracy
Encoding
Radiometric
accuracy
Cloud
%
Satellite, Landsat 1 3,
MSS
Satellite, Landsat 4
and 5, TM
Satellite, Landsat 4
and 5, TM
MSS (4-band)
57 m
F 100 m RMSE
8 bits
NA
< 20%
TM (7-band)
28.5 m
F 50 m RMSE
8 bits
NA
< 20%
TM (3-band)
28.5 m
F 50 m RMSE
8 bits
NA
< 20%
X-SAR
2.5 m
8 bits
NA
NA
DEM
5 10 m
F 2.5 m RMSE
RMSE (1j)
F 2.5 m horizontal
F 1 3 m vertical
RMSE (1j)
32 bits
NA
NA
MS1
0.7 m
F 100 m (center pt)
8 bits
F 10% absolute,
F 5% relative
< 10%
MS1
0.7 m
Relief dependent,
F 12.2 m (benign),
F 50 m (extreme)
8 bits
NA
< 10%
Pan
1m
F 250 m (std.),
F 3 m (precision)
(CE 90%)
F 250 m (std.),
F 5 m (precision)
(CE 90%)
F 12.2 m (std.),
F 2 m (precision)
(CE 90%)
F 12.2 m (std.),
F 5 m (precision)
(CE 90%)
F 25.4 m horizontal,
7 m vertical (CE 90%)
11 bits
F 10% absolute,
F 5% relative
< 10%
11 bits
F 10% absolute,
F 5% relative
< 10%
11 bits
F 10% absolute,
F 5% relative
< 10%
11 bits
F 10% absolute,
F 5% relative
< 10%
11 bits
NA
< 10%
Data provider: DigitalGlobe/Intermap

Orthorectified
Airborne,
radar image
STAR-3i IFSAR
DEM
Airborne,
STAR-3i IFSAR
Data provider: Positive Systems

IM-R1I-55
Airborne,
ADAR 5500
single-frame
imagery
MOS-G1
Airborne,
georeferenced
ADAR 5500
mosaic
Data provider: Space Imaging
Original Pan
Satellite, IKONOS-2
(georectified)
Original MS
(georectified)
Satellite, IKONOS-2
MSa
4m
Master Pan
(orthorectified)
Satellite, IKONOS-2
Pan
1m
Master MS
(orthorectified)
Satellite, IKONOS-2
MSa
4m
Model stereo pair

and digital
elevation model
Satellite, IKONOS-2
Pan/MSa, DEM
1 4 m, 15 m
MS = Multispectral, Pan = panchromatic, RMSE = root mean square error, CE = circular error.
a
Multispectral data emulating first four TM bands.
curement and evaluation of additional sources of commercial remote sensing data. DigitalGlobes QuickBird
high-resolution optical imagery is being evaluated as a
source of information for Earth science research and
applications. NASA contracted with EarthSat to provide
a global database of orthorectified scenes from Landsat
7 and mosaics made from these scenes. These scenes
were orthorectified using the TM dataset procured in the
NASA SDP Part I as a baseline to provide for easy
change detection between the periods. NASA procured
commercial LIDAR data from TerraPoint; unmanned
aerial vehicle based multispectral data from AeroVironment; and additional Intermap Technologies radar data
and Positive Systems multispectral data for a variety of
research and applications projects, including NASA Solid

Earth and Natural Hazards program research (TerraPoint),
agricultural research (AeroVironment, Positive Systems),
and aviation safety (Intermap). NASA also purchased
SPOT 4 Polar Ozone and Aerosol Measurement III
(POAM III) sensor data from Computational Physics
to support NASA-funded atmospheric research. Additionally, NASA supported a number of competitively
awarded research projects with funding from the SDP
Part II for procurement of various commercial data
products (including Radarsat, radar data; SPOT Image
multispectral data; Spencer B. Gross, LIDAR data; and
hyperspectral data from a variety of commercial sources).
Table 4 provides an overview of the data products that
10
Table 4
Overview of NASA Scientific Data Purchase Part II products
Image data product
Platform and sensor
Data type
Pixel size
Positional accuracy
Encoding
Radiometric
accuracy
Cloud
%
Data provider: EarthSat

Circa 2000 orthorectified
scenes
Satellite, Landsat 7,
ETM +
Pan (1-band)
MS (8-band)
NA
< 20%
ETM +
PS MS (3-band)
8 bits
NA
< 20%
Circa 2000 orthorectified,

pan-sharpened, mosaics
ETM +
PS MS (3-band)
14.25 m
F 75 m absolute
F 40 m relative
1990 RMSE
F 75 m absolute
F 40 m relative
1990 RMSE
F 75 m absolute
F 40 m relative
1990 RMSE
8 bits
Circa 2000 orthorectified,

pan-sharpened scenes
14.25 m
28.5 m
57 m
14.25 m
8 bits
NA
< 20%
MSa
0.7 m
F 100 m (center pt)
8 bits
< 10%
MSa
0.7 m
F 12.2 m (benign)
F 50 m
(extreme relief)
RMSE
8 bits
F 10% absolute
F 5% relative
NA
< 10%
X-SAR
2.5 m
8 bits
NA
NA
DEM
5-10 m
F 2.5 m horizontal
RMSE (1j)
F 2.5 m horizontal
F 1 3 m vertical
RMSE (1j)
32 bits
NA
NA
Pan
MSa
Pan
MSa
0.62 0.82
2.48 3.28
0.62 0.82
2.48 3.28
NA
11 bits
< 20%
F 23 m (CE 90%)
11 bits
F 10% absolute
F 5% relative
F 10% absolute
F 5% relative
Pan
MSa
0.62 0.82 m
2.48 3.28 m
11 bits
F 10% absolute
F 5% relative
< 20%
11 bits
NA
< 20%
NA
NA
NA
NA
Data provider: Positive Systems

IM-R1I-55 single-frame
Airborne, ADAR
imagery
5500
MOS-G1 georeferenced
Airborne, ADAR
mosaic
5500
Data provider: DigitalGlobe/Intermap

Orthorectified radar image
Airborne, STAR-3i
IFSAR
Digital elevation models
Airborne, STAR-3i
IFSAR
Data provider: DigitalGlobe

Basic image (full scene)
Radiometric correction
Standard image
Radiometric and geometric
correction
Orthorectified image
Radiometric correction
1:50,000
1:25,000
1:24,000
1:12,000
1:10,000
Pan-sharpened
Data provider: TerraPoint
Bald Earth digital terrain
model
Highest surface digital
surface model
Satellite, QuickBird
m
m
m
m
F 15.2 m (CE 90%)

F 7.6 m (CE 90%)
F 7.3 m (CE 90%)
F 6.1 m (CE 90%)
F 5.0 m (CE 90%)
Varies with product
< 20%
MS PS
0.62 0.82 m
Airborne, LIDAR
DTM
1.83 m
Airborne, LIDAR
DSM
1.83 m
MS
1m
NA
8 bits
NA
0 10%
MS
0.5 m
NA
8 bits
NA
0 10%
Vertical
profiles
Horizontal
resolutionb
200 km
NA
15
Estimated
retrieval errorsc
NA
Data provider: AeroVironment

Unmanned aerial vehicle
DuncanTech MS3100
Level A, calibrated and
digital camera
ground-registered image
Unmanned aerial vehicle
Kodak DCS
Level A, calibrated and
Hasselblad 555ELD
ground-registered image
digital camera
Data provider: Computational Physics
No image product solar
Polar Ozone and
occultation measurements Aerosol Measurement
of O3, NO2, H2O, total
(POAM III) on
density, and aerosol
SPOT 4
extinction
x,y = 1.83 m (CE 90%)

zbald = 0.3 m (CE 90%)
zveg = 0.9 m (CE 90%)
x,y = 1.83 m (CE 90%)
zbald = 0.3 m (CE 90%)
zveg = 0.3 m (CE 90%)
have been delivered for Part II of the Scientific Data

Purchase project.
11
to identify overlapping requirements in advance of data

procurement.
3.3. Verification and validation
3. USGS commercial imagery initiatives

3.1. Introduction
The USGS uses remote sensing technology as a fundamental source of data, products, and tools for mapping, for
making scientific observations of the Earths land surface,
and for a host of other vital uses ranging from resource
assessments and emergency management to homeland
security. Remote sensing from aircraft and space systems
has been a basic USGS tool for more than half a century.
The USGS commitment to acquire, archive, distribute, and
apply remotely sensed data has evolved as a result of its
mission history.
The recent growth of the commercial remote sensing
industry, both within the U.S. and abroad, offers a wealth of
new sources to the USGS for potential applications across a
dozen or more of its Earth science programs. The USGS Land
Remote Sensing Program (LRSP) has created a Commercial
Remote Sensing Project to evaluate commercial remotely
sensed data and to develop contracts to procure data for The
National Map, for other Department of the Interior (DOI)
Bureaus, and for other Federal civilian agencies.
The USGS needs to understand the performance characteristics of the sensors and geospatial data used to
support its mapping and science missions. For many
years, the USGS has calibrated the Nations analog aerial
mapping cameras. With the advent of a new generation
of digital cameras and high-resolution commercial satellite systems, the USGS is establishing a digital camera
laboratory and is upgrading field-based in-situ test ranges
to allow for the characterization of new products. In
addition, the USGS is contracting with academic institutions and is partnering with NASA to develop new
digital calibration methods and capabilities. For purposes
of assessing the geometric properties of remotely sensed
data, the USGS maintains and utilizes surveyed field test
sites located at or near its regional mapping centers and
has characterized optical, radar, and LIDAR technologies.
The test sites possess a significant number of accurate
ground control points that can be compared to locations
in satellite imagery or to elevation data to substantiate
vendor product specifications. The USGS utilizes partner
sites for testing radiometric properties of remote sensing
data and participates with NASA and NIMA in the Joint
Agency Commercial Imagery Evaluation team.
3.2. Commercial imagery acquisition

3.4. Application examples
The USGS is developing contracts to obtain a broad
range of commercially available satellite and airborne
data, including panchromatic, multispectral, hyperspectral,
LIDAR, and radar data. These contracts are distinguished
by their focus on off-the-shelf commercial sources and on
flexible data use and distribution terms to allow the data
to be shared with USGS partners and through The
National Map. Across Federal civil agencies using remotely sensed data, experience shows that centralized
procurement offers considerable cost savings to the Government through volume discounts, through reduction of
redundant contract administration costs, and through
avoidance of duplicate purchases. The USGS has provided
commercial data procurement assistance and expertise for
other agencies for several decades. It is expected that the
terms negotiated by the USGS on behalf of civil government agencies will be advantageous for the purchasing
organization. Consolidated contracts can also simplify the
ordering process for all parties and provide an opportunity
Remotely sensed data and technology have been widely

used within the DOI for decades, beginning with the
availability of aerial photography in the 1930s. During
the ensuing decades, USGS applications have encompassed
topographic mapping, land characterization, watershed
analysis, environmental assessments, mineral exploration,
global change research, geologic structure identification for
hazard assessment, floodplain assessment, biological resource and habitat analyses, and inventory and monitoring.
Various sensors have been deployed to generate panchromatic, multispectral, hyperspectral, infrared, and radar data
products. The successful launches of commercial highresolution satellites significantly expand the Nations imaging capacity and have stimulated the USGS to evaluate
these data for programmatic use. Within the USGS, IKONOS-derived digital elevation models were characterized,
and the use of IKONOS data has contributed significantly
to the agencys National Land Cover Characterization
Notes to Table 4:
MS = Multispectral, Pan = panchromatic, PS MS = pan-sharpened multispectral, CE = circular error, RMSE = root mean square error.
a
Multispectral data emulating first four TM bands.
b
Vertical resolution: O3 = 1 km (10 50 km), NO2 = 1.5 2.5 km (20 40 km), H2O = 1 2 km (10 40 km), aerosols = 1 1.5 km (10 30 km).
c
O3 = 3 5% (10 60 km), NO2 = 5 10% (20 45 km), H2O = 5 8% (10 50 km), aerosols = 10 20% (10 30 km).
12
Project4 and to research on forest structure and stand

density (Zanoni et al., 2003). Commercial satellite imagery
has the potential to contribute to The National Map, which
includes orthorectified imagery,5 elevation data, and selected data for transportation, hydrography, boundaries and
structures, land cover delineations, and geographic names.
The USGS believes that it will be possible to make
substantial improvements in the availability of current
orthorectified imagery by sharing the available high-resolution photography that is collected for state and local
government agencies, by directing the contracting for
additional aerial imagery, and by purchasing satellite imagery. Orthorectified imagery will be collected and maintained using the most efficient and effective combination of
airborne and spaceborne remote sensing capabilities.
4. NIMA Commercial Imagery Program

The NIMA CIP was established in early 1998 to assist
NIMA and its customers in the acquisition and use of
commercial imagery to support their missions. Because
the CIP came into being prior to the successful launch of
the first licensed 1-m commercial system, the early emphasis of the CIP was on educating the customer community on
the capabilities of the soon-to-be-available imagery and on
working with the licensed vendors to minimize the interface
issues associated with adding their data as a new source type
to be utilized by NIMA and its customers.
4.1. Commercial imagery acquisition
The provisions of DoD Directive 5105.60 (U.S. Department of Defense, 1996) designate NIMA as the action
agency for purchases of commercial imagery for the DoD
and, upon request, for other Federal agencies. This direction led NIMA to establish the Commercial Imagery
Program to focus on the establishment of cost-effective
purchasing contracts with commercial vendors. An area of
groundbreaking work had to be focused on the development and successful negotiation of detailed licensing
concepts to address the costs associated with a variety
of single- and multi-user scenarios. The licensing of
imagery provided by commercial remote sensing companies has evolved along with other working relationships
between vendors and the government. The evolution will
continue as government and industry continue to learn
more each day about the complexities and capabilities of
the industry. The emphasis has been on keeping the
environment truly commercial and on not dictating
NIMA-specific requirements.
Additional emphasis early-on was on educating the

potential customers and on assisting them in the selection,
ordering, and purchasing of appropriate commercial imagery sources to satisfy their requirements. By taking the
program office approach, NIMA brought together the capability to address commercial imagery from end-to-end in
one organizational element as opposed to working the issues
in each of the business-as-usual elements. This organizational approach is typically followed for quick-start, highvisibility efforts in new or emerging areas not yet mature
enough to integrate fully into the business-as-usual elements. NIMAs CIP has responsibility across the full Tasking, Processing, Exploitation, and Dissemination (TPED)
cycle (Fig. 4) and has tried to establish minimal capabilities
in each functional area.
Part of the early emphasis with any new data source is on
understanding the quality of the data, and commercial
satellite imagery shares this requirement. NIMA participates
in the JACIE team activities along with NASA and the
USGS. The initial JACIE evaluation of IKONOS took
longer than a year to accomplish. Additionally, JACIE has
continued evaluations of IKONOS for temporal stability.
Future evaluations of new systems and periodic re-evaluations of existing systems are required to be accomplished in
less time.
4.2. Evolution of NIMAs CIP
At the inception of the CIP, NIMA was faced with
taking on a new responsibility without the benefit of
additional funding, so the agency assigned modest resources to address the requirements at an acceptable level of
risk. Limited resources became a pacing factor of how
fast? and to what extent? NIMA could move to
expanded use of commercial imagery. NIMA is now
http://www.mrlc.gov.
A subset of The National Map layers corresponds to the National
Spatial Data Infrastructure primary data layers of orthoimagery, elevation,
hydrology, and transportation.
5
Fig. 4. NIMA TPED information cycle.
acquiring IKONOS imagery from Space Imaging, QuickBird imagery from DigitalGlobe, and a variety of other
spaceborne and airborne data types in response to diverse
customer requirements.
Since the events of 9/11, emphasis on the use of
commercial imagery has been growing for a variety of
reasons, not the least of which is the need to have the
flexibility of sharing imagery and imagery-derived products
with a diverse group of customers ranging from foreign
coalition partners to domestic first responders. This attribute
of commercial imagery was recognized from the beginning
of NIMAs involvement in the Commercial Imagery Program, but it has become much more important after 9/11 and
the birth of the Homeland Security mission.
Not only has there been increasing demand for unclassified imagery and its derived products, there has also been a
demand for improved spatial and temporal resolution, causing NIMA to expand its use of airborne data, such as
imagery with resolution of less than 1 m and LIDAR data
to capture very high levels of elevation detail over discrete
areas.
Concurrent with these increasing demands has been a
significant increase in funding that will facilitate increased
use and will address a more robust system engineering of
appropriate capabilities across the TPED cycle. In the same
time period, the number of operational spaceborne systems
has grown from one to two; a third system is planned to be
deployed and operational during calendar year 2003.
These factors have all come together to drive a maturing
of the industry and NIMAs usage of commercial remote
sensing data. NIMAs Commercial Imagery Program is
reflecting this maturity by moving each of the functions
gathered under a general program office umbrella out into a
business-as-usual operational scenario.
4.3. NIMA applications of commercial imagery
It is safe to say that commercial imagery is finding usage
in an expanding number of traditional imagery, imagery
intelligence, and geospatial intelligence applications. It is
also true that NIMA has been given clear guidance by those
responsible for the agencys oversight to expand the use of
commercial imagery.
Without going into specific details of where or why
commercial imagery has been collected in support of
Operation Enduring Freedom and the Global War on Terrorism, it is impressive to know that since 9/11, NIMA and
the commercial imagery industry have worked together to
achieve the following:

Commercial sources have collected and delivered more

than 400,000 km2 of imagery to NIMA.
NIMAs Commercial Satellite Imagery Library (CSIL)
data holdings have increased by more than 37%.
NIMA has disseminated each new image from the CSIL
an average of four times.
13
NIMA has disseminated over 35,000 images electronically or on CD-ROMs.
4.4. Future of commercial imagery

Commercial imagery utilization by NIMA and its customers is progressing at a significant pace. As with most
emerging technologies, a measure of success comes when
new technology matures to the point of no longer being
regarded as new or is dealt with outside of business-as-usual
processes. NIMA is working to improve responsiveness and
contract efficiency by pushing hard to evolve from simply
the delivery of pixels to the delivery of value-added products that directly meet its customers needs. NIMA continues to increase its use of commercial imagery and to seek
value-added production solutions that employ both commercial data and government-provided data.
5. Joint Agency Commercial Imagery Evaluation

As discussed earlier, the Joint Agency Commercial
Imagery Evaluation (JACIE) team includes NASA, the
USGS, and NIMA through an interagency government
partnership that works with industry and university affiliates
to validate and characterize data from commercial satellite
vendors (Zanoni et al., 2003). The cooperation between
government and industry has benefited all parties, including
the broader user community, in understanding the actual
characteristics and performance of remote sensing instruments with respect to both the specifications (verification)
and the use in geospatial solutions (validation). Each of the
partners brings different requirements and capabilities to the
joint evaluation process, allowing the partners to stay
focused on their own unique core functions and requirements while benefiting from the contributions and the
strengths of other team members. This approach reduces
the cost of a full evaluation by minimizing duplication of
effort by the government and industry. To date, this joint
effort has addressed a variety of independent verification
and validation areas including radiometric calibration, image quality, and geometric calibration.
The JACIE team is established under a NASA Space
Act Agreement to establish interagency collaboration for
characterizing commercial imagery. The JACIE team provides a single government interface for data characterization to commercial remote sensing companies. When
working with Space Imaging, the JACIE working group
characterized the IKONOS system and coordinated its
efforts to obtain appropriate system information from the
company. The initial effort focused on IKONOS because it
was the first commercial remote sensing satellite to achieve
operations on orbit, but as each new system becomes
available, similar evaluations will be needed. Likewise,
periodic re-evaluations will be required so that government
and industry can continue to use the systems with the
14
confidence of knowing that the systems are performing

according to expectations.
6. Lessons learned
The following summarizes findings and lessons learned
by the three agencies in the course of developing and
implementing programs to procure and use geospatial data
products delivered from commercial sources of remote
sensing systems.

NIMA, the USGS, and NASA have determined that the

commercial sector can deliver remote sensing data
products to serve our Nations economic, homeland,
and national security interests. Procuring commercial
remote sensing data from vendors through contracts with
terms of payment based on delivery of products makes
information and data products available to the government with no associated risk of system development and
deployment (i.e., for government procurements of
commercial geospatial data, the taxpayer does not
assume the burden of risk associated with the development and deployment of the remote sensing systems).
As the commercial remote sensing industry continues to
develop its capacity to meet Earth science and defense
requirements, findings from all three agencies indicate
that there are opportunities for the companies to enhance
customer service, product delivery times, and optimization of tasking over identified areas of interest. Data
specifications, minimum order sizes, licensing, pricing,
distribution policies, and acquisition windows are areas
that can be evolved to serve government purposes better.
As performance requirements are diverse across the three
agencies, solutions may need to be tailored to individual
agency missions and needs. There does not appear to be a
one size fits all solution.
NASA, the USGS, and NIMA have found that the
government requires data products to be characterized,
verified, and validated for use in Earth science research
and national defense applications. In contrast to government-owned systems and in the interest of protection of
intellectual property, commercial providers do not readily
release detailed engineering descriptions of their systems.
The commercial sectors limited descriptions of component performance or of the specific processes employed
to produce data products introduce a new paradigm for
the government community, which historically has had
significant insight into sensor design and operational
characteristics through direct oversight of contractors and
systems engineers. Commercial providers may not
characterize systems in the manner needed to support
government applications because of differences in
expectations for performance.
The JACIE approach to collaboration on geospatial data
product and remote sensing system performance provides
an effective solution to the challenge of diverse requirements for verification and validation. The JACIE
approach has been so successful that companies have
proposed using the university teams participating in the
JACIE to support their in-house performance characterizations. The JACIE team recognizes that if university
teams participated in in-house calibrations for companies, that it would severely impact the governments
ability to perform independent assessments of the
companies data. It is apparent that more teams and
methods for calibration and validation are needed.
The commercial sector plans to expand the availability of
data sources and types of remote sensing data products
with the potential to serve even broader needs of Earth
science and mapping communities within the U.S.
Government. The successful launches of Space Imagings IKONOS, DigitalGlobes QuickBird 2, and OrbImages OrbView-3 are providing increasing capacity for
the delivery of high-spatial resolution optical imagery
and derived products.
New licenses issued by the Department of Commerce
authorize on-orbit commercial systems to provide data
with spatial resolutions of 0.5 m. High-spatial resolution
products correlate to higher market shares, which may
correlate to increased interest by remote sensing
companies to provide higher spatial resolution datasets
on future systems.
Remote sensing companies will continue to be a primary
source of high-spatial resolution products available to the
Earth science and mapping communities supported by
federally funded acquisitions and/or by land remote
sensing programs, given current policies restricting the
U.S. Government from competing with the private sector.
The majority of the datasets offered by the commercial
sector have been in the form of terrestrial imagery
products and digital elevation model products. Government policies to populate the National Spatial Data
Infrastructure with the primary framework data layers
and other national priority applications projects continue
to increase demand for these types of commercial
geospatial data products.
Advancements in remote sensing technologies and
systems indicate increased capacity in currently deployed
approaches and project a trend for additional types of
data products from both airborne and spaceborne
platforms.
NASA assessed the performance of the Scientific Data

Purchase through surveys distributed with each dataset
delivered to participating scientists. The survey evaluated
the usefulness of the data and the quality of the services
provided and reported a favorable response rate of approximately 90%. Survey results indicate that the members
of the Earth science research community who participated
in the program realized benefits from provision of the
private-sector geospatial data delivered through the pro-
gram. The licensing and distribution practices negotiated

between NASA and the companies participating in the
Scientific Data Purchase were determined to be effective
in meeting a subset of the governments needs for geospatial data products to contribute to Earth science research and applications.
In response to the lessons learned in the Scientific Data
Purchase experiment, NASA has implemented a policy to
include a provision in each solicitation for Earth science
research and applications proposals for the option to use
commercial remote sensing data products.
7. Recommendations
Specific recommendations associated with U.S. Government agencies regarding the future of commercial data
purchases to serve national purposes include the following:

To facilitate widespread use and acceptance of commercial sources of remotely sensed data in the science and
operational communities, government-funded data purchases need to ensure adequate data rights for all
anticipated distribution and sharing.
The continuity of measurements must be considered
when evaluating the use of commercial data for Earth
science, monitoring, and mapping applications. Where
data is deemed necessary for public research or
operations, reliance on commercial assets comes with
the risk that the data stream may not be sustained or
remain consistent. Government policies must ensure the
compatibility of data from different sources through
rigorous calibration and validation activities.
Increase collaboration between and among Federal
agencies to promote data product standards to the
commercial remote sensing industry, facilitating economies of scale.
Maintain coordinated joint activities, such as the JACIE
team, to provide independent performance assessments
that combine the interests of several agencies while
providing a single interface to the commercial remote
sensing industry. Benefits include improved communications with industry, greater government and consumer
confidence in commercial products, and improved
techniques within industry. As the number of sources
increases for any given product type, the importance of
calibration and characterization against a baseline
performance standard becomes a critical element in
establishing the capacity of multiple sources to serve
specific applications on an interchangeable basis.
Increase cooperation with state and local governments in
the area of validating new datasets and sensors by sharing
infrastructures, experiences in methods development, and
empirical knowledge from real-world applications.
Create a national civil remote sensing strategy that
establishes a framework for government requirements
15
and funding to meet the range of common and disparate

needs for geospatial information to serve science and
operational uses. Historically, commercial data purchases
have been implemented on an ad hoc basis throughout
the U.S. Government. The private sector has difficulty in
responding to the range of mission requirements with
product lines that can serve multiple applications and
agency needs. National market surveys indicate that a
dependable U.S. Government market is critical to the
viability of the commercial remote sensing industry.
Benchmark the use of commercial data in decision support
systems administered by U.S. Government Federal
agencies to expand applications and demand; this is key,
as indicated by the Civil Imagery and Remote Sensing
Task Force of the Federal Geographic Data Committee in
its report to the Office of Science and Technology Policy
and Office of Management and Budget (Federal Geographic Data Committee, 2002). The Task Force has
assessed the importance of a systematic evaluation of civil
data acquisition, management, and distribution requirements along with the legislative, management, and budget
options needed to develop a national civil imagery and
remote sensing strategy.
References
Andrefouet, S., Muller-Karger, F., Palandro, D., Hu, C., Carder, K., Hochberg, E., & Maeder, J. (2001). High resolution IKONOS data for coral
reefs studies. Proceedings of the 2001 High Spatial Resolution Commercial Imagery Workshop, March 19 21, Greenbelt, MD, USA.
(Sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery
Evaluation Team, CD-ROM).
Ashton, M. S., Thenkabail, P. S., Enclona, E., Hall, J., Stucky, N., Harris,
D., Van Der Meer, B., Lin, T., & Griscam, B. (2002). Characterization
of Humid-Forest and Savanna Ecoregions of West and Central Africa
Using Satellite Sensor Data of Three Eras: Characterization of Eco
Regions in Africa (CERA) Annual Report for the Period February 28,
2001 to February 18, 2002 (p. 19). New Haven, CT: Center for Earth
Observation and School of Forestry and Environmental Studies, Yale
University.
Capolsini, P., Andrefouet, S., Rion, C., & Payri, C. (2003). A comparison
of Landsat ETM+, SPOT HRV, Ikonos, ASTER, and airborne MASTER data for coral reef habitat mapping in South Pacific islands. Canadian Journal of Remote Sensing, 29(2), 187 200.
Chang, J., Clay, D. E., Dalsted, K., & ONeill, M. (2002). The influence of
remote sensing data on predicting corn yield. Pecora 15/Land Satellite
Information IV Conference in Conjunction with International Society
for Photogrammetry and Remote Sensing (ISPRS) Commission I Midterm Symposium, November 10 15, Denver, CO, USA.
Clinton, W. J. (1994). Coordinating geographic data acquisition and access:
The national spatial data infrastructure. Executive Order 12906, The
White House, April 11. Federal Register, 59(71), 17671 17674.
Available at: http://www.fgdc.gov/publications/documents/geninfo/
execord.html (accessed July 18, 2003).
Driscoll, S., & Lawrence, R. L. (2002). Effects of color balancing of airborne multispectral imagery on invasive plant mapping. 2002 ASPRS/
ACSM Annual Conference, April 22 26, Washington, DC.
Federal Geographic Data Committee (2002). Report to the administration:
Value of civil imagery and remote sensing. FGDC Civil Imagery and
Remote Sensing Task Force. http://www.fgdc.gov/cirs/cirs.pdf.
Gieck, R. E., Kane, D. L., Hinzman, L. D., Overduin, P. P., McNamara,
16
J. P. (2002). Measurement of solid-state precipitation at the watershed

scale. EOS, Transactions, American Geophysical Union, 83(47) (Fall
Meeting Supplement, Abstract H11D-0875) Available at: http://www.
agu.org/meetings/fm02/fm02-pdf/fm02_ H11D.pdf.
Hansen, M. C., DeFries, R. S., Townshend, J. R. G., Marufu, L., & Sohlberg, R. (2002). Development of a MODIS tree cover validation data
set for Western Province, Zambia. Remote Sensing of Environment,
83(1 2), 320 335.
Hipple, J. D. (2001). Using hyperspectral remote sensing to quantify within-field spatial variability. Proceedings 3rd International Conference on
Geospatial Information in Agriculture and Forestry, Denver, CO.
Hipple, J. D., & Daugherty, D. J. (2000). Urban validation site for testing
impervious surface models derived from remotely sensed imagery.
IEEE International Geoscience and Remote Sensing Symposium Proceedings, V, 2074 2076.
Homer, C., Huang, C., Yang, L., & Wylie, B. (2002). Development of a
circa 2000 landcover database for the United States. 2002 ASPRS/
ACSM Annual Conference, April 22 26, Washington, DC.
Jenkins, C., & Anderson, A. B. (2002). Using conservation priorities to
design a biological corridor in the Atlantic forest of Brazil. 16th Annual
Society for Conservation Biology, Canterbury, England.
Koeln, G., Jones, T., & Melican, J. (2000). GeoCover LCTM: Generating
global land cover from 7600 frames of Landsat TM data. Proceedings
of ASPRS 2000 Annual Conference, May 22 26, Washington, DC.
Masuoka, P., Claborn, D., & Andre, R. (2001). Use of IKONOS and Landsat to estimate size of mosquito habitat for malaria control in South
Korea. Proceedings of 2001 High Spatial Resolution Commercial Imagery Workshop, March 19 21, Greenbelt, MD, USA. (Sponsored by
NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation
Team, CD-ROM).
Mercuri, P. A., Engel, B. A., & Johannsen, C. J. (2000). Using high-accuracy digital elevation model for agricultural applications. American
Society of Agronomy Abstracts, 85 (Minneapolis, MN).
Muskett, R. R., Lingle, C. S., Tangborn, W. V., & Rabus, B. T. (2003). Multidecadal elevation changes on Bagley Ice Valley and Malaspina Glacier,
South Central Alaska. Tohoku Geophysical Journal, 36(4), 422.
NASA Office of Earth Science (2000). Understanding earth system change:
NASAs Earth science enterprise research strategy for 20002010.
Available at: http://www.earth.nasa.gov/visions/researchstrat/
Research_Strategy.htm.
NASA Office of Earth Science (2002). Earth science enterprise applications

strategy for 20022012. Available at: http://www.earth.nasa.gov/
visions/appstrat2002.pdf.
National Research Council (2002). Toward New Partnerships in Remote
Sensing: Government, The Private Sector, and Earth Science Research.
Steering Committee on Space Applications and Commercialization,
Space Studies Board. Washington, DC: The National Academies Press.
Nolan, M., & Prokein, P. (2003). Evaluation of a new DEM of the
Putuligayuk Watershed for Arctic hydrological applications. 8th International Conference on Permafrost, July 21 25, Zurich, Switzerland.
(In press).
OPS (1994). Foreign access to remote sensing space capabilities: Fact
sheet. The White House, Office of the Press Secretary, Presidential
Decision Directive 23, March 10. http://www.fas.org/irp/offdocs/
pdd23-2.htm, accessed July 18, 2003.
OSTP (2003). U.S. commercial remote sensing policy: Fact sheet. Executive Office of the President of the United States, Office of Science and
Technology Policy, April 25.
Palandro, D., Andrefouet, S., Dustan, P., & Muller-Karger, F. E. (2003).
Change detection in coral reef communities using Ikonos satellite sensor imagery and historic aerial photographs. International Journal of
Remote Sensing, 24(4), 873 878.
Sever, T. L., & Irwin, D. E. (2003). Landscape archeology: Remote-sensing
investigation of the ancient Maya in the Peten rainforest of northern
Guatemala. Ancient Mesoamerica, 14(01), 113 122.
Thenkabail, P. S., Lin, T., Hall, J., Ashton, M., & Harris, D. (2003).
Detecting floristic changes across topographic gradients and moisture
regimes in a Central African rain forest using IKONOS and ETM+
satellite imagery. Forest Ecology and Management (submitted for
publication).
U.S. Department of Defense (1996). National Imagery and Mapping
Agency (NIMA). Department of Defense Directive 5105.60, October 11.
U.S. Government (1992). Land Remote Sensing Policy Act of 1992. 15
U.S.C. 5601 et seq.; Public Law 102 555, 106 Stat. 4163.
Snyder, G., & Lee, G. (2003). The joint agency commercial imagery
evaluation (JACIE) team: Overview and IKONOS joint characterization
approach. Remote Sensing of Environment, 88, 17 22 (this issue).

The Joint Agency Commercial Imagery Evaluation team:

overview and IKONOS joint characterization approach
Vicki Zanoni a,*, Tom Stanley a, Robert Ryan b, Mary Pagnutti b, Braxton Baldridge c,
Spencer Roylance d, Greg Snyder e, George Lee e
a
NASA Earth Science Applications Directorate/MA20, Stennis Space Center, MS 39529, USA
Lockheed Martin Space OperationsStennis Programs, Bldg. 1105, Stennis Space Center, MS 39529, USA
c
Booz Allen Hamilton, NIMA Commercial Imagery Program, now with Research Systems, Inc., Vienna, VA 22180, USA
d
Booz Allen Hamilton, NIMA Commercial Imagery Program, USA
e
U.S. Geological Survey, USA
b
Received 15 July 2002; received in revised form 17 April 2003; accepted 9 July 2003
Abstract
The Joint Agency Commercial Imagery Evaluation (JACIE) team was formed to leverage government agencies capabilities for the
characterization of commercial remote sensing data. The team is composed of staff from the National Aeronautics and Space Administration
(NASA), the National Imagery and Mapping Agency (NIMA), and the U.S. Geological Survey (USGS). Each JACIE agency has a vested
interest in the purchase and use of commercial imagery to support its research and applications. It is critical that this imagery be assessed for
its accuracy and utility. Through JACIE, NASA, NIMA, and USGS jointly characterized image products from Space Imagings IKONOS
satellite. The JACIE team acquired IKONOS imagery of several study sites to perform the assessments. Each JACIE agency performed an
aspect of the characterization according to its area of expertise. NASA and its university partners performed a system characterization
focusing on radiometric calibration, geopositional accuracy, and spatial resolution assessment; NIMA performed image interpretability and
feature extraction evaluations; and the USGS assessed the geopositional accuracy of several IKONOS products. The results of the JACIE
teams IKONOS effort, which were discussed with Space Imaging and then presented at an industry government workshop, ensured and
improved overall product quality and benefited both the commercial industry and the government. Additional JACIE activities include the
characterization of other commercial products such as those from the DigitalGlobe QuickBird and the Orbital Imaging (Orbimage) OrbView3 satellites.
D 2003 Published by Elsevier Inc.
Keywords: Joint Agency Commercial Imagery Evaluation team; IKONOS; Remote sensing
1. Introduction
The purchase of commercial remote sensing data by U.S.
federal agencies is a relatively new way of doing business
that has been and is continuing to be implemented within
government agencies. Commercial data has been purchased
to support the fields of Earth science research, civil government applications, and defense-related intelligence gathering. Strategic partnerships that capitalize on common
commercial data purchase goals can be beneficial. Both
the National Aeronautics and Space Administration (NASA)
and the National Imagery and Mapping Agency (NIMA)

have purchased large amounts of commercial satellite data.
The U.S. Geological Survey (USGS) is also planning to
implement commercial satellite data purchase activities.
Because these agencies experience common issues dealing
with the purchase of commercial imagery, they have formed
a Joint Agency Commercial Imagery Evaluation (JACIE)
team. The JACIE team leverages capabilities for commercial data characterization, and provides a unified government voice when interfacing with industry.
2. Background
E-mail address: vicki.m.zanoni@nasa.gov (V. Zanoni).
doi:10.1016/j.rse.2003.07.005
In 1997, the NASA Earth Science Enterprise (ESE)

initiated the Scientific Data Purchase (SDP), a US$50
18
V. Zanoni et al. / Remote Sensing of Environment 88 (2003) 1722
million project to ascertain the utility of commercial remote

sensing datasets for Earth science research and applications.
Through the SDP, NASA purchased commercial remote
sensing data products from four data providers: Earth
Satellite, Positive Systems, DigitalGlobe (formerly EarthWatch)/Intermap Technologies, and Space Imaging. The
NASA Stennis Space Center (SSC) Earth Science Applications (ESA) Directorate implemented the SDP and was
responsible for the verification and validation of the delivered datasets. A more detailed description of the SDP is
available in Birk et al. (2003).
NIMA established the Commercial Imagery Program
(CIP) in 1998 to support the acquisition and exploitation
of commercial imagery for its own use and for use by its
Department of Defense (DoD) customers. The CIP purchases imagery products from Space Imaging, SPOT Image,
DigitalGlobe, and Intermap Technologies. Imagery purchases through the CIP also include license upgrades so
the data can be shared across all DoD/Intelligence activities.
The CIP also purchases Landsat 5 and Landsat 7 data from
the U.S. Geological Survey (USGS) EROS Data Center.
Assessments of image quality and of the utility of commercial imagery sources are performed by NIMAs Civil and
Commercial Applications Project, which is part of the
agencys Imagery Support and Assessments Branch (NIMA,
2000).
The USGS expects to increase its use of commercial
geospatial data in its science and mapping programs. Toward this end, the USGS Land Remote Sensing Program is
exploring contractual mechanisms and agreements to facilitate cost-effective access to a wide range of commercial offthe-shelf remote sensing data sources. The USGS is also
interested in validating the accuracy and utility of such data
sources, both U.S. and foreign. Planned contracts would be
available for optional use by other civil agencies to help
meet their commercial imagery needs. This work is a
continuation and expansion of the USGSs historical and
long-standing service to the civil community.
Because Space Imaging, was the first company to launch
a commercial high-resolution remote sensing system successfully, NASA, NIMA, and the USGS were interested in
understanding IKONOS utility for science research and for
civil- and defense-related applications. Commercial data
products must be highly characterized to be useful to the
governments science and applications communities. However, commercial providers might not characterize systems
in the manner desired by scientific researchers because
commercial and science requirements often differ. This
paradigm is new to the governments imagery user community, which historically has had significant insight into
sensor design and operation. Today, government-built satellite systems cannot compete with commercial industry
(National Research Council, 2002). Thus, commercial providers must fill government requirements for high-resolution data. This procurement of data from the commercial
sector requires independent data characterization.
3. JACIE formation and activities

As part of its SDP program, NASAs conducted an
extensive effort to independently characterize the performance of the IKONOS system. Because NIMA also procures IKONOS data and because the USGS has interest in
procuring IKONOS data, the three agencies formed the
JACIE team in February 2000 to capitalize on mutual
interests and to leverage resources for the characterization
of commercial remote sensing data. NASA, NIMA, and the
USGS signed a JACIE Interagency Space Act Agreement in
June 2002 (NASA, 2002).
Each JACIE team member brings different strengths to the
characterization activity, minimizing duplication of effort by
the government and industry. The USGS, with its long history
in cartography, conducts IKONOS geometric accuracy
assessments that include evaluation of IKONOS georectified,
orthorectified, digital elevation model, and stereo image
products. NIMA and its customers perform DoD and intelligence-based product evaluations, including image interpretability, feature extractions, and photogrammetry assessments.
NIMAs customers also evaluate the utility of IKONOS data
for DoD and intelligence use. NASA performs overall system
performance characterization by evaluating SDP products
against contract specifications for spatial resolution, geometric accuracy, and radiometric accuracy. Additional NASA
efforts include evaluations of IKONOS data processing
techniques and comparisons with other systems, such as
Landsat 7 and the Moderate Resolution Imaging Spectroradiometer (MODIS). NASA has partnered with experts
from the Earth Science community to support system characterization. These experts from South Dakota State University, the University of Arizona, and the University of
Maryland have years of experience in characterizing and
validating NASA-developed sensors. Each agencys customers also play a valuable role in the characterization effort
as they use IKONOS data to support science research and
applications.
Multiple team members independently perform each
aspect of the characterization effort as illustrated in Table
1, thus strengthening the overall result. JACIE team
Table 1
JACIE characterization team members and roles
Organization
USGS
NIMA
NASA
University of
Arizona
South Dakota State
University
University of
Maryland
Government User
Community
Space Imaging
Spatial
X
X
X
X
Geometric
X
X
X
Radiometric
Applications
X
X
X
X
X
X
X
members have worked together to plan IKONOS data

acquisitions, to conduct field measurement campaigns,
and to analyze IKONOS imagery.
19
4.3. Stereo imagery mapping accuracy
JACIE-led vicarious calibration efforts determined that

Space Imagings radiometric calibration coefficients were
inconsistent with those produced by the JACIE team (Pagnutti et al., 2003). The JACIE team collaborated with Space
Imaging to investigate the inconsistency. As a result, the
JACIE-produced radiometric calibration results were incorporated into an updated set of Space Imaging calibration
coefficients (Peterson, 2001). This adjustment to IKONOS
radiometry ensured the validity of science results derived
from the IKONOS data in applications such as land cover
change detection, agriculture, NASA Earth Observing System (EOS) product validation, carbon sequestration, and
habitat mapping.
As part of a collaborative effort undertaken by NIMA,

the USGS, and Space Imaging, a comparison of Space
Imagings rigorous sensor model and an adjustable form
of the rational polynomial coefficient (RPC) model was
conducted through a large area block adjustment evaluation.
The objective was to determine if the adjustable RPC model
implemented by NIMA could be used to triangulate multiple
IKONOS stereo strips to NIMA mapping accuracy specifications. Test cases included individual stereo pairs as well as
triangulated blocks of overlapping strips. Image strips were
triangulated with varying numbers of ground control points.
During the course of the evaluation, it was revealed that a
two-pixel bias on the western edge of the stereo strips was
producing a cumulative error in the block adjustment (Ager
& Bresnahan, 2002). This error was investigated and
subsequently corrected by Space Imaging, and Space Imaging has been able to perform a similar block adjustment
that exceeded NIMAs accuracy requirement (Dial & Grodecki, 2002). The detection of this pixel bias, the open
discussion of its possible causes and corrections, and Space
Imagings response to its evaluation partners at NIMA and
the USGS to remedy the error occurred within the JACIE
mechanism invoked by the principal investigators at NIMA
and the USGS.
4.2. Spatial resolution
4.4. Digital elevation model vertical accuracy
Space Imaging applies an image restoration algorithm

called Modulation Transfer Function Compensation
(MTFC) that sharpens the imagery and improves feature
identification. Application of this algorithm effectively
enhances the imagerys spatial resolution but increases
noise. JACIE investigations (Pagnutti et al., 2003; Ryan et
al., 2003) discovered that the MTFC algorithm, or kernel,
was incorrectly applied by Space Imaging, creating an overcompensation in the image cross-track direction and an
under-compensation in the along-track direction. Space
Imaging subsequently rotated the MTFC kernel to correct
the error.
The JACIE teams analysis of MTFC allowed NASA
SDP personnel to advise science customers whether to
apply MTFC to imagery for specific study sites. For
quantitative studies requiring high signal-to-noise ratio
(SNR), MTFC should not be applied. For example, radiometric and spectral analyses in sites with high spatial
variability should not include MTFC. When visual inspection is of high importance, MTFC can improve image
quality and can be applied with no significant adverse
effects. In a specific example where MTFC was applied in
support of coral reef assessments by Andrefouet and
Muller-Karger (submitted for publication), NASA SDP
personnel advised that the data be reprocessed without
MTFC to ensure the validity of radiometric values in these
spatially variant regions.
The USGS evaluated digital elevation models (DEMs)

produced from IKONOS stereo pairs and found that the
vertical accuracy of the DEMs in some cases exceeded the
expected error limit of 7 m. A DEM over one study site
revealed a systematic vertical bias of about 15 m, double the
expected limit. Because of this finding, Space Imaging
changed its DEM production procedures to require more
than one stereo-image pair in the DEM block adjustment
(Constance, 2002).
4. Results from JACIE analyses

JACIE data evaluation results have been produced for
each aspect of the characterization, resulting in several key
improvements to IKONOS data quality as described below.
4.1. Radiometry
5. Agency user input

NASAs Earth science community has also played a
key role in the data characterization effort. As SDP
customers, scientists assess the performance of IKONOS
data in various research areas. IKONOS data has supported research in precision agriculture (Seelan, Laguette,
Casady, & Seielstad, 2003), in coral reef assessment
(Andrefouet et al., 2003), in Antarctic ice shelf investigation (Bindschadler, 2001), in urban environments
(Small, 2003), and many other areas (see other papers
in this issue). In many cases, scientists uncovered data
issues that were detectable only by performing specific
scientific studies.
Likewise, NIMA customers within the DoD conduct
defense and intelligence applications using IKONOS data.
20
Uses of IKONOS imagery by NIMA include supporting

civil authorities in disaster response (Sidor, 2001), updating theater and tactical maps for U.S. Armed Forces
(NIMA, 2001), and producing specialized imagery and
map products for the U.S. Naval fleet and special forces
(Ecuyer, 2001).
Within the USGS, the use of IKONOS data has
significantly contributed to the agencys National Land
Cover Characterization Project. Through use of IKONOS
data provided by the NASA SDP and use of other highresolution data, an approach was developed to quantify
impervious surfaces and forest canopy density. The practicality and affordability of the proposed method for
large-area mapping was successfully tested over several
geographic areas. The approach has been implemented
and recommended for mapping two ecologically important
parameters for the entire United States. The IKONOS data
acquired from diverse biome and ecoregions are especially valuable for training and testing land-cover algorithms
and for mapping processes where other high-resolution
imagery is not available or is of low quality (Yang,
Huang, Homer, Wylie, & Coan, 2003). In another USGS
application, Chen et al. (2002) used IKONOS panchromatic and multispectral imagery, Landsat 7 imagery, and
airborne scanning LIDAR imagery to estimate forest
structure and stand density in the Black Hills of South
Dakota. The results of this study indicated that IKONOS
data can make the important distinctions between tree
canopy coverage and exposed understory grasses near
peak summertime greenness.
6. Communicating JACIE results

Results of the JACIE characterization efforts are presented at annual JACIE High Spatial Resolution Commercial Imagery Workshops (NASA, NIMA, & USGS,
2001; NASA, NIMA, & USGS, 2002; NASA/NIMA/
USGS, 2003). The annual workshop communicates the
results of the JACIE characterization effort and provides a
forum for data user and industry interaction. The workshop is unique in its focus on the use of commercially
purchased, high-spatial-resolution remote sensing data
within the government. IKONOS characterization results
and government applications were presented at the 2001
workshop held in Greenbelt, MD, and at the 2002 and
2003 workshops held in Reston, VA. More than 150
representatives from NASA, NIMA, the USGS, and the
agencies user communities attended each workshop.
Representatives from Space Imaging also participated in
the workshops and presented results of the companys
internal IKONOS calibration efforts. Feedback from
workshop participants and Space Imaging representatives
indicated that the meetings were a success and that
similar workshops should continue to be held in the
future.
7. Benefits and conclusions from the JACIE experience

The JACIE team collaboration has produced benefits
beyond the positive results stemming from the data characterization. The group provides a single government interface
to Space Imaging, not only for characterizing the IKONOS
system but also for obtaining system information from
Space Imaging.
Through JACIE technical interchange meetings, several
previously unknown features concerning IKONOS data
were uncovered, including the use of MTFC as described
above. For instance, the JACIE team learned that the
IKONOS system employs lossless compression onboard
the spacecraft for telemetry purposes. This compression
minimally affects research because the 11-bit data provides
an increased dynamic range-almost an order of magnitude
better than existing 8-bit systems.
Through JACIE interaction, NASA also learned that
NIMA purchased certain data products from Space Imaging
that were not available to NASA at that time. As a result,
NASA modified the Space Imaging SDP contract so that
these products would also be made available to NASAs
Earth science customers.
The relationship that developed between government and
industry played a key role in the success of the JACIE
effort. It is important to note that the JACIE effort is not
intended to discredit industry or to cast suspicion on a
commercial providers ability to provide high-quality data.
The joint government agency activity to conduct independent assessments helps the government become a smart
buyer of the data for its users. Any discrepancies in
IKONOS data that were uncovered through JACIE efforts
were discussed openly and constructively with Space Imaging. The interactions between Space Imaging, NASA,
NIMA, and the USGS established a spirit of trust and
cooperation toward improving IKONOS data quality for
all users.
The JACIE team provides a valuable service to both
the remote sensing industry and the government. By
performing independent evaluations of commercial systems,
the JACIE team provides a non-biased assessment of commercial data. These assessments can only help the industry
to gain credibility in both government and non-government
markets.
Data characterization is of vital importance to the government. NASA scientists use of commercial data can
potentially influence decisions concerning global change
issues and related public policies. NASA must ensure that
the delivered data products possess the accuracies promised
by the data provider. It is imperative that these sources of
data be well characterized in well-understood ways so they
are used appropriately.
Likewise, NIMA supports the DoD intelligence community through the distribution and management of remotely sensed data. Commercial data is used by NIMA
users to support intelligence and mapping tasks that are
driven by NIMA/DoD/Intelligence requirements and standards. Commercial imagery can best be used to answer an
intelligence problem if the imagerys quality and utility are
known and trusted. NIMAs quantification of imagery
interpretability is typically expressed through a National
Imagery Interpretability Rating Scale, or NIIRS (Leachtenauer, Malila, Irvine, Colburn, & Salvagio, 1997), a form
of assessment not normally performed by either the civil
community or the vendor. Similarly, commercial imagery
intended for the production of maps must meet NIMA
accuracy standards, requiring an independent verification
of accuracy. Thus, it is critical that commercial data be
independently characterized to ensure that defense and
intelligence decision makers use accurate and reliable
information.
The USGS also has a vital interest in understanding the
qualities and characteristics of the imagery that it applies
in its scientific investigations and in the mapping products
that it provides to the Nation. Furthermore, many Federal
and State government agencies rely on the USGS to
assess commercial products for their mapping applications. All of these uses of IKONOS data strongly justify
the need for an independent commercial data characterization effort.
8. Future role of JACIE

The JACIE team plans to evaluate other commercial
satellite systems, as they become operational. The team
has already performed an initial characterization of DigitalGlobes QuickBird image products (NASA/NIMA/USGS,
2003). This system provides 0.6-m panchromatic and 2.44m multispectral imagery and provides the best non-classified spatial resolution imagery currently available from
space. The team will also characterize the 1-m panchromatic
and 4-m multispectral imagery from the Orbimage OrbView-3 satellite system. Other potential roles for the JACIE
team include the following:

Develop data requirements by interacting with user

communities
Research and develop new data characterization
techniques
Develop standard characterization methods and
procedures
Support the establishment of guidelines and standards for
procuring remotely sensed imagery
Support the development of data policies
Design and implement industry government workshops
The release of the U.S. Commercial Remote Sensing

Policy in April 2003, defined new guidance for U.S.
government use of spaceborne commercial remote sensing
capabilities. The policy directs the U.S. government to rely
on spaceborne commercial remote sensing capabilities, to
21
the maximum practicable extent, for meeting imagery and

geospatial needs (OSTP, 2003). This encouraged use of
commercial data within the government necessitates continued independent characterizations, such as those performed
by the JACIE team.
9. Summary
The NASA, NIMA, and USGS JACIE team was established in 2000 and formalized in June 2002 with a goal of
performing independent characterization of commercial data
used by the three agencies. The team has collaborated to
characterize the 1-m panchromatic and 4-m multispectral
image products available from the Space Imaging, IKONOS
satellite. Characterizations focused on radiometric accuracy,
spatial resolution, and geopositional accuracy assessments.
The JACIE effort and collaborative interactions with Space
Imaging have resulted in several improvements to IKONOS
data quality. Both Space Imaging and the government have
benefited from product improvements and from the trust that
has evolved from this government industry collaboration.
NASA, NIMA, and USGS data users have performed a
variety of research and applications projects using IKONOS
data. JACIE characterization and user research results were
presented at JACIE-sponsored workshops in 2001, 2002,
and 2003. JACIE team activities also include characterization of DigitalGlobes QuickBird and Orbimages OrbView3 high-resolution panchromatic and multispectral data.
Acknowledgements
The authors would like to thank Marcia Wise for her
support in the editing of this manuscript.
References
Ager, T., & Bresnahan, P. (2002). IKONOS stereo imagery accuracy assessment. Proceedings of the 2002 High Spatial Resolution Commercial
Imagery Workshop, March 25 27, Reston, VA, USA, sponsored by
Team, CD-ROM.
Andrefouet, S., Kramer, P., Torres-Pulliza, D., Joyce, K. E., Hochberg,
E. J., Garza-Perez, R., Mumby, P. J., Riegl, B., Yamano, H., White,
W. H., Zubia, M., Brock, J. C., Phinn, S. R., & Muller-Karger, F. E.
(2003). Multi-sites evaluation of IKONOS data for classification of
tropical coral reef environments. Remote Sensing of Environment, 88,
127 142. (doi:10.1016/j.rse.2003.04.005)
Andrefouet, S., & Muller-Karger, F. (2003). Influence of pre-processing of
Ikonos images on texture characterization of bottom types in shallow
coral reefs. IEEE Transactions on Geosciences and Remote Sensing,
(submitted for publication).
Bindschadler, R. (2001). Ice dolines on Larsen Ice Shelf, Antarctica: An
application of IKONOS data. Proceedings of the 2001 High Spatial
Resolution Commercial Imagery Workshop, March 19 21, Greenbelt,
MD, USA, sponsored by NASA/NIMA/USGS Joint Agency Commercial
Imagery Evaluation Team, CD-ROM.
22
uses of commercial remote sensing data. Remote Sensing of Environment, 88, 3 16. (doi:10.1016/j.rse.2003.07.007)
Chen, X., Vierling, L., Rowell, E., Dykstra, D., Capehart, W., & DeFelice,
T. (2002). Relationships among IKONOS imagery, airborne scanning
LIDAR, and ground-based tree inventory data in a Ponderosa Pine
forest: A multiple endmember approach. Proceedings of the 2001 High
Spatial Resolution Commercial Imagery Workshop, March 19 21,
Greenbelt, MD, USA, sponsored by NASA/NIMA/USGS Joint Agency
Commercial Imagery Evaluation Team, CD-ROM.
Constance, E. (2002). IKONOS DEM evaluation. Proceedings of the 2002
High Spatial Resolution Commercial Imagery Workshop, March 25
27, Reston, VA, USA, sponsored by NASA/NIMA/USGS Joint Agency
Commercial Imagery Evaluation Team, CD-ROM.
Dial, G., & Grodecki, J. (2002). Block adjustment with rational polynomial
camera models. ACSM-ASPRS 2002 Conference and Technology Exhibition Proceedings, April 22 26, Washington, DC.
Ecuyer, D. (2001). Use of IKONOS data in littoral battlespace. Proceedings
of the 2001 High Spatial Resolution Commercial Imagery Workshop,
March 19 21, Greenbelt, MD, USA, sponsored by NASA/NIMA/USGS
Joint Agency Commercial Imagery Evaluation Team, CD-ROM.
Leachtenauer, J. C., Malila, W., Irvine, J., Colburn, L., & Salvagio, N.
(1997). General Image-Quality Equation: GIQE. Applied Optics: Information Processing, 36(32), 8322 8328.
NASA (2002). Domestic Non-reimbursable Space Act Agreement between
National Aeronautics and Space Administration John C. Stennis Space
Center, National Imagery and Mapping Agency Commercial Imagery
Program, and United States Geological Survey Land Remote Sensing
Program for the Joint Agency Commercial Imagery Evaluation (JACIE)
Team. SSC/FED-02-001-06, John C. Stennis Space Center, MI.
NASA, NIMA, & USGS (2001). Proceedings of the 2001 High Spatial
MD, USA, sponsored by NASA/NIMA/USGS Joint Agency Commercial
Imagery Evaluation Team, CD-ROM.
Resolution Commercial Imagery Workshop, March 25 27, Reston, VA,
USA, sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team, CD-ROM.
NASA/NIMA/USGS (2003). Proceedings of the 2003 High Spatial Resolution Commercial Imagery Workshop, May 19 21, Reston, VA, USA,
sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery
Evaluation Team, CD-ROM.
National Research Council (2002). Toward New Partnerships in Remote
Sensing: Government, the Private Sector and Earth Science Research.
Washington, DC: National Academy Press.
NIMA (2000). Commercial Imagery Program Brochure.
NIMA (2001). Commercial Imagery Uses Brochure, Version 2.0.
OSTP (2003). U.S. Commercial Remote Sensing Policy: Fact Sheet. Executive Office of the President of the United States, Office of Science and
Technology Policy, April 25.
Pagnutti, M., Ryan, R., Kelly, M., Holekamp, K., Zanoni, V., Thome, K., &
Schiller, S. (2003). Radiometric characterization of IKONOS multispectral imagery. Remote Sensing of Environment, 88, 52 67. (doi:10.1016/
j.rse.2003.07.008)
Peterson, B. (2001). IKONOS Relative Spectral Response and Radiometric
Calibration Coefficients, Space Imaging. Doc. #SE-REF-016, Revision
A, April 26.
Ryan, R., Baldridge, B., Schowengerdt, R., Choi, T., Helder, D., & Blonski, S. (2003). IKONOS spatial resolution and image interpretability.
Remote Sensing of Environment, 88, 37 51. (doi:10.1016/j.rse.2003.
07.006)
Remote sensing applications for precision agriculture: A learning community approach. Remote Sensing of Environment, 88, 156 168.
(doi:10.1016/j.rse.2003.04.007)
Sidor, J. (2001). Utility of IKONOS data for disaster response. Unpublished presentation at the 2001 High Spatial Resolution Commercial
Imagery Workshop, March 19 21, Greenbelt, MD, USA, sponsored
by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation
Team.
Small, C. (2003). High resolution spectral mixture analysis of urban reflectance. Remote Sensing of Environment, 88, 169 185. (doi:10.1016/
j.rse.2003.04.008)
Yang, L., Huang, C., Homer, C. G., Wylie, B. K., & Coan, M. J. (2003). An
approach for mapping large-area impervious surfaces: Synergistic use
of Landsat-7 ETM+ and high spatial resolution imagery. Canadian
Journal of Remote Sensing, 29(2), 230 240.

IKONOS satellite, imagery, and products

Gene Dial *, Howard Bowen, Frank Gerlach, Jacek Grodecki, Rick Oleszczuk
Space Imaging, 12076 Grant St., Thornton, CO 80031, USA
Received 9 December 2002; received in revised form 10 July 2003; accepted 12 August 2003
Abstract
The IKONOS satellite simultaneously collects 1-m panchromatic and 4-m multispectral images, providing the commercial and
scientific community with a dramatic improvement in spatial resolution over previously available satellite imagery. The sun-synchronous
IKONOS orbit provides global coverage, consistent access times, and near-nadir viewing angles. The system is capable of 1:10,000 scale
mapping without ground control and 1:2400 scale mapping with ground control. The IKONOS ground station produces radiometrically
corrected images, georectified images, orthorectified images, stereo pairs, and digital elevation models (DEMs) for image analysis,
photogrammetric, and cartographic applications. This article provides an overview of the IKONOS satellite, ground systems, products, and
applications.
Keywords: IKONOS satellite; Panchromatic; Imagery
1. Introduction
Space Imaging was formed to explore the commercial
prospects of high-resolution satellite imagery. Space Imaging contracted Lockheed Martin to develop the Space
Segment and Raytheon to develop the Ground Segment of
the Commercial Remote Sensing Satellite (CRSS) now
known as IKONOSR. The IKONOS satellite was launched
September 24th, 1999 to provide global, accurate, highresolution imagery to individuals, organizations, and governments for mapping, monitoring, and development.
The design of a high-resolution satellite imaging system
proceeds from a series of design trades between spatial
resolution and swath width, revisit time and off-nadir
viewing angle, image compression and data transmission
rates, and other performance specifications. Higher resolution results in a narrower swath width. Waiting for nearnadir opportunities results in long revisit times but short
revisit times can be obtained with off-nadir viewing geometries. The vast quantity of collected data and available
communications bandwidth require that the imagery be
compressed for transmission.
Space Imaging provided high-resolution imagery to
NASA under the Scientific Data Purchase (SDP) contract
* Corresponding author. Tel.: +1-303-254-2058; fax: +1-303-254-2214.
E-mail address: gdial@spaceimaging.com (G. Dial).
doi:10.1016/j.rse.2003.08.014
so that NASA could test the utility of commercial imagery

for research purposes. The SDP contract licensed NASA to
freely redistribute IKONOS imagery to scientists and
researchers. SDP products are described later in this article
and results of the NASA SDP research are published
elsewhere in this journal issue.
2. Basic IKONOS system design

IKONOS performance is summarized in Table 1. The
panchromatic sensor with 82-cm ground sample distance
(GSD) at nadir provides high resolution, intelligence-quality
imagery. Simultaneously, the multispectral sensor collects
blue, green, red, and near-infrared (NIR) bands with 3.28-m
nadir resolution, providing natural-color imagery for visual
interpretation and color-infrared imagery for remote sensing
applications. The pan GSD increases to 1 m at 26j obliquity
or 60j elevation angle. (Obliquity is the satellite-centric
angle from nadir to the line-of-sight. Elevation is the targetcentric angle from the horizon to the line-of-sight. These
angles would be complementary but for the curvature of the
earth.)
On-board gyros, star trackers, and GPS receivers measure attitude and ephemeris for medium-scale mapping
without requiring Ground Control Points (GCPs). Higher
accuracy mapping can be done with GCPs.
24
G. Dial et al. / Remote Sensing of Environment 88 (2003) 2336
Table 1
IKONOS performance summary
Specification
Value
Orbit height
Orbit inclination
Descending node time
Field of regard
Revisit time at mid-latitude
681 km
98.1j, sun synchronous
f 10:30 a.m., local solar time
Up to 45j off nadir
3 days at 60j elevation
Panchromatic and multispectral
13,816 pixels
3454 pixels
11 km at nadir
11 km
1000 km
400 km
11 bits
0.82 m at nadir
17%
Blue, green, red, NIR
3.28 m at nadir
445 516 nm
506 595 nm
632 698 nm
757 853 nm
Image sensors
Width of panchromatic arrays
Width of multispectral arrays
Field of view
Minimum image length
Maximum mono image length
Maximum stereo image length
Radiometric resolution
Panchromatic ground sample distance
MTF at Nyquist
Multispectral bands
Multispectral GSD
Blue bandpass
Green bandpass
Red bandpass
NIR bandpass
A global network of regional ground stations enable

international control and access to IKONOS imagery. Those
ground stations automatically process the huge volumes of
IKONOS imagery collected daily.
Collection capacity, image quality, radiometric accuracy,
mapping accuracy, and ground processing requirements
were defined before satellite design. Image quality requirements include modulation transfer function (MTF), signalto-noise ratio (SNR), and co-registration of the panchromatic and multispectral image detectors. Radiometric accuracy
requirements include the absolute calibration of the four
multispectral bands and the relative calibration of all pixels.
Mapping accuracy requirements include the metric accuracy
of stereo models, digital elevation models (DEMs), and
orthophotos produced with and without ground control.
Ground processing requirements include functional capabilities to operate the satellite and produce products as well as
capacity requirements to process large volumes of data per
day. These requirements will be described separately in the
sections that follow.
3. IKONOS sensor imagery

The IKONOS sun-synchronous orbit provides global
coverage with frequent revisit opportunities between
F 82j latitude. The IKONOS satellite is highly agile. It
can pitch and roll to acquire images far away from nadir, so
that it can collect long strip images, multiple small images,
or even stereo images of the same area on the same orbital
pass. The orbit provides daily access to sites within 45j of
nadir, 3-day revisit within 26j of nadir, and 141-day revisit
within 1j of nadir.
3.1. Image collection

The IKONOS orbit altitude is approximately 681 km and
inclined 98.1j to the equator, providing sun-synchronous
operation. The orbit parameters were chosen to provide midlatitude areas with daily revisit at 45j obliquity, 3-day
revisit at 26j obliquity, 11-day revisit at 10j obliquity,
and 141-day revisit at 1j obliquity. Ground tracks subdivide
the earth as follows: The ground between two successive
Fig. 1. Relationship between IKONOS revisit intervals, latitude, sensor elevation angles, and GSD.
25
Fig. 2. Area collection rates for monoscopic and stereo images.
passes on 1 day are divided in thirds on days 2 and 3,

providing a 1-m GSD every 3 days. These passes are further
divided approximately in quarters during an 11-day period,
providing 11-day revisit with obliquity less than 10j. This
pattern of 3- and 11-day cycles walks around the earth to
provide access within 1j of nadir every 141 days. This
sophisticated orbit provides a workable compromise between desires for short revisit times and low obliquity
angles at regular intervals. The 10:30 a.m. local mean solar
time of the descending node provides good illumination and
shadow definition. Fig. 1 shows the relationship between
revisit time, latitude, sensor elevation angle, and GSD.
The IKONOS satellite is a monolithic pointing body. The
camera is rigidly attached to the satellite bus. The entire
satellite rotates from image to image without articulation,
scanning mirrors, or other moving parts. The optical fieldof-view of the telescope provides a nadir swath of 11 km
with an 82-cm GSD. The sensor line rate produces an area
collection rate at 1-m GSD of almost 5390 km2/min in the
strip mode, the greatest and most efficient method of
imaging. For large, contiguous area collection the satellite
must maneuver between image strips, causing the imaging
efficiency to decrease. Finally, small image collection is the
least efficient due to the small area per image and maneuver
time from image to image. Fig. 2 illustrates the area
collection rates for the various imaging modes along with
the average historical collection rate and the collection rate
achieved for large area mapping. During a single orbital
pass, IKONOS can image a contiguous area of 4700 km2 at
a 1-m GSD or 10,000 km2 with GSD < 1.2 m.
The maneuver capability of the IKONOS satellite allows
for same pass stereo collection of significant area, a feature
that differentiates IKONOS in the high-resolution, commercial satellite-imaging arena. Fig. 2 also shows the stereo area
rate capability. Note that IKONOS can image a strip as long
as 450 km and then rotate back to collect the entire area in
stereo on the same pass.
From launch to August of 2003, Space Imaging collected
over 162,000,000 km2 of the earths surface and archived
over 400 terabytes of data.
3.2. Image quality
Image quality can be described by SNR, MTF, and
National Imagery Interpretability Rating Scale (NIIRS)
ratings (Ryan et al., 2003). SNR and MTF were measured
during On Orbit Acceptance Tests (OOAT) under conditions
that approximated a 10% reflective target with a 30j sun
elevation angle. Table 2 shows the OOAT measurements of
SNR and MTF. As expected, the system was found to be
shot-noise limited. MTF was measured with the edge target
shown in Fig. 3 with the results shown in Fig. 4. Exoatmospheric measurements of IKONOS MTF are discussed
in Bowen and Dial (2002).
Table 2
IKONOS SNR and MTF measurements
Band
SNR
Nyquist MTF
Pan
Blue
Green
Red
NIR
89
94
143
103
67
17%
26%
28%
29%
28%
26

Table 3
Radiometric performance requirements and results
Fig. 3. IKONOS image of edge target used for modulation transfer function
measurements (Imageryn Space Imaging).
For image analysts, numerical measurements such as

GSD, SNR, and MTF are not as significant as the ability
to extract information from images. In an effort to quantify
image interpretability, a U.S. Government/Contractor team
created the NIIRS (IRARS, 1996). The visible NIIRS
rating scale consists of 10 levels from 0 to 9 defined by
interpretation tasks or criteria, with 9 being the most
demanding and requiring the highest-quality imagery. For
Description
Requirement
OOAT
results
Post OOAT
results
Absolute radiometric accuracy

(multispectral bands only)
Relative radiometric accuracy
(all bands)
Tap-to-tap (banding)
Pixel-to-pixel (streaking)
Linearity
Stability
Failed pixels
F 10%
F 10%
F 10%
F 5%
F 5%
F 5%
F 10%
Compensation
F 5%
F 5%
F 5%
NA
None
< 1%
< 1%
F 1%, 1-r
F 3%, 1-r
None
example, analysts should be able to count railroad tracks in

a NIIRS 4 image and be able to identify the type of rail car
(flat car, box car, etc.) and engine (diesel, steam, etc.) in a
NIIRS 5 image. NIMA has rated IKONOS panchromatic
imagery at NIIRS 4.5 (Baldridge, 2002). That rating would
have been higher if the evaluation had been restricted to
imagery with a GSD less than 1 m; as it happens, oblique
imagery with GSD up to 1.2 m was included in the
evaluation.
The optical stability and focus are crucial to image
quality. Trained image analysts can detect small focus
differences when comparing side-by-side image samples
collected under good conditions from a well-behaved imaging system. Periodically, a through focus set of images is
collected with IKONOS and evaluated to determine if focus
adjustment is required to maintain image quality.
Because IKONOS is an agile imaging satellite that
collects imagery at varying obliquities and target distances
Fig. 4. Modulation transfer function measurements during OOAT.

Table 4
Radiometric calibration coefficients ((cm2 sr DN)/mW)
Product type
Production date
Bands
Blue
Green
Red
NIR
11-bit
Pre 22 Feb. 2001

Post 22 Feb. 2001
Pre 22 Feb. 2001
Post 22 Feb. 2001
633
728
79
91
649
727
81
91
840
949
105
119
746
843
93
105
8-bit
or slant ranges, analysts often ask if theres a defocus effect.

The difference in focus for an image at nadir vs. one at 60j
elevation is a small fraction of the depth of focus.
The angular relationship between the collector, the area
of interest, and the sun has a dramatic impact on overall
image quality (Gerlach, 2001). Specular reflections of the
sun off water bodies, tin roofs, and other reflectors can
saturate the sensor. Acquiring images with the sun on the
opposite side of the target results in forward scatter that fogs
the image. Images at low elevation angles have long
atmospheric path lengths and large GSD. Space Imaging
schedules satellite image collection to avoid specular reflections from water bodies, to avoid looking towards the sun,
and typically above 60j elevation angle (or less than 26j
off-nadir) to reduce atmospheric path effects.
3.3. Radiometry
The IKONOS satellite was designed to meet radiometric
requirements for calibration to an absolute standard, uniform
relative response across the scene, linearity, and stability
over time. These same general specifications were established several decades ago when remote sensing systems
were first being implemented. IKONOS radiometric calibration is discussed in Bowen (2002) and Bowen and
Oleszczuk (2002). The importance of radiometry to auto-
27
mated classification is discussed in Dial, Gibson, and

Poulsen (2001). Table 3 compares the radiometric calibration requirements, set before satellite design, with performance measurements during and after OOAT. Banding and
streaking were within specification but visually distracting.
Streaking is pixel-to-pixel variation in detector response.
Banding is variation between detector arrays or between
electronic readouts, called taps, within those arrays. Calibration improvements have reduced banding and streaking
to be virtually imperceptible.
3.3.1. Absolute radiometric accuracy
Within the first year after launch, with the assistance of
the NASA Commercial Remote Sensing Program (now
called the Earth Science Applications Directorate), several
vicarious ground calibration activities were conducted to
independently verify the radiometric accuracy of the IKONOS system (see Pagnutti, 2002; Pagnutti et al., 2003). In
February 2001, a change was made to increase the gain of
each multispectral array to eliminate saturation banding in
images produced by extremely bright scenes. The calibration
coefficients were adjusted to compensate for this gain
change. Table 4 shows IKONOS radiometric calibration
coefficients.
3.3.2. Relative radiometric uniformity
The requirement for relative radiometric uniformity was
taken directly from the specifications for the Landsat systems. This is a reasonable specification for large area data
capture with a broad range of image content and a dynamic
range of only 256 counts. However, at the completion of
OOAT, it was clear that this requirement was not sufficient
for IKONOS 11-bit imagery.
Fig. 5 shows a uniform scene of an Antarctic snowfield
with an average response of ~1200 counts and a range of
Fig. 5. Relative radiometric uniformity requirements demonstrated on Antarctic snowfield uniform scene. Left: Visible banding artifacts still visible after
radiometric correction to better than 10 counts or 0.8%. Right: Radiometric uniformity is now better than 0.5% over the dynamic range of the detectors
(Imageryn Space Imaging).
28
Fig. 6. IKONOS radiometric linearity.
less than F 50 counts. Radiometric calibration that just

meets the F 5% specification could result in variations of
as much as F 60 counts. The left image shows that
radiometric correction to better than 10 counts or 0.8% still
leaves clearly visible banding artifacts. With continued
relative radiometric uniformity assessments and updates, it
has been possible to reduce and eliminate the perceived nonuniform variations in IKONOS imagery as shown in the
image on the right. Relative radiometric uniformity is now
better than 0.5% over the dynamic range of the detectors.
3.3.3. Linearity
The response of the IKONOS system has been shown to
be linear throughout the dynamic range of the detectors. The
technique used to demonstrate this capability is to image
several stable, radiometrically characterized stars. The ataperture radiance for each star is calculated. The star is
imaged and the total DN generated by the star is measured.
Fig. 6 shows the calculated versus measured response
values. This technique shows the linearity of the IKONOS
system to be better than 1% of full scale.
Fig. 7. IKONOS radiometric stability.

Table 5
Geometric performance requirements and OOAT results
Product
Requirement
OOAT results
Stereo without GCP

(level 2 stereo)
Stereo with GCP
(level 3 stereo)
Ortho from one stereo pair
without GCP (level 4a)
Ortho from three stereo pairs
without GCP (level 4a)
Ortho from stereo with GCP
(level 4b)
DEM from stereo without GCP
(level 5b)
DEM from three side-by-side
stereo pairs without GCP
(level 5c)
DEM from stereo with GCP
(level 5e)
25.4 m CE90,
22.0 m LE90
2.0 m CE90,
3.0 m LE90
25.4 m CE90
12.2 m CE90
4.8 m, 7.8 m CE90,

5.2 m, 1.3 m LE90
1.5 m, 1.7 m CE90,
1.5 m, 1.6 m LE90
6.9 m, 12.3 m,
18.8 m CE90
8.8 m CE90
2.0 m CE90
1.4 m CE90
30 m CE90
5.8 m LE90
7 m RMSE
4.3 m RMSE
3 m LE90
2.0 m, 2.1 m LE90
Multiple OOAT results shown for multiple test cases.
3.3.4. Radiometric stability

Radiometric stability ensures that the imagery collected
over a period of time is consistent. Annual measurements by
Space Imaging verify that the same stars produce the same
response from year to year. Fig. 7 shows the results of the
stellar imaging that occurred in 2001 as solid markers and
the results 2002 as cross hairs. The results show that the
IKONOS system has been stable within F 3%, 1-r since
launch.
3.3.5. Failed pixels
The 923,250 detectors that make up the IKONOS panchromatic and multispectral arrays function without a single
29
failure. If a pixel were to fail, ground software can replace

that pixel with interpolated data.
3.4. Geometric accuracy
3.4.1. Interior and exterior orientation
Interior and exterior orientation of the IKONOS satellite
are derived from sophisticated attitude and ephemeris determination systems, a stable optical assembly, and a solid
state focal plane, enabling IKONOS to achieve high geometric accuracy with or without ground control.
The Field Angle Map (FAM) describes the interior
orientation of the IKONOS camera. FAM allows one to
determine the line-of-sight vector in the camera coordinate
system for each image pixel.
The on-board star trackers and gyros determine satellite
attitude. The exterior orientation of the camera is defined by
the interlock angles, which relate camera orientation angles
to the satellite attitude. The initial interlock angles were
determined by pre-launch assembly measurements and further refined by in-flight calibrations.
3.4.2. RPC camera model
Photogrammetric users require a camera model to make
measurements from the imagery. A camera model provides
a mathematical relationship between object and image
coordinates. For high-resolution pushbroom sensors such
as IKONOS, the physical camera model, based on the
interior and the exterior geometry and other physical properties of the sensor, becomes complicated. To simplify
interface with the end users of IKONOS imagery, Space
Imaging distributes the Rational Polynomial Camera (RPC)
model (Grodecki, 2001) with IKONOS photogrammetric
Fig. 8. IKONOS geometric accuracy testing range over San Diego consisting of 140 ground control points over a 22 22-km area (Imageryn Space Imaging).
30
Fig. 9. Post-FAM-calibration residual errors over the Denver test range displayed against the GCP layout; no significant systematic error pattern.
products. To ensure that using the RPC camera model in lieu

of the physical model causes no discernable accuracy loss to
the end user, Space Imaging performed rigorous accuracy
analysis, validation, and testing of the RPC model. The
evaluation methodology and the results of the RPC accuracy
analysis are given in Grodecki (2001) and Grodecki and
Dial (2001), and show that, for all possible imaging scenarios, the RPC camera models do not differ from the physical
camera model by more that 0.04 pixel. Rational function
camera models are also discussed in Tao and Hu (2001).
3.4.3. Pre-operational (OOAT) calibration
Table 5 compares IKONOS geometric accuracy requirements, set prior to satellite design, with accuracy test results
during OOAT. Circular error at 90% confidence (CE90)
describes horizontal accuracy, while linear error at 90%
confidence (LE90) describes vertical accuracy. The IKONOS geometric accuracy was tested during the OOAT, using
the San Diego test range consisting of 140 GCP over a
22 22-km area shown in Fig. 8 (Grodecki & Dial, 2001).
GCPs were derived from large-scale controlled aerial photography and have 0.5-m CE90 horizontal and 1-m LE90
vertical accuracy. As shown in Cook et al. (2001), Grodecki
and Dial (2001), and Fraser, Hanley, and Yamakawa (2002),
accuracy requirements have been met or exceeded for all
IKONOS geometric products.
calibrations was to reduce systematic errors of the interior

orientation, improve panchromatic-multispectral registration, and improve accuracy of the interlock angles.
3.4.5. FAM calibration
The in-flight FAM calibration was performed over the
Denver test range consisting of 33 GCPs surveyed with
differential GPS to 0.5-m CE90/0.9-m LE90 accuracy. The
FAM parameters determined over the Denver test range
were later verified over the Southern test range, comprising
over 200 GCPs and covering 100 100-km area in southwestern Australia. It is seen that the post-FAM-calibration
residual errors over the Denver test range, shown in Fig. 9
against the GCP layout, exhibit no significant systematic
error pattern. Moreover, they are all within F 1 pixel, which
is the expected combined GCP and pick accuracy. The
residual errors over the Southern test range are shown in
Fig. 10. As before, no significant systematic error pattern
can be observed in the residual plot.
3.4.6. Interlock calibration
Space Imaging used a set of independent images, taken
repeatedly over the Space Imaging MTF target, to improve
interlock angle calibration. The interlock angle errors for
3.4.4. In-flight calibrations

Space Imaging endeavored to further improve geometric
accuracy through FAM and interlock in-flight calibrations
(Grodecki & Dial, 2002b). The main goal of the in-flight
Fig. 10. Post-FAM-calibration residual errors over the southern test range
displayed against the GCP layout; no significant systematic error pattern.
Fig. 11. Error distribution after corrections applied to improve interlock

angle calibration.

Table 6
Mississippi block adjustment results
GCP
None
1
Average
error / (m)
2.9
1.5
Average
error k (m)
Average
error h (m)
CE90
(m)
LE90
(m)
2.5
0.7
1.0
0.0
4.4
2.7
2.7
2.5
each image were determined by block adjusting the IKONOS images in our ground station and later used to compute
the mean interlock angle corrections. Fig. 11 shows the error
distribution after applying the interlock angle corrections. It
is seen that the resulting geometric accuracy of an uncontrolled IKONOS imagery is better than 10-m CE90.
3.4.7. Validation
A Space Imaging project located in a relatively flat area
of Mississippi, with six stereo strips and a large number of
well-distributed GCPs, was used to validate the FAM and
interlock calibrations. Each of the 12 source images was
produced as a georectified image with RPC camera model
data. The images were then loaded onto a SOCET SETR
workstation running a Space Imaging developed RPC block
adjustment model, described in more detail in Dial and
Grodecki (2002a) and Grodecki and Dial (2002a). To
quantify block adjustment accuracy, GCPs were selectively
changed between control and check points. Horizontal and
vertical accuracy for the case with no GCP and the case with
one GCP are given in Table 6. Vertical and horizontal error
distribution for a case with a single GCP is shown in Fig.
12. More detailed results can be found in Grodecki and Dial
(2002b).
Several statistical tests of IKONOS accuracy have been
performed since the completion of OOAT. IKONOS mono
accuracy, exclusive of terrain displacement, was measured
31
at 4-m RMS per axis (9-m CE90) with RMS relative errors
of 50 parts per million (Dial & Grodecki, 2002b). IKONOS
stereo accuracy without control has been reported at 6.2-m
CE90 horizontal with 10.1-m LE90 vertical by Dial and
Grodecki (2003a) and 7.9-m CE90 horizontal with 7.6-m
LE90 vertical by Ager (2003).
4. IKONOS system products and applications

After successful completion of satellite launch, calibration, and test, Space Imaging began meeting the needs of
government and commercial customers by delivering the
products that interface the IKONOS satellite to the user
community. These products meet customer needs for image
analysis, cartography, and photogrammetry.
4.1. NASA Scientific Data Purchase
The NASA Scientific Data Purchase contract specified
the products described in Table 7. Accuracy of the Standard
Original, specified pre-launch at 250-m CE90, turned out to
be considerably better with an accuracy of 15-m CE90
exclusive of terrain displacement. The Precision Original
product was problematic. Intended only for flat ( F 3 m)
regions where orthorectification would not be necessary, it
was sometimes ordered in hilly regions where large terrain
displacement errors were not corrected by orthorectification
but small satellite positioning errors were corrected with
ground control. Both the Precision Original and the Precision Master products required high satellite elevation angles
leading to long collection times. Space Imaging does not
market an equivalent to the Precision Original Product
because of these issues. Instead of Precision Plus orthorectified products with 2-m CE90 accuracy, Space Imaging
Table 7
Scientific data purchase products
SDP product
name
Description
Nearest commercial
product name
Standard
Original
Georectified image with 250 m

CE90 accuracy exclusive of
terrain displacement.
Georectified image collected
above 80j elevation with ground
control and 3 m CE90 pan
accuracy, 5 m CE90 MSI
accuracy exclusive of terrain
displacement.
Orthorectified image, 12.2 m
CE90 accuracy.
Orthorectified image, 2 m CE90
pan accuracy, and 5 m CE90 MSI
accuracy.
Bundle with a stereo pair (25 m
CE90 horizontal, 22 m LE90
vertical accuracy) and a DEM
(7 m RMSE).
Geo
Precision
Original
Standard
Master
Precision
Master
Model
Fig. 12. Validation of FAM and interlock calibrations, and quantification of
block adjustment accuracy: vertical and horizontal errors shown by dotted
and solid arrows, respectively, for a case with a single GCP.
None
Pro (10 m CE90)

Precision Plus
Reference Stereo +
IKONOS Terrain
Model
32
Table 8
Scientific data purchase parameters
Parameter
Value
Bits
DRA
MTFC
Mosaic
Pan sharpen
Resampling
GSD
Bands
Datum
Projection
Format
Media
11
No
Yes
No
No
Cubic convolution
1 m pan, 4 m MSI
Pan + blue, green, red, and NIR
WGS84
UTM
GeoTIFF
CD-ROM
recommends Precision orthorectified products at 4-m CE90

because they can be delivered much more quickly.
One area where the SDP product specifications excelled
was the uniformity of product delivery parameters. Where
commercial products can be delivered in a variety of bit,
band, processing, and format options, the SDP established
consistent standards shown in Table 8. The uniform SDP
delivery parameters facilitate testing algorithms and software across the entire SDP archive. This is a model for
future archives of satellite image data.
4.2. Image analysis
IKONOS Geo images are designed to meet the image
analysis needs of customers on a budget. Geo images are
collected with sensor elevation angles as low as 60j,
providing 3-day revisit cycle for rapid image collection.
Geo images are rectified to a map projection at a constant
height without the use of a DEM. Not requiring a DEM
speeds delivery but hurts accuracy, so Geo images are

suitable for image analysis applications, that do not require
a positional accuracy, but do require fast response or low
price. Typical applications are emergency response, national defense/intelligence, and environmental monitoring
applications.
Geo images are produced in a variety of formats according to the needs of the application and data handling
capabilities of the user. Different band combinations are
available: panchromatic (black and white), multispectral,
pan-sharpened (color), or a bundle of panchromatic and
multispectral imagery. Only three bands can be simultaneously displayed on computer monitors for visual interpretation. Common band combinations are natural color
red green blue (RGB) displays and color infrared (CIR)
with sensor NIR displayed as red, sensor red displayed as
green, and sensor green displayed as blue. Natural color
may be preferred for general use and color infrared preferred
for remote sensing applications.
Imagery can be delivered in 8-bit formats for ease of use
on small computer systems or 11-bit formats for full
dynamic range. Dynamic range adjust (DRA) modifies
grayscale values to enhance visual interpretability, while
multispectral imagery with DRA off maintains absolute
radiometric accuracy (Oleszczuk, 2000). Automatic machine classification favors 11-bit DRA-off products, while
human visual interpretation favors DRA-on products with
8 or 11 bits. Fig. 13 illustrates stretching an 11-bit image to
reveal shadow detail on an 8-bit display.
4.3. Cartography
Cartographers have long used aerial photography for
map making. Making maps requires the extraction of
Fig. 13. Extracting detail from shadows in 11-bit imagery. Left: Image displayed with full radiometric range. Right: Same image stretched to show detail in
shadows.
topography, hydrology, transportation, and other features

from imagery. Two approaches are employed-monoscopic
and stereoscopic.
Fig. 14 diagrams stereoscopic extraction. Stereoscopic
extraction provides three-dimensional feature coordinates
for 3-D Geographic Information System (GIS) systems
and databases. Working with softcopy stereo display systems, the cartographer extracts contours and 3-D feature
coordinates directly from the stereo imagery, assuring correct cartographic relationship of contours and features.
While stereo workstations are more expensive than monoscopic systems, stereo imagery costs less than the combination of DEM and ortho images derived from that stereo,
so economics may favor stereo extraction. The stereo
products required for stereo map making will be discussed
in the next section with photogrammetric products.
Fig. 15 diagrams monoscopic extraction from DEM and
ortho imagery. Automatic software generates contour lines
from DEM and the cartographer extracts features from the
ortho image. Feature extraction uses low-cost, eye-friendly
monoscopic displays. DEM and ortho products to support
monoscopic extraction are discussed in this section.
4.3.1. Orthorectified images
Orthorectified images are the fundamental products for
image maps, GIS image base, and cartographic extraction.
Image maps show imagery with marginalia, scale, and grids,
just like a regular map, except the image itself is shown
instead of features represented by icons and vectors. GIS
systems commonly show hydrographic, transportation, and
other information as vector layers. Displaying those vectors
on top of a base image adds context to the vector information.
Image features in orthorectified imagery are terrain
corrected to planimetrically correct coordinates. Map vectors such as transportation, vegetation, and cultural features
can be digitized from the ortho image.
Positional accuracy is increasingly important in todays
GPS-enabled world. IKONOS orthorectified images are
commonly produced in accuracy grades shown in Table 9
so that clients can order accuracy appropriate to their needs.
33
Fig. 15. Cartographic extraction from ortho images and DEM.
IKONOS accuracy without control is sufficient to make

Reference products from a single stereo pair. Pro products
can be produced from a multi-image block adjust solution or
from a single image with ground control. Precision products
require ground control. IKONOS product accuracy is controlled by CE90; corresponding RMS accuracy and scale
values are shown for reference. The standardization of
accuracy grades simplifies ordering, collection, processing,
and quality control of ortho images.
4.3.2. Digital elevation models
A DEM representing the surface of the earth is used to
assess flood dangers, communication lines of sight, airport
safety, and other applications. An important application of
DEM is to correct imagery for terrain displacement during
orthorectification. DEM can be created from IKONOS
stereo imagery in various accuracy grades and resolutions
corresponding to DTED-1, DTED-2, USGS 7.5Vquad
DEM. IKONOS best DEM is specified at 3-m LE90.
Automatic software generates the DEM that is then edited
by operators at stereo workstations. DEM can be edited to
the terrain surface, the top surface, or for orthorectification.
Terrain DEMs, edited to the surface of the earth without trees
and buildings, are best for hydrological studies. Top surface
DEMs, edited to building and treetops, are best for line of
sight analysis and aircraft navigation. DEMs for orthorectification are terrain edited except that they are edited to
correctly show roads and bridges during orthorectification.
IKONOS DEMs can be terrain edited, top surface edited, or
Table 9
IKONOS commercial ortho accuracy specifications
Fig. 14. Cartographic extraction from stereo images.
Product name
CE90 (m)
RMS (m)
Scale
GCP
Standard
Reference
Pro
Precision
Precision Plus
50
25
10
4
2
25
12
5
2
1
1:100,000
1:50,000
1:12,000
1:4800
1:2400
No
No
Optional
Yes
Yes
34
edited for orthorectification according to the application

requirement. The lower accuracy DEM grades require less
editing to meet their accuracy requirements.
4.4. Photogrammetry
The Ortho-Kit and Stereo image products include camera
model data along with the imagery. The camera model data
in RPC format enables users with suitable software to
photogrammetrically process the imagery. Accuracy can be
improved with ground control. Images can be orthorectified
with external DEMs or stereo images can be used to create
DEMs. Monoscopic images with RPC camera model are
called Ortho-Kit images. The principal application of OrthoKit images is to allow users to orthorectify IKONOS images
with their own DEM. Precision and Reference Stereo images
refer to stereo images processed respectively with and
without ground control. Image orientation is described by
RPC data for both Precision and Reference Stereo, the
difference being that RPC coefficients for Precision Stereo
images have been updated by use of ground control and so
are more accurate.
Fig. 16 shows an anaglyph visualization made from an

IKONOS stereo pair. Both images of an IKONOS stereo
pair are collected on the same orbital pass as illustrated in
Fig. 17. Typically, one image is collected above 72j
elevation angle and the other above 60j with 30 45j
convergence angle between the two images. The convergence angle facilitates three-dimensional measurement,
while the high-elevation leg is suitable for orthorectification.
Same-pass stereo collection results in the identical scene
content and lighting conditions for both images and facilitates automatic terrain extraction of DEMs. After DEM
extraction, the image of the stereo pair with the higher
elevation angle can then be orthorectified with the DEM.
Stereo accuracy can be improved with ground control, either
by Space Imaging with Precision Stereo or by the users with
commercial software.
Space Imaging has certified several commercial software
packages for photogrammetric processing of IKONOS data:
ERDAS IMAGINER, OrthoBase ProR, and Stereo AnalystR; PCI OrthoEngineR; INTERGRAPH ImageStationR;
and BAE SOCET SETR. Used appropriately, these software
packages can accurately perform block adjustment, terrain
Fig. 16. Anaglyph visualization made from an IKONOS stereo pair (Imageryn Space Imaging).
35
Fig. 17. Same-pass collection of IKONOS stereo images.
extraction, orthorectification, and cartographic extraction

from IKONOS imagery (Dial & Grodecki, 2003b).
5. The space imaging IKONOS Block II system: the next

step
Having described the IKONOS satellite performance and
products, we conclude with a glimpse into plans for the next
generation, high-resolution commercial image satellite system, described for now simply as Block II.
The planned orbit for the IKONOS Block II satellites is
681 km, the same as Block I. The present license is for a
nadir GSD of 0.50 m. However, Space Imaging has applied
for a license that would permit 0.40 m at nadir so that we
could provide a 0.5-m GSD over a field of regard of 700
km centered about the ground track. IKONOS today is 0.8m GSD at nadir increasing to 1.0 m at 26j obliquity for a
700-km field of regard. So Block II provides a 2
improvement in GSD from the same orbit as Block I with
the same high perspective, 141-day repeat cycle and 3- and
11-day subcycles.
The satellite and payload design have been significantly
improved to leverage advances in technology since the
IKONOS design and build timeframe of 1994. The payload
field of view and line rate have been increased and the
detector size decreased. This increases the area collection
rate to twice that of IKONOS today. A 2 decrease in GSD
increases the data rate by 4 . Combined with the doubled
area collection rate, Block II will generate pixels at a rate
eight times that of Block I. This will provide a significant
advantage in reducing the cost to produce products and
shortening delivery timelines. The large contiguous coverage
will image areas faster when clear weather occurs.
The agility of the satellite will be significantly increased
to both improve the area collection efficiency and point
target collection. It is expected that Block II will collect

point targets at a rate nearly twice that of Block I. The nadir
swath will be increased from 11 to 15 km. The area rate
collection for Block II in the strip mode will be >12,000
km2/min at a GSD of 0.5 m. The higher collection rates and
maneuvering capability will allow the Block II satellite to
completely image a one degree cell of 110 110 km in
mono on a single pass all at a GSD < 0.55 m. The Block II
system will provide a significant improvement over Block I
in image resolution, large area mapping, and point target
collection.
References
Ager, T. (2003). Evaluation of the geometric accuracy of Ikonos imagery.
SPIE 2003 AeroSense Conference, April 21 25, Orlando, FL.
Baldridge, B. (2002). Civil and Commercial Applications Project (CCAP):
Evaluation of imagery interpretability for IKONOS pan, MSI, and pansharpened imagery. Proceedings of the 2002 High Spatial Resolution
Commercial Imagery Workshop, March 25 27, Reston, VA, USA.
(CD-ROM; sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team).
Bowen, H. S. (2002). Absolute radiometric calibration of the IKONOS
sensor using radiometrically characterized stellar sources. Proceedings of the ISPRS Commission I Mid-Term Symposium/Pecora 15Land Satellite Information IV Conference, November 10 14, Denver, CO.
Bowen, H. S., & Dial, G. (2002). IKONOS calculation of MTF using
stellar images. Proceedings of the 2002 High Spatial Resolution Commercial Imagery Workshop, March 25 27, Reston, VA, USA. (CDROM; sponsored by NASA/NIMA/USGS Joint Agency Commercial
Imagery Evaluation Team).
Bowen, H. S., & Oleszczuk, R. (2002). IKONOS radiometric stability and
relative calibration. Proceedings of the 2002 High Spatial Resolution
Commercial Imagery Workshop, March 25 27, Reston, VA, USA.
(CD-ROM; sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team).
Cook, M. K., Peterson, B. A., Dial, G., Gibson, L., Gerlach, F. W., Hutchins, K. S., Kudola, R. S., & Bowen, H. S. (2001). IKONOS technical
36
performance assessment. In S. S. Shen, & M. R. Descour (Eds.), Proceedings of SPIE: Algorithms for Multispectral, Hyperspectral, and
Ultraspectral Imagery VII, 4381(10), pp. 94 108.
Dial, G., Gibson, L., & Poulsen, R. (2001). IKONOS satellite imagery and
its use in automated road extraction. In Baltsavias, Gruen, & Gool
(Eds.), Automatic extraction of man-made objects from aerial and space
images (III). A.A. Balkema Publishers.
Dial, G., & Grodecki, J. (2002a). Block adjustment with Rational Polynomial Camera models. Proceedings of ASPRS 2002 Conference, April
22 26, Washington, DC.
Dial, G., & Grodecki, J. (2002b). IKONOS accuracy without ground
control. Proceedings of ISPRS Commission I Mid-Term Symposium,
November 10 15, Denver, CO.
Dial, G., & Grodecki, J. (2003a). IKONOS stereo accuracy without ground
control. Proceedings of ASPRS 2003 Conference, May 5 9, Anchorage, Alaska.
Dial, G., & Grodecki, J. (2003b). IKONOS applications. Proceedings of
ASPRS 2003 Conference, May 5 9, Anchorage, Alaska.
Fraser, C., Hanley, H., & Yamakawa, T. (2002). High-precision geopositioning from IKONOS satellite imagery. Proceedings of ASPRS 2002
Conference, April 22 26, Washington, DC.
Gerlach, F. (2001). How collection geometry affects specular reflections.
Imaging NOTES, March/April.
Grodecki, J. (2001). IKONOS stereo feature extraction-RPC approach.
Proceedings of ASPRS 2001 Conference, April 23 27, St. Louis, MO.
Grodecki, J., & Dial, G. (2001). IKONOS geometric accuracy. Proceedings
of Joint Workshop of ISPRS Working Groups I/2, I/5 and IV/7 on High
Resolution Mapping from Space 2001, September 19 21. Hannover,

Germany: University of Hannover.
Grodecki, J., & Dial, G. (2002a). Block adjustment of high-resolution satellite images described by rational polynomials. PE&RS (January 2003).
Grodecki, J., & Dial, G. (2002b). IKONOS geometric accuracy validation.
Proceedings of ISPRS Commission I Mid-Term Symposium, November
10 15, Denver, CO.
Imagery Resolution Assessments and Reporting Standards (IRARS)
Committee (1996, March). Civil National Image Interpretability Rating
Scale (NIRRS) Reference Guide (http://www.fas.org/irp/imint/niirs_c/
index.html).
Oleszczuk, R. (2000). To DRA or not DRA. Imaging NOTES, September/
October.
Pagnutti, M. (2002). NASA IKONOS radiometric characterization. Proceedings of the 2002 High Spatial Resolution Commercial Imagery
Workshop, March 25 27, Reston, VA, USA. (CD-ROM; sponsored
by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation
Team).
Schiller, S. (2003). Radiometric characterization of IKONOS multispectral imagery. Remote Sensing of Environment 88, 52 67 (this issue).
Ryan, R., Baldridge, B., Schowedgerdt, R., Choi, T., Helder, D., & Blonski,
S. (2003). IKONOS spatial resolution and image interpretability. Remote Sensing of Environment 88, 37 51 (this issue).
Tao, V., & Hu, Y. (2001). A comprehensive study of the rational function
model for photogrammetric processing. Photogrammetric Engineering
and Remote Sensing, 67(12), 1347 1357.

IKONOS spatial resolution and image interpretability characterization

Robert Ryan a,*, Braxton Baldridge b, Robert A. Schowengerdt c, Taeyoung Choi d,
Dennis L. Helder d, Slawomir Blonski a
a
Lockheed Martin Space Operations-Stennis Programs, Remote Sensing Directorate, Bldg. 1105, John C. Stennis Space Center, MS 39529, USA
b
Booz Allen Hamilton, now with Research Systems, Inc., Vienna, VA, USA
c
Department of Electrical and Computer Engineering, University of Arizona, Tucson, AZ, USA
d
Department of Electrical Engineering, South Dakota State University, Brookings, SD, USA
Received 7 January 2003; received in revised form 18 July 2003; accepted 30 July 2003
Abstract
Five individual projects characterized the spatial performance of the IKONOS commercial imaging sensor. The result was determination
of the spatial image quality of IKONOS data products in terms of the National Imagery Interpretability Rating Scale (NIIRS), the system
Modulation Transfer Function (MTF), the system stability over its first year of operation, the characteristics of the Space Imaging MTF
Compensation (MTFC) procedure, and the application-specific capabilities of IKONOS imagery. Both panchromatic and multispectral
imagery were evaluated. Major conclusions of this work are that the system was stable in imaging performance during the first year of
operation, that its MTF meets the specification for the NASA Scientific Data Purchase program, that the initial MTFC processing appears to
be transposed in the in-track and the cross-track directions, that the MTFC results in a noise amplification of 2 to 4 in addition to
sharpening the imagery, and that IKONOS panchromatic imagery achieves an average NIIRS rating of 4.5.
Keywords: IKONOS; Modulation Transfer Function; National Imagery Interpretability Rating Scale
1. Introduction
The pre-eminent characteristic of IKONOS is its significantly higher spatial resolution compared to other nonmilitary satellite remote sensing systems. Both the National
Aeronautics and Space Administration (NASA) and the
National Imagery and Mapping Agency (NIMA) have purchased IKONOS 1- and 4-m ground sample distance (GSD)
imagery for various purposes. NIMA acquires commercial
satellite imagery as part of its mission to provide geospatial
information to the Department of Defense (DoD) and to the
national intelligence community. NIMAs vehicle for assessing the image quality and utility of commercial imagery is
the Civil and Commercial Applications Project (CCAP).
NASA, through its Scientific Data Purchase (SDP), purchased imagery primarily for land use research. NASA
Stennis Space Center and its academic partners at the
University of Arizona and South Dakota State University

E-mail address: Robert.Ryan@ssc.nasa.gov (R. Ryan).
doi:10.1016/j.rse.2003.07.006
are responsible for assessing image quality for NASA as part

of the SDP programs Verification and Validation activity.
Although NASA and NIMA address significantly different sets of problems, both agencies are using this highspatial-resolution imagery in similar ways. While the multispectral aspects of satellite systems have historically been
exploited by NASA researchers, in the case of IKONOS, the
1-m panchromatic and pan-sharpened multispectral imagery
have often been used for visual inspection and mapping
applications, similar to the way such imagery is used by the
defense communities (Garvin, Mahmood, & Yates, 2002;
Tucker, 2002). NASA scientists rely upon the IKONOS
spatial and geolocation characteristics primarily to detect
and to identify small features.
The spatial resolution of most remote sensing systems is
described in terms of the sensor MTF and GSD. In the case of
IKONOS, the imagery NASA purchased was specified to
have a GSD at nadir of 0.82 m in the panchromatic band and
of 3.24 m in the multispectral bands. In addition, the
minimum allowable MTF at the Nyquist frequency was
specified to be 0.1 in the panchromatic band and 0.24 in
the multispectral bands. These specifications, however, are
for raw data and not for the products available to both NASA
38
R. Ryan et al. / Remote Sensing of Environment 88 (2003) 3752
and NIMA. Because it is difficult to perform a direct

measurement of MTF on orbit, measurements of edge response or full width half maximum (FWHM) of a line-spread
function (LSF) are often performed. These values are related
to the MTF as follows: The derivative of the edge response is
the LSF. The Fourier transform (FT) of the LSF provides an
estimate of the MTF in one orientation. The FWHM of the
LSF is another measure of the edge quality, which is often
compared with the GSD. In cases where the GSD becomes
large, such as with the multispectral bands, the pulse spread
function, derived from the image of a strip or pulse target, can
be used to estimate the MTF or the LSF.
The work presented here illustrates some of the complexities encountered when dealing with an emerging commercial
product. While not commonly known, NASA negotiated its
contract and data specifications with Space Imaging, LLC,
before the system was on-orbit and before all of the present
product options were available. NIMA negotiated its contracts with Space Imaging later, resulting in different products
being available to NASA and to NIMA. One of the significant
differences in offerings is that NIMA could purchase imagery
that is radiometrically corrected without geometric correction, while NASA could purchase only imagery that has been
both radiometrically and geometrically corrected. The NIMA
contract with Space Imaging calls the pure radiometric
products TIFF and calls the resampled products GeoTIFF. At the time of the NASA contract negotiations, only
cubic convolution resampling was available. NIMA contracted with Space Imaging to produce a nearest neighbor
resampled product that later was made available to the NASA
community. Users of IKONOS imagery should investigate
the relative benefits of each post-processing option.
NASA was also initially offered imagery with Modulation
Transfer Function Compensation (MTFC) applied. MTFC is
a form of image sharpening that attempts to correct the
inherent Modulation Transfer Function (MTF) roll-off with
spatial frequency caused by finite detector size, spacecraft
motion, diffraction, aberrations, atmospheric scattering, turbulence, and electronic effects (Holst, 1995). MTFC often is
used to boost the National Imagery Interpretability Rating
Scale (NIIRS) rating and image interpretability. MTFC generally increases the sharpness and interpretability of the
imagery, but it also introduces several artifacts, such as
ringing at edges and increased noise. Applications that
depend on spectral analysis generally require higher signalto-noise ratio (SNR) than do purely visual applications. For
both types of analyses, it is important to understand the effect
of MTFC on SNR, as well as the spatial frequency content of
the scene being studied, before selecting the MTFC processing option. In the sections below, the magnitude of effects
produced by MTFC processing is estimated. Pagnutti et al.
(2003), in this issue, discusses the effects of MTFC on
radiometry in various scene types.
Another fact not generally known is that all IKONOS
data are compressed off the focal plane, using a Kodak
proprietary compression technique, from 11 bits/pixel to 2.6
bits/pixel for transmission to the ground. In the strict sense,

this nonlinear compression violates the linear shift invariant
requirement for MTF analysis. For these reasons, the results
described in this paper should be interpreted as productspecific rather than the true fundamental engineering performance of the system.
The intent of the spatial characterization effort performed
by NASA and by NIMA is as follows:
Evaluate the usefulness of IKONOS for image interpretation tasks

Understand the effect of spatial processing available from
Space Imaging
Evaluate the on-orbit spatial imaging performance of
IKONOS
Determine if any degradation has occurred during the
first year of IKONOS operation
Research results that address these goals are presented in
this paper. Various approaches are used, including visual
inspection, modeling, noise analysis, image gradients, and
MTF. In Section 2, the relationships between image quality,
edge response, and MTFC to the National Imagery Interpretability Rating Scale are described. Section 3 discusses
IKONOS MTFC analysis. Section 4 describes relative
image quality analysis using the image digital number
(DN) gradient as a sharpness measure. Section 5 describes
MTF analysis using rectangular pulse targets. In Section
6, MTF analysis using an edge target at the Stennis Space
Center is described. Section 7 describes the application of
IKONOS imagery in standard image interpretation tasks and
NIIRS estimation with certified analysts. Finally, the various
analyses are tied together and conclusions are drawn for
future sensor characterizations.
2. Relationship between NIIRS, edge response, GSD,

and MTFC
Image quality is the result of a complex relationship
between GSD, MTF, MTFC, and SNR. Visual interpretability ratings such as NIIRS can be estimated from the edge
response, ringing overshoot, and SNR using the empirically
based General Imagery Quality Equation (GIQE) (Leachtenauer, Malila, Irvine, Colburn, & Salvaggio, 1997). NIIRS
is a graduated, criteria-based, 10-point scale used to indicate
the amount of information that can be extracted by imagery
(IRARS, 1996). A commonly accepted form of the GIQE
that accounts for the effects listed above follows:
NIIRS 10:251 alog10 GSDGM blog10 RERGM
0:656HGM
0:344G
SNR
where GSDGM is the geometric mean of the ground

sampled distance, RERGM is the geometric mean of the
relative edge response, HGM is the geometric mean-height

overshoot caused by MTFC (Leachtenauer et al., 1997),
and G is the noise gain associated with MTFC. In the
current form of the GIQE, SNR is estimated for differential
radiance levels from Lambertian scenes with reflectances
of 7% and 15% with the noise estimated from photon,
detector, and uniformity noise terms. If the RER exceeds
0.9, then a equals 3.32 and b equals 1.559; otherwise, a
equals 3.16 and b equals 2.817.
The GSD is computed in both ground plane directions in
inches, from which GSDGM is then calculated. Similarly, the
RERGM is the geometric mean of the RERs computed in the
orthogonal image directions. An edge response is determined from an image of extended bright and dark uniform
rectangular areas of at least 10 20 pixels in extent or is
estimated from a pulse or other target. The edge response is
normalized such that the asymptotic dark and bright values
are scaled to zero and unity, respectively. The RER is the
slope of a normalized edge, measured at F 0.5 pixels from
the edge center location. The GIQE overshoot H accounts
for the ringing associated with the MTFC and is measured
over 1.0 3.0 pixels from the edge in 0.25-pixel increments.
In most cases, the overshoot H is the maximum value over
this range. However, if the response is monotonic over this
range, the overshoot H is taken as the value at 1.25 pixels
from the edge. Again, H is estimated in orthogonal image
directions and the geometric mean is calculated. The noise
gain term is defined in Eq. (2):
"
N X
N
X
G
MTFCKernelij 2
# 12
2
i1 j1
Noise gain associated with the MTFC processing was

investigated with both expression (2) and with simulated
scenes. Simulated scenes, with and without MTFC processing, were generated with the expected noise properties
of IKONOS imagery. The noise gain in the simulated
scenes was estimated from the ratio of the standard
deviation of image DNs with MTFC applied to that
without MTFC applied. In both the simulation and expression (2) methods, the MTFC kernels provided by Space
Imaging for the different bands were used. Both methods
produced similar noise gain estimates and are listed in
Table 1. The panchromatic noise gain is approximately 4,
being twice that of the multispectral bands, with the
multispectral band noise gain increasing with wavelength.
Table 1
Noise gain associated with MTFC processing
Band
Noise gain
Blue
Green
Red
NIR
Pan
1.59
1.63
1.68
1.81
4.16
39
The simulation results and expression (2) are in good

agreement. Although Space Imaging implements its MTFC
option in the spatial domain, the MTFC is described here
in the frequency domain as provided by Space Imaging to
NASA for analysis. MTFC functions are designed to boost
the higher spatial frequencies without affecting the zero
spatial frequency. The IKONOS panchromatic MTFC
function is a monotonic function increasing from unity at
zero spatial frequency to over 6 at the Nyquist frequency.
This function is calculated by zero-padding a fast Fourier
transform of the MTFC kernel. The function is a relatively
symmetrical flower petal shape, but it is slightly stronger
in the cross-track direction than in the in-track direction.
Although not shown here, NASA also estimated the
MTFC kernel by estimating the transfer function from
several georeferenced scenes, with and without MTFC
applied (Pagnutti et al., 2003). The magnitude and shape
of the estimated MTFC functions qualitatively agreed with
the functions provided by Space Imaging, but NASA
found that the transfer functions were always rotated from
true north by several degrees or more because Space
Imaging used a kernel that combines the resampling and
MTFC processing.
3. IKONOS MTFC analysis

Several of the effects of IKONOS MTFC processing
were investigated through simulation. Simulated edge
responses were generated assuming a symmetrical Gaussian MTF with an MTF at the Nyquist frequency of 0.1, the
IKONOS MTF specification for panchromatic imagery. A
Gaussian MTF function was chosen because many system
MTF functions are approximately Gaussian in shape. The
expected asymmetries were ignored because sufficient
system information was not available to model them
accurately. The simulations are thus approximations, but
they do illustrate several features seen in other sections of
this paper. The simulations start with an 8 over sampled
edge response using an MTF of 0.1 as mentioned above.
This edge was then decimated to the supplied sampling. In
the case of panchromatic imagery, a simulated edge is
generated at a 0.108-m GSD for the simulations and then
resampled to 0.82 m to allow examination of the effects of
sampling and aliasing on the edge response. A second set
of edge responses was generated by applying the MTFC
processing. In the real imaging process, the individual
detectors sample a continuous function with the relative
sampling controlled by the focal length, slant range, and
detector spacing, while the absolute position relative to an
edge is a random process. This level of over sampling
minimizes any aliasing effects in the simulation. The
magnitude of the ringing depends on the phasing of the
sampling. Simulations showed that MTFC processing
resulted in an edge response overshoot of about 10%
and an RER improvement of approximately 50%.
40
Analysis of the amplitude of the Fourier transform of

multispectral MTFC kernels showed that they are not as
strong as the panchromatic MTFC with their peak values at
Nyquist being less than one half the panchromatic MTFC
peak. This is not surprising since the IKONOS multispectral
specification for MTF at the Nyquist frequency is 0.24.
Also, the MTFC functions are stronger as wavelength
increases. The overshoot is under a few percent for all
bands. Each MTFC function should be different for each
multispectral band, since both diffraction and charge coupled device MTF functions are wavelength dependent. The
multispectral MTFs are also far more asymmetric than the
panchromatic MTFC function. The near-infrared (NIR)
MTFC function is the most symmetric and the blue band
is the most asymmetric, with each MTFC function being
stronger in the cross-track direction than in the in-track
direction as seen in the blue band edge response simulations. This calculation is similar to panchromatic calculations but with the multispectral MTF at the Nyquist
frequency set to 0.24 for the simulations. The MTFC
processed edge showed only minimal change in the in-track
direction. Because the MTF typically rolls off more quickly
in the in-track direction, the MTFC was expected to be
stronger in the in-track direction. At the time of this writing,
NASA and Space Imaging had discussed this finding, and
Space Imaging had rotated the kernels.
4. Multispectral and panchromatic band relative image

analysis
4.1. Introduction
Constructed and cultural targets can be used for MTF
analysis of satellite imaging sensors (Rauchmiller & Schowengerdt, 1988; Schowengerdt, Archwamety, & Wrigley,
1985; Storey, 2001). Examples of target use for MTF
analysis of IKONOS are included in Sections 5 and 6 of
this paper. However, logistical difficulties with constructed
targets and lack of control of cultural targets, such as bridges,
roads, and other linear features, make target-free approaches
to image-quality evaluation desirable. This section describes
such an approach undertaken from July 2000 to July 2001 to
determine IKONOS image quality stability.
4.2. Method
Two images of the same area taken at different times
will have unique characterizations due to changes in
surface cover, solar illumination angles, sensor view angles,
atmospheric conditions, and sensor performance. If the
non-sensor-related factors are negligible for a given pair
of images, then it would be possible to use the image pair
for an evaluation of change in sensor performance. The DN
gradient (i.e., the discrete derivative between neighboring
pixels) is proposed as an image quality metric. It is well
known that the DN gradient is directly related to the

sharpness of an image and is in some ways analogous
to the RER discussed previously. The image pair to be
compared is first processed to extract the same ground
region and to match the DN histograms on a global basis
(Schowengerdt, 1997). The latter step removes global
(consistent over the whole image) solar irradiance variations due to different solar angles and global atmospheric
transmittance or path radiance differences on the two dates.
The DN gradient calculation was performed using the
Roberts gradient, the magnitude of the vector sum of
neighboring pixel differences in the F 45j directions,
which is a long-established method of obtaining an image
gradient (Castleman, 1996).
4.3. Imagery
Two examples will be used as reference benchmarks for
this image comparison technique. Both are panchromatic
images processed without MTFC and with MTFC. One
image is of an MTF target maintained by Space Imaging
for IKONOS evaluation, and the other image is of Tucson,
AZ. The differences are visually substantial in both cases
as shown in Fig. 1. The target image represents a relatively
simple scene consisting almost entirely of edges and lines.
The Tucson image is a more complex scene consisting of
cultural features. Because the image collects are the same
in both cases, there are no differences other than those
caused by MTFC. Comparison of the average gradient
yielded the results in Table 2. Therefore, average DN
gradient differences on the order of 30 45% are expected
in comparison of MTFC-On and MTFC-Off processed
images, which is consistent with the simulations discussed
in Section 3.
4.4. Image quality stability
IKONOS collects were obtained over Tucson, AZ on
July 23, 2000 and on July 15, 2001. The solar and sensor
angles for these collects are given in Table 3. Solar azimuth
and elevation affect the length and direction of shadows and
affect the overall irradiance of level terrain by the cosine of
the solar zenith angle (Schowengerdt, 1997). The effect of
sensor azimuth and elevation on the recorded image is more
complex. For example, in the case of high-resolution sensors such as IKONOS, sensor azimuth and elevation can
determine whether one side of building is seen. In addition,
specular reflection can occur from small, directional objects,
such as vehicle windshields and metal roofs. In this case
(Table 3), the cosine irradiance factor between the collects in
2000 and 2001 is 1.0348, or only about 3%. This factor was
applied as a gain to the 2000 image to normalize it to the
same average irradiance as the 2001 image.
Two areas were extracted from each years image and
the Roberts gradient was applied to each area. The areas
were chosen to avoid significant land cover changes (e.g.,
41
Fig. 1. Space Imaging target (August 5, 2001) with MTFC-Off (left) and with MTFC-On (right). The Roberts gradient magnitude images are shown directly
below. Note the sharper gradient in the MTFC-On case. The amplitude of the gradient is greater in the latter case.
new pavement on a parking lot or road) and were

registered visually when cut from the images. One of the
characteristics of this technique for image comparison is
that it is not particularly sensitive to image misregistration
of a few pixels. Since a large area is used, the average
gradient magnitude is little affected by a slight offset in
either image. The average DN gradient magnitudes calculated from the image regions are given in Table 4. The
average percentage difference on the two dates with
comparably processed imagery was 4.6% for Area 1 and
8.9% for Area 2. In both cases, these differences are much
less than the nearly 50% difference between the MTFCOff and MTFC-On images for the Space Imaging target
and Tucson data (Table 2).
4.5. Summary and conclusions

A simple relative-analysis technique for measuring
sharpness was applied to two IKONOS collects of Tucson,
AZ, taken approximately a year apart. Using the average
gradient magnitude as a measure of image sharpness, the
two images differ by less than 10%. In comparison to the
gradient magnitude difference of 30 45% between images
processed with and without MTFC, the two images acquired
approximately a year apart have the same image quality.
Other than imaging system performance changes, factors
that could cause the 10% difference include differences in
sensor look angles, changes in land cover, and differences in
solar irradiance angles. These factors were minimized in this
study by using images with high sensor elevation angle, by
selecting image regions with little land cover change, and by
Table 2
Reference image comparisons for sharpness quality metric
Location
Processing
Average DN
gradient
Average %
difference
Space Imaging
target
Tucson, AZ
MTFC-Off
MTFC-On
MTFC-Off
MTFC-On
54.2
78.2
60.97
80.38
44.3
31.8
Table 3
Anniversary image pair solar and sensor angles
Date
Solar
azimuth (j)
Solar
elevation (j)
Sensor
azimuth (j)
Sensor
elevation (j)
July 23, 2000

July 15, 2001
113.8
117.7
65.4
70.2
136.1
276.8
84.2
84.1
42
Table 4
Average DN gradient magnitudes for Tucson sites
Date
Area 1
(residential)
499 406 pixels
Area 2
(road/industrial)
424 451 pixels
July 23, 2000

July 15, 2001
167.2
159.5
129.7
118.1
using image collects with solar elevation and azimuth angles

within 5j of each other.
5. On-orbit MTF measurement by South Dakota State

University
5.1. Introduction
This section describes a procedure that was developed for
estimating the edge spread function (ESF) and MTF of high
spatial resolution imaging sensors while in orbit. No underlying mathematical model is assumed since complete system
descriptions for typical sensors are rarely available.
5.2. Method description
The primary target consisted of a set of four 3 30 m
blue tarps placed in a relatively uniform grassy field and
oriented in a pattern representing a rectangular shape
12 60 m in size. The 60-m length extended from north
to south as shown in Fig. 2. As shown in the figure, these
tarps are quite bright in the blue wavelength range with
approximately 0.4 for reflectance. As wavelength increases,
the reflectivity decreases until reflectances in the red wavelengths are nearly the same as vegetation. However, in the
NIR, reflectance again increases to nearly 0.4. Thus, this
target is particularly well suited for MTF evaluation of
multispectral imagery in the blue and in the near-infrared.
Tarp 1 and tarp 2 (T1 and T2) were selected as reference

tarps. They were aligned by surveyors transit at an angle of
8j east of true north to obtain as straight an edge line as
possible. In addition, all seams were aligned by transit to
maintain straight edges. To understand better the importance of target angle, an example ESF is shown in Fig. 3.
All pixel centers are shown as dotted angled grids. The
dashed lines indicate the phasing of the pixel center
locations as the edge location changes with each row of
pixels. The horizontal axis is scaled in units of pixels
corresponding to one ground sample interval in the output
image. The vertical axis, in units of digital number, represents the value of each pixel. The output edge function is
then sampled at a resolution of 20 points per pixel. As the
orientation of the angle changes, the sampling changes as
well, becoming either coarser or finer. Optimal angles exist
that place the subpixel sample locations on a uniform grid
with the limited target length. Because IKONOS imagery is
resampled so that true north is up, an optimal target angle
for both panchromatic and multispectral bands was found to
be 8j east of true north.
This methods first step involves determining exact edge
location. Edge positions are determined on a line-by-line
basis using available pixel information. For example, a
blurred edge is shown in Fig. 4. Simple digital differentiation has been applied to detect maximum slope. The subpixel edge points are determined by fitting a cubic
polynomial to the edge data using four values around the
maximum slope point. The zero crossing location of the
second derivative indicates the curves inflection point,
which is then assumed to be the subpixel edge location.
An underlying assumption is that the edge of the target
lies in a straight line. Any deviations from a straight edge
represent errors in the geometry of the image and a potential
contribution to the overall MTF of the system. With this
thought in mind, all edge cross-sections were forced to lie
along a straight line by fitting a least squares line through
Fig. 2. Illustration of tarp orientation with respect to true north and corresponding example image.
43
line. All vertical rows were used to estimate one averaged

spline as shown in Fig. 6.
The tarp width should be carefully chosen because of
the zero crossing points in the sinc function. This type of
target effectively deals with the size that would be required
to estimate the ESF and LSF directly from an edge target.
In the cases of 4-m GSD systems, the edge would
probably need to exceed 40 m in length. This is especially
important since many imaging systems are required to
meet minimum specifications of MTF at the Nyquist
frequency. A tarp width of one pixel does not include
any zero crossing points before the Nyquist frequency;
however, the output signal from such an input is too small
and may likely be affected by noise. With tarp widths of
two, four, or six pixels, the Nyquist frequency occurs at
the zero crossing point; when the output FT is divided by
the input FT, the MTF value at the Nyquist frequency
cannot be determined. To compute MTF values at the
Nyquist frequency, a tarp width of three pixels appears to
be optimal. Although a three-pixel wide input contains one
zero crossing point between zero and the Nyquist frequency, the width is large enough to produce a well-defined
image target and the output signal is strong enough to be
minimally affected by noise.
Fig. 3. Edge spread function projection from angled ground sample interval
points.
5.3. Results
the subpixel edge locations obtained from the previous step

and then declaring that the actual edge locations lie on that
line. In Fig. 5, the circles on the edge show the input edge
positions for individual rows of pixels. The line represents
the least squares estimate for all edge positions. Cubic
splines were then used to interpolate each usable horizontal
row of aligned edge data. Twenty values were interpolated within one pixel point to build a pseudo-continuous
As shown in the imagery acquired on August 13, 2001

(Fig. 7), the width of the pulse was determined by the actual
tarp width. Edge detection was applied on every row and the
subpixel edge positions were adjusted by the least squares
fitting line. The average profile was normalized by the
difference between the mean grass DN value and the mean
tarp DN value. The input pulse and the adjusted output were
then Fourier transformed (Fig. 8); the curve denoted with
stars is the discrete Fourier transform of the output and the
Fig. 4. A blurred edge and differentiation of the edge to locate the point of maximum slope.
44
Fig. 5. IKONOS multispectral image of blue tarps deployed at Brookings, SD on August 13, 2001.
plot with circles is the Fourier transform of the input pulse.

Finally, the MTF was calculated by dividing each output
value by its corresponding input and by normalizing the
result by the spatial average component magnitude. The
resulting MTF, shown in Fig. 9, has a value at the Nyquist
frequency of approximately 0.31.
One of the drawbacks of working with pulse type
targets is that in Fourier space, the corresponding function
is a sinc function that has a number of zero crossings. In
a noiseless system, the output response would also be
zero at these frequencies. However, because noise is
always present in real systems, significant errors will
often be introduced at frequencies near where these zero
crossings occur. Because the input function is rapidly

approaching zero near the zero crossing frequency, division in Fourier space will often produce noticeable errors
in MTF at frequencies close to where the zero crossings
occur. An apparent anomaly is observable in Fig. 11 at
the normalized frequency 0.3 where the MTF function
value is larger than it should be as suggested by the
overall MTF curve.
Figs. 10 and 11 exhibit all four pulse spread functions
and MTFs obtained during the summer 2001 season. Fig.
10 shows a consistent estimate of the pulse spread function
with undershoot visible on each side of the pulse. Fig. 11
shows good repeatability of MTF estimation with the
Fig. 6. Pulse response function obtained from blue tarps on August 13,
2001.
Fig. 7. Input pulse function and output pulse response.
Fig. 8. Input sinc function and output response in Fourier space.
single exception noted above. Estimates near the input sinc

function zero crossing frequency (i.e., between f = 0.3 and
f = 0.4) show more variability.
5.4. Summary
This work was an attempt to characterize the performance of a high-spatial-resolution imaging system in orbit
by estimating its frequency response to ground inputs. The
tarp-based target provided a usable pulse input for the 4-m
multispectral bands. The physical layout of the target was
found to be critical for a reasonable MTF estimation;
proper orientation of the target enabled reasonable determination of subpixel edge locations. The MTF results
obtained from IKONOS images suggested that the minimum value at the Nyquist frequency for the multispectral
bands was 0.25, exceeding NASAs Scientific Data Purchase specifications.
Fig. 9. MTF function for the blue tarp target on August 13, 2001.
45
Fig. 10. Overplot of four estimates of blue band pulse response functions
from blue tarps.
6. MTF analysis using the Stennis Space Center target

6.1. Introduction
Spatial resolution of image products is affected by characteristics of the satellite camera and by processing of the
images after reception at a ground station. Image processing
may include such steps as geometric correction and geographic registration as well as image sharpening based on
MTF compensation. Because the spatial resolution characterization is conducted for the on-orbit satellite, the imaging
process is also affected by transmission of radiation through
the Earths atmosphere. Therefore, characterization of the
spatial resolution of image products must also account for
atmospheric effects.
Fig. 11. Overplot of four estimates of blue band MTF functions from blue
tarps.
46
6.2. Method
Full width at half maximum of a LSF is used as a measure
of spatial resolution of the images. Before LSFs are derived
from edge responses by numerical differentiation, the edge
responses are measured and analyzed using a modified knifeedge technique (Tzannes & Mooney, 1995). Adjacent black
and white square panels, either painted on a flat surface or
deployed as tarps, form a ground-based edge target used in
the tests. During the measurements, the edge target is intentionally oriented so the image of the edge is aligned slightly
off-perpendicular to a pixel grid direction. The tilted-edge
modification to the original knife-edge method allows properly sampled edges to be obtained, minimizing aliasing
(Reichenbach, Park, & Narayanswamy, 1991). To measure
an edge response, a rectangular region containing the tilted
edge is extracted from an image of the edge target as shown in
Fig. 12. In such a region, each line across the edge forms an
approximate edge response. Exact edge responses (in the
direction perpendicular to the edge) are obtained when
distances are additionally scaled by cosine of the tilt angle.
The distance correction is usually small, but it becomes
important when results from measurements with different
edge orientations are to be compared.
Size of the edge target panels is a critical factor in spatial
resolution measurements of satellite images. To accommodate the 1-m GSD of the IKONOS panchromatic images,
panels 20 20 m in size are used. Larger panels would
provide even more accurate measurements, but deployment
of such large targets becomes extremely difficult. Therefore,
the number of edge response samples available for analysis is
still limited by the target size even with the additional
sampling provided by the edge tilt. For panels of given size,
sampling of the spatial response is also affected by the edge
tilt angle. When deployable tarps are used in the measure-
ment, orientation of the edge can be optimized for maximum

oversampling. Based on pre-test simulations, a tilt angle of 5j
was selected and used in measurements conducted with a set
of specially coated, reflective tarps owned and operated by
the NASA Earth Science Applications Directorate at Stennis
Space Center.
With a limited number of the available edge response
samples, image noise may significantly affect the results of a
spatial resolution characterization. Not only does numerical
differentiation of an edge response amplify the noise present
in the data and produce a spurious LSF, but also determination of the edge response slope becomes less accurate. To
mitigate adverse effects of image noise and limited sampling,
a smooth, analytical function is fitted to the edge responses.
In the present approach, a superposition of three sigmoidal
functions is utilized. Only three functions are used because
too many components could cause the analytical function to
fit the data points too well and to reproduce the noise and
errors rather than the actual information. Thus, the fitting is
performed for all the edge responses simultaneously using the
formula in Eq. (3):
ei x d
3
X
a
k

x b1 Di b2

k1 1 exp

ck
The distance x is measured in the direction perpendicular

to the edge. The nonlinear least-squares optimization is
conducted for nine parameters: a1, a2, a3, c1, c2, c3, b1, b2,
and d. Expanding on the work of Tzannes and Mooney,
position and orientation of the edge are found simultaneously
with the parameters characterizing spatial resolution in one
computational process of a nonlinear least-square fit of the
two-dimensional analytical function to the intensities in the
edge image. The parameters ak, b1, b2, and ck are common for
Fig. 12. IKONOS panchromatic images of the edge target tarps deployed at Stennis Space Center, MS, for the easting direction measurement on January 15,
2002 (left) and for the northing direction measurement on February 17, 2002 (right). Gray rectangular frames overlaid on the tarp images show the areas
selected for the edge response analyses.
47
Fig. 13. Measured edge responses (left column) and the best fits to them with superposition of three sigmoidal functions (right column). Data are for the
measurements in the northing direction using the IKONOS image acquired on February 17, 2002 and processed either with MTFC-On (bottom row) or with
MTFC-Off (top row).
all the edge responses, while the difference in the edge

position is introduced by the edge response index (i) multiplied by image GSD (D). Because all of the edge positions are
located on a straight line, they are specified with the simple
formula b1Di + b2. Tangent of the tilt angle is equal to the

absolute value of the parameter b1. To further suppress noise
artifacts, all three sigmoidal functions are restricted to the
same positions of the edge specified by the parameters b1 and
Fig. 14. Superimposed edge responses and the fitted sigmoidal functions for the IKONOS images processed either with MTFC-On (bottom row) or MTFC-Off
(top row). Northing direction (February 17, 2002) is the left column and easting direction (January 15, 2002) is the right column.
48
Fig. 15. Line spread functions derived from the fitted edge responses for the IKONOS images processed either with MTFC-On (bottom row) or with MTFCOff (top row). Northing direction (February 17, 2002) is the left column and easting direction (January 15, 2002) is the right column.
b2. This assumption also ensures that the analytical edge

response function is symmetrical.
6.3. Results
Examples of the measured edge responses and the
analytical functions fitted to them are shown in Fig. 13.
The presented analyses were conducted for IKONOS panchromatic image products that were georeferenced using the
cubic convolution resampling and the Universal Transverse
Mercator projection with the WGS-84 datum. The satellite
acquired two source images on different dates: one for
measurement of the edge response in the easting direction
(along the rows of image pixels) and the other for measurement of the edge response in the northing direction (along
the columns of the image pixels). Each of the source images
was used to create two different image products by processing it either with MTFC-On or MTFC-Off. In this way,
effects of MTFC on spatial resolution of the images can be
directly evaluated by studying this set of four images.
Finding the parameters b1 and b2 during the curve-fitting
Table 5
Line spread function FWHM for IKONOS panchromatic images (January/
February 2002)
MTFC-Off
MTFC-On
Easting (m)
Northing (m)
1.45
0.85
1.27
0.52
process is equivalent to shifting the edge responses to a

single reference location so that all the edge points are
aligned. Superimposing all the shifted edge responses creates a new one with a finer spatial sampling as shown in Fig.
14. Comparison of the measured edge responses with the
fitted, analytical functions indicates that the edge responses
are asymmetric. The asymmetry is noticeable even in the
images processed without the MTF compensation. For the
images with MTFC, the measured edge responses additionally contain apparent overshoots and undershoots. The
analytical function does not fit those secondary features
exactly, but it does reflect the general shape of the measured
edge response. Nevertheless, it is evident that the MTFC
greatly improves spatial resolution of the IKONOS images
and makes the edge responses much steeper.
After an analytical edge response function is obtained
from the best fit, it is differentiated numerically to derive
LSF and its FWHM as shown in Fig. 15. The edge
responses were extracted five times from each of the
images by independent selection of the analysis area.
Mean results of the FWHM measurements obtained by
averaging the five samples are listed in Table 5. To
characterize spatial resolution of the IKONOS panchromatic images, the FWHMs should be compared with the
1-m GSD of the resampled image products. Such comparison shows once again that the MTFC clearly increases
sharpness of the IKONOS images. In absolute terms, both
images acquired in the early months of 2002 have rather
narrow LSFs even without the MTFC applied. These
results confirm high spatial resolution of the IKONOS

panchromatic images.
7. Assessing imagery utility for NIMA: IKONOS

panchromatic image interpretability study
7.1. Introduction
The NIMA Civil and Commercial Applications Project
performed an interpretability study of IKONOS panchromatic imagery to evaluate the information content of IKONOS imagery in support of standard image interpretation
tasks supporting military and civilian applications. The
metrics used to gauge utility in this study were the NIIRS
and task satisfaction of Essential Elements of Information
(EEI). An EEI represents a request for intelligence information. The EEIs were restated in terms of image observables
and related tasks. For example, an EEI might ask the
intelligence analyst to determine the number of long-range
tactical aircraft. The observable might be large camouflaged
fighter aircraft. The task is detect large camouflaged
fighter aircraft. Image task EEIs are associated with a
Visible NIIRS level, e.g., a NIIRS 3 image would be able to
satisfy all level 3 EEIs. EEIs are derived from a variety of
sources, such as NIIRS criteria and the NIMA Community
Needs Forecast. The EEIs chosen for the IKONOS panchromatic evaluation addressed image tasks for NIIRS
levels 3 through 6.
7.2. Methods
Twenty-four level 1 TIFF panchromatic images were
acquired from Space Imaging. TIFF products are radiometrically corrected only. Sensor arrays are joined and contrast
balanced; geometric distortions are not eliminated and the
GSD for each pixel increases with distance from nadir
(mean collected GSD of 0.97 m for level 1 imagery). Fifteen
archived IKONOS level 2 GeoTIFF panchromatic images
were acquired from the DoDs Commercial Satellite Imagery Library. GeoTIFF image products have been radiometrically and geometrically corrected and resampled to the
Universal Transverse Mercator projection, hence each pixel
has been resampled to a uniform 1-m GSD (mean collected
GSD of 0.99 m for level 2 imagery). Imagery collection
dates ranged from November 1999 to July 2000.
The image matrix used in this evaluation was intended to
provide scene coverage of tropical, arid, northern temperate,
and southern temperate climate regimes within a single
season. Images were examined for Order of Battle content,
and image subsets were selected or chipped out. A total
of 72 image chips were used in the evaluation: 46 image
chips were created from the 24 TIFF images and 26 image
chips were created from the 15 GeoTIFF images. A chipping routine was used to generate the image chips and a
two-power (2 ) enlargement. The 2 version was created
49
for each chip using nearest-neighbor resampling and applying an identical histogram stretch as the 1 version. Each
chip pair (1 and 2 ) was rotated to the appropriate
cardinal direction that best aligned image obliquity to the
top of the display.
The evaluation was conducted at the NIMA Imagery
Support and Assessment Branchs softcopy evaluation
facility. All evaluation participants used the same workstation with a precision color monitor, which was calibrated
before the start of the evaluation. The monitors were set to
a minimum luminance response of 0.10 fl and a maximum
luminance response of 35.0 fl. Evaluation participants were
free to roam and zoom at 1 or 2 magnification within
the image. All ratings were made at 2 . No interactive
enhancement of the imagery was allowed and image chips
were rendered with no additional processing.
Participants for this evaluation consisted of eight NIMA
Visible NIIRS-certified imagery analysts. Evaluation participants were experienced with assessing Order of Battle as
well as natural and cultural features on panchromatic
imagery. The evaluation procedure consisted of each participants reviewing a sequence of panchromatic (pan)
scenes on the softcopy workstation and responding with
both NIIRS ratings and task satisfaction confidence ratings
for each scene. Examples of specific questions used are as
follows:

Determine the NIIRS rating for this image.

What is your confidence in your ability to identify areas
suitable for use as light fixed-wing aircraft (e.g., Cessna,
Piper Cub, Beechcraft) landing strips?
NIIRS ratings and EEI responses were given for each

panchromatic scene before advancing to the next scene. The
Visible NIIRS manual was available for reference during the
evaluation. The task satisfaction scores are given on a 0
100 confidence rating scale, where 0 means the task cannot
be performed and 100 means the task is certain to be
performed. The analysts were instructed to assume that
normal collateral imagery and information about the target
were available. All ratings and responses were entered by
way of a graphical user interface sliding bar.
7.3. Results
Eight imagery analysts completed the evaluation over an
11-day period. The final dataset consisted of 72 NIIRS
ratings and 250 confidence ratings of the EEI tasks for 72
image chips. Statistical analysis of the data began with the
standard assessment of reliability and consistency. The interrater correlation, rater-group correlation, and Cronbachs a
were computed to examine consistency and reliability
among the imagery analysts. The rater-group correlations
ranged from 0.68 to 0.75 and the a was 0.89, indicating a
high degree of consistency among the raters. Analysis of
Variance was used to identify outliers in the datasets. One
50
image was removed from the NIIRS dataset and two images
were removed from the EEI confidence ratings. These
cleaned datasets were used for all further analyses. The
average NIIRS value for TIFF pan images was 4.65 based
on a population of 46 image chips obtained from 24 image
products. The average NIIRS value for GeoTIFF pan images
was 4.41 based on a population of 26 chip sets from 15
image products.
For these levels of NIIRS for the TIFF imagery, the RER
is approximately 0.7. Approximately a 0.4 NIIRS improvement is expected with MTFC processing based on the
GIQE.
7.4. NIIRS ratings
The first step in obtaining NIIRS ratings was to calculate
descriptive statistics. Then an analysis of variances was
conducted to examine the difference in ratings for format
(GeoTIFF vs. TIFF) and climate (Arid, Tropical, Temp (N),
Temp (S)), including GSD as a covariate. For all analyses
that include GSD, the collected GSD was used (as opposed
to the resampled GSD) and was transformed to log10.
Climate was not found to be significantly different, so the
variable was dropped from further analyses. An Analysis of
Covariance was conducted to determine the impact of
format on mean NIIRS. The analysis included log10GSD
as a covariate, as well as an interaction term between
log10GSD and format. The analysis revealed that format
was not a significant main effect, but log10GSD ( p = 0.02)
and the interaction term ( p = 0.07) were significant predictors of NIIRS (R2 = 0.19). These results indicate that format
does not directly affect NIIRS ratings. The 0.24 difference
in mean NIIRS between TIFF and GeoTIFF is based on the
average differences in GSD and the interaction between
GSD and format. TIFF imagery has a lower average GSD
(0.931 m) than GeoTIFF (0.993 m), thus accounting for
higher NIIRS ratings. These results were used to derive
simplified regression Image Quality Equations, predicting
NIIRS from log10GSD for both formats:
GeoTIFF predicted NIIRS 4:41 0:77 log10 GSD
TIFF predicted NIIRS 4:52 5:05 log10 GSD
However, because of the limited range of GSD (0.8 1.3

m), it was not possible to fit an IQE to predict NIIRS with
great accuracy. The predicted NIIRS was the same for both
formats at a GSD of 1.06 m. It is useful to compare Eqs. (1)
and (2) to the GIQE derived to predict NIIRS based on
electro-optical parameters and system design (NIMA,
1996). The slope for the TIFF equation is much steeper
than the GIQE-obtained 3.32; however, a 95% confidence interval includes 3.32, indicating that GSD is a
significant predictor of NIIRS ratings for TIFF imagery. On
the other hand, the confidence interval for GeoTIFF regres-
sion slope includes zero, indicating that GSD may not be a

significant predictor of NIIRS for GeoTIFF imagery.
If the slope were indeed zero, then the NIIRS value for
this imagery does not depend on the collected GSD. Since
GeoTIFF images have been resampled to have a processed
GSD of 1 m regardless of collected GSD, a slope of zero is
not unreasonable. However, the zero slope can be true only
over a very limited range of GSD. For example, an image
with a collected GSD of 2 m that has been resampled to 1 m
clearly will not be as good as an image with a true GSD of 1
m. Thus, it is hypothesized that over a larger range of GSD,
Eq. (4) would be a broken line with two slopes instead of a
straight line, with the break occurring at or near 1 m. The
slope for GSD less than 1 m would be zero or possibly
somewhat negative. The slope of GSD greater than 1 m
would be comparable to EO panchromatic imagery characterized by the GIQE, i.e., about 3.31.
7.5. Essential elements of information
Means and standard deviations were calculated for the
250 confidence ratings for various EEIs. Each EEI had an
associated NIIRS level obtained from a previous evaluation.
Analyses of covariances were conducted to compare confidence ratings by climate, NIIRS level, and format, including
log10GSD as a covariate. NIIRS levels were determined by
rounding the NIIRS requirement for each task to the lower
level (i.e., a 3.2 and a 3.8 rating would both be NIIRS Level
3). Mean EEI confidence ratings were analyzed by climate,
format, log10GSD, and NIIRS level and their interactions.
Format, level, and log10GSD were found to be significant
(R2 = 0.476, p < 0.04). Unlike the NIIRS ratings, the interaction between log10GSD and format was not significant.
Next, the main effects for the two categorical variables were
explored. TIFF had an average of 6.3 points higher than
GeoTIFF, indicating that raters had more confidence in their
ratings for TIFF images.
A Tukey post-hoc test (Wigglesworth, 2002) was conducted to determine which NIIRS levels differed significantly from one another. The test indicated that all paired
comparisons were significantly different (R 2 = 0.45,
p < 0.01). Because NIIRS level 3 EEIs are the easiest to
answer, they had the highest mean EEI confidence ratings,
while level 6 EEIs had the lowest mean confidence ratings.
A confidence rating of 75% indicates that the task can be
satisfied with reasonable confidence. The results indicated
that GSD and the interaction of the related resampling were
significant in the predicted NIIRS ratings. Thus, the 0.24
mean NIIRS difference between TIFF and GeoTIFF was
due to the lower average GSD for TIFF imagery. The
analyses of the EEI confidence ratings showed two main
effects for format and NIIRS level. TIFF imagery had higher
mean EEI ratings than did GeoTIFF imagery. Also, level 3
NIIRS had the highest EEI ratings (indicating the most
confidence) while level 6 NIIRS had the lowest EEI ratings
(indicating the least confidence).
7.6. Conclusion
An average NIIRS rating of 4.5 was achieved with
IKONOS pan imagery, so this imagery should be able to
satisfy intelligence tasks for NIIRS levels 3 and 4. Although
the highest NIIRS and EEI values were obtained with the
TIFF products, the level of processing was not found to be a
significant predictor of NIIRS in the sample size used. As
might be expected, GSD was found to be significant
predictor of NIIRS for both formats. The range in average
NIIRS values for the entire sample of TIFF and GeoTIFF
images was 3.61 5.28, with NIIRS values increasing as
collected GSD decreased. A statistical analysis shows that
the IQE for two formats had statistically different slopes. It
is suspected that a broken-line IQE is more appropriate for
the GeoTIFF images, but such a model could not be
distinguished from a single linear equation with the data
available. Because TIFF images are higher NIIRS than
GeoTIFF images when the collected GSD is actually less
than 1 m, the TIFF image format is recommended for
intelligence EEI-type application if the user has the option.
This study indicates that IKONOS panchromatic products
can satisfy requirements for NIIRS level 4 imagery. Where
the best possible satisfaction of image tasks is required, the
highest collected resolution (i.e., at nadir) IKONOS panchromatic imagery should be acquired. For imagery where
coordinate embedded pixels are not required (as in GeoTIFF) and submeter GSD is assured, TIFF products should
be obtained.
8. Conclusions and summary

The wide variety of products available from Space
Imaging and the lack of access to raw imagery make
comparison of image quality analyses with system-level
specifications difficult. First, the product is compressed
onboard the spacecraft. Although little degradation in the
imagery is noticeable, classic linear system analysis
definitions do not necessarily hold. Second, the imagery
is processed with several proprietary algorithms. The
georeferenced imagery is also resampled with two different kernels: cubic convolution and nearest neighbor.
MTFC processing, when performed, is integrated with
the various resampling methods. For these reasons, product characterization should to be considered instead of
system characterization.
Several different types of Space Imaging IKONOS products were evaluated for their spatial resolution. Overall, the
image quality has been excellent and as advertised, but the
variety of product options was new to many of the scientists
using the data. The spatial resolution was also relatively
stable over the period of the work.
MTFC processing improves imagery quality but increases
noise. The panchromatic imagery noise level is increased by
over a factor of 4. Although not many truly dark targets have
51
been available for examination, imaging over water bodies,

forest, and other dark targets without MTFC is probably the
best option for further MTFC analyses. Native GSD imagery
has better interpretability than resampled imagery; however,
NIMA is presently the only government agency that can
obtain this product.
Comparing results between different groups is challenging because all groups did not evaluate the same products. In
addition, no common method exists to evaluate image
quality. A more intuitive approach is to follow the GIQE,
in which edge response, GSD, SNR, and processing methods
are used to define the spatial resolution requirements in terms
of the MTF at the Nyquist frequency. Although the GIQE has
not been fully tested for the various IKONOS product
options, developing specifications in terms of the GIQE
parameters could be beneficial.
Currently, it is very difficult to measure MTF at high
spatial frequencies on-orbit because the SNR of such
measurements degrades with increasing spatial frequency.
Because these measurements are often noisy and difficult to
interpret, other simpler and more intuitive metrics, such as
the width of the point spread function, the line spread
function, or the slope of the edge response, may be more
practical and useful. These three characteristics are all
related by Fourier transforms, so measuring any one characteristic allows determination of the other two.
Acknowledgements
This work was supported by the NASA Earth Science
Applications Directorate under contract number NAS 13650 at the John C. Stennis Space Center, MS and by NIMA.
The authors thank Marcia Wise and Denise Jarrell for their
assistance in preparing this manuscript. Special thanks go to
Vicki Zanoni for her support, careful reading, and comments
on this manuscript.
References
Castleman, K. R. (1996). Digital image processing ( p. 667). Upper Saddle
River, NJ: Prentice Hall.
Garvin, J. B., Mahmood, A., & Yates, J. (2002). Monitoring landscapes on
oceanic islands: Sub-meter imaging from IKONOS in the context of
RADARSAT SAR. Proceedings of the 2002 High Spatial Resolution
Commercial Imagery Workshop, March 25 27, Reston, VA, USA (sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team, CD-ROM).
Holst, G. C. (1995). Electro-optical imaging system performance
( pp. 146 147). Bellingham, WA: SPIE Optical Engineering Press.
IRARS (1996). Civil NIIRS Reference Guide. Imagery Resolution Assessment and Reporting Standards Committee, http://www.fas.org/irp/imint/
niirsM_c/guide.htm, accessed June 12, 2003.
Leachtenauer, J. C., Malila, W., Irvine, J. M., Colburn, L. P., & Salvaggio,
N. L. (1997). General Image-Quality Equation: GIQE. Applied Optics,
36(32), 8322 8328.
NIMA (1996). General Image Quality Equation Users Guide, Version 4.0.
National Imagery and Mapping Agency.
52
Pagnutti, M., Ryan, R., Kelly, M., Holekamp, K., Zanoni, V., Thome, K.,
& Schiller, S. (2003). Radiometric characterization of IKONOS multispectral imagery. Remote Sensing of Environment. (doi:10.1016/
jrse.2003.07.008)
Rauchmiller, R. F., & Schowengerdt, R. A. (1988). Measurement of the
Landsat Thematic Mapper MTF using an array of point sources. Optical
Engineering, 27(4), 334 343.
Reichenbach, S. E., Park, S. K., & Narayanswamy, R. (1991). Characterizing digital image acquisition devices. Optical Engineering, 30,
170 177.
Schowengerdt, R. A. (1997). Remote sensingmodels and methods for
image processing ( p. 522). San Diego, CA: Academic Press.
Schowengerdt, R. A., Archwamety, C., & Wrigley, R. C. (1985). Landsat
Thematic Mapper image-derived MTF. Photogrammetric Engineering
and Remote Sensing, 51(9), 1395 1406.
Storey, J. C. (2001). Landsat 7 on-orbit modulation transfer function estimation. In H. Fujisada, J. B. Lurie, & K. Weber (Eds.), Proceedings of SPIE:
Sensors, Systems, and Next-Generation Satellites, 4540 (pp. 50 61).
Tucker, C. J. (2002). From Agamemnon to IKONOS In search of the
Trojan War. Proceedings of the 2002 High Spatial Resolution Commercial Imagery Workshop, March 25 27, Reston, VA, USA (sponsored by
Team, CD-ROM).
Tzannes, A. P., & Mooney, J. M. (1995). Measurement of the modulation transfer function of infrared cameras. Optical Engineering, 34,
1808 1817.
Wigglesworth, J. K. (2002). What is a Post-Hoc Analysis? Ithaca College,
Survey of Statistical Methods, http://www.ithaca.edu/jwiggles/stats/
notes/notes23.htm, accessed July 30, 2003.

Radiometric characterization of IKONOS multispectral imagery

Mary Pagnutti a,*, Robert E. Ryan a, Michelle Kelly b, Kara Holekamp a, Vicki Zanoni c,
Kurtis Thome d, Stephen Schiller e
a
Remote Sensing Directorate, Lockheed Martin Space Operations-Stennis Programs, Building 1105, John C. Stennis Space Center, MS 39529, USA
b
Lockheed Martin Space Operations-Stennis Programs, USA
c
NASA Earth Science Applications Directorate, John C. Stennis Space Center, USA
d
Remote Sensing Group, Optical Sciences Center, University of Arizona, USA
e
Physics Department, South Dakota State University, USA
Received 2 May 2002; received in revised form 25 July 2003; accepted 31 July 2003
Abstract
A radiometric characterization of Space Imagings IKONOS 4-m multispectral imagery has been performed by a National Aeronautics
and Space Administration (NASA) funded team from NASA Stennis Space Center, the University of Arizona Remote Sensing Group (U of A
RSG), and South Dakota State University (SDSU). Both intrinsic radiometry and the effects of Space Imaging processing on radiometry were
investigated. Relative radiometry was examined with uniform Antarctic and Saharan sites. Absolute radiometric calibration was performed
using reflectance-based vicarious calibration methods on several uniform sites imaged by IKONOS coincident with ground-based surface and
atmospheric measurements. Ground-based data and the IKONOS spectral response function served as input to radiative transfer codes to
generate a top-of-atmosphere (TOA) radiance estimate. Calibration coefficients derived from each vicarious calibration were combined to
generate an IKONOS radiometric gain coefficient for each multispectral band assuming a linear response over the full dynamic range of the
instrument. These calibration coefficients were made available to Space Imaging, which subsequently adopted them by updating its initial set
of calibration coefficients. IKONOS imagery procured through the NASA Scientific Data Purchase was processed with or without a
Modulation Transfer Function Compensation (MTFC) kernel. The radiometric effects of this kernel on various scene types were also
investigated.
Keywords: IKONOS; Space Imaging; Multispectral imagery
1. Introduction
The National Aeronautics and Space Administration
(NASA) Earth Science Enterprise (ESE) initiated the
Scientific Data Purchase (SDP) to evaluate a new way
of obtaining remote sensing datasets for meeting its
research objectives. In the SDP, NASA purchased data
without the up-front investment required with a traditional
approach where a remote sensing system is built and
operated by the U.S. Government. The NASA Earth
Science Applications (ESA) Directorate at Stennis Space
Center (SSC) implemented the SDP and administered the

E-mail address: Mary.Pagnutti@ssc.nasa.gov (M. Pagnutti).
doi:10.1016/j.rse.2003.07.008
contracts with the commercial companies that were competitively selected to provide the programs remotely
sensed image products: Earth Satellite (EarthSat), Space
Imaging, DigitalGlobe (formerly EarthWatch Incorporated)/Intermap Technologies, Positive Systems, and AstroVision International. Image products are also being
procured from DigitalGlobe with additional SDP funds.
The ESA Directorate was also responsible for verification
and validation of products delivered under the SDP (Birk
et al., 2003).
The U.S. Government is prohibited from competing
with U.S. private industry (National Research Council,
2002); therefore, several federal government agencies,
including NASA, the National Imagery and Mapping
Agency (NIMA), the U.S. Geological Survey (USGS),
and the National Oceanic and Atmospheric Administration (NOAA), as well as state and local governments, all
54
M. Pagnutti et al. / Remote Sensing of Environment 88 (2003) 5368
purchase commercial imagery for use in a wide range of

applications.
The IKONOS satellite, owned and operated by Space
Imaging of Thornton, CO, was the first of a new class of
commercial remote sensing satellites with unprecedented
spatial resolution (4-m multispectral ground sample distance (GSD) and 1-m panchromatic GSD). The unique
imagery available from IKONOS and from other highspatial resolution systems becoming available has great
potential for research sponsored by the U.S. Government.
IKONOS datasets open new opportunities for virtual
ground truthing by validating conclusions derived from
the Governments coarser spatial resolution systems. For
the first time, phenomena inferred by spectral processing
of coarse-resolution data can be validated using this highspatial resolution data. In addition, other Earth science
research can now be supported by data with significantly
greater fidelity. IKONOS can resolve features in scenes
that Landsat 7, with its 30-m GSD and 8-bit dynamic
range, cannot resolve. In addition, spatial structure of
dark scenes, such as dense forest canopies, can now be
investigated from space-derived IKONOS data (Sawaya et
al., 2003). IKONOS radiometric calibration and its associated accuracy are critical for many of these studies,
particularly when comparing data from other systems,
when performing atmospheric correction, or when estimating various vegetative indices (Justice, Eck, Tanre, &
Holben, 1991).
Commercial data products must be completely characterized to be useful to NASAs broad science community
and to other government users. Because commercial and
science requirements often differ, commercial providers do
not necessarily characterize systems as NASA researchers
desire or need. Commercial imagery is typically highly
processed, being radiometrically and geometrically corrected and possibly being sharpened before it is delivered.
This is a new paradigm for the NASA science community,
which historically has had significant insight into sensor
design and operation. This new way of doing business,
therefore, requires independent data characterization and a
thorough understanding of how product processing affects
the imagery.
Three independent groups performed absolute radiometric assessments of IKONOS 4-m multispectral imagery: NASA SSC, the University of Arizona Remote
Sensing Group (U of A RSG), and South Dakota State
University (SDSU). NASA SSC also performed a relative radiometric assessment. Each group used some
variant of a reflectance-based vicarious radiometric characterization method in which surface and atmospheric
measurements taken at the time of the satellite overpass
are used to predict top-of-atmosphere (TOA) radiance.
These results were combined to derive an IKONOS
radiometric gain coefficient for each multispectral band
assuming a linear response over the full dynamic range
of the IKONOS multispectral imagery. The gain coef-
ficients were then compared to those provided by Space

Imaging.
A secondary objective of this investigation was to
evaluate the radiometric effects of image enhancements
routinely provided by Space Imaging to its customers.
Space Imagings commercial Carterrak product is processed with a Modulation Transfer Function Compensation (MTFC). This imagery is visually superior to
nonenhanced imagery but is subject to some artifacts,
such as ringing near high-contrast edges. MTFC is an
edge sharpening technique that is used to partially restore
image degradation due to finite detector size, optical
aberrations, motion, diffraction, and electronic effects.
See Ryan et al. (2003), in this issue, for more information on MTFC. It has been used to sharpen Cathode Ray
Tube displays, and the intelligence community has used it
to improve manmade object recognition (Holst, 1995;
Leachtenauer, Malila, Irvine, Colburn, & Salvaggio,
1997). MTFC can take many forms and is usually
implemented in software.
During the initial phase of the SDP, imagery with
MTFC applied (MTFC-On) was the only option provided
by Space Imaging. Therefore, a significant portion of the
vicarious calibration imagery being analyzed was with
MTFC-On. Several months into the contract, Space
Imaging provided an option for NASA to procure
MTFC-Off imagery. The fundamental question arose of
whether MTFC should be applied to data being procured
under the NASA SDP. It should be noted that Space
Imagings standard Carterrak products are typically provided with MTFC-On. In the future, researchers may not
have the option to procure non-enhanced (MTFC-Off)
imagery directly from Space Imaging. If MTFC significantly affects the results of some scientific investigation,
for better or worse, the form of the imagery that most
benefits the researcher should be procured. Therefore,
MTFC effects on various scene types were investigated
to provide some guidance about when this type of
processing is appropriate.
2. IKONOS satellite background

A technical description of the IKONOS system can be
found in Dial, Bowen, Gerlach, Grodecki, and Oleszczuk
(2003) in this issue. The IKONOS multispectral bandpasses are similar to the first four Landsat Thematic
Mapper (TM) bands and are shown in Fig. 1. The
near-infrared (NIR) band is a slightly modified Landsat
TM band that minimizes atmospheric water absorption.
Dielectric stack interference filters and the inherent detector response define the multispectral response. The
digital number (DN) dynamic range of the system is
approximately 1800 with 11-bit quantization. The data are
compressed onboard from 11 to 2.6 bits/pixel for data
transmission and are restored on the ground. The effects
55
Fig. 1. IKONOS spectral response curves for panchromatic and multispectral bands.
of these compressions appear to be minimal (Brower et

al., 2000); however, small effects are observed for low
dynamic range scenes.
3. NASA SDP IKONOS data products

The NASA SDP provided several types of IKONOS
data products to the ESE and affiliated researchers.
NASA procured only products that were both radiometrically and geometrically corrected, similar to the Space
Imaging commercial Carterrak products. The IKONOS
SDP imagery was divided into Original and Master
classes. Original imagery is georectified and Master
imagery is orthorectified. All data were projected to a
UTM/WGS84 Datum. Both Original and Master imagery
could be ordered through the SDP with either Standard
or Precision geometric accuracy. Standard products were
georectified without the use of surveyed ground control
points while Precision products utilized ground control
points. Most of the radiometric analysis discussed in this
paper was performed on Standard Original imagery. Raw
intrinsic panchromatic and multispectral imagery was
spatially resampled to 1 and 4 m, respectively, and was
georeferenced using either a cubic convolution or a
nearest neighbor interpolation algorithm. Off-nadir viewing was typically limited to 26j off-axis for the NASA
SDP products to guarantee 1 m or better intrinsic data. In
addition, IKONOS imagery was processed with or without MTFC.
The absolute radiometric accuracy of the IKONOS data
product is specified by the manufacturer to be better than
10% and the relative radiometric accuracy is specified to be
better than 5% (Dial et al., 2003, this issue). Absolute
radiometric values are defined in terms of a National

Institute of Standards and Technology (NIST) or other
traceable source. In the context of this paper, relative
radiometric characterizations compare radiometry between
pixels or arrays observing the same radiance. In a pushbroom system, such as IKONOS, each pixel in the array must
be corrected for gain and offset variation. In addition to
inherent detector-to-detector differences, vignetting and
spatial filter variations must also be corrected. Space Imaging precorrects each pixel so that a single radiometric
coefficient can be applied to all pixel DNs within each band
to determine the radiance value, which significantly simplifies the calibration for the user.
3.1. Relative radiometric accuracy assessment
NASA performed a relative (pixel-to-pixel within a
single band) radiometric accuracy assessment using large
uniform scenes. Several sites, including Antarctic ice sheets
from the SDP archive and selected Saharan sites, were used
for this assessment. The Saharan sites were identified by
searching through EarthSat GeoCover Landsat 5 TM imagery acquired through the SDP. These 30-m GSD Landsat
images were evaluated, and the most uniform areas within
these images were chosen. Both the Antarctic and Saharan
sites were extremely uniform, so that slight differences
between the three IKONOS focal plane array calibrations
were observed in the imagery. Previous studies have used
the Advanced Very High Resolution Radiometer (AVHRR)
and MeteoSat to identify these types of sites (Kaufman &
Holben, 1993). However, sites identified with the large
kilometer class GSD of these systems were not uniform
enough at the 4-m class GSD IKONOS scale and could not
be used.
56
Fig. 2. IKONOS blue band of the uniform Saharan scene (stretched).
For pushbroom systems such as IKONOS, variations

between detectors appear in the in-track direction of
scanning. Bad detectors manifest themselves as dark or
white lines in the in-track direction on the imagery, and
slight differences in calibration between arrays show up
as banding. Fig. 2 shows the blue multispectral IKONOS
band for one of the uniform Saharan scenes located in
Libya. This image has a DN range of 590 765 and has

been stretched to visualize the in-track banding resulting
from the three separate IKONOS detector arrays. The
figure also shows a thin white vertical line, indicating a
faulty detector.
In general, variations within a single detector array are
not easily observed because image processing attempts to
Fig. 3. Cross-sectional plot of pixels summed in the in-track direction of the uniform Saharan scene.
minimize deleterious image artifacts. Therefore, to visualize array and detector defects better, columns of pixels in
the image were summed and normalized. The summing
process, in essence, increases the signal-to-noise ratio and
amplifies any image defects. Fig. 3 shows a cross-sectional
plot of this summing, indicating that the differences
between arrays (banding) are less than 1.5% for each
band, which is well within the NASA SDP IKONOS
relative radiometric accuracy specification of 5%. The plot
also shows that a few pixels have DN values slightly less
than their adjacent pixels. This is most clearly seen in the
blue band on the first detector array. The single white
stripe shown in Fig. 2 appears as a single spike in the
cross-sectional summation plot (Fig. 3), and the array
transitions appear as step discontinuities. Similar results
were obtained with the other Saharan and Antarctic scenes.
Overall, the imagery is well corrected and our analysis
illustrates how Space Imaging can provide a single calibration coefficient instead of a set of coefficients, one for
each detector.
3.2. Absolute radiometric accuracy assessment
NASA SSC, the U of A RSG, and SDSU all used a
reflectance-based vicarious calibration method to evaluate
the absolute IKONOS multispectral imagery radiometry
(Dinguirard & Slater, 1999; Kaufman & Holben, 1993;
Meygret, Briottet, Henry, & Hagolle, 2000; Slater et al.,
1987; Slater, Biggar, Thome, Gellman, & Spyak, 1994;
Slater, Biggar, Thome, Gellman, & Spyak, 1996; Thome
et al., 1993). This method involves imaging a relatively
uniform area of interest coincident with taking ground-based
measurements. TOA spectral radiance is estimated from
ground reflectance, atmospheric pressure, water, and aerosol
profile measurements. The sensor spectral response function, provided by Space Imaging from laboratory measurements, is used to estimate the measured in-band radiance.
The sites used in this investigation are listed in Table 1,
which summarizes the location, collection dates and times,
and satellite elevation and azimuth angle. Each teams
approach and results are discussed and a compilation of
57
all the measurements is presented. These results are compared to the initial Space Imaging calibration coefficients
that were provided with each image. Space Imaging
obtained its calibration coefficients by imaging well-known
stellar sources primarily found in the Gunn and Stryker
catalog (Bowen, 2002). Measured integrated DN values
were compared against these known sources to determine
these initial calibration coefficients (Peterson, 2001). This
procedure was repeated several times over several years to
determine radiometric stability.
4. NASA SSC vicarious characterization

Lunar Lake playa in Nevada was chosen for the
NASA SSC IKONOS vicarious characterization due to
its high surface albedo, which exceeds 30% for all four
IKONOS multispectral bands, and due to its location in a
high-altitude desert region approximately 1.7 km above
sea level. The site receives little precipitation and has a
high percentage of cloud-free days. The playa is actually
a dry lakebed consisting of hard-packed clay that is
uniform over a large area. Based on previous measurements, the U of A RSG has found that the surface of the
playa is approximately Lambertian at viewing angles up
to 30j off nadir. NASA scientists routinely utilize the
Lunar Lake site for radiometric calibration of airborne
and spaceborne sensors (Nandy, Thome, & Biggar, 2001;
Thome & Nandy, 2000).
The IKONOS image shown in Fig. 4 was acquired on
June 7, 2000. The zoomed inset shows the area of playa
used by NASA SSC. The NASA SSC site is marked by
four Tracer 8 8-m tarps provided by Boeing, which
were used for Airborne Terrestrial Applications Sensor
(ATLAS) (DaMommio & Kuo, 1992) vicarious calibrations being performed at the same time. The 8 8-m
tarps are too small for the IKONOS 4-m GSD multispectral radiometric characterization other than to serve as
fiducial markers. Each tarp was surveyed to a few meters
accuracy using the Global Positioning System and can be
easily found in the georeferenced IKONOS image. The
Table 1
Vicarious calibration acquisition summary
Date
Location
Overpass
time (UTC)
Solar
zenith (j)
Solar
azimuth (j)
Sensor
nadir (j)
Sensor
azimuth (j)
Nominal
GSD (m)
May 26, 2000
White Sands
Missile Range, NM
Lunar Lake, NV
Railroad Valley, NV
Lunar Lake, NV
Railroad Valley, NV
Ivanpah, CA
Brookings, SD
Brookings, SD
17:38
21.9
116.9
12.1
346
0.85
18:16
18:14
18:23
18:23
18:20
17:12
17:13
24.5
24.3
22.9
22.8
24.5
26.5
25.8
123.2
123.8
126.3
126.9
122.6
136.1
135.9
19.0
11.6
17.0
9.5
6.4
17.9
4.4
158
144
351
267
284
0.45
302
0.90
0.85
0.89
0.84
0.83
0.87
0.83
June 7, 2000
June 7, 2000
June 10, 2000
June 10, 2000
July 21, 2000
June 30, 2000
June 30, 2000
58
Fig. 4. IKONOS image of Lunar Lake Playa (RGB) acquired on June 7, 2000.
vans used for transportation to the playa are visible to the

left of the tarps in the image. It is interesting to note that
the vans tire tracks are also visible from space. The
reflectance of the tire tracks is higher than the rest of the
playa by a few percent because the surface cracks that
are found naturally throughout the playa were compressed
and filled in by the weight of the vans as they were
driven over the playa.
The NASA SSC ground reflectance measurements of
Lunar Lake playa were taken on three separate transect
lines using two hand-held Analytical Spectral Device
FieldSpecR Full Range spectroradiometers (ASDs). These
commercially available spectroradiometers have a spectral
resolution of 1.4 nm in the 350 1000-nm spectral range
and have a spectral resolution of 10 nm in the 1000
2500-nm spectral range. The output from the spectroradiometers is interpolated within ASD software to report
data at 1 nm sampling across the entire spectral range.
Both ASDs were optimized at the start of each transect
line before any measurements were taken. ASD optimization sets the integration time for the visible NIR
detector and performs a dark count for the existing light
level that is present during the measurement.
Each transect line was divided into 11 different discrete
ASD measurement stations. Reflectance measurements
along two of the transect lines were entirely of playa ground
surface with sampling distances of 12.5 m between measurement stations. Ground reflectance measurements along
the third line were composed of both playa ground surface
and tarps. A distance of 12.5 m (approximately 3 pixels)
existed between each measurement station and 25 m (approximately 6 pixels) existed between the tarp center points.
Analysis has shown that tarp effects on the measurement are
minimal at 3 pixels away from the edge. Before each station
measurement, a 99% reflectance Labsphere SpectralonR
panel measurement was taken to generate reflectance values.
The ASD instruments were programmed to take 250 spectra

(10 measurements with 25 spectra averages) at each station.
In addition, the ASDs were carefully moved back and forth
while measuring the surface at each measurement station to
increase the spatial averaging at each location. Dark current
readings were also taken at each station with one of the
ASDs.
Vicarious calibrations were performed at Lunar Lake
playa on two separate days: June 7 and June 10, 2000.
There was little to no cloud cover on either day. The
NASA SSC team fielded a Yankee Environmental Systems Multi-Filter Rotating Shadowband Radiometer
(MFRSR) at Lunar Lake but it failed to record data
because of a memory card failure. The U of A RSG
team, however, provided solar radiometer (Ehsani, Reagan, & Erxleben, 1998) measurements taken at Railroad
Valley playa on June 7. Railroad Valley playa is approximately 30 km northeast of Lunar Lake playa. The U of A
RSG acquired solar radiometer data at Lunar Lake on
June 10. ASD measurements were taken along two
transect lines (playa only and tarp/playa transect) on June
7 and along all three lines on June 10. A radiosonde
balloon was also launched prior to each IKONOS overpass. IKONOS acquired data on June 7 at 18:15 coordinated universal time (UTC) at an elevation angle of 71.0j
and an azimuth angle of 157.9j, and again on June 10 at
18:23 UTC at an elevation angle of 73.0j and an azimuth
angle of 350.7j as shown in Fig. 5.
The NASA SSC TOA radiance analysis approach is a
two-step process. The first step creates a representative
atmosphere to estimate aerosol and molecular constituents
accurately for radiative transfer calculations at the time of
the satellite collection. The second step uses the representative atmosphere, location, time, sensor viewing geometry,
sensor spectral response, and reflectance data to estimate
TOA radiance.
59
Fig. 5. IKONOS Lunar Lake acquisition geometry.
The representative atmosphere was obtained by using

the calibrated sun photometer optical depth measurements
at the time of the satellite acquisition along with radiosonde data, site latitude and longitude, and time of
acquisition. The radiosonde data were used to create input
atmospheric layers for Moderate Resolution Transmittance
(MODTRAN) (Berk et al., 1999) radiative transport
calculations. The pressure, temperature, and humidity
radiosonde profile was used to estimate Rayleigh scattering and water vapor absorption. The amount of aerosol in
the atmosphere was estimated by varying the surface
range of the boundary layer in MODTRAN until the
predicted extinction at each sun photometer channel best
agreed in a least squares sense with the sun photometer
measurements. For this analysis, a mid-latitude summer
aerosol was used.
For June 10, the optimal surface range for the
boundary layer was found to be 70 km. Typical surface
range values often exceed 100 km, but a few thin cirrus
clouds were noted on that day at Lunar Lake. Comparisons between the solar radiometer transmission measurements and MODTRAN predictions are shown in Fig. 6.
There is excellent agreement between the solar radiometer transmission measurements and the MODTRAN
predictions.
The same atmospheric profile developed for June 7 was
used for June 10 because of the MFRSR instrument failure
at Lunar Lake playa. A sensitivity analysis was performed,
which showed that TOA radiance was largely insensitive to
changes in the boundary layer surface range when these
values exceeded 50 km. TOA radiance varied only a few
percent when the surface range was varied between 50 and
150 km. This result was due to the high reflectance of the
Lunar Lake playa and low aerosol loading. Because June 7

was very clear (and comparable to June 10), the surface
range very likely exceeded 70 km, especially since the
surface range exceeded 70 km at nearby Railroad Valley
playa as evidenced by the U of A RSG sun photometer
measurements.
The playa surface reflectance at each measurement
station was estimated by measuring both the soil and a
99% reflectance SpectralonR reference panel and by incorporating Labspheres SpectralonR hemispherical reflectance
calibration curve and the modeled SpectralonR panel bidirectional reflectance distribution function (BRDF) effects.
The SSC panel BRDF effect was estimated using the
Jackson model (Jackson, Clarke, & Moran, 1992). Reflectance values taken on June 7 and on June 10 were used to
form spectral albedo files as input into MODTRAN. Once
the aerosol estimates and spectral reflectance values were
determined, they were combined with the radiosonde profile, site latitude and longitude, satellite acquisition time,
and sensor viewing geometry to estimate the TOA radiance.
IKONOS spectral response curves were provided by Space
Imaging to calculate in-band TOA radiance values. These
radiance values were then compared with IKONOS sensor
measurements.
Images acquired by IKONOS on June 7 and 10, 2000
were analyzed to determine the measured radiance values.
For each band, DN values along each transect line were
averaged and converted to mean radiance values. These
mean values were compared to the MODTRAN TOA
estimates. Table 2 shows NASA SSCs predicted radiance
values compared to Space Imagings calibrated radiance.
The Space Imaging blue, green, red, and NIR mean radiances averaged between June 7 and 10 differed from NASA
60
Fig. 6. Comparison of solar radiometer measurements and MODTRAN prediction for June 10, 2000.
SSCs modeled radiance by 4%, 15%, 27%, and 49%,

respectively.
A first-order estimate of the error in the TOA radiance
estimate is 5 7% (Biggar, Slater, & Gellman, 1994) for
any single measurement. Additional processing effects not
considered in the Biggar paper were reviewed. The original NASA SDP IKONOS imagery was ordered with
MTFC applied (MTFC-On). Additional imagery was ordered without MTFC (MTFC-Off) to assess the effects of
this sharpening algorithm. Analysis showed that the root
mean square (RMS) difference in the radiance between
imagery processed with and without MTFC was less than
2% for the spatially uniform Lunar Lake playa. Since
averaging along transect lines further reduces MTFC
effects, MTFC processing effects were ignored in this
analysis. The effect of georeferencing interpolation was
also ignored for Lunar Lake and for other uniform scenes
Table 2
NASA SSC TOA radiance results
Site
Date
Band
DN
TOA
radiance
[W/(m2 sr)]
IKONOS
radiance
[W/(m2 sr)]
Lunar
Lake
6/7/00
Lunar
Lake
6/10/00
Blue
Green
Red
NIR
Blue
Green
Red
NIR
828
1178
1174
1083
811
1093
1108
1028
12.56
16.987
13.531
14.153
12.841
17.28
13.642
14.053
12.991
20.551
17.711
21.536
12.029
19.075
16.712
20.442
Percent
difference
3.43
20.98
30.89
52.17
6.32
10.39
22.50
45.46
because of the highly uniform nature of the scene and

because of pixel averaging. In addition, any residual
effects of compression would be minimized for uniform
scenes. Therefore, with approximately a 5 7% error, the
differences between the Space Imaging calibrated radiance
and the NASA SSC predicted radiance in the green, red,
and NIR bands were significant.
5. U of A RSG vicarious characterization

The U of A RSG performed several independent
characterizations of IKONOS radiometry using methods
similar to those used in the RSGs work with the Landsat
sensors (Thome, 2001). These methods are identical in
basic approach to those described above but with some
differences in implementation. Solar radiometer data are
collected as described above, along with atmospheric
pressure and other meteorological variables. An ASD and
99% SpectralonR reference panels were used to provide
the reflectance data.
For the U of A RSG reflectance collection, the user
walked a path parallel to the along-track direction of
IKONOS for a distance of 80 m. Fifteen of these 80-m
lines were walked with the along-track lines separated by 20
m. The site was separated into sixty 20-m pixels (15 4),
and ASD data were collected continuously while the user
walked the 15 transects. The spectroradiometer was configured to average 20 spectra per sample, and 8 samples were
collected within a single pixel, resulting in a total of 9600
spectra and 480 samples collected over the site. It took less
than 30 min to collect the data when the reference measurements are included. The reference measurements were taken
at every two transects and at the beginning and end of the
collection, giving a total of 9 reference sets (or 72 reference
samples).
The reflectance of the site was determined by ratioing
the measurements described above to those of the reference panel for which the bidirectional reflectance factor
has been determined in the laboratory. The sampling level
of the reference, relative to the test site surface, reduces
uncertainties due to changes in instrument response with
time and changing atmospheric conditions while keeping
the data collection time to a reasonable level. Knowing the
BRDF of the reference allows the reflectance of each
sample to be computed, taking into account effects due
to sun-angle changes and reflectance panel BRDF over the
30 min of data collection. Once each of the spectral
samples of the site was converted to reflectance, and all
480 data points were averaged to give a single spectral
reflectance for the entire site. The same sampling approach
and test site size were used for all four sites used by the U
of A RSG.
A critical part of this reflectance retrieval is the characterization of the reference panel in the laboratory. The
calibration of this panel was done with reference to a
standard made from a pressed polytetrafluoroethylene
panel based on a prescribed approach defined by NIST.
The calibration reference is a directional-to-hemispheric
reflectance standard provided by NIST. Polynomial fits
were made to the measured data to calculate the reflectance of the field standard for the sun-view geometry and
wavelengths for a given set of field measurements. The
SpectralonR sample was calibrated for a nadir viewing
sensor with a variety of solar illumination angles. The
measurements were made in only one plane, but this does
not create problems in the field because the SpectralonR
panel was aligned in the field such that the solar incident
angle stayed in this same plane. Two different 18-in.,
monolithic SpectralonR panels and two ASDs were used
for this work (U of A RSG serial numbers 6 and 7 for the
SpectralonR panels with the #6 panel paired with ASD687
and the #7 with the ASD614).
Rayleigh scattering, water extinction, and ozone and
aerosol distribution were derived from the solar radiometer data and from the meteorological data. The aerosol
distribution was defined in terms of a Junge distribution
(Biggar, Gellman, & Slater, 1990). These atmospheric
and surface data were inputs to a radiative transfer code
that computes hyperspectral, at-sensor radiances based on
a Gauss-Seidel iteration approach (Thome, GustafsonBold, Slater, & Farrand, 1996). The code assumes a
plane-parallel, homogeneous atmosphere and divides this
atmosphere into layers to account for the vertical distribution of scatterers and weak absorption due to ozone in
the visible and near infrared (approximately the 400 800nm spectral range known as the Chappuis absorption
61
band). The Junge parameter described above that was

derived from the solar radiometer measurements was used
to compute Mie scattering phase functions used in the
code. While the radiative transfer code can include nonLambertian effects of the surface, bidirectional reflectance
measurements of the surface were not available for this
work, so the surfaces were assumed to be Lambertian.
Past BRDF measurements of the U of A sites (Nandy et
al., 2001; Thome & Nandy, 2000) and sensitivity analysis
indicate that less than a 0.5% uncertainty is caused by
this assumption.
Strong gaseous absorption effects due to water vapor
were determined using MODTRAN, which computes transmittance for the sun-to-surface-to-satellite path for 1-nm
intervals from 350 to 2500 nm (Berk et al., 1999). This sunto-ground-to-sensor transmittance was multiplied by the atsensor radiance output from the radiative transfer code to
correct the radiances for this strong absorption. While this
approach is an approximation that excludes interactions
between diffusely scattered radiances and absorption, it
does not cause large uncertainties for application to IKONOS because of the small effect of absorption within the
IKONOS bands and the high surface reflectance of the test
sites used in this work.
The relative radiances that are the output of the radiative
transfer code were converted to absolute radiance by multiplying by a supplied solar irradiance curve corrected for
changes in Earth sun distance. The solar irradiance stan-
Table 3
U of A RSG TOA radiance results
Site
Date
Band
DN
TOA
IKONOS
Percent
radiance
radiance
difference
[W/(m2 sr)] [W/(m2 sr)]
Lunar
Lake
6/7/00
Lunar
Lake
6/10/00
Railroad
Valley
6/7/00
Railroad
Valley
6/10/00
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
851
1216
1216
1123
791
1137
1151
1062
734
1012
968
875
667
914
876
787
1422
1826
1694
1564
642
935
999
929
13.155
18.287
13.509
15.293
12.884
17.738
13.272
15.140
9.953
13.431
9.258
10.513
10.352
13.990
9.718
11.009
22.160
28.236
18.641
20.749
9.668
13.946
10.956
12.745
White
5/26/00
Sands
Missile
Range
Ivanpah
7/21/00
13.359
21.222
18.341
22.326
12.418
19.843
17.360
21.113
11.523
17.661
14.600
17.400
10.471
15.951
13.212
15.646
22.323
31.867
25.551
31.093
10.078
16.318
15.068
18.469
1.56
16.05
35.77
45.99
3.62
11.87
30.81
39.45
15.77
31.49
57.70
65.47
1.14
14.02
35.95
42.12
0.74
12.86
37.07
49.85
4.24
17.01
37.53
44.91
62
dard used in this work is that of MODTRAN and is identical

to the spectrum selected for NASAs Landsat 7 project.
These hyperspectral radiances were then integrated and
band-averaged using the supplied spectral response curves
from Space Imaging to obtain both the band-integrated and
band-averaged spectral radiance at the sensor.
The U of A RSG performed ground truthing at four
different sites. Lunar Lake playa and Railroad Valley
playa in Nevada were both imaged twice, and the White
Sands Missile Range, NM and Ivanpah playa, CA were
each imaged once. All images used in this analysis were
processed with MTFC-On. Table 3 summarizes the U of
A RSG TOA radiance results. The differences between
the U of A RSG vicarious calibration and the Space
Imaging results were similar to the NASA SSC TOA
radiance estimates. The red and NIR bands had significant differences between the IKONOS measurement and
predictions of the ground measurement team.
Table 4
SDSU TOA radiance results
Site
Date
Band
DN TOA
IKONOS
Percent
radiance
radiance
difference
[W/(m2 sr)] [W/(m2 sr)]
Brookings 6/30/00 Blue 249

Grass
Pass 1 Green 289
pixels
Red
225
NIR
577
Dark
Pass 1 Green 149
pixels
Red
71
NIR
60
Grass
Pass 2 Green 304
pixels
Red
239
NIR
589
Dark
Pass 2 Green 147
pixels
Red
70
NIR
40
3.618
4.168
2.7
7.72
2.439
1.806
0.652
0.654
3.679
4.219
2.718
7.741
2.488
1.839
0.660
0.651
3.910
5.044
3.394
11.471
2.889
2.600
1.071
1.193
4.097
5.305
3.605
11.710
2.920
2.566
1.056
0.795
8.04
21.01
25.69
48.59
18.43
43.98
64.25
82.39
11.37
25.75
32.63
51.27
17.36
39.50
59.97
22.15
6. SDSU vicarious characterization

SDSU performed an IKONOS radiometry characterization in Brookings, SD. The IKONOS sensor collected two
images of Brookings on June 30, 2000. One was a standard
original product collected at 17:12 UTC with a collection
elevation of 73.12j (collection azimuth of 0.45j) and the
other was a precision original product collected 1 min later
with a collection elevation of 85.59j (collection azimuth of
302.13j). Within the Brookings scene, a 180 180-m
uniform grass field ( < 10% variability) was chosen as the
ground truth target. Extensive ground truth data were
recorded simultaneously with the overpass, including measurements of surface reflectance, atmospheric transmittance,
diffuse-to-global ratio of the hemispherical sky irradiance,
and sky path radiance. A radiosonde vertical profile was
also acquired at the time of the satellite overpass. The
ground truth data were used to calibrate a MODTRAN
model predicting the at-sensor total radiance of the target
and the sky path radiance.
The reflectance data were taken with an ASD in a manner
similar to both NASA SSC and U of A RSG measurements.
There were 108 spectra collected by walking 6 transects
over the grass target area, 30 m apart. Each spectrum was an
average of 50 spectral scans recorded in continuous motion
to increase spatial averaging. ASD measurements of a
SpectralonR panel were also collected to transform the
target radiance spectra to reflectance. The panel was calibrated relative to a NIST reflectance standard in the U of A
RSG calibration lab, which included a correction for the
BRDF of the panel. Because of greater aerosol loading at
this site, a more extensive aerosol estimation technique than
described above was used to set up MODTRAN for
calculating the TOA radiance. The surface range and aerosol
distribution were adjusted to agree with the solar radiometer
data recorded at the ground, simultaneous with the IKONOS
overpass, and from an aircraft above the boundary layer.

Global-to-diffuse irradiance ratios from an MFRSR were
used to define the scattering phase function asymmetry
factor in the MODTRAN calculations. Rayleigh scattering
and water vapor absorption were estimated using radiosonde
data. In addition, the radiance values derived from the
SpectralonR panel measurements were also checked for
consistency.
The IKONOS scenes were searched for dark targets that
could be assumed to have zero reflectance. A zero reflectance surface will have an at-sensor signal equal to the sky
path radiance. For the blue and green bands, these pixels
came from shadows in tree strips. For the red and NIR
bands, the dark target pixels were from water bodies.
MTFC-Off imagery was chosen to avoid any MTFC effects.
By extracting, from both scenes, DNs from the target area
and surfaces that were identified as having zero reflectance,
the gain and bias for transforming at-sensor radiance to
image DNs were determined and compared to the gain and
bias provided by Space Imaging. Table 4 summarizes the
SDSU TOA radiance estimate results. There is more scatter
in this data than in the other results because of the more
challenging environment (dark target with increased atmospheric contribution), but the results show the same trends
as the previous results.
7. Compilation of NASA SSC, U of A RSG, and SDSU

results
Each of the radiance estimates provided by NASA SSC,
U of A RSG, and SDSU were plotted against IKONOS
image DNs on a single plot for each multispectral band.
With each measurement equally weighted, a straight line
least squares fit was performed to generate a composite
estimate of the gain and offset. The error of this composite

measurement should be significantly smaller than the error
of each individual measurement. The measurements cover a
significant portion of the entire dynamic range of the
system. Two forms of least squares fitting were performed:
one generating both gain and offset parameters and the other
generating only a gain parameter (offset set to zero). Since
very little difference was found between the two approaches, and since Space Imaging routinely measures dark
63
counts to subtract out the offset, the offset was set to zero
and the least squares approach generating only a gain
parameter was used.
The composite radiometric calibration curves and the
Space Imaging calibration curve are shown in Fig. 7. Table
5 lists both the Space Imaging and the NASA team gain
coefficients for each of the multispectral bands and the ratio
of the two sets of coefficients. Only the Space Imaging
IKONOS blue band calibration was in excellent agreement
Fig. 7. NASA team radiometric compilation calibration curves.
64
Table 5
NASA team gain coefficients
Table 7
U of A RSG TOA radiance results using updated SI calibration coefficients
Band
Space
Imaging
gain
NASA team
gain
(NASA team gain)/

(Space Imaging gain)
Site
Date
Band
DN
TOA
IKONOS
Percent
radiance
radiance
difference
[W/(m2 sr)] [W/(m2 sr)]
Blue
Green
Red
NIR
63.7
57.3
66.3
50.3
64.1 F 2.6
65.4 F 2.7
87.7 F 3.1
75.8 F 2.9
1.01
1.14
1.32
1.51
Lunar
Lake
6/7/00
Lunar
Lake
6/10/00
Railroad
Valley
6/7/00
Railroad
Valley
6/10/00
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
Blue
Green
Red
NIR
851
1216
1216
1123
791
1137
1151
1062
734
1012
968
875
667
914
876
787
1422
1826
1694
1564
642
935
999
929
13.155
18.287
13.509
15.293
12.884
17.738
13.272
15.14
9.953
13.431
9.258
10.513
10.352
13.99
9.718
11.009
22.16
28.236
18.641
20.749
9.668
13.946
10.956
12.745
with NASAs composite measurement; the others were

significantly different. As a result of this calibration activity,
Space Imaging updated its published coefficients to incorporate the coefficients derived by the NASA team (Dial et
al., 2003, this issue). Tables 6 8 show how the three teams
vicarious calibration TOA radiance estimates compare with
the Space Imaging updated calibrated radiance values.
White
5/26/00
Sands
Missile
Range
Ivanpah
7/21/00
8. MTFC processing
The effect of MTFC on radiometry for vicarious calibrations and for imagery in general was evaluated as part of this
investigation. To facilitate this effort, Space Imaging provided a 5 5-km scene over Phoenix, AZ, with and without
MTFC processing taken on October 12, 1999 at 17:46 UTC.
The image contained a large urban area having a high spatial
frequency content because of the many edges associated
with buildings and roads. The image also contained some
vegetative areas with relatively low spatial frequency content. First-order statistics, means, standard deviations, histograms, RMSs, and peak differences between MTFC-On and
MTFC-Off imagery were calculated for both the panchromatic and the multispectral radiometrically corrected, georeferenced imagery. In addition, all image pairs (processed
with and without MTFC) were visually inspected and the
differences between them were displayed visually as a
difference image. MTFC estimates were made, and
inverses to these functions were performed to restore
compensated imagery back to its MTFC pre-processed state.
Table 6
NASA SSC TOA radiance results using updated SI calibration coefficients
Site
Date
Band
DN
TOA
radiance
[W/(m2 sr)]
IKONOS
radiance
[W/(m2 sr)]
Lunar
Lake
6/7/00
Lunar
Lake
6/10/00
Blue
Green
Red
NIR
Blue
Green
Red
NIR
828
1178
1174
1083
811
1093
1108
1028
12.56
16.987
13.531
14.153
12.841
17.28
13.642
14.053
13.081
18.151
13.976
14.517
12.812
16.841
13.19
13.78
Percent
difference
4.148
6.852
3.289
2.572
0.226
2.541
3.313
1.943
13.444
18.737
14.476
15.054
12.496
17.519
13.702
14.236
11.596
15.593
11.524
11.729
10.537
14.083
10.429
10.550
22.464
28.136
20.167
20.965
10.142
14.407
11.893
12.453
2.20
2.46
7.16
1.57
3.01
1.23
3.24
5.97
16.50
16.10
24.47
11.57
1.79
0.67
7.31
4.17
1.37
0.36
8.18
1.04
4.90
3.31
8.55
2.29
A small 1024 1024 pixel section from the upper left

corner of each original panchromatic and multispectral
image was chosen for the analysis. This size image was
chosen to enable a quick look and to take advantage of basetwo fast Fourier transforms. A visual inspection of each
image pair shows that buildings and other high-spatialcontent objects are more discernible in the MTFC-processed
image. MTFC broadened the histogram on both the low and
high DN ends of the MTFC-On image. Some pixels were
driven to zero or saturated by the MTFC processing. FirstTable 8
SDSU TOA radiance results using updated SI calibration coefficients
Site
Date
Band
DN TOA
IKONOS
Percent
radiance
radiance
difference
[W/(m2 sr)] [W/(m2 sr)]

Grass
Pass 1 Green 289
pixels
Red
225
NIR
577
Dark
Pass 1 Green 149
pixels
Red
71
NIR
60
Grass
Pass 2 Green 304
pixels
Red
239
NIR
589
Dark
Pass 2 Green 147
pixels
Red
70
NIR
40
3.618
4.168
2.7
7.72
2.439
1.806
0.652
0.654
3.679
4.219
2.718
7.741
2.488
1.839
0.66
0.651
3.934
4.453
2.679
7.735
2.907
2.296
0.845
0.804
4.123
4.684
2.845
7.895
2.938
2.265
0.833
0.536
8.72
6.84
0.79
0.19
19.18
27.12
29.64
22.98
12.07
11.02
4.68
2.00
18.10
23.17
26.26
17.64

Table 9
First-order panchromatic statistics summary of Phoenix site
Statistic
Mean DN
Standard deviation DN
Minimum DN
Maximum DN
Maximum difference DN
Minimum difference DN
RMS difference DN
Panchromatic
MTFC-Off
Panchromatic
MTFC-On
598.2
183.6
149
1983
844
781
0.085
598.2
205.2
0
2047
844
781
0.085
order statistics for the two Phoenix images are shown in

Tables 9 and 10. The image mean DN value was not
affected by MTFC processing; however, the standard deviation between pixels increased by about 10%. Cross-sectional plots of the Phoenix MTFC-On and MTFC-Off
imagery were made to examine pixel-by-pixel variations.
The difference between these plots indicates the magnitude
of edge and structure sharpening that has occurred as a
result of MTFC. It was not uncommon, in a given row, to
observe differences exceeding 50% in areas where there
were large gradients. The RMS difference in DNs between
panchromatic imagery processed with and without MTFC
was nearly 8.5% and was between 4.7% and 7.6% for all
four multispectral bands.
Differencing the two images pixel by pixel throughout
the scene generated a difference image. Fig. 8 shows the
original MTFC-On image and the difference image between
MTFC-On and MTFC-Off for the Phoenix scene. For
display reasons, absolute values of these differences were
used. The difference image shows up bright in areas of
many edges because of building and roads, indicating large
differences between the imagery processed with and without
MTFC in those areas. The difference image shows up darker
in more uniform areas, indicating smaller differences between the two images in those regions.
The Space Imaging MTFC function was estimated in the
frequency domain by calculating the ratio of the Fourier
transform of the MTFC-processed image to the Fourier
transform of the non-MTFC-processed image. To improve
the estimate of the transfer function, the image was broken
down into sixteen 256 256 point transforms, which were
65
averaged to estimate the MTFC function for each small

section of the image. The panchromatic MTFC function
appeared to be a flower petal in shape, indicating relative
symmetry between the in-track and cross-track directions.
The multispectral MTFC functions appeared to be more
cylindrical in shape and boosted the high spatial frequencies
by half as much as in the panchromatic case. Additionally,
the cross-track direction is more strongly compensated than
the in-track direction. Similar results were obtained for each
multispectral band, but in general the effects of MTFC
increased in going from the blue to NIR band. Contour
plots of the two-dimensional MTFC functions show that
they have been rotated relative to the image coordinate
system, apparently because of geometric referencing. Space
Imaging performs the MTFC and georeferencing simultaneously with a combined convolution kernel.
To test the validity of the estimated MTFC function, a
256 256-pixel MTFC-processed image provided by Space
Imaging was processed again with the inverse of the MTFC
function estimate. Before the MTFC function estimate was
inversed, it was further smoothed with a 5 5 convolution
kernel-averaging filter. If a good estimate of the MTFC
function was found, the resulting restored image should be
close to the original image that was not MTFC processed.
The RMS difference between imagery that was not processed with MTFC and imagery that had been restored back
to its pre-MTFC processed state was less than 1%. Applying
the same inverse function to other images did not always
produce ideal results. This indicates that the MTFC process
is only partially reversible because georeferencing is performed at the same time that MTFC processing occurs.
Another interesting observation is that the SNR is lowered with MTFC processing. Simulated scenes representing
shot noise limited imagery show a significant increase in
noise (20 40%) for MTFC-processed multispectral imagery
compared to non-MTFC processed imagery. This observation is not surprising because the shot noise process will
manifest itself as a two-dimensional white noise process
with significant high spatial frequency components that will
be amplified when MTFC is applied. This increased noise is
evident in both simulated and acquired scenes containing
known uniform areas. This likely explains why acquired
scenes such as Lunar Lake show nearly a 2% RMS
Table 10
First-order multispectral statistics summary of Phoenix site
Blue
MTFC-Off
Mean DN
S.D. DN
Minimum DN
Maximum DN
Maximum difference DN
Minimum difference DN
RMS difference DN
328.3
89.46
153
2047
425
327
0.047
Green
MTFC-On
328.3
97.89
0
2047
MTFC-Off
412.5
133.45
135
2047
444
307
0.059
Red
MTFC-On
412.5
147.29
0
2047
MTFC-Off
397.8
146.0
83
2047
454
327
0.070
NIR
MTFC-On
397.8
161.5
0
2047
MTFC-Off
426.9
126.2
65
2047
427
378
0.0761
MTFC-On
426.8
145.84
0
2047
66
Fig. 8. Phoenix NIR MTFC-On image (top) and difference between MTFC-On and MTFC-Off NIR imagery (bottom).
difference in pixel values between MTFC-Off and MTFCOn imagery in the NIR band. Simulations of ideal uniform
scenes that do not contain noise show no difference in pixel
value between MTFC-Off and MTFC-On imagery.
9. Summary and conclusions

NASA performed a relative radiometric accuracy assessment of IKONOS imagery using large uniform Antarctic
and Saharan scenes and found that differences between
detector arrays were well within the Space Imaging data

specification. Uniform scenes for this analysis were found in
part by searching through the EarthSat GeoCover Landsat 5
TM imagery purchased through the NASA SDP. The
relative radiometric accuracy assessment was conducted
by comparing values obtained by summing the pixels of a
uniform image in the column direction. This method, which
amplifies detector defect effects, found that differences
between summed pixels were less than 1.5%. This analysis
illustrates that Space Imaging has properly calibrated the
detectors in a relative sense, and that a single calibration
coefficient for each band can be used to calibrate the entire

image.
The first independent absolute radiometric characterization of a new class of commercial imagery was conducted.
Several different vicarious calibrations from three different
teams (NASA SSC, the U of A RSG, and SDSU) were
combined to assess the IKONOS radiometric performance.
Although each team used unique radiative transfer methods
and slightly different approaches, the results were in excellent agreement. In the green, red, and NIR bands, significant
discrepancies between the vicarious calibration and Space
Imagings calibration were found. As a result of this activity,
Space Imaging updated its IKONOS multispectral radiometric calibration coefficients. This NASA team performed
additional vicarious calibrations to ascertain radiometric
stability of the IKONOS system over time.
The results of these radiometric characterizations as well
as results from other characterizations performed by NASA,
NIMA, and USGS teams were presented at annual JACIE
High Spatial Resolution Commercial Imagery Workshops
(NASA, NIMA, & USGS, 2001, 2002).
A first look at the question Should MTFC be applied to
imagery procured under the NASA Scientific Data Purchase
(SDP)? was taken. At this stage, it is recommended that
MTFC be applied to the imagery when the imagery will be
used primarily to inspect features visually. Over relatively
uniform scenes such as vegetative areas, little difference was
observed with either option. In scenes with many edges and
point sources, individual pixel values were observed to
differ by more than 50% between the two options with
scene RMS differences typically less than 8.5% for both the
panchromatic and multispectral products. In these cases and
when radiometry is important, imagery without MTFC
processing is preferable.
Acknowledgements
This work was supported by the NASA Earth Science
Applications Directorate under contract number NAS 13650 at the John C. Stennis Space Center, MS.
References
Berk, A., Anderson, G. P., Acharya, P. K., Chetwynd, J. H., Bernstein, L.
S., Shettle, E. P., Matthew, M. W., & Adler-Golden, S. M. (1999).
MODTRAN4 users manual. Hanscom AFB, MA: Air Force Research Laboratory, Space Vehicles Directorate, Air Force Materiel
Command.
Biggar, S. F., Gellman, D. I., & Slater, P. N. (1990). Improved evaluation of
optical depth components from Langley plot data. Remote Sensing of
Environment, 32, 91 101.
Biggar, S. F., Slater, P. N., & Gellman, D. I. (1994). Uncertainties in the inflight calibration of sensors with reference to measured ground sites in
the 0.4 1.1 Am range. Remote Sensing of Environment, 48, 245 252.
67
uses of commercial remote sensing data. Remote Sensing of Environment,

88, 3 16. (doi=10.1016/j.rse.2003.07.007)
Bowen, H. S. (2002). Absolute radiometric calibration of the IKONOS sensor using radiometrically characterized stellar sources. In Proceedings of
the ISPRS Commission I Mid-Term Symposium/Pecora 15-Land Satellite Information IV Conference, November 10 14, Denver, CO.
Brower, B. V., Cosgrove, M. A., Lewis, D. H., VanLare, G. R., Fisher, T. J.,
& Charneski, D. M. (2000). Advanced space-qualified downlink image
compression ASIC for commercial remote sensing applications. In A.
G. Tescher (Ed.), Proceedings of SPIE: Applications of digital image
processing XXIII, vol. 4115 (pp. 311 319).
DaMommio, T., & Kuo, S. (1992). Optical design for the ATLAS multispectral scanner. In D. M. Aikens, V. L. Genberg, G. C. Krumweide, &
M. J. Thomas (Eds.), Proceedings of SPIE: Design of optical instruments, vol. 1690 (pp. 56 69).
IKONOS satellite, imagery, and products. Remote Sensing of Environment, 88, 23 36. (doi:10.1016/j.rse.2003.08.014)
Dinguirard, M., & Slater, P. N. (1999). Calibration of space-multispectral imaging sensors: A review. Remote Sensing of Environment, 68,
194 205.
Ehsani, A. R., Reagan, J. A., & Erxleben, W. H. (1998). Design and performance analysis of an automated 10-channel solar radiometer instrument.
Journal of Atmospheric and Oceanic Technology, 15, 697 707.
Holst, G. (1995). Electro-optical imaging performance ( pp. 146 147).
Bellingham, WA: SPIE Optical Engineering Press.
Jackson, R. D., Clarke, T. R., & Moran, M. S. (1992). Bidirectional calibration results for 11 spectralon and 16 BaSO4 reference reflectance
panels. Remote Sensing of Environment, 40, 231 239.
Justice, C., Eck, T., Tanre, D., & Holben, B. (1991). The effect of water
vapor on the normalized difference vegetation index derived for the
Sahelian region from NOAA AVHRR data. International Journal of
Remote Sensing, 12(6), 1165 1187.
Kaufman, Y. J., & Holben, B. N. (1993). Calibration of the AVHRR visible
and near-IR bands by atmospheric scattering, ocean glint and desert
reflection. International Journal of Remote Sensing, 14(1), 21 52.
Leachtenauer, J. C., Malila, W., Irvine, J., Colburn, L., & Salvaggio, N.
(1997). General image-quality equation: GIQE. Applied Optics, 36(32),
8322 8328.
Meygret, A., Briottet, X., Henry, P., & Hagolle, O. (2000). Calibration of
SPOT 4 HRVIR and vegetation cameras over Rayleigh scattering. In
W. L. Barnes (Ed.), Proceedings of SPIE: Earth observing systems V,
vol. 4135 (pp. 302 313).
Nandy, P., Thome, K., & Biggar, S. (2001). Characterization and field use
of a CCD camera system for retrieval of bidirectional reflectance distribution function. Journal of Geophysical Research-Atmospheres,
106(D11), 11957 11966.
MD, USA (sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team, CD-ROM).
Resolution Commercial Imagery Workshop, March 25 27, Reston,
VA, USA (sponsored by NASA/NIMA/USGS Joint Agency Commercial Imagery Evaluation Team, CD-ROM).
National Research Council (2002). Toward new partnerships in remote
sensing: Government, the private sector and earth science research.
Peterson, B. (2001). Space Imaging IKONOS radiometric characterization.
In Proceedings of the 2001 High Spatial Resolution Commercial Imagery Workshop, March 19 22, Greenbelt, MD, USA (sponsored by
Team, CD-ROM).
Ryan, R., Baldridge, B., Schowengerdt, R. A., Choi, T., Helder, D. L., &
Blonski, S. (2003). IKONOS spatial resolution and image interpretability characterization. Remote Sensing of Environment, 88, 37 51.
(doi:10.1016/j.rse.2000.07.006)
68
Sawaya, K., Olmanson, L., Heinert, N., Loeffelholz, B., Rich, R., Brezonik,
P., & Bauer, M. (2003). Extending satellite remote sensing to local
scales: Land and water resource management using high-resolution
imagery. Remote Sensing of Environment, 88, 143 155. (do:10.1016/
j.rse.2003.04.006)
Slater, P. N., Biggar, S. F., Holm, R. G., Jackson, R. D., Mao, Y., Moran,
M. S., Plamer, J. M., & Yuan, B. (1987). Reflectance- and radiancebased methods for the in-flight absolute calibration of multispectral
sensors. Remote Sensing of Environment, 22, 11 37.
Slater, P. N., Biggar, S. F., Thome, K. J., Gellman, D. I., & Spyak, P. R.
(1994). In-flight radiometric calibration of ASTER by reference to wellcharacterized scenes. In W. L. Barnes, & B. J. Horais (Eds.), Proceedings of SPIE: Platforms and systems, vol. 2317 (pp. 49 60).
Slater, P. N., Biggar, S. F., Thome, K. J., Gellman, D. I., & Spyak, P. R.
(1996). Vicarious radiometric calibration of EOS sensors. Journal of
Atmospheric and Oceanic Technology, 13, 349 359.
Thome, K., & Nandy, P. (2000). Accuracy of ground-reference calibration
of imaging spectroradiometers at large sensor view angles. In M. R.

Sargent, & I. M. Sargent (Eds.), Proceedings of SPIE: Imaging spectrometry VI, vol. 4132 (pp. 260 268).
Thome, K. J. (2001). Absolute radiometric calibration of Landsat-7 ETM+
using the reflectance-based method. Remote Sensing of Environment,
78, 27 38.
Thome, K. J., Gellman, D. I., Parada, R. J., Biggar, S. F., Slater, P. N., &
Moran, M. S. (1993). In-flight radiometric calibration of Landsat-5
Thematic Mapper from 1984 to present. In P. S. Chavez, & R. A.
Schowengerdt (Eds.), Proceedings of SPIE: Recent advances in sensors, radiometric calibration, and processing of remotely sensed data,
vol. 1938 (pp. 126 131).
Thome, K. J., Gustafson-Bold, C. L., Slater, P. N., & Farrand, W. H.
(1996). In-flight radiometric calibration of HYDICE using a reflectance-based approach. In S. S. Shen (Ed.), Proceedings of SPIE: Hyperspectral remote sensing and applications, vol. 2821 (pp. 311 319).

IKONOS geometric characterization

Dennis Helder a,*, Michael Coan b, Kevin Patrick b, Peter Gaska c
a
Electrical Engineering Department, South Dakota State University, PO Box 2220, Brookings, SD 57007, USA
b
USGS EROS Data Center, Sioux Falls, SD 57105, USA
c
NIMA, 3838 Vogel Road, L-85, Arnold, MO 63010, USA
Received 2 June 2002; received in revised form 24 January 2003; accepted 22 April 2003
Abstract
The IKONOS spacecraft acquired images on July 3, 17, and 25, and August 13, 2001 of Brookings SD, a small city in east central South
Dakota, and on May 22, June 30, and July 30, 2000, of the rural area around the EROS Data Center. South Dakota State University (SDSU)
evaluated the Brookings scenes and the USGS EROS Data Center (EDC) evaluated the other scenes. The images evaluated by SDSU utilized
various natural objects and man-made features as identifiable targets randomly distribution throughout the scenes, while the images evaluated
by EDC utilized pre-marked artificial points (panel points) to provide the best possible targets distributed in a grid pattern. Space Imaging
provided products at different processing levels to each institution. For each scene, the pixel (line, sample) locations of the various targets
were compared to field observed, survey-grade Global Positioning System locations. Patterns of error distribution for each product were
plotted, and a variety of statistical statements of accuracy are made. The IKONOS sensor also acquired 12 pairs of stereo images of globally
distributed scenes between April 2000 and April 2001. For each scene, analysts at the National Imagery and Mapping Agency (NIMA)
compared derived photogrammetric coordinates to their corresponding NIMA field-surveyed ground control point (GCPs). NIMA analysts
determined horizontal and vertical accuracies by averaging the differences between the derived photogrammetric points and the fieldsurveyed GCPs for all 12 stereo pairs. Patterns of error distribution for each scene are presented.
Keywords: IKONOS; Brookings; Space Imaging
1. Introduction
As part of the Joint Agency Commercial Imaging Evaluation (JACIE, which includes NASA, NIMA, and the
USGS), several institutions evaluated the geometric registration of Space Imagings IKONOS imagery using data
purchased under the NASA Scientific Data Purchase (SDP)
Program and the NIMA Commercial Imagery Program
(CIP). The results of the various registration efforts, using
definitions set up for the SDP over a year prior to the launch
of IKONOS, are reported in this article. This paper is neither
an endorsement nor promotion of Space Imaging products
by any of the agencies listed above. It is a report that
summarizes evaluations of registration accuracy of the
images purchased from Space Imaging by the JACIE
agencies. The current data products publicly available from
E-mail address: dennis_helder@sdstate.edu (D. Helder).
doi:10.1016/j.rse.2003.04.002
Space Imaging follow a different nomenclature from what

was used to establish the specifications for the SDP;
however, by comparing the results from registration analyses of the different products that were evaluated, one should
be able to get an idea of the accuracy of the commercial
products relative to one another.
With the increasing commercialization of space, there are
more products becoming available that have higher spatial
resolution than Landsat data. Space-based commercial highresolution imagery with spatial resolution less than 4 m has
been available to the general public for a short time. The
JACIE team was established to perform analysis and verification of products from commercial sources to ensure that
the quality of the commercial venders products matches the
potential needs of government customers. These test and
analysis activities were designed to determine whether the
data sets meet contract specifications. Multiple governmental agencies are working in cooperation with the NASA
Stennis Space Center in the validation and verification tests
performed by JACIE.
70
D. Helder et al. / Remote Sensing of Environment 88 (2003) 6979
Verification and validation analyses involved the assessment of geolocation, spatial resolution, radiometric
response functions, and the creation of a digital elevation
model (DEM) using multiple IKONOS data sets prepared
by Space Imaging. The geometric evaluation sites at
Brookings, SD, the EDC near Sioux Falls, SD, and the
12 NIMA sites distributed worldwide comprise the test
site locations where the registration evaluations were
performed.
The Space Imaging Carterra commercial products and
their relevant registration accuracies are:
Product
Accuracy
Geo
Reference
Precision
F 50 m CE90, 23.6 m RMS

F 25.4 m CE90, 11.8 m RMS
F 4.1 m CE90, 1.9 m RMS
CE90, or circular error 90%, refers to how large a circle

would be if, on average, it were to enclose 90% of the GCP
geolocation errors. For example, 50 m CE90 indicates that
geodetic locations derived from that product would be
within 50 m of the true location 90% of the time.
The CE90 specifications as listed in the NASA SDP
contract with Space Imaging (RFO 13-SSC-O-97-21) are:
Product
Accuracy
Original Standard
Original Precision
F 250 m horizontal accuracy

F 3 m pan and F 5 m multispectral, horizontal
accuracies (exclusive of terrain effects)
F 12.2 m horizontal accuracy
F 2 m pan and F 5 m multispectral, horizontal
accuracies
Master Standard
Master Precision
Original Standard products do not incorporate the use of

ground control points (GCPs) or terrain correction; Original
Precision products utilize ground control points; Master
Standard products use terrain correction; and Master Precision products use both ground control points and terrain
correction.
The IKONOS spacecraft is in a sun synchronous, 98.1j
inclined, 681 km orbit, with an equatorial crossing time of
10:30 a.m. The platform is a pointable, along-track and
cross-track system enabling ground imaging repeat times of
1.5 days for multispectral data and 2.9 days for panchromatic data. Nadir repeat is similar to that of the repeat cycle
for the Landsat system: 16 days. The ground resolution at
nadir is 1 m for panchromatic and 4 m for multispectral
images.
The imaging sensor was developed by Kodak for Space
Imaging. The system consists of an anastigmat, three-lens
mirrors, 10-m focal length with focal ratio of F14.3, all of
which are contained within a 1.5-m long by 0.79-m wide
enclosure. The push-broom focal plane sensor contains
13,500 pixels in the panchromatic array and four bands of
color pixels at 3375 pixels per band. The approximate

spectral bandwidths and spatial resolutions are:
Band
Spectral bandwidth
(Am)
Spatial resolution
(m)
PAN
Blue
Green
Red
Near IR
0.45 0.90
0.45 0.52
0.52 0.60
0.63 0.69
0.76 0.90
1
4
4
4
4
Although the IKONOS sensor has been collecting data

for only a short period of time, many reports have been
generated describing the geometric features of the instrument (cf. Baltsavias, Pateraki, & Zhang, 2001; Davis &
Wang, 2001; Dial, 2000; Fraser, Hanley, & Yamakawa,
2001; Gerlach, 2000; Hanley & Fraser, 2001). Fraser, for
example, indicated that 1-m accuracies were possible with
the standard product if good ground control points were
available. The purpose of this paper is to report on the
geometric accuracies of the Space Imaging products as they
specifically relate to the NASA Scientific Data Purchase.
The images were evaluated in accordance with the
guidelines given in the Federal Geographic Data Committee
standard (FGDC-STD-007.3, 1998). Appendix 3C of that
section describes a three-step process: well-defined points,
data acquisition for the independent source of higher
accuracy, and check point location. Features clearly identifiable in the imagery were chosen. The feature locations
were measured via a GPS-based survey and the UTM
coordinates from the image data were recorded. The horizontal root-mean square error (RMSE), and then circular
errors were calculated in accordance with FGDC-STD007.3 Appendix 3A.
This paper also presents the methodology employed by
NIMA analysts in a geometric characterization of 12 IKONOS stereo pairs. The characterization involved comparing
the accuracy of each GCP with the calculated horizontal and
vertical accuracies of the respective IKONOS stereo pairs
for each scene. NIMAs evaluation of the geometric accuracy of each stereo pair is presented and conclusions about
the average accuracy for each of the 12 scenes are discussed.
Lastly, we discuss some extrapolations of the effects of
enhancements to Space Imaging interior and exterior orientation that were investigated subsequent to this study.
2. Description/comparison of test sites

Three types of sites were used: the Brookings/SDSU site
used GCPs based on available natural and man-made
features that were time-invariant; the EDC site used highly
accurate GCPs (artificial targets) deployed during the overpass; and the NIMA stereo sites were scattered worldwide
with GCPs at each site.
2.1. Brookings/SDSU site

As part of the JACIE effort to verify geometric accuracy
of commercial imagery obtained through the NASA Scientific Data Buy Program, a set of GCPs was established in
and around Brookings, SD (Fig. 1). This site was chosen
because of relatively level terrain, the presence of a small
city in a rural landscape, a pronounced grid system established by the network of roads typical in the Midwest, and
convenience to the South Dakota State University Image
Processing Laboratory which is located in Brookings. Additionally, it was desired to develop a site that could be used
for a number of years, for multiple sensors, and that would
require minimal maintenance. For this reason, it was decided to identify GCPs that would be as time-invariant as
possible and did not require deployment of artificial targets
for each satellite overpass. This was achieved relatively
easily due to the presence of the aforementioned road
system. A total of 49 usable GCPs were obtained and the
majority of them consisted of intersections of roads, sidewalks, or railroad lines. These points were chosen to be well
scattered throughout an 11 11-km region. In addition,
other features were used such as airport runway centerline
markings and a helicopter landing pad near a hospital.
In order to obtain an accurate set of GCPs, the Ground
Reference Information Team (GRIT) from Stennis Space
Center performed a GPS-based survey. Base stations were
set up at two National Geodetic Survey (NGS) points: one at
the local airport (PID PR1201) and another near Interstate
29 (PID PR1044), the major highway in this region. All
points were checked for accuracy via triangulation with a
maximum error of 5 cm. Since these GCPs were based on
the NAD83 datum and Space Imaging data is obtained in
ITRF96/WGS84 (G873), a datum conversion was necessary
71
(Strange, 2000). This was accomplished using NGS-supplied software for horizontal time dependent position
(HTDP) giving correction factors near scene center of
+ 0.733m UTM X and 0.838m UTM Y.
2.2. EROS Data Center site
The USGS/EROS Data Center is located about 15 miles
northeast of Sioux Falls, SD (Fig. 1). The area around the
center is mostly gently rolling farmland, planted with corn,
soybeans, or left fallow for grazing, and intersected by a
rectangular system of straight, unpaved roads. The roads are
a result of the Public Lands Surveys, typically of 1-mile2
sections, used to establish land ownership in this region.
Over an approximately 10 by 10 km2 area suitable for a
single IKONOS scene acquisition, a system of artificially
marked points were established to augment the few existing photo-identifiable man-made features proximal to the
EROS Data Center. Marker panels were constructed and
placed near the road intersections to create a nearly linear
grid of points. The panels were nominally aligned with the
orbital path of the spacecraft such that the long axis
approximated the along-track direction of the satellite.
The middle of the center panel, a 3-ft2, was the registration
point (Fig. 2).
These panel markers provided clearly identifiable points
for which to compare the image pixel locations to the GPS
coordinates. The temporary targeted panels on the EDC site
served as highly accurate GCPs. The EDC team provided
the Stennis GRIT crew with two National Geodetic Survey
Cooperative Base Network (CBN) stations (PIDs OQ1033
and OQ1245) as the basis for establishing onsite control for
registration of the target ground control points. The simultaneous observations between the two NGS CBN stations
Fig. 1. Location of Brookings and Sioux Falls, South Dakota test sites.
72

Table 1
Listing of dates and satellite view angles for Summer 2001 Brookings site
data sets
Summer 2001 Brookings collects
Fig. 2. EDC artificial GCP panels. The long, rectangular panel legs, and
spacing were designed for accurate location of the center square.
and the two new onsite control points agreed within 0.020
m. These new control points were simultaneously occupied
and used to locate all panel points, within agreements of
0.050 m. Because the initial control of the field survey was
based on the NAD83 datum, all of the resulting GPS point
locations supplied by the Stennis crew were in the NAD83
datum. However, the coordinate system of the images
supplied by Space Imaging is based on the WGS84 datum.
The NGS software was used to compute the required datum
transformation from WGS84 to NAD83. For a point near
the center of the scene acquired over EDC, these values
were computed to be: + 0.756m UTM X, 0.808 m UTM
Y. All data points were transformed to NAD83 prior to
calculation of RMSE.
2.3. NIMA sites
The Precision Engagement Divisions Office of Targeting
and Navigation (PTNT) at NIMA ordered IKONOS stereo
pairs over Test and Evaluation (T&E) sites. The T&E
feature coordinates provide highly accurate photogrammetric control data for several locations worldwide for use in
metric assessment and calibration activities for National
Technical Means. The control data are derived from GPS
ground truth surveys. Each T&E site contains at least 10
distinct photo-identifiable features. Test and Evaluation
graphics depict each of the cluster points along with their
associated geodetic coordinates and accuracies.
3. Procedures
3.1. Brookings/SDSU
During the summer of 2001, four IKONOS images were
collected of the Brookings site. All imagery was the Space
Imaging Standard Original product, i.e. no ground control
Date
Azimuth (j)
Elevation (j)
July 3
July 17
July 25
August 13
281.85
289.94
269.50
353.76
76.19
60.80
73.20
82.54
points or terrain correction were used to create this product.

Acquisition dates and satellite view angles are listed in
Table 1. It should be noted from this table that the satellite
was viewing from northwest to southeast when the images
were collected. Also, the satellite elevation angle was fairly
low on several dates, notably July 17. Once the imagery was
obtained, six individuals in the SDSU Image Processing
Laboratory evaluated each scene for geometric accuracy.
These individuals ranged from new students who had never
experienced this activity before, to those who had done this
several times. Each individual identified and recorded the
location of 49 GCPs. Analysis indicated that the error
contribution due to the operators and their ability to identify
the type of GCPs was, at most, F 1 m RMSE. The RMSE
values were calculated in the northing and easting direction
as well as the total RMSE vector magnitude (calculated as
the orthogonal sum of the two components), along with
several other commonly used measures of image accuracy.
Lastly, CE90 values were calculated since the SDP specified
geometric accuracy of the imagery was in these units. While
both panchromatic and multispectral bands were analyzed,
only slight decreases in geometric accuracy were noted
when comparing the multispectral band results to the
panchromatic band. These decreases in geometric accuracy
were attributed to the increased pixel size of the multispectral bands, and therefore only the panchromatic band results
are presented here. Overlay checks of the multispectral data
against the panchromatic data showed no apparent displacement that could not be accounted for by the difference in the
size of the pixels (1 m pan, 4 m multispectral).
Table 2
Geometric accuracy results for all four dates, panchromatic band,
Brookings site
RMSEx
RMSEy
RMSEmin/RMSEmax
RMSE vector
Accuracy
Circular standard error
(39%)
probable (50%)
Circular map accuracy
standard (90%)
July 3
July 17
July 25
August 13
3.458
0.804
0.233
3.551
5.217
2.131
15.663
1.774
0.113
15.763
21.341
8.719
12.755
1.288
0.101
12.820
17.186
7.021
2.681
1.693
0.632
3.171
5.353
2.187
2.509
10.265
8.267
2.575
5.388
23.921
19.454
4.812
73
Fig. 3. Displacement vectors indicating GCP errors for the panchromatic band on July 3.
3.2. USGS/EDC
Both the Original Standard and Original Precision products were initially evaluated by the EDC using the same
image acquisition. The 11-bit image of each Space Imaging
panchromatic product contained a great deal of information
at the high end of the dynamic range, enabling distinction
between bright objects such as graveled roads, concrete, and
metal roofs, as well as the Tyvek target material. In normal
screen display, however, the Tyvek material used in the

panels was so bright that it also increased the brightness of
adjacent pixels. For this reason, a square-root reduction of
the 11-bit data was performed, and the result was rescaled
into an 8-bit image. This successfully lowered the dynamic
range, while preserving the identities present at the highest
end. After doing so, the target materials were readily
identifiable, with an apparent size and shape that more
accurately reflected their actual configuration.
Fig. 4. Displacement vectors indicating GCP errors for the panchromatic band on July 17.
74
Fig. 5. Illustration of terrain elevation correction.
The effects of data quantization and resampling during

the data acquisition, preprocessing, and georegistration
stages combined with various field effects (i.e. panel orientation, steep sideslopes, or panels partially obscured by dirt
blown off of plowed fields) made some ground control point
targets difficult to identify. In all cases, the panel legs
helped to establish the apparent center of the target panel. In
those cases when the center panel was excessively oversampled or obscured, knowing the relative geometry of the
legs and center panel especially helped to identify the target
point on the image.
3.3. NIMA
IKONOS stereo imagery was delivered in National
Imagery Transmission Format (NITF) 2.0 on CD-ROM
media. Corresponding right and left images were acquired
within one minute of each other on the same orbital pass.
Both images were resampled to a uniform pixel size of 1 m
and epipolar rectified to the WGS84 ellipsoid. This process
orients the imagery for optimal stereo display. The NITF is
available only to specific United States and international
government agencies.
NIMA ingested the IKONOS stereo pairs and mensurated each photo-identifiable point using the commercially
available SOCET Set software, version 4.3.1, licensed by
British Aerospace Engineering (BAE) Systems. NIMA
analysts used T&E point graphics to locate the points in
each scene. The T&E graphics products consist of annotated
panchromatic imagery and diagrams that locate each GCP
and provide the surveyed geodetic coordinates, elevations,
and accuracies for each GCP. Analysts measured T&E

points on the stereo pairs, computed ground coordinates,
and tabulated values for latitude, longitude, and elevation
for each extracted T&E point. These values were compared
to the published T&E coordinates and elevations and
entered into a computer spreadsheet for statistical analysis.
Tools within Socet Set enable the analyst to photogrammetrically mensurate geographic coordinates and elevation
from the encoded pixels in the stereo image. The NIMA
analysts measured latitudes and longitudes and imported
these into the NIMA-developed software program GeoDiff,
which calculates the difference between the photogrammetrically derived geographic coordinates and the published
T&E coordinates for a given GCP and displays the difference in a prescribed unit of measure.
Table 3
Geometric accuracy results for all four dates, panchromatic band,
Brookings site, after terrain correction
RMSEx
RMSEy
RMSEmin/RMSEmax
RMSE vector
Accuracy
(39%)
probable (50%)
Circular map accuracy
standard (90%)
July 3
July 17
July 25
August 13
4.011
0.879
0.219
4.107
5.986
2.445
9.249
2.370
0.256
9.548
14.219
5.809
6.191
3.063
0.495
6.907
11.326
4.627
4.281
1.842
0.430
4.661
7.494
3.062
2.879
6.840
5.448
3.605
6.232
14.488
10.482
7.073

Table 4
EDC scene Standard Original product results
NSSDA: RMSEx
NSSDA: RMSEy
NSSDA: RMSEmin/RMSEmax
NSSDA: RMSEr
NSSDA: accuracy (95% confidence)
NMAS: circular standard error (39%)
NMAS: circular error probable (50%)
NMAS: circular map accuracy standard (90%)
5.885 m
3.819 m
0.649 m
7.015 m
11.876 m
4.852 m
5.712 m
10.646 m
The rational polynomial coefficients (RPC) provided in

the NITF header data constitute a mathematical rational
function that effectively models the sensors interior and
exterior orientation at the time of image acquisition. This
information is used in stereo intersection associated with
precision image exploitation procedures. Space Imaging
provided stereo pairs with two versions of the RPC tags,
RPC00A (eight cases) and RPC00B (four cases). Space
Imaging implemented RPC00B to improve precision of
some parameters. The RPC00B format represents a revision
made during the Civil and Commercial Applications Project
(CCAP) evaluation performed by NIMA. Space Imaging
now sells a Geo Ortho Kit that enables anyone to exploit the
RPC data using GeoTIFF.
4. Results
4.1. Brookings/SDSU
Table 2 shows the accuracy results for the panchromatic
band on all four dates. For each date, the observer whose
75
RMSE vector result was the median value (third largest for
the case of six observers) is reported. All other values were
within F 0.5 m. These results indicate excellent geometric
accuracy in the northing direction, down to the 1-m level for
the July 3 date. However, in the easting direction, accuracies
are significantly larger on two dates. For the July 3 and
August 13 dates, easting RMSEs are at the 3-m level.
However, on the July 17 and July 25 dates, easting RMSEs
increase dramatically to 13 m and more. It is clear that there
is a strong correlation between the RMSE vector magnitude
and the satellite elevation angle. Higher elevation angles give
smaller RMSE vector magnitudes. When RMSEx does not
equal RMSEy, geolocational errors are not circular. Since,
from a practical point of view, this will rarely occur in
practice, a threshold was established for determining if errors
are indeed circular. Established by the NSSDA, a threshold
of RMSEmin/RMSEmax > 0.6 is used as an indicator of
circularity, where RMSEmin is the smaller of RMSEx or
RMSEy, and RMSEmax is the larger of the two. For the
present results, the ratio of RMSEmin/RMSEmax is only
greater than 0.6 on August 13 indicating that this is the only
date where the errors can be considered circular in nature.
Even though these ratios do not satisfy the criteria in FGDCSTD-007.3 Appendix A, circular errors were calculated for
the sake of completeness using the methods outlined in this
appendix. On the July 3 and August 13 dates CE90 hovers
around 5 m. However, on the other two dates, values of 20 m
or more were obtained. These results indicate the presence of
a systematic error. But, it should be noted that all CE90
values are still within specification for this data product.
Vector representation of these results (panchromatic
band) for each GCP is shown for the July 3 and 17 images
Fig. 6. Displacement vectors of the Standard Original product of the EDC scene.
76
Table 5
Three observers of the EDC Precision Original product
NSSDA: RMSEx
NSSDA: RMSEy
NSSDA: RMSEr
NSSDA: accuracy
(95% confidence)
NMAS: circular standard
error (39%)
NMAS: circular error
probable (50%)
NMAS: circular map accuracy
standard (90%)
Table 6
Height adjusted values of EDC Precision Original product
Obs1 (m)
Obs2 (m)
Obs3 (m)
3.297
0.750
0.227
3.381
4.952
3.235
0.794
0.245
3.331
4.930
3.292
0.722
0.219
3.371
4.914
2.023
2.014
2.007
2.382
2.372
2.364
5.130
5.055
5.115
in Figs. 3 and 4. In these plots, the magnitude of the error

has been scaled by a factor of 100. In both of these plots, as
well as plots for the other two dates, virtually all vectors
point in the same westerly direction. However, the magnitude of the error is much larger on July 17 and 25.
Standard Original products as obtained from Space
Imaging are not corrected for terrain effects. Thus, a
significant portion of the easting errors may have been
due to variations in satellite elevation, especially when the
IKONOS satellite imaged at low elevation angles. All of
the data had been geometrically registered to a constant
elevation of 464.5 m above the ellipsoid. Since the
elevation of all the GCPs was known, a correction for
displacement in terrain height was performed. The entire
range in elevation for the Brookings GCPs was only 30
m. The procedure for terrain correction is illustrated in
Fig. 5. Differences in elevation, along with the satellite
NSSDA: RMSEx
NSSDA: RMSEy
NSSDA: RMSEr
NSSDA: accuracy
(95% confidence)
NMAS: circular standard
error (39%)
NMAS: circular error
probable (50%)
NMAS: circular map accuracy
standard (90%)
Obs1 (m)
Obs2 (m)
Obs3 (m)
1.930
0.605
0.313
2.023
3.103
1.996
0.636
0.319
2.095
3.221
1.930
0.585
0.303
2.017
3.079
1.268
1.316
1.258
1.493
1.549
1.481
3.070
3.179
3.061
elevation angle, were used to calculate a displacement

error. This error was then resolved into northing and
easting components.
Results obtained after correction for terrain effects are
shown in Table 3. The RMSE values in the easting direction
have been reduced to half what they were before terrain
correction for the July 17 and 25 dates. This also reduced
the RMSE magnitude on those dates to 9.5 and 6.9 m,
respectively. However, the RMSE ratio still indicates that
the errors are not circular in nature. The CE90 values were
also reduced to 14 and 10 m. However, it is interesting to
note that the RMSE in the northing direction increased
slightly, as did the RMSE in the easting direction on July
3 and August 13. This resulted in increases in CE90 of 1 2
m on those dates. While this increase was not expected, it
may be due in part to the degree of accuracy possible with
the type of GCPs and product being used.
Fig. 7. Displacement vectors of EDC scene in the Precision Original product with terrain effects.
77
Fig. 8. Terrain corrected displacement vectors of EDC scene in the Precision Original product.
4.2. USGS/EDC
The equations in FDGC-STD-007.3 Appendix A were
applied to the observers tagging of the paneled points in the
panchromatic image. One experienced observers results for
the Standard Original product are shown in Table 4. Notice
the ratio of the root-mean-square errors, RMSEmin/RMSEmax
is greater than 0.6 indicating, as mentioned above, the
suitability of the data for approximation of circular errors.
The drift pattern for this image is show in Fig. 6.
For the EDC scene processed to a Precision Original
product, three observers were used. The first two observers
had no prior knowledge of the experimental procedures, and
after a brief introduction, were left alone to generate their
results. The third observer was the same individual who
evaluated the Standard Original image in its delivered form.
Table 5 shows their individual findings. Notice the remarkTable 7
Results for stereo pairs provided in RPC00A format
Stereo pair accuracy
Antananarivo, MA
Miami, FL
Sioux City, IA
Fallon, NV
Abu Musa, TC
Hickam AFB, HI
St. Simons Is., GA
Villa Delores, AR
Utapao, TH
Acquisition date
HA (CE90)
(m)
VA (LE90)
(m)
29 April 2000
28 May 2000
28 May 2000
27 April 2000
3 October 2000
15 May 2000
3 June 2000
23 September 2000
16 May 2000
6.27
2.57
4.60
7.98
10.32
10.95
11.15
7.08
8.92
8.58
8.21
15.19
2.00
3.28
16.11
5.90
5.00
13.39
Abu Musa, TC, and Villa Delores, AR, repeats in both tables having
different RPC tags.
ably high agreement of RMSE components. These components are of the most significance because they are used to
derive all other statistical statements. The maximum difference between observers is expressed by the differences in
the RMSEr statistic (between observers 1 and 2) as (3.381
3.331 m) = 0.050 m. This is believed to be a direct consequence of using the pre-marked panels of known geometric
and spectral characteristics. Also notice the difference of the
Precision Original product from the Standard Original
products ratio of RMSE. Here, in all three cases, it falls
below the guidelines for FGDC case 2 by being less than the
threshold of 0.6. This raises a flag: errors may not be
normally distributed, or independent in x and y components.
Despite the indicated unsuitability of these data for generating further circular error statistics as per the FGDC
standard, circular error statistics were generated for consistency in reporting. The implications of this are noticeable by
looking at the drift patterns (Fig. 7). The drifts all tend to be
horizontal in nature. In other words, these errors may be
systematic.
It was not until after these results were reported at the
first JACIE Workshop on Commercial High-Resolution
Table 8
Results for stereo pairs provided in RPC00B format
Stereo pair accuracy
Abu Musa, TC
Villa Delores, AR
Christchurch, NZ
Keflavik, IC
Sunnyvale, CA
Acquisition date
HA (CE90)
(m)
VA (LE90)
(m)
3 October 2000
23 September 2000
27 April 2000
13 Mar 2001
17 April 2001
10.09
6.79
6.18
8.35
9.11
2.69
5.36
9.80
3.79
6.16
78
Table 9
Summary of accuracies for stereo products
Stereo RPC00A
Horizontal
accuracy (m)
Vertical
accuracy (m)
(n = 9)
Maximum
difference
Minimum
difference
Mean
11.15
16.11
2.57
2.00
7.76
8.63
Maximum
difference
Minimum
difference
Mean
Removing the terrain effects at each point, the ratio of

RMSEmin/RMSEmax was recalculated. The resultant was
still less than the threshold of 0.6 for case 2, indicating
the remaining errors are indeed systematic. However, for
consistency of reporting, the circular error statistics are
generated. The new RMSEr values show a great improvement over the non-elevation corrected values, resulting in a
drop to almost 2 m for panchromatic data.
Stereo RPC00B
Horizontal
accuracy
(n = 5)
4.3. NIMA
Maximum
difference
Minimum
difference
Mean
10.09
9.80
6.18
2.69
8.10
5.56
Maximum
difference
Minimum
difference
Mean
Accuracies of T&E points vary with the particular target,

but are generally within 1 m in latitude, longitude, and
elevation coordinates. (The specific accuracies and locations
of the T&E points used in this evaluation are classified.)
Tables 7 9 display the results from the accuracy analyses of
the T&E points for each stereo pair. The points for all 12
scenes fall within Space Imagings stated stereo accuracy
specification of 25 m for the horizontal component. Ninety
percent of the horizontal measurements are within 9.1 m.
With respect to the vertical component, all 12 elevation
measurements fall within Space Imagings stated vertical
accuracy specification of 22 m (IKONOS Imagery Products
and Product Guide, Space Imaging, 2002, p. 5). Ninety
percent of the elevation measurements have vertical accuracies better than 13.1 m (Fig. 9).
Vertical
accuracy
Imagery in 2001 that the USGS/EROS Data Center was

informed that the Precision Original product was reduced to
an average elevation, and any terrain effects were not
removed. Upon close inspection of the metadata files, the
RPC text file was found to contain a variable named
HEIGHT_OFF, with a value of + 0423.000 m. This
was confirmed to be the ellipsoidal height used as the
average elevation for the reduction, by inspection of the
ellipsoidal heights of the control points used to determine
the line of minimum displacement as sketched on the prior
displacement diagram, and verified by the engineering staff
at Space Imaging.
A delta height was computed between each control
points height above ellipsoid and this mean height. Using
this delta height value and the nominal acquisition elevation
angle, a magnitude of displacement was computed. This
magnitude was assumed to be oriented along the nominal
acquisition azimuth, which was broken into both UTM X
and UTM Y components. These components were applied
to the previously observed displacements, and these new
results are in Table 6 and the new drift plotted in Fig. 8.
5. Conclusions
In all cases reported, the Space Imaging products met
specifications. Excellent geometric accuracy can be obtained with the Standard Original producton the order
of 5 7 m RMSE for relatively flat terrain. There appears to
be a relationship between satellite elevation angle and
geometric accuracy. It was found that higher elevation
angles tend to result in lower RMSE. The relationship
suggests that satellite elevation angles above 75j tend to
Fig. 9. Stereo pair circular error (CE) and linear error (LE).
maximize the geometric accuracy of the Standard Original

product. Terrain correction, even for a relatively flat site,
improved RMSE values at lower satellite elevation angles.
However, in all cases, there seems to be a systematic error
that prevents measurements that are circular in nature, that
is, there appears to be a significantly larger easting error as
compared to the northing error.
The analysis of the Standard Original and Precision
Original data products showed the multispectral and panchromatic products to be well coregistered. In evaluating the
panchromatic images, it was found that the elevations
changes across the EDC scene, along with the small pixel
size for the panchromatic data, were enough to cause a
noticeable shift in the RMSE values between non-terrain
corrected data and data that were corrected for terrain
height. The evaluations also brought out the need for the
FDGC to consider addressing accuracy issues when the
RMSEmin/RMSEmax falls below the 0.6 threshold for case 2
of FDGC-STD-007.3. Since the first case in the standard
deals with the situation when RMSEx = RMSEy, and the
second case deals with situations where the ratio of these
two lies from 0.6 to 1.0, a third case should be developed for
those cases where the ratio is less than this value and noncircular errors clearly are present.
Lastly, the IKONOS stereo pairs investigated in this
study fall well within Space Imagings stated specifications
for absolute horizontal and vertical accuracy. Note that this
study addresses the accuracy of stereo images, in this case,
verification of the stated absolute accuracies. Space Imaging
has since enhanced interior orientation and exterior orientation. The interior orientation enhancement was applied to
all imagery processed after 18 November 2001 and improved the accuracy on the left edge of the imagery by 2 m.
The authors believe this enhancement would not affect the
results of the accuracy study reported herein as the measurements for all 12 scenes were near the center of their
respective images. The exterior orientation enhancement
79
was also applied to all imagery processed after 31 January

2002 and this improved the horizontal and vertical accuracy
by approximately 4 m. The authors believe that this exterior
orientation enhancement would change the results from this
study and the accuracies could be improved. None of the
stereo images evaluated for this study had either enhancement applied. Space Imaging could reprocess the existing
imagery with their new exterior orientation enhancement
and the images could be reevaluated. This is the only
methodology that would allow us to determine the veracity
of Space Imagings exterior orientation enhancement.
References
Baltsavias, E., Pateraki, M., & Zhang, L. (2001). Radiometric and geometric evaluation of IKONOS GEO images and their use for 3D building modeling. Proc. Joint ISPRS Workshop High Resolution Mapping
from Space 2001, Hannover, 19 21 September 2001. Pages on CDROM.
Davis, C. H., & Wang, W. (2001). Planimetric accuracy of IKONOS 1-m
panchromatic image products. Proc. ASPRS Annual Conference, St.
Louis, 23 27 April 2001 (p. 14), On CD-ROM.
Dial, G. (2000). IKONOS satellite mapping accuracy. Proc. ASPRS
Annual Conference, Washington, DC, 22 26 May 2000 (p. 8), On
CD-ROM.
Fraser, C. S., Hanley, H. B., & Yamakawa, T. (2001). Sub-metre geopositioning with IKONOS GEO imagery. Proc. Joint ISPRS Workshop High
Resolution Mapping from Space 2001, Hannover, 19 21 September
2001 (p. 8), On CD-ROM.
Geospatial Positioning Accuracy Standards Part 3: National Standard for
Spatial Data Accuracy, FGDC-STD-007.3-1998.
Gerlach, F. (2000). Characteristics of Space Imagings one-meter resolution
satellite imagery products. International Archives of Photogrammetry
and Remote Sensing, vol. 33 (Part B1) (pp. 128 135).
Hanley, H. B., & Fraser, C. S. (2001). Geopositioning accuracy of IKONOS imagery: Indications from 2D transformations. Photogrammetric
Record, 17(98), 317 329.
Strange, W. (2000 (October)). Do you really have WGS 84 coordinates?
Professional Surveyor, 20(9), 37 38.

Empirical comparison of Landsat 7 and IKONOS multispectral

measurements for selected Earth Observation System (EOS)
validation sites
Samuel N. Goward *, Paul E. Davis, David Fleming, Lee Miller, John R. Townshend
Department of Geography, University of Maryland, 2181 LeFrak Hall, College Park, MD 20782, USA
Received 21 June 2002; received in revised form 4 March 2003; accepted 2 July 2003
Abstract
The Space Imaging IKONOS observatory may provide an important benefit in terrestrial scientific research. The five-band, 1 m
panchromatic and 4 m multispectral measurements have the potential to provide a source of measurements to evaluate subpixel land cover
variability in measurements from observatories such as Landsat 7 Enhanced Thematic Mapper Plus (ETM+) and Terra MODIS sensor. The
IKONOS observations are at a spatial scale equivalent to field measurements typically carried out in ecological and land cover research. As
such, the IKONOS observations may serve as a source of virtual ground measurements, for the lower spatial resolution, global
observatories.
In this study we examine how well IKONOS sensor observations replicate Landsat 7 ETM+ visible/near infrared observations for selected
Earth Observation System (EOS) validation sites in the United States. The sites examinedBeltsville, MD, Konza Prairie, KS, and Sevilleta,
NMsample the east west moisture gradient across the United States. Observations for each site were acquired, as nearly time-coincident as
possible, from ETM+ and the IKONOS sensor, several times over the growing season. This was done to insure that we compared these two
sensors over the widest range of observing conditions possible.
We also examined IKONOS imagery from Phoenix, AZ, where Space Imaging had and had not applied a modulation transfer function
compensation (MTFC) process. The MTFC is their standard product. We found that this product, at the original 4 m spatial resolution,
appears to have minor radiometric artifacts as a result of the process. When the IKONOS observations were aggregated to 30 m, this problem
was essentially absent, allowing us to proceed with the remainder of our study.
We processed the IKONOS sensor and ETM+ measurements to produce close approximates of each other. Our processing steps included
ortho-rectification, calibration to planetary reflectance, pixel alignment and pixel aggregation. We initially found radiometric differences
between the two sensors that increased with increasing wavelength. Space Imaging updated their calibration information, based on analyses
from NASA Stennis Space Center staff, which removed much of this discrepancy. We now find that the IKONOS red and near infrared
measurements differ between the two sensors, with IKONOS generally producing higher reflectance in the red band and lower reflectance in
the near infrared band than the Landsat 7 ETM+ sensor. This results in the IKONOS sensor producing lower spectral vegetation index
measurements, for the same target, than ETM+, a measurement variation that has been observed between other sensors.
We also encountered far more cirrus cloud (and shadow) contamination in these paired observations that we had expected. After careful
initial selection, we lost over half of our image pairs from the analysis because of cirrus cloud contamination. We do not know whether this is
simply because of the paired, comparative design of this study or whether it relates to the increased spatial and radiometric resolution of the
IKONOS sensor.
The results of this study not only provide a baseline assessment of IKONOS versus Landsat 7 ETM+ visible and near infrared
measurements but also suggest some of the issues that need more attention when comparing other sensor systems as well as developing the
design of future land observatories.
Keywords: Landsat 7; IKONOS; ETM+

E-mail address: sgoward@umd.edu (S.N. Goward).
doi:10.1016/j.rse.2003.07.009
S.N. Goward et al. / Remote Sensing of Environment 88 (2003) 8099
1. Introduction
The value of multi-scale land imaging is well understood
(Colwell, 1960; Reeves, 1975). Examining land heterogeneity at spatial resolutions between a meter and a kilometer
reveals many of the artifacts that originate from natural
processes and human activities (Hall, Strebel, & Sellers,
1988; Miller, 1978; Woodcock & Strahler, 1987). For example, such multi-scale observations will permit relating regional and global trends in vegetation growth to local scale
processes such as fire and forest clearing by local human
populations (Kasischke, French, & Bourgeau-Chavez, 1999).
Such observations are now available from spacecraftdeployed, digital sensors. The recent innovation of high
spatial resolution observations from the Space Imaging
IKONOS1 sensor, combined with imagery from sensors such
as Moderate Resolution Imaging Spectrometer (MODIS) on
the NASA Terra platform and the Enhanced Thematic
Mapper Plus (ETM+) onboard Landsat 7, provides multiscale measurements from 1 m to 1 km.
Unlike previous satellite land observatories, which have
predominately been government funded, the recent high
spatial resolution systems, such as Space Imaging s IKONOS satellite and Digital Globes Quickbird, are commercially supported observatories. This provides a novel
opportunity for US federal agencies such as NASA, to explore the use of commercial vendors to support their scientific
goals. To explore this potential, NASAs Earth Science
Enterprise conducted a Scientific Data Purchase that included
IKONOS observations from Space Imaging. This study was
undertaken in support of this evaluation process. That is: do
the commercially supplied IKONOS sensor observations
compliment and compare favorably with current NASAfunded observatories such as the ETM+ on Landsat 7?
81
calibration, in order to convert each basic data set to

equivalent radiometric units. This study focuses on such a
comparison between the Landsat 7 ETM+ instrument and
the IKONOS sensor (Table 1).
We have taken the approach in this study of assuming that
we represent the average, reasonably well-educated, remote
sensing data user, who understands that they need to address
time coincidence, radiometric calibration and geo-registration but may not understand new image pre-processing
procedures, such as modulation transfer function (MTF)
compensation. Our primary application goal is to evaluate
whether and how these new observations compliment our
other multi-scale data sources. While pursuing this goal, our
team confronted questions, which we pursued at a technical
level in order to understand the differences we found. The
degree of our success can be viewed, to some degree, as an
ease-of-use metric.
We explored the following questions in this study:
How do the near-coincident Landsat ETM+ and

IKONOS multispectral measurements compare?
Can 4 m IKONOS sensor multispectral measurements
serve as virtual ground observations for 30 m ETM+
visible and near infrared measurement?
To address these questions, we attempted to acquire
paired, nearly time coincident, essentially cloud-free
IKONOS and ETM+ observations for selected locations in
the United States. We then conducted an empirical comparative analysis of the observations, to determine whether they
provided similar measurements of the landscapes under
evaluation when they were processed to approximate each
other.
2.1. The observatories
2. Approach
Over the last half century, terrestrial applications of
remote sensing have shifted from qualitative, visual analysis
(Colwell, 1960) to quantitative, spectro-radiometric evaluations (Swain & Davis, 1978). Derivation of similar quantitative spectro-radiometric measurements from two or more
different sensors can be difficult. When multiple sensors are
used to examine land characteristics, a number of uncertainties can be introduced into the analysis problem (Duggin
& Philipson, 1985; Pease & Pease, 1972; Williams, 1991).
Factors such as time of year, time of day, solar illumination
conditions, viewing angle, spectral band passes, and radiometric precision can and do vary between comparative
observations. Typically, instrument observations cannot be
compared at the simple digital number (DN) level but
require careful attention to instrument characterization and
1
Despite reviewer concerns, Space Imaging states that IKONOS rather
than Ikonos is the correct form.
The IKONOS sensor has several similar measurement

characteristics to the Landsat 7 ETM+. These include a
nominal descending orbit at 10 AM local solar equatorial
crossing time on the sunlit side of the Earth and multispectral bandwidths measuredat full width at half-maxthat
are similar to the four shortest wavelength bands of ETM+
(Table 1).
However, in several respects, the two observatories differ.
First, the IKONOS sensor provides a nearly factor of 10 finer
spatial resolution at 4 m (multispectral) versus 30 m for
ETM+. Second, the IKONOS sensor acquires 11-bit digiTable 1
Comparison of IKONOS and Landsat visible and NIR spectral band passes
Band
IKONOS spectral
range (nm)
Landsat spectral
range (nm)
(1)
(2)
(3)
(4)
445 516
506 595
632 698
757 853
450 515
525 605
630 690
775 900
Blue
Green
Red
NIR
82
Fig. 1. Study site locations relative to MODIS June 9 25, 2000 spectral vegetation index measurements (Normalized Difference Vegetation Index).
tized radiometry which is compressed for telemetry, versus

8 bits, in two alternate gains, direct telemetry from ETM+.
Third, the Landsat platform is in a fixed nadir ( F 7.5j scan
angles) viewing position, whereas the IKONOS platform
may be positioned over a wide range of viewing positions,
including looking quite close to the Earths horizon. The
NASA Scientific Data Purchase staff restricted IKONOS
acquisitions to view zenith angles of less than 30j, to limit
observation far-off nadir viewing variations. We assume that
such near-nadir IKONOS observations are comparable to
the Landsat near-nadir views.
At the start of this study, we knew much more about the

ETM+ instrument (http://landsat7.usgs.gov/) and Landsat
platform than we did about IKONOS and its sensor, in
particular, ETM+ sensor performance relative to ground
measurements. We therefore used the ETM+ observations
as our reference data set. Please note that comparing observations from any two sensors does not necessarily prove that
either sensor is providing an accurate assessment of
ground conditions. However, the analysis does provide a
relative comparison between the two sensors, an important
assessment if the IKONOS sensor was used to evaluate
Table 2
Scene pairs examined at primary and secondary study sitesa
Scene pair
Abbreviation
IK date
ETM+
date
Days
apart
IK
VAA
(j)
ETM+
VAA
(j)
IK
VZA
(j)
ETM+
VZA
(j)
IK
SAA
(j)
ETM+
SAA
(j)
IK
SZA
(j)
ETM+
SZA
(j)
Vegetation
type
Pair
accepted
Beltsville 1
Beltsville 2
Beltsville 3
Konza 1
Konza 2
Konza 3
Sevilleta 1
Sevilleta 2
Sevilleta 3
Sevilleta 4
Sevilleta 5
Phoenix 1
BEL1
BEL2
BEL3
KON1
KON2
KON3
SEV1
SEV2
SEV3
SEV4
SEV5
PHO1
06/03/00
09/29/00
09/10/00
05/11/00
07/16/00
09/15/00
05/12/00
07/23/00
09/18/00
10/02/00
12/04/00
10/12/99
06/12/00
10/02/00
10/02/00
06/07/00
07/09/00
09/11/00
05/09/00
07/28/00
09/14/00
09/30/00
12/03/00
10/10/99
9
4
23
27
7
4
3
5
4
3
1
3
339.59
333.39
264.03
68.51
151.25
268.25
104.28
258.94
76.70
19.32
137.72
27.82
98.2
98.2
98.2
98.2
98.2
98.2
278.2
278.2
278.2
278.2
278.2
98.2
22.46
15.96
25.97
24.68
19.69
28.18
20.04
27.32
23.46
15.00
14.67
29.28
0.13
0.13
0.13
1.13
1.13
1.13
2.27
2.27
2.27
2.27
2.27
1.80
130.1
157.2
153.7
129.5
124.5
156.2
120.7
124.2
143.4
152.4
159.8
148.5
123.1
152.0
152.0
124.5
122.5
144.9
124.9
119.0
141.6
148.0
157.4
150.7
66.99
46.02
53.03
61.08
63.25
51.46
63.56
67.45
51.57
48.33
30.76
44.06
65.39
43.43
43.43
65.30
63.81
50.14
63.83
62.74
52.23
47.38
29.74
45.83
Forest, crops
Forest, crops
Forest, crops
Grass, crops
Grass, crops
Grass, crops
Semi-desert
Semi-desert
Semi-desert
Semi-desert
Semi-desert
Urban
No
Yes
No
No
No
Yes
No
No
Yes
No
Yes
Yes
Scene pair includes both an IKONOS and Landsat image. Dates are for acquisition; time of day is always late morning. VAA is the sensor viewing
azimuth angle, VZA is the sensor viewing zenith angle, and SZA is the solar zenith angle. Azimuths are solar angles at time of acquisition; sensor azimuths not
listed here. Pair accepted indicates whether scene pair was used in final study (Yes) or was disqualified (No).
subpixel land cover conditions measured by the ETM+

instrument.
2.2. Study design
Our primary interest was in evaluating the comparative
performance of the multispectral measurements, specifically
with respect to land vegetation conditions. This follows a
long tradition in land satellite remote sensing, begun with
Landsat, which finds such multispectral measurements
uniquely suited to monitoring the status and quantity of
vegetation green foliage across the globe (Billings & Morris,
1951; Curran, 1981; Daughtry & Ranson, 1986; Goward &
Williams, 1997; Holben, Tucker, & Fan, 1980; Ives, 1939;
Jordan, 1969; Krinov, 1947; National Research Council,
1986; Sellers & Schimel, 1993; Smith, Ustin, Adams, &
Gillespie, 1990; Suits, 1972; Tucker, 1980). We therefore
83
focused our analysis goal on examining the comparative

performance of selected spectral vegetation indices and
developed the design of the study to insure that we covered
a reasonable, if not all encompassing, range of vegetation and
observing conditions.
Our study design primarily focused on three NASA
Earth Observing System (EOS) Land Validation sites
(http://www-eosdis.ornl.gov/eos_land_val/valid.html) distributed east west across the United States (Fig. 1). We
also later added an additional site, Phoenix, AZ, to evaluate
the MTF compensation Space Imaging applied to their
imagery. Use of the EOS sites, which included two US
National Science Foundation Long-Term Ecological Research (LTER) sites (Sevilleta, NM, and Konza Prairie,
KS), provided us access to considerable additional ground
information about these locations (Franklin, Bledsoe, &
Callahan, 1990) This access proved useful, resolving uncer-
Fig. 2. Scatter plots at (a) 30 m (b) 60 m, (c) 90 m and (d) 120 m of IKONOS versus ETM+ measurements from Konza Prairie site for the green spectral band.
84
tainties encountered in the analysis. For example, in the

IKONOS imagery, we noted high spatial frequency patterns
in the Sevilleta, NM landscape. In talking with the NSF
LTER staff, this was attributed to a particular kangaroo rat
that is resident in this desert landscape.
The east west distribution of the validation sites was
selected to follow the major precipitation gradient across the
United States. Annual precipitation declines from relatively
high levels in the eastern US, to quite low levels in the west.
This gradient produces a change in natural vegetation from
eastern forest cover in the Maryland site to grasslands at the
mid-continent Kansas location and desert scrub in the
western, New Mexico site. This change in vegetation cover
types co-varies with declining vegetation amounts (biomass,
leaf area index) from east to west (Fig. 1).
For each of the NASA EOS sites, we acquired image
pairs at multiple times through the growing season, from
May through December (Table 2). These paired observations from different times of the year were acquired to
permit evaluation of the impact of varying solar illumination
as well as vegetation conditions, both of which vary with the
seasons.
2.3. Image selection
Our goal was to acquire image pairs as nearly time
coincident as possible, constrained by the requirement that
each image needed to be essentially cloud-free in order for
the comparison to work. The 26j view angle constraint on
IKONOS reduces the IKONOS local observation repeat
frequency to, on average, 4 days, whereas the Landsat repeat
frequency is 16 days for the US. This suggests that there
should be a useable IKONOS/Landsat 7 image pair within 4
days of each other, at least for the United States, assuming
that Space Imaging acquires all possible cloud-free imagery. Unfortunately, this hypothetical, technical assessment
does not consider cloud contamination of actual Space
Imaging mission operations decisions. The details of the
mission operations were not available to the science users in
this NASA SDP program.
Because of the uncertainty of when acceptable IKONOS
observations would take place, we waited for the IKONOS
acquisition and then sought out the nearest time coincident
ETM+ observation. Most of our image pairs are within 4
days of each other (Table 2). Unfortunately, not all Landsat
acquisitions are cloud-free either, which led to nearly a
month observation time difference in two of the image
pairs.
Reduced resolution, browse imagery, provided by Space
Imaging and the US Geological Survey EROS Data
Center, was used for image selection. This browse imagery
limited assessment of cloud contamination, particularly
thin cirrus clouds. This limitation resulted in the later
rejection of 6 out of the 12 image pairs initially selected
for analysis (Table 2). A seventh image pair was additionally rejected because of phenological changes that had
occurred over the 27 days between the two acquisitions

(Konza 1) (Fig. 2).
3. Methods
Effective comparison of the images required several
processing steps noted in Table 3. These steps were carried
out to insure that the measurements from each sensor were
as closely comparable as possible.
3.1. Ortho-rectification
Examining the relative performance of IKONOS and
ETM+ requires, as much as possible, that individual compared pixels are for the same location on the Earths surface.
One of the largest potential sources of error is relief displacement caused by the two sensors viewing the same variable
elevation terrain from two different look angles (Bernstein,
1983; Slama, 1980). We found, particularly for the Konza
Prairie site, that relief displacement created significant registration problems. Note that, on average, the difference in view
zenith angle between the two sensors was approximately 25j
(Table 2). To address this problem, we subjected our primary
site observations to consistent ortho-rectification procedures.
Geographic reference points were extracted by locating
clearly observed points in the Landsat image, IKONOS
image, and US Geological Survey digital orthoquadrangles
(DOQs) (USGS, 1996). The US Geological Survey 1:24,000
DEM data, available for the United States, provided the
terrain information (USGS, 1993) and ortho-rectification
was carried out for both sensor data sets using the PCI
Geomatics OrthoEngineR IKONOS and Landsat 7 Models
(Lee & Bethel, 2001). Nearest neighbor re-sampling was
employed in the ortho-rectification process to preserve as
much as possible the radiometry of the imagery. Some
residual misalignment between Landsat and IKONOS is still
visibly evident in the resultant products, probably the result of
the differing spatial resolution of the IKONOS sensor and
Landsat 7 ETM+ instrument.
3.2. Spatial aggregation
Comparison of the observations radiometry required
aggregation of the IKONOS measurements to spatial resolution equivalent to the Landsat spectral measurements.
Numerous aggregation methods have been used in remote
Table 3
Processing methods
Process
Goal
Topographic ortho-rectification
Masking clouds and cloud shadow
Radiometric calibration
Pixel aggregation
Pixel-to-pixel registration
View of same target
Equivalent physical units
Measurement from same
ground area.
sensing, including averaging all values, sampling every nth

pixel and choosing the dominant value (Bian, 1997). In
this case, since we are attempting to approximate the
integrated reflected energy that the Landsat 7 ETM+ instrument measures, we employ the averaging approach.
The differing spatial resolutions of ETM+ and IKONOS
sensors also makes image alignment a potentially significant
problem since individual instantaneous field of views
(IFOVS) are not perfectly aligned in an x y domain between
the two sensors. Further, the image-processing software we
employed (including Imagine, PCI and ENVI) all treat pixel
locations in simple integer units, within which it is not
possible to define a common framework for the 4 and 30
m pixels produced by the IKONOS sensor and the ETM+
sensor, respectively. This led to use of subpixel sampling to
resolve the basic image alignment problem.
After experimenting with 1, 0.5 and 0.25 m subsampling,
we selected the 0.5 m IKONOS pixel subsample approach,
as an effective tradeoff between precision and execution
time. To align specific IKONOS pixels with their ETM+
equivalent, the top left corner of the 0.5 m subsampled
IKONOS pixel that best aligned with the top left corner of
the ETM+ 30 m pixel was identified through visual inspection. This pixel alignment served as the starting point for
aggregating the IKONOS observations to 30 m. The aggregated 30 m IKONOS images, when compared visually to
the ETM+ images, indicate slight residual misalignments of
not more than 0.25 m.
3.2.1. The MTF factor
Initially, one might assume that compilation of Landsatequivalent measurements from the IKONOS 4 m IFOV
pixels would simply require aggregation to 30 m. However,
sensor radiometry is not this simple (Schott, 1997). In a
typical sensor design, only about 50% of the spectral
radiance detected comes from the pixel dimension describing the sensor (e.g. 30 m for Landsat 7 ETM+ and 4 m for
IKONOS) or its modulation transfer function (MTF). However, the proportion of the radiance that originates from
outside the nominal IFOV also depends upon atmospheric clarity and adjacent target reflectance.
Aggregation produces a nearly perfect square-wave
30 m point spread function (PSF) for the aggregated
observations, so that IKONOS 4 m observations aggregated
to 30 m are not directly comparable to Landsat ETM+ 30 m
measurements. One approach that might be taken is to
either apply Landsat MTF characteristics to the IKONOS
measurements or attempt to process the Landsat observations through compensation of the Landsat MTF characteristics (Slater, 1980).
We selected to take the alternate, simpler approach of
aggregating both the IKONOS and ETM+ measurements to
coarser spatial resolutions (60, 90 and 120 m) to evaluate
whether there are factors beyond the MTF that might need
to be considered (Fig. 2). We would expect, for example,
that if MTF was the major unknown for the Landsat, then
85
the IKONOS and ETM+ observations would closely approximate each other at 60 m. Other factors such as variable
atmospheric attenuation and scene differences would come
into play beyond this aggregation level.
3.3. Cloud exclusion
Some scenes we selected for analysis had clearly visible
cloud and cloud shadow, typical when cumulus-type clouds
are present. Pixels from such targets, which are not shared
across the images pairs, confuse comparative analysis and
therefore needed to be excluded. We visually identified
these features in the images and produced masks to exclude
them from analysis. This task was harder than we had
anticipated, requiring reassessment several times before
we were satisfied with the results. This masking effort did
not extend to our discovery of considerable cirrus cloud
contamination in half of the scene pairs we had originally
selected for analysis. There was simply too much uncertainty in how we might approach this problem. We simply
rejected such image pairs from further analysis.
3.4. Radiometric calibration
Although digital remotely sensed data are widely distributed and analyzed in the form of digital numbers (DN), DNs
are not suitable for comparative analysis, particularly when
comparing the output from two sensors. The IKONOS/
ETM+ is a case in point. For the ETM+ sensor, the DNs
are the same values that were produced by the A-to-D
converter onboard the satellite. For IKONOS observations,
the DNs made available to the public are produced as the
result of internal Space Imaging pre-processing. Space Imaging re-maps the original instrument DN values so that the
calibration-offset term is removed from the consumer-supplied values. The ETM+ DN values are not comparable to the
IKONOS DN values. In general, the use of DNs for comparative analysis purposes is not valid and can produce to
seriously misleading conclusions (Goward, Dye, Turner, &
Yang, 1993). Conversion to calibrated physical units is
required.
3.4.1. Spectral radiance
Conversion to spectral radiance is a substantial improvement over use of DNs in comparative analysis. When transformed, all individual sensor measurements are in comparable physical units. This is generally accomplished through
information supplied by the instrument developer in the form:
Lk gain DN offset
1
2
where Lk = sensor-measured spectral radiance W/m /sr/Am),

gain = change in radiance as a function of digital number,
and offset = the DN value where zero radiance is detected
(Table 4).
Information for conversion of ETM+ sensor DN values
spectral radiance is found in the Landsat Data Users
86
Table 4
ETM+ spectral radiance range (W/m2-sr-Am)
Band
number
Low gain
High gain
LMIN
LMAX
LMIN
LMAX
1
2
3
4
6.2
6.4
5.0
5.1
293.7
300.9
234.4
241.1
6.2
6.4
5.0
5.1
191.6
196.5
152.9
157.4
From Landsat 7 Science Data Users Handbook.
Handbook (LPSO, 1998). Space Imaging supplies equivalent information for the IKONOS sensor (Peterson, Gerlach,
& Hutchins, 2001).
There are limitations with use of spectral radiance measurements for comparative purposes. If two sensors do not
have identical sensor spectral band passes, the sensors will

not provide comparable spectral radiance values (Fig. 3).
This problem of differing spectral band passes is definitely
true for ETM+ versus IKONOS and it is generally true, even
within the same series of sensors, such as the Thematic
Mapper on Landsat (Slater, 1979) and the Advanced Very
High Resolution Radiometer (AVHRR) sensors aboard the
National Oceanic and Atmospheric Administrations
(NOAA) polar orbiting satellites (Trishchenko, Cihlar, &
Li, 2002). Second, for observations taken at differing times
of day, differing times of the year, or differing latitudes, the
radiance measurements will inversely vary with solar zenith
angle. Again, it is difficult to compare these measurements
because they vary as a function of both ground conditions
and solar illumination conditions.
Fig. 3. Relative spectral response functions for Landsat 7 ETM+ and IKONOS (a) blue spectral band, (b) green spectral band, (c) red spectral band and (d) near
infrared spectral band. Note the differences particularly for the red and near infrared bands.
3.4.2. Planetary spectral reflectance

Planetary reflectance is more easily compared between
sensors, as well as to ground measurements, than spectral
radiance because it adjusts for incident solar irradiance.
Planetary reflectance is the ratio of sensor-measured spectral
radiance from the Earth divided by spectral radiance incident at the sensor altitude (Eq. (5)) (LPSO, 1998).
qpk
pLk d 2
ESUNk coshs
87
Table 6
IKONOS bandwidth characteristics ( 016, 1999)
Band
Pan
MS-1
MS-2
MS-3
MS-4
(Blue)
(Green)
(Red)
(VNIR)
Lower
50% (nm)
Upper
50% (nm)
Bandwidth
(nm)
Center
(nm)
525.8
444.7
506.4
631.9
757.3
928.5
516.0
595.0
697.7
852.7
403
71.3
88.6
65.8
95.4
727.1
480.3
550.7
664.8
805.0
where qpk = planetary reflectance (%), Lk = spectral radiance

at sensors aperture, either Landsat or IKONOS (W/m2-Amsr), ESUNk = band dependent mean solar exo-atmospheric
irradiance (W/m2-Am), hs = solar zenith angle (radians),
d = earth sun distance (astronomical units) (Tables 5 and 6).
The exo-atmospheric spectral radiance for Landsat
(ESUN) is provided in the Landsat Data Users Handbook
(LPSO, 1998) and is based on the MODTRAN-3 solar
spectrum. For IKONOS, the MODTRAN-3 solar spectrum
was generated using SBDART (Ricchiazzi et al., 1998) and
used to compute ESUN values using the spectral filter
response function of the IKONOS system (Peterson et al.,
2001) (Table 7).
Planetary reflectance, as a top of the atmosphere measurement, includes the impacts of the atmosphere and thus
will vary with atmospheric conditions even when nothing
changes on the land surface observed.
3.5. Spectral vegetation indices
Comparison of Landsat ETM+ and IKONOS sensor
observations on the per-spectral-band basis provides a firstorder assessment of their relative measurement performance.
However, further insights into the relative measurement
performance are revealed through inspection of multi-band
combinations such as spectral vegetation indices (SVI).
There have been a multitude of transformations proposed
Table 5
IKONOS original, pre and post CalCoefk
Spectral
band
Originala
CalCoefk,
DN [mW/
cm2-sr] 1
Pre 2/22/01b
CalCoefk,
DN [mW/
cm2-sr] 1
Post 2/22/01b
CalCoefk,
DN [mW/
cm2-sr] 1
Full-scale
dynamic
range (mW/
cm2-sr)
MS-1
(Blue)
MS-2
(Green)
MS-3
(Red)
MS-4
(VNIR)
637
633
728
2.98
573
649
727
3.32
663
840
949
2.87
503
746
843
3.75
(3) The pre and post CalCoefk values are to be used respectively for images
that have image production (creation) dates before and after February 22,
2001. The original CalCoefk values for the blue, green, red, and visible near
infrared are no longer used for calibration.
a
Space Imaging Document Number SE-REF-016, N/C.
b
Space Imaging Document Number SE-REF-016, Rev. A.
for visible-near infrared spectral measurements for monitoring of vegetation (Deering, Rouse, Haas, & Schell, 1975;
Huete, 1988; Jackson, 1983; Kaufman & Tanre, 1992; Kauth
& Thomas, 1976; Wiegand & Richardson, 1984). The purpose of these indices is to compensate for variable background (e.g. soil and litter) reflectance and some forms of
atmospheric attenuation, while emphasizing vegetation spectral features (Trishchenko et al., 2002).
In this study, we examined three alternative SVIs that
combine the spectral measurements in differing algebraic
forms:
1. The simple ratio (SR, Eq. (6)) originally used by Jordan
(1969) to examine the ratio of light at 800 Am to that at 675
Am to measure leaf area index underneath forest canopies.
SR
qnir
qvis
where qnir = near infrared reflectance and qvis = visible

reflectance.
2. Normalized Difference Vegetation Index (NDVI, Eq. (7))
first used by Deering et al. (1975) to characterize grazing
pastures in the western US
NDVI
qnir qvis
qnir qvis
3. Atmospheric Resistant Vegetation Index, ARVI, was

originally devised for use with the MODIS sensor, but
can be used with other sensors. ARVI is resistant to
atmospheric effects (in comparison to the NDVI) and is
accomplished by a self-correcting process for the
atmospheric effect in the red channel, using the difference in the radiance between the blue and the red
channels (Kaufman & Tanre, 1992)
ARVI
qnir 2qred qblue

qnir 2qred qblue
where qblue = blue reflectance.

Table 7
Landsat 7 and IKONOS ESUNk values (W/m2-Am)
Band
Landsat 7
IKONOS
Blue
Green
Red
NIR
1970
1843
1555
1047
1939
1847
1536
1148
88
3.6. Data analysis

We focused on descriptive statistics to evaluate the comparative performance of the sensors. These methods include
histograms, scatter plots, explained variance (r2), and a twodirectional estimated line fit. The latter method was employed
because the assumptions of the standard ordinary, one-direction, least square method (OLS) are not met when comparing
images. Biases in these OLS line fits are due to measurement
error in both data sets. Press, Teukolsky, Vetterling, and
Flannery (1992) present a solution to this problem and note
that fitting a straight line model to data that are subject to error
in both coordinates is considerably more difficult. The
solution used for the scene comparisons was derived from
algorithms supplied by Peltzer (2000) (http://www.mbari.
org/~etp3/regressindex.htm) that are based on original derivation by Pearson (1901). This solution provides the major
axis as the best fit line that is also the first component in
principal component analysis.
In addition, we computed image difference and image
ratio maps to evaluate whether any regional biases where
present in the observations (e.g. atmospheric variability). Our
expectation, if the two sensors perform identically, is that
there should be a nearly perfect relation observed between the

two observation sets. Any deviations observed would then be
subject to further analysis in an effort to determine why they
differ.
4. Results and discussion

Our analysis initially diverted to comparing an IKONOS
scene where MTF compensation had been applied versus the
same scene where it had not been applied. Our major
conclusion from this analysis was that the MTFC impact on
science uses was modest although not negligible. We concluded that we should proceed with our proposed study using
the MTFC scenes that Space Imaging would normally supply
to their customers.
4.1. MTF compensation study
Early in this study, researchers at NASA Stennis Space
Center discovered that the IKONOS data product Space
Imaging was supplying to the NASA Scientific Data Purchase had been subjected to a modulation transfer function
Fig. 4. Scatter plots of blue (a), green (b), red (c) and near infrared (d) planetary reflectance, from 4 m measurements with MTFC-on versus MTFC-off for the
October 12, 1999 scene. Note the reflectance bias introduced by brightening the bright pixels and darkening the dark pixels.
89
Fig. 5. Scatter plots of NDVI (a), SR (b) and ARVI (c) spectral vegetation indices from 4 m measurements with MTFC-on versus MTFC-off for the October 12,
1999 scene. Note the outlier measurements in the MTFC-on data set.
compensation process (Ryan et al., 2003). Since the potential

impact of this processing step on scientific applications was
relatively unknown, we undertook a brief assessment of a
scene (Phoenix, AZ) where MTFC had been applied versus
the same scene where MTFC had not been applied.
Comparison of the Phoenix imagery, with and without
MTF compensation, revealed that in general the MTFC
image visually appears sharper, as expected. This image
sharpening also alters the basic structure of the multispectral
measurements, for all spectral bands, darkening dark pixels
and brightening light pixels (Fig. 4). Since spectral vegetation indices contrast the visible and near infrared measurements, the MTF compensation amplifies the SVI measurements considerably (Fig. 5). In the NDVI measurements
(Fig. 5a), the low, negative values decrease from 0.2 to, on
average, 0.5 and the large positive values increase from
0.5 to 0.7. Note the outliers also produced. The simple ratio
measurements (Fig. 5b) show similar departures but with the
largest deviations at high values, where values typically
increase from 4.0 to >6.0, again with outliers above this.
The ARVI impacts (Fig. 5c) are larger still, causing outliers
in excess of 4 (Table 8).
The occurrence of significant SVI outliers (i.e. values far
in excess of expected values) in the MTF compensated
measurements suggests that the MTFC approach used by
Space Imaging will alter the results of certain vegetative
Table 8
Blue/red band reflectances and ratios for dark targets
Location
Target
IKONOS
Landsat
Blue Red Ratio Blue Red Ratio

Beltsville, MD
Konza, KS
Phoenix, AZ
Forest
8.2 4.2 1.95
Forest
9.1 5.9 1.54
Dark
12.9 11.3 1.14
vegetation
Sevilleta, NM (Sept.) Shadow
9.5 5.2 1.83
Sevilleta, NM (Dec.) Shadow
10.0 4.4 2.27
9.0 4.1
9.8 5.6
11.8 9.6
2.20
1.75
1.23
8.3 4.5
9.1 4.3
1.84
2.11
indices. This means that scientists should be careful to

examine and exclude image pixels where over-enhancement has occurred. Without an aggressive study, probably
including field measurements, it would be difficult to determine precisely where to truncate the observations. Nevertheless, some effort should be made to exclude such data
values prior to any 4 m pixel level analysis with the IKONOS
MTFC measurements.
For this study, the question of whether MTFC might
impact our IKONOS/ETM+ comparison was also of concern. We therefore compiled aggregated 30 m observations
for MTFC applied and not applied data. Our conclusion is
that there are small random differences of F 1% planetary
reflectance between these two data sets, in each of the multispectral bands (Fig. 6). This translates to about F 0.05
NDVI and similar relative values for the SR and ARVI (Fig.
7). This size of error is above the basic noise level of these
sensors but we do not anticipate that this level of uncertainty
is an important factor in this current study.
4.2. IKONOS comparison with Landsat 7
4.2.1. Constraints on comparison
Initially, we were concerned that bi-directional reflectance
might be a factor producing differences in the observations
from the IKONOS sensor and the ETM+ (Walthall, Norman,
Welles, Campbell, & Blad, 1985). Whereas the Landsat 7
ETM+ sensor has a fixed viewing geometry, scanning
F 7.5j in view zenith from nadir and in view azimuth to
the east at 98j and to the west 278j, the IKONOS sensor
may be positioned over a wide range of view directions. The
view zenith angle for IKONOS observations examined in
this study varied between 15j and 30j and the view azimuth
between 20j and 340j. In addition, the view direction varied
between 19j and 259j (Table 2). This raised the possibility
that BDR variations might cause IKONOS and ETM+ to
observe differing scene reflected radiance. Our inspection of
90
Fig. 6. Scatter plots of blue (a), green (b), red (c) and near infrared (d) planetary reflectance measurements from aggregated 30 m measurements with MTFC-on
versus MTFC-off for the October 12, 1999 scene. Note that minor residual differences are still present at this nearly factor of 10 averaging.
the viewing geometries revealed that with the BEL2, KON3

and PHO1 pairs, both sensors look in the forward scatter
direction, whereas in the SEV3 and SEV5 pairs, both sensors
look toward the back-scatter direction. However, only in the
case of BEL2 does the IKONOS view near the principal
plane, in this case within 4j of the forward scatter direction

from the sun, whereas the paired Landsat scene is more than
40j away from the backscatter direction. In all other cases,
both sensors are 30j or more away from the principal plane.
We therefore did not expect to see substantial BDR effects in
Fig. 7. Scatter plots of NDVI (a), SR (b) and ARVI (c) spectral vegetation indices from the aggregated 30 m measurements with MTFC-on versus MTFC-off
for the October 12, 1999 scene.
these image pairs. Paired imagery, others might examine,

may still encounter BDR effects. We simply, by luck, did not
acquire such a pair in this case.
Our initial study design had identified 12 image pairs,
across the three US EOS validation sites, to pursue our analysis. Although we exercised great care in selecting specific
images for analysis, we have been quite surprised to find,
over time, that we have had to eliminate seven pairs from
consideration. Of the seven pairs eliminated, one (May June
2000, Konza Prairie, KS) was removed because a major
spring green-up occurred between the May 11 IKONOS
observation and the June 7 Landsat ETM+ acquisition. The
91
two images recorded such differing vegetation conditions that

it is not possible to derive a simple relation between the two
scenes.
The remaining six pairs were removed from consideration due to sub-visible cloud and cloud shadow contamination. We did not detect this contamination in the
browse imagery provided by Space Imaging and the USGS
EROS Data Center, nor did we see this contamination in our
quick visual comparison of the images when we first
received them. Only after intensive comparative analysis,
in which we could not explain our results, did we return to
visual inspection of the imagery. Only by flickering be-
Fig. 8. Sevilleta, NM May 2000 scene pair revealing cirrus cloud presence in the IKONOS image. Obvious cloud contamination is present in the upper left corner
of the IKONOS image (a) versus no cloud contamination in the ETM+ image (b). A ratio of the IKONOS scene/ETM+ scene (c) reveals further cloud (cirrus) and
shadow contamination in the IKONOS scene stretching diagonally across the center of the IKONOS image. The lower right corner is mountain and its shadows.
92
Fig. 9. Comparison of IKONOS and ETM+ 120 m aggregations for five study images (Table 2) (a) blue band, (b) green band, (c) red band and (d) near infrared band. Note that the Sevilleta results deviate from the
other sites. Why this is the case is not certain.
tween the two images were we able to detect the subtle

contamination produced by cirrus clouds and their shadows
(Fig. 8). We found similar problems in more than 50% of
the image pairs we had selected for analysis. There is
clearly a need for a much more refined cloud detection
methodology than we used in this study. Hopefully, our
experience will act as a caution for other remote sensing
researchers to look very carefully at their imagery for cirrus
contamination.
4.2.2. Early calibration results
Our early analysis, carried out during 2000, employed the
calibration information initially supplied by Space Imaging
(Peterson, Gerlach, & Hutchins, 2000). Our results indicated
that the initial calibration information produced a radiance
and reflectance bias for IKONOS measurements. The bias,
compared to Landsat 7 ETM+ measurements, was smallest at
the blue wavelength (band 1) and largest for the near infrared
wavelength (band 4). In all cases, the IKONOS calibration
information produced radiance measurements that were
greater than the equivalent Landsat spectral radiance measurements. These findings were consistent with the results of
other studies presented at the High Spatial Resolution
Commercial Imagery Workshop held in March of 2001 in
Greenbelt, Maryland. These findings along with Space
Imagings ongoing IKONOS calibration efforts led to a
revision of the calibration coefficients in May 2001 (Peterson
et al., 2001). We employed the revised calibration information in our analysis.
4.2.3. Comparative spectral radiometry
In general, the observations from Beltsville, MD, Konza
Prairie, KS, and Phoenix, AZ produce equivalent results
(Fig. 9). The blue and the green band measurements are
nearly identical between the two sensors. In the red band
IKONOS records a slightly higher reflectance (f 2%) than
Landsat and in the near infrared IKONOS records a reflectance f 5% lower than Landsat ETM+. The observations
from Sevilleta, NM deviate from the other three sites and
between each other. In all bands, the Sevilleta measurements
from IKONOS tend to be higher than Landsat measurements
and the difference increases with increasing reflectance. The
December Sevilleta measurements record larger differences
from the ETM+ measurements than the September Sevilleta
measurements.
4.2.3.1. Spectral band passes. The observed differences in
red and near infrared measurements appear to originate from
differences in the relative spectral response (RSR) functions
of the two sensors (Fig. 3). Note in particular that the red and
near infrared RSR features of the IKONOS sensor more
strongly overlap the so-called vegetation red edge at
approximately 720 nm, where pigment absorption ceases
and is replaced, at longer wavelengths by cellular scatter at
longer wavelengths (Wooley, 1971). This should cause the
IKONOS to record higher red reflectance and lower near
93
infrared reflectance than those of the equivalent spectral

measurements from the ETM+ instrument.
We evaluated the impact of the differing RSR functions of
the two sensors by combining them with the TOA incident
solar spectrum and the spectral reflectance of typical land
materials including green vegetation, to produce the same
outcome as we observe in our comparative analysis of
IKONOS and ETM+ measurements (Fig. 10). Note that
based on the differing sensor RSR functions that the IKONOS tends to record higher red reflectance and lower near
infrared reflectance than ETM+, particularly for green vegetation spectra.
4.2.3.2. Atmospheric attenuation. The deviations in the
observed relations between scene pairs, in particular for the
Sevilleta, NM measurements, may originate from variable
atmospheric conditions. For selected visible wavelength
dark targets, such as shadows and forest canopies, we
extracted sample blue and red wavelength reflectance measurements. Following the principles behind the atmospherically resistant vegetation index (Kaufman & Tanre, 1992),
we computed the blue/red ratio observed for these dark
targets in each scene examined. Basically, the assumption
is that at the Earths surface such targets will have nearly
equal, low reflectances, typically in the 2 4% range. As
atmospheric scatter increases, the scattered light added to the
surface-leaving signal will preferentially increase in the blue
wavelengths.
The computed blue/red (BR) ratio shows that the Beltsville, MD and Sevilleta, NM December scenes record the
highest atmospheric scatter (Fig. 11, Table 8). The Phoenix,
AZ observations are the clearest. However, it is also important to note that for the Phoenix, AZ, Konza, KS and
Beltsville, MD, the Landsat scene has a higher BR ratio
than the IKONOS scene. For Sevilleta, NM in September,
they are nearly equal and for December, the IKONOS scene
records a higher ratio.
Atmospheric variability is one possible explanation but
there are others. More analysis will be needed to fully
understand why paired comparison varies between the
Sevilleta and other image pairs. For example, it is not clear
why the offset between Sevilleta December and the other
scenes is approximately the same for all four of the
multispectral wavelengths. If this were strictly an atmospheric scattering variation, one would expect the difference
to be larger at shorter wavelengths. We initially considered
possible BRDF effects, but as noted previously, we do not
believe that this factor is important in these particular scene
pairs.
4.2.4. Spectral vegetation indices
Confirmation that the shifts seen in the individual spectral
bands between Sevilleta and the other scene pairs are consistent across all wavelengths is found in the spectral vegetation indices results (Fig. 12). For all the scenes compared
the results are consistent, showing an offset of + 0.1 for
94
95
Fig. 11. Comparative blue/red spectral band ratios for five image pairs examined in this study. Note that the band ratios vary between 1.2 and 2.3, suggesting
wide variations in atmospheric attenuation between scene pair. However, for any given scene pair, the atmospheric conditions are more closely related.
NDVI measurements from ETM+ versus NDVI measurements from IKONOS. This is the equivalent of f 12%
ground cover (or absorbed PAR) (Goward & Huemmrich,
1992; Huemmrich & Goward, 1992). Similarly, The simple
ratio results indicate that IKONOS proportionately underestimates Landsat measurements by f 28%, approximately
the same as underestimating leaf area index (LAI) by the
same proportion (e.g. Landsat LAI of 3.0 would be 2.2 in the
IKONOS measurements). The ARVI results are a bit more
complicated because of the blue band inclusion. However,
they, in general, confirm the NDVI results and further suggest
problems in the December Sevilleta image pair.
These results confirm that the spectral band passes of
the two sensors differ (Fig. 3). Basically, the IKONOS
measures lower SVI measurements for the same land
surface conditions than ETM+. One interesting conclusion
is that despite the differences noted in the individual
spectral band measurements, particularly between sites,
the spectral vegetation index calculations show considerable consistency between all the scene pairs, across all
sites. This may be important in ultimately explaining the
potential source of the observed variations noted within
specific spectral wavelengths (Fig. 9). For example, perhaps some calibration variation over time has not been yet
identified.

Overall, this study has found that the Space Imaging
IKONOS sensor produces spectro-radiometric measurements that are similar in value to Landsat ETM+ imagery.
There are, however, several issues relative to the IKONOS
imagery that are worth considering when used for scientific
comparison to ETM+ measurements:
1. The MTF compensation, at least as originally applied by
Space Imaging to most IKONOS imagery, appears to
have an impact on the radiometrically calibrated measurements. Caution should be exercised in numerical applications of these MTFC measurements with respect to
presence of anomalous, outlier values. These outlier
values, for example, produce erroneous spectral vegetation index measurements.
2. The individual spectral band measurements vary between
IKONOS and Landsat 7 ETM+ particularly in the red
and near infrared spectral regions. These differences
appear to originate primarily from differences in the RSR
functions of the two sensors.
3. In this study we found substantial offsets in the spectral
band measurements from the Sevilleta study site relative
to the other study locations. We do not understand why
Fig. 10. Scatter plots of modeled IKONOS and ETM+ measurements for the (a) blue, (b) green, (c) red, and (d) near infrared spectral regions based on the
relevant sensor, solar and ground information. These results confirm the proposed impact of the IKONOS relative spectral response functions versus the
Landsat 7 ETM+ RSR functions.
96
97
Fig. 12. Scatter plots of NDVI (a), SR (b) and ARVI (c) of IKONOS and ETM+ aggregated to 120 m for five study sites. Note that in all cases the IKONOS
measurements are lower than the same Landsat 7 ETM+ measurement, but that they vary as a function of the SVI used.
this occurs. It does however suggest that there are

additional differences between the IKONOS and ETM+
measurements that are still not understood.
4. Despite the individual spectral band variations noted,
the results from the spectral vegetation indices analysis
indicate that the difference between the sensors is
consistent from site to site. Overall, the IKONOS
sensor measurements underestimate spectral vegetation
indices relative to Landsat 7 ETM+ sensor. These
differences appear to result from differences in the
relative spectral response functions of these two
systems.
5. Although in the NASA SDP acceptable view zenith angle
was restricted to < F 30j, we found that relative relief
displacement between the two sensors is a significant
problem for geo-registration. Despite the time and effort
we dedicated to this problem, our results were of limited
success. Pixel-to-pixel geo-registration between the two
observations sets still contained undesirable, residual
error.
An additional conclusion, which comes as a surprise to
this research team, is the magnitude of the cloud contamination we missed initially in both image data sets. Perhaps
the higher spatial resolution or the improved radiometry of

the IKONOS sensor makes this problem more obvious.
There is clearly more work needed to identify cloud contamination in terrestrial observations. Whether this requires
better cirrus cloud detection or more comprehensive data
analytical approaches is not certain. For example, the cirrus
band on MODIS, which is also specified for the Landsat
Data Continuity Mission (LDCM), may be vital to the
development of more automated approaches to exploiting
regional and global remote sensing systems. We also believe
that variable atmospheric scatter is an important unknown in
comparative analysis, particularly within specific individual
spectral measurements.
Analysts interested in adding IKONOS, or other new
sensor systems, to their suite of multi-scale, multispectral
observing systems need to exercise caution. Pay careful
attention to the comparative differences between sensor
systems and be exceptionally careful in the way that both
data sets are treated. For example, be sure to convert to
planetary reflectance prior to combined analysis, using the
most recent radiometric calibration information available for
each sensor.
Successful use of remotely sensed measurements from
more than one sensor (or even the same sensor over time)
98
requires paying careful attention to numerous details. These

include:

Interactions between sensor viewing geometry and local

terrain.
Applications of post-acquisition image enhancements,
such as MTF compensation.
Comparability of radiometric and spectral calibration
approaches.
Comparative performance of relative spectral response
functions for differing sensors.
Presence of variable atmospheric scattering and absorption and cloud contamination in measurements.
This comparative analysis of Space Imaging IKONOS

measurements and Landsat 7 ETM+ measurements has
revealed many underlying, critical questions that need to
be addressed in future Earth observing remote sensing
missions. In most cases, we already understand how these
factors can be addressed and compared across variable
sensor systems. In those cases where we are less certain,
for example, in cloud cover contamination, major research
initiatives should be undertaken to resolve the uncertainties.
One of the real lessons learned is that, for broad-band
multispectral sensors, it is exceptionally difficult if not
impossible to design systems in which the spectral measurements are identical between sensors. This is simply a
problem of detector and filter manufacturing, producing
identical systems is nearly impossible (Slater, 1980). Given
the large number of visible/near infrared multispectral sensors that been used for earth observations over the last three
decades, we need to pay more attention on how to compare
measurements between such sensors (Teillet, Staenz, &
Williams, 1997; Trishchenko et al., 2002).
As we move toward future observatories, finding solutions to the problems noted in this study should serve as
major drivers for the design of the observatories.
Acknowledgements
This study was carried out with support from the NASA
Stennis Space Center, Earth Science Applications Directorate under NASA Grant NAG 1398001. Various members of
the NASA Stennis Staff, including Vicki Zanoni, Tom
Stanley, Bob Ryan and Mary Pagnutti have provided major
assistance in completing the study. We would also like to
acknowledge the value reviews we received from the
external reviewers. Their patience with our draft manuscript
made this a much clearer and better paper upon revision.
Thanks for your help.
References
Bernstein, R. (1983). Image geometry and rectification. In R. N. Colwell
(Ed.), Manual of remote sensing ( pp. 873 922). Falls Church, VA:
American Society of Photogrammetry.
Bian, L. (Ed.) (1997). Multiscale nature of spatial data in scaling up

environmental models. CRC/Lewis Publishers, Boca Raton, FL.
Billings, W. D., & Morris, R. J. (1951). Reflection of visible and infrared
radiation from leaves of different ecological groups. American Journal
of Botany, 38, 327 331.
Colwell, R. N. (Ed.) (1960). Manual of photographic interpretation (p. 868).
Washington, DC: Amer. Soc. Photogrammetry.
Curran, P. J. (1981). Multispectral remote sensing for estimating biomass
and productivity. In H. Smith (Ed.), Plants in the daylight spectrum
( pp. 65 99). London: Academic Press.
Daughtry, C. S. T., & Ranson, K. J. (1986). Measuring and modeling
biophysical and optical properties of diverse vegetative canopies,
Rep. No. 043086. Laboratory for Applications of Remote Sensing, Purdue University, West Lafayette, IN.
Deering, D. W., Rouse, J. J. W., Haas, R. H., & Schell, J. S. (1975).
Measuring forage production of grazing units from Landsat MSS data.
Proceedings 10th International Symposium Remote Sensing Environment ( pp. 1169 1178). Ann Arbor, MI: Environmental Research Institute of Michigan.
Duggin, M. J., & Philipson, W. R. (1985). Relating ground, aircraft and
satellite radiance measurements: Spectral and spatial considerations.
Remote Sensing Letters, 6, 1665 1670.
Franklin, J. F., Bledsoe, C. S., & Callahan, J. T. (1990). Contributions
of the Long-Term Ecological Research Program. Bioscience, 40(7),
509 523.
Goward, S. N., Dye, D. G., Turner, S., & Yang, J. (1993). Objective
assessment of the NOAA Global Vegetation Index data product. International Journal of Remote Sensing, 14(18), 3365 3394.
Goward, S. N., & Huemmrich, K. F. (1992). Vegetation canopy PAR absorptance and the normalized difference vegetation index: An assessment
using the SAIL model. Remote Sensing of Environment, 39, 119 140.
Goward, S. N., & Williams, D. L. (1997). Landsat and earth systems
science: Development of terrestrial monitoring. Photogrammetric Engineering & Remote Sensing, 63(7), 887 900.
Hall, F. G., Strebel, D. E., & Sellers, P. J. (1988). Linking knowledge
among spatial and temporal scales: Vegetation, atmosphere, climate
and remote sensing. Landscape Ecology, 2(1), 3 22.
Holben, B. N., Tucker, C. J., & Fan, C. J. (1980). Assessing soybean leaf
area and leaf biomass with spectral data. Photogrammetric Engineering,
46, 651 656.
Huemmrich, K. F., & Goward, S. N. (1992). Spectral vegetation indices
and the remote sensing of biophysical parameters. 1992 International
Geoscience and Remote Sensing Symposium ( pp. 1017 1019). Arlington, TX: Insititute of Electrical and Electronic Engineers.
Huete, A. R. (1988). A soil-adjusted vegetation index (SAVI). Remote
Sensing of Environment, 25, 295 309.
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. Remote Sensing of
Jordan, C. F. (1969). Derivation of leaf area index from quality of light on
the forest floor. Ecology, 50(4), 663 666.
Kasischke, E. S., French, N. H. F., & Bourgeau-Chavez, L. L. (1999). Using
satellite data to monitor fire-related processes in boreal forests. In E. S.
Kasischke, & B. J. Stocks (Eds.), Fire, climate change and carbon
cycling in the boreal forest ( pp. 406 422). New York: Springer-Verlag.
Kaufman, Y., & Tanre, D. (1992). Atmospherically Resistant Vegetation
Index (ARVI). Institute for Electrical and Electronics Engineers Transactions on Geosciences and Remote Sensing, 30(2), 261 270.
Kauth, R. J., & Thomas, G. S. (1976). The tasseled capa graphic description of the spectral temporal development of agricultural crops as seen
by Landsat. 10th Symposium on Machine Processing of Remotely Sensed
Data ( pp. 41 51). West Lafayette, IN: Purdue University.
Krinov, E. L. (1947). Spectral Reflectance of Natural Formations (Transl.
by NEC of Canada, T1439, G. Belkov). Akad. Nank, USSR, Laboratorica Aerometodov, Moscow.
Lee, C., & Bethel, J. (2001). Georegistration of airborne hyperspectral

image data. IEEE Transactions on Geoscience and Remote Sensing,
39(7), 1347 1351.
LPSO (1998). Landsat-7 science data users handbook. Greenbelt, MD:
Landsat Project Science Office.
Miller, D. H. (1978). The factor of scale: Ecosystem, landscape mosaic and
region. In K. A. Hammond, G. Macinko, & W. B. Fairchild (Eds.),
Sourcebook on the environment: A guide to the liturature ( pp. 63 88).
Chicago: The University of Chicago Press.
National Research Council (1986). Remote sensing of the biosphere.
Pearson, K. (1901). On lines and planes of closest fit to systems of points in
space. Philosophical Magazine, 6(2), 559 572.
Pease, S. R., & Pease, R. W. (1972). Photographic films as remote sensors
for measuring albedos of terrestrial surfaces, Rep. No. 14-08-000111914. California State at Riverside, California.
Peltzer, E. T. (2000). Matlab shell-scripts for linear regression analysis. Model II Regressions, vol. 2002. http://www.mbari.org/~etp3/
regressindex.htm.
Peterson, B., Gerlach, F., & Hutchins, K. (2000). IKONOS relative spectral
response and radiometric calibration coefficients, Rep. No. SE-REF016. Space Imaging LLC, Thorton, CO.
Peterson, B., Gerlach, F., & Hutchins, K. (2001). IKONOS relative spectral
response and radiometric calibration coefficients. Thornton, CO: Space
Imaging.
Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. (1992).
Numerical recipes in fortran 77, the art of scientific computing (2nd
ed.). Cambridge: Cambridge Univ. Press.
Reeves, R. G. (Ed.) (1975). Manual of remote sensing (1st ed.) (p. 2123).
Falls Church, VA: American Society of Photogrammetry.
Ricchiazzi, P., Yang, S., et al (1998). SBDART: A research and teaching
software tool for plane-parallel radioactive transfer in the Earths atmosphere. Bulletin of the American Meteorological Society, 79(10),
2101 2114.
S. (2003). IKONOS spatial resolution and image interpretability. Remote Sensing of Environment 88, 37 51 (this issue).
Schott, J. R. (1997). Remote sensing: The image chain approach (1st ed.).
New York: Oxford Univ. Press.
Sellers, P. J., & Schimel, D. S. (1993). Remote sensing of the land biosphere and biogeochemistry in the EOS era: Science priorities, methods
and implementation. Global and Planetary Change, 7, 279 297.
99
Slama, C. C. (Ed.) (1980). Manual of photogrammetry. Falls Church, VA:

American Society of Photogrammetry.
Slater, P. N. (1979). A re-examination of the Landsat MSS. Photogrammetric Engineering & Remote Sensing, 45, 1479 1495.
Slater, P. N. (1980). Remote sensing: Optics and optical systems. Reading,
MA: Addison Wesley.
Smith, M. O., Ustin, S. L., Adams, J. B., & Gillespie, A. R. (1990).
Vegetation in deserts: A regional measure of abundance from multispectral images. Remote Sensing of Environment, 31, 1 26.
Suits, G. (1972). The calculation of the directional reflectance of a vegetative canopy. Remote Sensing of Environment, 2, 117 125.
Swain, P. H., & Davis, S. M. (Eds.) (1978). Remote sensing: The quantitative approach. New York: McGraw-Hill Book.
Teillet, P. M., Staenz, K., & Williams, D. J. (1997). Effects of spectral,
spatial, and radiometric characteristics of remote sensing vegetation indices of forested regions. Remote Sensing of Environment, 61, 139 149.
Trishchenko, A. P., Cihlar, J., & Li, Z. (2002). Effects of spectral response
function on surface reflectance and NDVI measured with moderate resolution satellite sensors. Remote Sensing of Environment, 81(1), 1 18.
Tucker, C. J. (1980). A critical review of remote sensing and other methods
for non-destructive estimation of standing crop biomass. Grass and
Forage Science, 35, 177 182.
USGS (1993). Digital elevation modelsdata users guide 5. Reston, VA:
U.S. Geological Survey.
USGS (1996). Standards for digital orthophotos, specifications. Reston,
VA: U.S. Geological Survey.
Walthall, C. L., Norman, J. M., Welles, J. M., Campbell, M., & Blad, B. L.
(1985). Simple equation to approximate the biodirectional reflectance
from vegetative canopies and bare soil surfaces. Applied Optics, 24(3),
383 387.
Wiegand, C. L., & Richardson, A. J. (1984). Leaf area, light interception
and yield estimates from spectral components analysis. Agronomy Journal, 76, 543 548.
Williams, D. L. (1991). A comparison of spectral reflectance properties at
the needle, branch and canopy level for selected conifer species. Remote
Woodcock, C. E., & Strahler, A. H. (1987). The factor of scale in remote
sensing. Remote Sensing of Environment, 21, 311 322.
Wooley, J. T. (1971). Reflectance and transmittance of light by leaves.
Plant Physiology, 47, 656 662.

High spatial resolution satellite observations for validation of MODIS land

products: IKONOS observations acquired under the NASA Scientific
Data Purchase
Jeffrey T. Morisette a,*, Jaime E. Nickeson b, Paul Davis c, Yujie Wang d, Yuhong Tian d,
Curtis E. Woodcock d, Nikolay Shabanov d, Matthew Hansen c, Warren B. Cohen e,
Doug R. Oetter f, Robert E. Kennedy f
a
NASA Goddard Space Flight Center, Code 923, Greenbelt, MD 20771, USA
b
Science Systems and Applications, Incorporated, USA
c
Department of Geography, University of Maryland, College Park, MD, USA
d
Department of Geography, Boston University, Boston, MA, USA
e
USDA Forest Service, USA
f
Forest Science Department, Oregon State University, Corvallis, OR, USA
Received 10 May 2002; received in revised form 26 February 2003; accepted 25 April 2003
Abstract
Phase II of the Scientific Data Purchase (SDP) has provided NASA investigators access to data from four different satellite and airborne
data sources. The Moderate Resolution Imaging Spectrometer (MODIS) land discipline team (MODLAND) sought to utilize these data in
support of land product validation activities with a focus on the EOS Land Validation Core Sites. These sites provide a globally distributed
network of sites where field, aircraft, and satellite data are being collected. As much as possible, uniform data sets useful for validation are
being made available for the core sites. The globally consistent, high-resolution imagery available from IKONOS are being used for their
geolocation accuracy and ability to characterize the landscape at the 1- to 4-m spatial resolution. This paper provides an overview of the
MODIS Land Teams validation strategy to incorporate high-resolution imagery and presents three case studies as examples of the use of
IKONOS data for MODIS land validation activities. We conclude that the globally consistent data from IKONOS, available through NASAs
SDP, have supplied critical validation data sets at a reasonable cost.
Keywords: MODIS; IKONOS; NASA
1. Introduction
The Moderate Resolution Imaging Spectrometer
(MODIS) is a key instrument onboard the EOS Terra
(formerly EOS-AM) satellite, successfully launched in December 1999. An additional MODIS sensor aboard the EOS
Aqua (formerly EOS-PM) platform, launched in May 2002,
now complements the first MODIS sensor. MODIS instrument data are converted on a systematic basis into derived
atmospheric, terrestrial, and oceanic products. Scientists

E-mail address: jeff.Morisette@gsfc.nasa.gov (J.T. Morisette).
doi:10.1016/j.rse.2003.04.003
responsible for the creation of MODIS land (MODLAND)

products have generated a suite of moderate (250 m 1 km)
spatial resolution land products designed to support global
change research and natural resource applications (Table 1;
Justice et al., 2002). The overall goal of NASAs Earth
Science Enterprise is to determine how the Earth is changing and what the consequences are for life on Earth (NASA,
2001). Years of research using the Advanced Very-HighResolution Radiometer (AVHRR) sensor have provided
much of the methodological experience for MODLAND
product development. However, spectral and geometric
constraints of the AVHRR sensor limit the development of
the range of products necessary to assess global change
(Cihlar, 1997). The MODLAND products will provide
products and information that address the broader range of
J.T. Morisette et al. / Remote Sensing of Environment 88 (2003) 100110

Table 1
MODLAND products (Justice et al., 2002)
101
2. Central coordination of commercial data acquisition

with the needs of the science team
Product suite
Product
DAAC
MOD09
MOD11
MOD43
MOD10
MOD29
MOD13
MOD15
MOD17
MOD12
MOD14
MOD44
Surface reflectance
Surface temperature and emissivity
BRDF/albedo
Snow cover
Sea ice extent
Vegetation indices
LAI/FPAR
Net primary vegetation production
Land cover and change
Thermal anomalies and fire
Vegetation cover conversion/continuous fields
EDC
EDC
EDC
NSIDC
NSIDC
EDC
EDC
EDC
EDC
EDC
EDC
Products related to the case studies presented in this paper are shown in
bold font. The Distributed Active Archive Center (DAAC) column
represents the distribution point, either Land Processes (LP) or National
Snow and Ice Data Center (NSIDC).
Additional details available for LP DAAC (http://edcdaac.usgs.gov/modis/
dataprod.html) and for NSIDC DAAC (http://nsidc.org/).
global change research questions associated with biogeochemical cycling, energy balance, land cover change, and
ecosystem health (Table 1).
Lessons learned from the previous generation of global
land imaging systems indicate that product validation is
critical for accurate and credible product usage (Cihlar,
Chen, & Li, 1997; Justice & Townshend, 1994). The
Committee on Earth Observing Satellites (CEOS) Working
Group on Calibration and Validation (WGCV) defines
validation as the process of assessing, by independent
means, the quality of the data products derived from the
system outputs (Justice et al., 2000). In this context, the
MODLAND validation activities are a means by which
independent field, airborne, and other satellite data are
collected and used to assess the accuracy of MODLAND
products. These validation activities will provide the user
community with quantitative estimates of uncertainty for
MODLAND products.
The objective of this paper is to demonstrate how
NASAs Scientific Data Purchase (SDP) has been used
to support MODIS land product validation. We start with
a description of the EOS Core Sites and the information
gathered to characterize these sites. The NASA SDP
program was utilized to acquire at least one fine-resolution
(1 4 m) IKONOS image per core site. These fine spatial
resolution images complement the high-resolution (15 30
m) Landsat 7 and ASTER data available for the EOS
Core Sites (Morisette, Privette, & Justice, 2002). To
provide specific examples of the use of IKONOS data,
we present three case studies. The case studies demonstrate the utility of fine-resolution data for validation of
moderate spatial resolution products. The MODLAND
products related to these case studies are listed in bold
font in Table 1. Lastly, we discuss the experience of
coordinating commercial data for the needs of a large
science team.
The case studies presented below represent only a small

portion of the IKONOS data that have been collected
under the NASA SDP in support of EOS validation
activities. NASAs Goddard Space Flight Center (GSFC)
coordinated the acquisition of these data from an allocation
set up by NASA-HQ within the SDP for IKONOS data.
Prelaunch validation planning required that the MODLAND team identify supporting data available and establish the infrastructure to acquire, archive, and distribute
these data. The SDP provided IKONOS data for all 26
EOS Core Sites as well as many additional validation sites.
Currently, at least one IKONOS scene has been acquired
for all core sites. Browse imagery, metadata, and ordering
instructions for core site IKONOS data are posted on the
EOS Core Sites web pages (http://landvol.gsfc.nasa.gov/
MODIS/coresite_gen.asp).
While centralized coordination of the IKONOS tasking
was designed to provide efficiencies in data use, it did
require an extensive amount of effort to submit and track the
150 IKONOS acquisitions tasked under the SDP. To help
manage the status, issues, and action items for these tasks,
periodic teleconferencing with the SDP personnel was
essential. Tasks were submitted via the SDP web site
interface, a process that became smoother and more efficient
over time, with added flexibility and features that were
lacking early on. Occasional communication directly with
Space Imaging, the IKONOS data provider, was very
helpful although we could not request Space Imaging to
modify a task directly, as this could only come from the
NASA SDP. Our centralized coordination of resources
helped to organize and optimize the collection of IKONOS
data for the MODLAND team; however, it created another
administrative layer between the end user and the data
provider, which was at times inefficient. The benefit of
centralized coordination was the prevention of duplication
of effort and the creation of the potential for data sharing
and multiple usage of a given acquisition. An example of
this was the use of a scene from Harvard Forest site by the
MODIS Vegetation Index team (University of Arizona) to
validate the vegetation index product, while the BigFoot and
Boston University teams used the same data to validate the
leaf area index (LAI) product.
3. EOS Land Validation Core Sites

The EOS Land Validation Core Sites facilitate land
product validation by focusing information and measurement resources at a limited number of established sites for
the validation of land products derived from moderateresolution remotely sensed imagery. Morisette et al.
(2002) provide details on goals, rational, and geographic
and biophysical information for the EOS Core Sites.
102
Under research activities being carried out at the University of Maryland with the support of the NASA Stennis
Space Center, comparative analysis of observations from the
EOS Landsat 7s Enhanced Thematic Mapper plus (ETM+)
and IKONOS has been conducted (Goward et al., 2004). To
support this Commercial Remote Sensing for Earth Science
(CRESS) activity, the principal investigators contributed to
the development of the EOS Land Validation Core Site
information database to assist with comparative analyses.
This resulted in CRESS staff developing site characterization layers for each of the EOS Core Sites. The layers
include maps and tables of land cover, soils, elevation,
climate, and population, as well as a list of contacts and
in situ measurements taken at each site. Since the core sites
are located globally in a variety of biomes (Morisette et al.,
2002), it was imperative to utilize standardized data sets
with global coverage to generate consistent characterization
layers. Thus, data sources were acquired from scientifically
acceptable data sets subject to: (1) availability in digital
form, and (2) consistent global coverage. The characterization output GIS layers, integrated reports, and JPEG maps
for each core site are available on-line or on CD-ROM
through the Global Land Cover Facility at the University of
Maryland (http://glcf.umiacs.umd.edu/; search under data). A summary of characterization layers is given in Table
2 with values for each layer given for the three sites
presented in the following three case studies.
The first case study, over Senanga, Zambia, demonstrates
the use of paired IKONOS and ETM+ to classify the
landscape into either crown cover or no crown cover.
The second case study is over a fairly homogenous landscape

of Sevilleta, NM. Here the paired IKONOS and ETM+ data
are used to assess the relationship between image spatial
resolution, spectral transformation, and spatial structure.
Spectral transformations include principal components, tasseled cap, and NDVI (Normalized Difference Vegetation
Index). The spatial structure is analyzed through empirical
semivariograms. The analysis of the scales of spatial variation guided the field sampling design. In the third case study,
both field sampling and IKONOS imagery show the high
spatial variability at the Maun, Botswana, site within a 30-m
ETM+ pixel. The objective of this study is to produce an LAI
surface at the ETM+ spatial resolution. Spectral patches,
as determined by ETM+ spectral values, are classified based
on interpretation of the IKONOS data. This classification
determines the proper algorithm to apply to each patch to
produce an LAI surface from the ETM+ data. Together, the
three case studies represent the general approach of collecting independent field and/or imagery to validate MODLAND products.
4. Case study 1: using IKONOS data to validate the

MODIS Vegetation Continuous Fields product
The MODIS Land Teams Vegetation Continuous Fields
are produced to meet the needs of global biogeochemical
and climate modelers. The layers included in this product
are percent bare ground, herbaceous, and tree cover. The
tree cover layer is further broken down by percent ever-
Table 2
Characterization layers available for each site
Global data layer
Source
Resolution
Measurement at
Maun
Measurement at
Senanga
Measurement at
Sevilleta
Reference
(ETM+ path/row
footprint, major
roads and water
bodies
Elevation
UMD land cover
Various
Scale varies
19.92 latitude,
23.59 longitude
15.86 latitude,
23.34 longitude
34.32 latitude,
106.8 longitude
USGS, 1996, GTOPO30

Hansen, DeFries, Townshend, &
Sohlberg, 2000; University of
Maryland
Loveland et al., 2000; EROS Data
Center
Knyazikhin, Martonchik, Myneni
et al., 1998; Boston University
DeFries, Hansen, & Townshend,
2000; University of Maryland
ESRI, 1998; ArcAtlas
1 km
1 km
929 m
Wooded grassland
1019 m
Cropland
1925 m
Open shrubland
1 km
Savanna
Open shrubland
1 km
Savanna
Cropland/natural
mosaic
Broadleaf crops
Shrubs
1 km
9%
14%
0.00%
Scale = 1:3 million
Alluvial subtropicals
Tobler, Deichman, Gottsegen, &

Maloy, 1995; Global Demography
Project
NASA CIDC, 1999
Original, varies by
country and unit
1 person per square

mile
Subtropical tropical
podzols
4 persons per square
mile
Calcichromic
cambisols
1 person per
square mile
NA
Jan pr: NA
Jul pr: NA
Jan temp: 25.2jC
Jul temp: 19.48jC
Jan pr: NA
Jul pr: NA
Jan temp: 25jC
Jul temp: 20.14jC
Jan pr: 0.61 mm

Jul pr: 2.17 mm
Jan temp: 0.6jC
Jul temp: 25.31jC
EDC land cover

BU biomes
UMD percent tree
cover
Soils
Population density
Climate; monthly
means
green, deciduous, needleleaf, and broadleaf. The product is

currently being derived on an annual basis from MODIS
surface reflectance values for a given year. The spatial
resolution is 500 m (Hansen et al., 2002). Considering that
the area covered by in situ measurements is extremely small
relative to the size of satellite pixels and the large spatial
variation in the landscape, it is unrealistic to expect to
collect sufficient field data to support a direct comparison
of field measurements with a large number of coarseresolution pixels. An alternative is to employ both field
measurements and high-resolution satellite data to derive an
accurate fine-resolution map over a sufficiently extensive
area, degrade it to the coarse resolution, and compare this
map with that derived from the coarse-resolution imagery.
The first and third case studies take place within the
context of the Southern Africa Fire and Atmosphere Regional Initiative 2000 (SAFARI 2000). This is an international research initiative, within which a major component is
remote sensing research and validation of NASA EOS data
products associated with EOS Terra and Landsat 7 (Privette
et al., 2002). As part of the SAFARI 2000 science initiative,
a validation exercise for the MODIS continuous field of tree
cover product was undertaken. Using field measurements,
IKONOS and ETM+ imagery, and ancillary map sources, a
tree cover map was produced and validated for two ETM+
scenes in Western Province, Zambia. The validation method
103
used was designed to be easily replicated, with the goal of

performing similar validation exercises in other biomes
across the globe.
Five IKONOS images were classified into crown cover/
no crown cover maps at 4-m resolution. These maps were
aggregated by counting the number of IKONOS crown
pixels within 30 m of ETM+ cells in order to create a
continuous training data set of percent crown cover. IKONOS data allow for the direct delineation of the variable of
interest, tree crown cover, and its subsequent scaling to
ETM+ and MODIS resolutions. The validation map was
created at the 30-m ETM+ resolution and then averaged to
250- and 500-m resolutions. Three dates of ETM+ data were
acquired to predict percent crown cover using a regression
tree algorithm. Single-date IKONOS was coupled with GPS
field observations to produce tree crown classifications at the
4-m spatial resolution. For single-date ETM+, it can be
difficult, if not impossible, to separate some tree covers from
grasslands in this area of Zambia. Multitemporal ETM+
imagery, however, does improve the ability to discriminate
categories that are inseparable in a single ETM+ image. Fig.
1 shows the process by which IKONOS data were characterized to create a MODIS scale map to be used in validation
exercises. The final validation map covers the entire area of
Landsat WRS path/rows 175/070 and 175/071, an area
covering almost 75,000 km2.
Fig. 1. Example of progression from IKONOS to MODIS for the Senanga IKONOS acquisition. (a) False-color infrared composite of IKONOS bands 4, 3, and
2. (b) Crown/no crown classified map. (c) Percent crown cover training from IKONOS aggregated to 30-m cells. (d) Result of ETM+ characterization of
percent crown cover. (e) ETM+ result aggregated to 250-m MODIS resolution cells. (f) Crown cover map from MODIS data using data from ETM+-aggregated
250-m map. Includes material from Space Imagingn.
104
Fig. 2. Plot of field-measured crown cover versus ETM+-predicted crown

cover for pixel where the center of site was located. For this plot, the solid
line is the 1:1 line. The dotted line is the derived relationship, describing the
reduced major axis, y = 0.95x + 0.38. R2 = 0.74. RMSE = 9.76%. Grey
squares represent the class boundaries of GOFC tree cover strata.
Comparisons of IKONOS-derived tree cover and field

data to ETM+ tree estimates yielded root mean square errors
(RMSEs) of f F 10% crown cover. Fig. 2 shows field data
plotted against ETM+ tree cover estimates. In this plot, the
utility of the simple RMSE measure in deriving meaningful
accuracy estimates is contrasted with a land cover classification approach. The RMSE measure allows the user to
determine the acceptability of the magnitude of errors
present more easily than with classification accuracy. For
example, if the measurements from the field data are
converted to the five-class Global Observations of Forest
Cover (GOFC, 1999) scheme (shown as gray boxes in Fig.
2), an accuracy of only 65% results. However, the RMSE of
9.76% reveals, as does Fig. 2, that the errors are most likely
in adjacent cover classes.
The ability to identify crown cover from IKONOS
data, its global availability, and the methods developed
to relate them to ETM+ and MODIS data indicates that
1 4 m of spatial resolution satellite imagery can be a
valuable tool for validation of a Vegetation Continuous
Fields product.
5. Case study 2: uses of Scientific Data Purchase

imagery in the BigFoot project
The overall goal of the BigFoot project is to provide
quality assessment of MODIS science products related to
the global carbon cycle, including land cover, LAI/fraction
of photosynthetically active radiation (FPAR), and net
primary vegetation production. To do so, BigFoot uses
ground measurements, remote sensing data, and ecosystem
process models at nine of the EOS Land Validation Core

Sites representing different biomes of the western hemisphere from Alaska to Brazil (http://www.fsl.orst.edu/larse/
bigfoot). Each BigFoot site is 5 5 km in size, surrounding
the relatively small footprint (approximately 1 km2) of a
CO2 flux tower. Multiyear in situ measurements of ecosystem structure and functional characteristics are collected at
each site. The sampling design and scaling logic allow
explicit examination of fine-grained spatial structure of
these measured properties and provide for both field- and
image-based ecological characterizations of the greater flux
tower footprint. Multiyear measurements ensure that interannual consistency of MODIS products can be assessed. To
help associate field and point measurements to the larger
MODIS pixel, the BigFoot team uses higher-resolution
imagery to explore the variation within the MODIS pixel.
Specifically, IKONOS imagery is used for the georeferencing of Landsat data, field reconnaissance, and interpretation
of Landsat spectral patterns; it has also been used to
examine spatial patterns in the landscape. This provides
insight into the design of a field sampling strategy.
Field sampling for remote sensing studies is a complex
endeavor, with multiple competing objectives and constraints. The initial BigFoot sampling design (Campbell,
Burrows, Gower, & Cohen, 1999) was based almost exclusively on logistical considerations (given the high cost of
field sampling) and a desire to saturate the flux tower
footprint with plot-based measurements. Due to reliance
on Landsat ETM+ for mapping, each plot was 25 25 m to
match the 25-m pixel size of the resampled ETM+ imagery.
The design was a nested spatial series (Burrows et al., in
press) to determine scales of spatial variation among the
measured variables. Only minimal attention was given to
sampling the spectral characteristics of a site, as provided by
Landsat imagery. Consequently, the full range of spectral
variability at a site was often undersampled, and occasionally minority land cover types were not included. Moreover,
the goal of using field data to determine spatial autocorrelation structure required that some field plots be autocorrelated. From the perspective of developing remote sensing
models from field data, however, such autocorrelation is an
inefficient use of resources (Curran, 1988). Additionally, if
there were important scales of variation below that of the 25
m captured within an individual plot, that variation would
be ignored. By incorporating remotely sensed imagery into
our sampling design, a first critical question can be
addressed: To what extent can imagery be used as a
surrogate for field measurements to determine the scale of
biophysical variation on a landscape? This question was
answered in another study using semivariograms, a geostatistical tool for determining scales of variation in spatial
data (Cohen, Spies, & Bradshaw, 1990; Woodcock, Strahler,
& Jupp, 1988). Initial results show that field measurements
of LAI and a first principal component of Landsat spectral
data had semivariogram ranges that were nearly identical
(US Forest Service, Warren Cohen, unpublished data). This
was true for sites in four separate biomes, where the ranges
(or scales of variation) were between 125 and 500 m.
Assuming that image data can adequately track general
spatial patterns of a biophysical variable, image data can be
used to ask whether there are scales of variation finer than
25 m. This can be determined using IKONOS data. Fig. 3
shows IKONOS imagery degraded from 4 to 512 m and
corresponding semivariograms calculated from that imagery
at a desert core site, the Sevilleta National Wildlife Refuge
in New Mexico. These semivariograms are derived from the
first principal component of the four IKONOS multispectral
bands. As theory suggests (Milne & Cohen, 1999), the
shape of the semivariogram changes with the support (or
pixel size) of the imagery used. In particular, the sill is
reduced as image grain size increases, due to the reduction
of image variance with each successive coarsening of the
imagery. Determination of the range of a semivariogram is
generally considered to be the lag at which 95% of the sill is
reached (Deutsch & Journel, 1998). Using this rule, the
105
ranges of these semivariograms decrease from about 800 to

500 m as the image is degraded from 4 to 512 m (Table 3).
Although this shows that spatial support of imagery, indeed,
has some effect on the observed scale of spatial variation,
the most important point is that even at 4 m of support, the
scale of variation is significantly greater than the spatial
resolution of Landsat. These results suggest that Landsat
ETM+ data can be used to determine the scale of variation
in the biophysical properties of this landscape.
If the scale of variation at other sites is smaller than the
grain of Landsat ETM+ imagery, then IKONOS imagery
would be an excellent alternative for use in determining the
scale of variation. However, IKONOS data contain only two
major axes of variation (visible and near-infrared), much
like the Landsat Multi-Spectral Scanner (MSS), while Landsat ETM+ data contain three basic spectral dimensions
(Crist & Cicone, 1984) due to its inclusion of shortwave
infrared reflectance. Before using IKONOS imagery as a
surrogate for Landsat ETM+ to determine the scale of
Fig. 3. First principal component of a four-band IKONOS image over a 3 3-km area of the Sevilleta National Wildlife Refuge in New Mexico spatially
aggregated from 4 to 512 m (top). Semivariograms calculated from the imagery at each spatial resolution or grain size (bottom). Includes material from Space
Imagingn.
106
Table 3
Semivariogram ranges for the IKONOS image of Fig. 3 spatially
aggregated from a support of 4 m to a support of 512 m
Support (m)
Range (m)
4
8
16
25
32
64
128
256
512
504
527
559
580
600
638
690
756
800
The range of 25-m support was calculated from Landsat ETM+ data.
spatial variation, the spectral dimensionality of image data

also needs to be considered. Additionally, the impact of a
specific spectral transformation on the observed scale of
variation needs to be considered.
First, a semivariogram based on the first principal component of the first four reflectance bands (roughly matching
the four multispectral IKONOS bands) from a 25-m ETM+
image of Sevilleta was examined. Here, the range fell
directly between those of the resampled IKONOS four-band
principal component imagery at 16 and 32 m (Table 3). This
suggests that, at this site, the difference between ETM+ and
IKONOS is not significant in determining the spatial structure of the landscape. As another check for the influence of
sensor characteristics and spectral transformation on the
observed scale of variation, NDVI was calculated from the
two sensors (Fig. 4). The ranges of the two NDVI semivariograms were essentially equal, further supporting the
notion that basic sensor characteristics are similar. However,
the range of the NDVI image is significantly different from
the range of the principal component image (1500 versus
580 m). This suggests that the scale of variation depends on
the parameter derived from the imagery. With that in mind,
the Tasseled Cap Wetness (Crist & Cicone, 1984) layer (a

transformation requiring a shortwave infrared band not
available from IKONOS) was examined to see if it exhibited
a different scale of variation than the first principal component or the NDVI. At this site, the range of the Wetness
semivariogram was nearly identical to the range of the
principal component semivariogram (Fig. 4). As such,
across the Sevilleta desert landscape, IKONOS was equivalent to Landsat ETM+ for characterizing the spatial structure of biophysical properties at the 30-m and larger spatial
resolution (to the extent that these are adequately represented by spectral data). In this case, Landsat ETM+
imagery was appropriate for determining the scale of spatial
variation. However, this was only confirmed by using finerresolution IKONOS data. Whether this is true at other sites
would need to be determined in a similar manner (i.e.,
analyzing both ETM+ and IKONOS data).
Similar exercises at other BigFoot sites indicate that the
scale of variation in imagery differs significantly with the
spectral transformation used. As such, it is prudent (when
designing a sampling scheme for field measurements to
support remote sensing analyses) to examine the relationships between the biophysical variables of interest and
different spectral transformations. If more than one biophysical variable is of interest, or if one seeks a more
general application of those relationships, it may be prudent
to include several different types of spectral transformations
in determining scales of spatial variation for application in
designing a field sampling scheme.
6. Case study 3: using IKONOS to assist in producing a

high-resolution LAI surface
Similar to case study 1, case study 3 combines field
measurements and high-resolution satellite data to derive an
Fig. 4. Normalized semivariograms for the first principal component and the NDVI of 4-m IKONOS imagery and the NDVI and Wetness of Landsat ETM+
imagery from a 3 3-km area of the Sevilleta National Wildlife Refuge in New Mexico.
accurate fine-resolution surface that can be degraded to

coarse resolution and compared to the MODIS product. In
case study 1, the surface was percent tree cover. Here the
surface is LAI. Due to complexities in the LAI algorithm
(Myneni, Nemani, & Running, 1997), it is important to
consider the spatial structure and spectral values of the highresolution satellite data. Unlike case study 2, the landscape
studied here shows high spatial variation at much smaller
distances, even less than the ETM+ pixel resolution.
As part of SAFARI 2000, an international group of
researchers completed an intensive field campaign in Botswana and Zambia between February 28 and March 18,
2000. These dates coincided with the first week of MODIS
Earth view. The field sites are located along the International
Geosphere Biosphere Program (IGBP) Kalahari Transect
(KT). The KT extends over a large rainfall gradient (200
1000 mm/year mean annual rainfall) in an area of uniform
soils, the Kalahari sands. The vegetation extends from
equatorial forest to subtropical, arid shrubland of the Kalahari desert (Dowty et al., 2000). In order to validate the
MODIS LAI/FPAR products, a team from Boston University participated in the SAFARI 2000 wet season campaign;
intensive measurements of LAI, FPAR, canopy reflectance,
and transmittance were made in four field sites (Pandamatenga, Maun, Okwa River Crossing, and Tshane, Botswana)
along the Kalahari Transect.
During the SARFARI 2000 wet season and Kalahari
Transect field campaign, conducted between February 28
and March 18, 2000, field measurements of LAI were
collected along three 750-m transects at a spacing of 25
m within a 1 1-km site near Maun, Botswana. LAI values
were measured using an LAI-2000 plant analyzer under
diffusely illuminated sky conditions. Evaluation of these
field measurements of LAI indicates a high degree of spatial
variability over relatively short distances (Fig. 5). From Fig.
5, it is clear that there is virtually no relationship between
adjacent measurements only 25 m apart. This result raises
questions about the area covered by individual LAI measurements, the heterogeneity of the surface, and how they are
Fig. 5. Spatial variability of field measurements of LAI along transects at

the Maun site. Three west east-oriented transects are named from north to
south as N, A, and B respectively.
107
Fig. 6. Variogram showing spatial variance as a function of distance, as

calculated from the panchromatic band of IKONOS with 1-m spatial
resolution. Notice the log scale for distance and the extremely rapid rise of
variance over very short distances.
related. IKONOS imagery, with its high spatial resolution, is

well suited for exploration of the spatial scales of variability
in landscapes. A variogram (Fig. 6) of the 1-m panchromatic
band from an IKONOS image of the Maun site shows a
dramatic rise in spatial variance over distances as short as
10 m. This suggests that the LAI values between the sample
points (25 m apart) are not spatially related, indicating a
high level of heterogeneity in the spatial distribution of LAI
as shown in Fig. 5. It also indicates the difficulty of using
single field measurements to characterize areas larger than
approximately 100 m2. This result implies that denser
sampling is needed to characterize the LAI of single Landsat ETM+ pixels (approximately 900 m2 in size). This fact
is further exacerbated by any possible misregistration between the field measurements and images.
One way to try to minimize the effects of sparse field
sampling of ETM+ pixels and possible geolocation errors is
to shift the scale of analysis from pixels to vegetation
stands, or patches (Tian et al., 2002a). We used an image
segmentation procedure to separate patches of vegetation to
serve as the basis of validation. The segmentation groups
pixels into patches based on their spectral similarity and
adjacency. Each patch may contain tens or hundreds of
pixels. The mean value of field samples within a patch is
then taken as an estimate of the true mean of the entire
patch. Fig. 7 shows the ETM+ image over the 1 1-km site
and the resulting patch map. There are 15 patches in total
(Tian et al., 2002a).
Biome type is an important information required by the
MODIS LAI algorithm. We used the six-class biome types
defined by Myneni et al. (1997). Each biome type defines a
certain vegetation soil pattern, which includes characteristics of canopy structure, brightness of canopy background,
understory vegetation, ground cover, crown shadowing, etc.
To run the MODIS LAI algorithm on ETM+ pixels (Knyazikhin, Martonchik, & Diner et al., 1998; Knyazikhin,
Martonchik, Myneni, Diner, & Runing, 1998), it is necessary to identify the biome type of each patch. The ETM+
108
Fig. 8. Color IKONOS RGB image from bands 4, 3, and 2 of a 1 1-km

region of the Maun site. Yellow + signs represent sampling points.
Includes material from Space Imagingn.
Fig. 7. (a) Color ETM+ RGB image from bands 4, 3, and 2 of a 1 1-km
region of the Maun site. (b) The patch map after image segmentation.
Patches 1, 2, 4, 7, 8, 9, 12, 13, and 15 are savannas. Patches 3, 5, 6, 10, 11,
and 14 are shrubs.
image was atmospherically corrected using the simplified

method for atmospheric correction (SMAC) algorithm
(Hame et al., 2001; Rahman & Dedieu, 1994). From the
ETM+ image alone, it is difficult to identify the biome type
within each patch. However, accurate subpixel information
can be obtained through the IKONOS image (Fig. 8). One
can see that patches 3, 5, 6, and 10 are mostly small bushes/
trees without understory vegetation and hence are identified
as shrubs; patches 7, 8, 9, and 12 are mostly small bushes/
trees with understory vegetation and hence are identified as
savannas. Using this IKONOS-derived biome classification
for each patch, the MODIS LAI algorithm was applied to
produce a high-resolution LAI map from the ETM+ surface
reflectances (Fig. 7a). Comparison of the mean LAI over
patches derived from field measurements and ETM+ data is
shown in Fig. 9. This result then formed the basis for
applying the MODIS LAI algorithm over a 10 10-km

area of Landsat ETM+ images for comparison with a
10 10 window of MODIS pixels.
Similar techniques have been applied to validation data
collected in conifer forests near Ruokolahti, Finland, during
June 14 21, 2000 (Wang et al., in review) and Harvard
Forest, between July 21 and 25, 2000 (Shabanov et al., in
press). The value of IKONOS images for these efforts lies in
its fine spatial resolution, which supports exploration of the
spatial scale of variability in landscapes and classification of
biome type for ETM+-derived spectral patches.
Fig. 9. Mean LAI over patches derived from field measurements and ETM+
data.
7. Conclusion
The validation of global products requires consistent
data with which to compare the global products. A main
concern for validating the relatively coarse-resolution products from the MODLAND team is that of relating field or
ground-based point measurements with the large area
represented by a MODIS pixel. The approach used by the
MODLAND team was to use imagery at intermediate
scales to bridge between the two. The 1-m panchromatic
and 4-m multispectral data from IKONOS, available in a
consistent format and quality throughout the globe, have
provided a complement to the 30-m Landsat 7 ETM+ data.
Without the high-resolution imagery available through the
SDP, the only other option for such data would have been
airborne imagery. Flight planning, digital image acquisition, and post-processing are extremely time-consuming
and expensive. Furthermore, companies or agencies supplying such imagery typically work within a finite range,
thus making it difficult (or impossible) to acquire consistent data over the globally distributed EOS Land Validation
Core Sites. Lastly, there are some regions of the world
where airborne image acquisition is either impossible or
unsafe. In light of these considerations, NASAs investment
in high-resolution imagery available through the SDP has
supplied the EOS Land Validation Core Sites with unique,
globally consistent, critical validation data sets at a reasonable cost.
One of the biggest concerns in the use of IKONOS from
the SDP was the uncertainty in the timing of acquisitions.
The SDP contract with Space Imaging required that we
accept data acquired within F 2 weeks of the requested
acquisition date. Attempting to acquire a scene for a narrow
date window to coincide with fieldwork was nearly impossible. In many instances, tasks required a formal extension
of the original date window specification in order to accept
data acquired outside of that original window. Tasking
priorities and overpass timing were largely unknown to
us. There is no doubt there are competing requests for
IKONOS acquisitions, and cloud cover also has to be taken
into account; however, explicit knowledge of how NASAs
SDP requests fit within Space Imagings tasking priority
system was never known. The uncertainty of acquisition
timing was difficult for those trying to couple IKONOS data
acquisitions with field activities. When relating satellite
spectral data and field measurements, it is desired that
ground data collection be nearly simultaneous with the
satellite overpass. Currently, for EOS platforms and several
other satellites, the EOS program provides a web-based tool
to obtain overpass information (http://earthobservatory.nasa.
gov/MissionControl/overpass.html). Such information for
IKONOS would be helpful to ascertain at least the potential
dates and times for IKONOS acquisitions. Knowledge of
the priority of SDP requests, and the likelihood of actual
collection from a list of possible acquisitions would be very
useful information to have.
109
Heretofore, f 1-m spatial resolution data available to

the research community were limited to those collected
from aircraft platforms. In light of this, we should consider
the cost and coordination effort required in the collection
of an equivalent amount of airborne data. We found that
to cover a 5.5-m2 area, an area not quite as large as the
11 11-km2 area of our IKONOS acquisitions, with color
IR orthorectified airborne digital imagery would cost anywhere between US$50,000 and US$75,000. This is assuming that the study area is within the range covered by
an aerial survey company, some of which only operate
regionally. Some of the core sites are international and
somewhat remote, which means that some sites might be
impossible or prohibitively expensive to survey by aircraft.
In contrast, the IKONOS data were available globally and
ranged in price from f US$6000 for imagery over North
America to f US$15,000 for international acquisitions.
MODLANDs coordination of IKONOS data through
NASAs SDP has required considerable time and effort.
However, the coordinated approach and core site infrastructure have resulted in efficiencies by allowing the same data
to be used to validate more than one product by more than
one team. The case studies presented here demonstrate the
use of IKONOS data in support of validation efforts for
some of the MODLAND products and provide examples of
how the spatial resolution of IKONOS data can lend insight
into the analysis of more coarse-resolution imagery.
Acknowledgements
MODIS land validation activities are primarily supported
through the MODIS land discipline team and EOS
validation, with special thanks to C. Justice, J. Privette,
and D. Starr. The MODLAND allocation of SDP resources
was made possible through NASA Headquarters and NASA
Stennis Space Center, with special thanks to Diane Wickland
and Fritz Policelli. Thanks also to NASA Code YS, through
Martha Maiden, for providing the ETM+ data used to scale
between IKONOS and MODIS. Case study 2 and three were
part of the Southern African Regional Science InitiativeSAFARI 2000. Thanks to three anonymous reviewers who
provided helpful suggestions that improved the paper.
References
Burrows, S., Gower, S., Clayton, M., Mackay, D., Ahl, D., Norman, J., &
Diak, G. (2002). Application of geostatistics to characterize LAI from
flux tower to landscape scales using a cyclic sampling design. Ecosystems 5, 667 679.
Campbell, J. C., Burrows, S., Gower, S. T., & Cohen, W. B. (1999). BigFoot: Characterizing land cover, LAI, and NPP at the Landscape Scale
for EOS/MODIS Validation. Field manual 2.1. Environmental Science
Division, vol. 4937. Oak Ridge, TN: Oak Ridge National Laboratory,
103 pp.
Cihlar, J. (1997). GCOS/GTOS plan for terrestrial climate related obser-
110
vations. Version 2. GCOS Report 32 (130 pp.). Geneva: World Meteorological Organisation.
Cihlar, J., Chen, J., & Li, Z. (1997). On the validation of satellite-derived
products for land applications. Canadian Journal of Remote Sensing,
23(4), 381 389.
Cohen, W., Spies, T., & Bradshaw, G. (1990). Semivariograms of digital
imagery for analysis of conifer canopy structure. Remote Sensing of
Crist, E. P., & Cicone, R. C. (1984). A physically-based transformation of
thematic mapper datathe TM tasseled cap. IEEE Transactions on
Geoscience and Remote Sensing, GE-22, 256 263.
Curran, P. J. (1988). The semivariogram in remote sensing: An introduction. Remote Sensing of Environment, 24, 493 507.
DeFries, R. S., Hansen, M. C., & Townshend, J. R. G. (2000). Global
continuous fields of vegetation characteristics: A linear mixture model
applied to multi-year 8 km AVHRR data. International Journal of
Remote Sensing, 21(6 7), 1389 1414.
Deutsch, C. V., & Journel, A. G. (1998). GSLIB geostatistical software
library and users guide. New York: Oxford University Press, 369 pp.
Dowty, D., Frost, P., Lesolle, P., Midgley, G., Mukelabai, M., Otter, L.,
Privette, J., Ringrose, S., Scholes, B., & Wang, Y. (2000). Summary of
the SAFARI 2000 wet season field campaign along the Kalahari Transect. Earth Observer, 12, 29 34.
ESRI (1998). ArcAtlas. 380 New York Street, Redlands, CA 92373-8100,
USA: ESRI.
GOFC Design Team (1999). A strategy for global observation of
forest cover: Version 1.2, http://www.fao.org/gtos/gofc-gold/docs/
execsum.pdf.
Goward, S. N., David, P. E., Fleming, D., Miller, L., & Townshend, J. R.
measurements for selected earth observation system (EOS) validation
sites. Remote Sensing of Environment, 88, 79 98 (this issue).
Hame, T., Stenberg, P., Andersson, K., Rauste, Y., Kennedy, P., Folving, S.,
& Sarkeala, J. (2001). AVHRR-based forest proportion map of the PanEuropean area. Remote Sensing of Environment, 77, 76 91.
Hansen, M. C., DeFries, R. S., Townshend, J. R. G., & Sohlberg, R. (2000).
Global land cover classification at 1 km spatial resolution using a classification tree approach. International Journal of Remote Sensing,
21(6/7), 1331 1364.
Hansen, M. C., DeFries, R. S., Townshend, J. R. G., Sohlberg, R., Dimiceli, C., & Carroll, M. (2002). Toward an operational MODIS continuous field of percent tree cover algorithm: Examples using AVHRR and
MODIS data. Remote Sensing of Environment, 83(1 2), 303 319.
Justice, C., Belward, A., Morisette, J., Lewis, P., Privette, J., & Baret, F.
(2000). Developments in the validation of satellite sensor products for
the study of land surface. International Journal of Remote Sensing,
21(17), 3383 3390.
Justice, C. O., & Townshend, J. R. G. (1994). Data sets for global remote
sensing: Lessons learnt. International Journal of Remote Sensing,
15(17), 3621 3639.
Justice, C., Townshend, J., Vermote, E., Masuoka, E., Wolfe, R., Saleous,
N., Roy, D., & Morisette, J. (2002). An overview of MODIS, its data
processing and products for terrestrial science applications. Remote
Sensing of Environment, 83(1 2), 3 15.
Knyazikhin, Y., Martonchik, J. V., Diner, D. J., Myneni, R. B., Verstraete,
M., Pinty, B., & Gorbon, N. (1998). Estimation of vegetation canopy
leaf area index and fraction of absorbed photosynthetically active radiation from atmosphere-corrected MISR data. Journal of Geophysical
Research, 103, 32239 32256.
Knyazikhin, Y., Martonchik, J. V., Myneni, R. B., Diner, D. J., & Runing,
S. (1998). Synergistic algorithm for estimating vegetation canopy leaf
area index and fraction of absorbed photosynthetically active radiation
from MODIS and MISR data. Journal of Geophysical Research, 103,
32257 32275.
Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. O., Zhu, Z., Yang, L.,
& Merchant, J. W. (2000). Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. International Journal of Remote Sensing, 21(6/7), 1303 1330.
Milne, B. T., & Cohen, W. B. (1990). Multiscale Assessment of Binary
and Continuous Land cover Variables for MODIS Validation Mapping, and Modeling Applications. Remote Sensing of Environment,
70, 82 98.
Morisette, J. T., Privette, J. L., & Justice, C. O. (2002). A framework for the
validation of MODIS land products. Remote Sensing of Environment,
83(1 2), 77 96.
Myneni, R. B., Nemani, R. R., & Running, S. W. (1997). Estimation of
global leaf area index and absorbed par using radiative transfer models. IEEE Transactions on Geoscience and Remote Sensing, 35,
1380 1393.
NASA (2001). Exploring our home planet: Earth Science Enterprise
strategic plan, http://www.earth.nasa.gov/visions/stratplan/
ese_strategic_plan.pdf.
NASA CIDC (1999). Climatology interdisciplinary data center.
NASA. Atmospheric soundings from TOVS, http://daac.gsfc.nasa.
gov/CAMPAIGN_DOCS/FTP_SITE/INT_DIS/readmes/tovs.html.
Privette, J. L., Myneni, R. B., Knyazikhin, Y., Mukufute, M., Robert, G.,
Tian, Y., Wang, Y., & Leblanc, S. G. (2002). Early spatial and temporal
validation of MODIS LAI product in Africa. Remote Sensing of Environment, 83, 232 243.
Rahman, H., & Didieu, G. (1994). SMAC: A simplified method for the
atmospheric correction of satellite measurements in the solar spectrum.
International Journal of Remote Sensing, 15, 123 143.
Shabanov, N. V., Wang, Y., Buermann, W., Dong, J., Hoffman, S., Smith,
G. R., Tian, Y., Knyazikhin, Y., Gower, S. T., & Myneni, R. B. (2003).
Effect of foliage spatial heterogeneity in the MODIS LAI and FPAR
algorithm over broadleaf forests. Remote Sensing of Environment 85(4),
410 423.
Tian, Y., Woodcock, C. E., Wang, Y., Privette, J. L., Shabanov, N. V., Zhou,
L., Zhang, Y., Buermann, W., Dong, J., Veikkanen, B., Hame, T., Andersson, K., Ozdogan, M., Knyazikhin, Y., & Myneni, R. B. (2002a).
Multiscale analysis and validation of the MODIS LAI product over
Maun, Botswana: II. Sampling strategy. Remote Sensing of Environment, 83, 431 441.
Tobler, W., Deichman, U., Gottsegen, J., & Maloy, K. (1995). The global
demography project. UC Santa Barbara: NCGIA.
USGS, United States Geological Survey (1996). GTOPO30: Global 30 Arc
second elevation data set. Sioux Fallo, South Dakota, USA: Land
Processes Distributed Active Archive Centre, http://edcdaac.usgs.gov/
gtopo30/gtopo30.html.
Wang, Y., Woodcock, C. E., Buermann, W., Stenberg, P., Voipio, P., Smolander, H., Hame, T., Tian, Y., Hu, J., Knyazikhin, Y., & Myneni, R. B.
(in review). Validation of the MODIS LAI product in coniferous forest
of Ruokolahti, Finland. Remote Sensing of Environment.
Woodcock, C. E., Strahler, A. H., & Jupp, D. L. (1988). The use of variograms in remote sensing: I. Scene models and simulated images. Remote Sensing of Environment, 25, 323 343.

IKONOS imagery for the Large Scale BiosphereAtmosphere Experiment

in Amazonia (LBA)
George Hurtt a,b,*, Xiangming Xiao a, Michael Keller a,c, Michael Palace a, Gregory P. Asner d,
Rob Braswell a, Eduardo S. Brondzio e, Manoel Cardoso a, Claudio J.R. Carvalho f,
Matthew G. Fearon a, Liane Guild g, Steve Hagen a, Scott Hetrick h, Berrien Moore III a,
Carlos Nobre i, Jane M. Read j, Tatiana Sa f, Annette Schloss a,
George Vourlitis k, Albertus J. Wickel f,l
a
Institute for the Study of Earth Oceans and Space, University of New Hampshire, Durham, NH 03824, USA
b
Department of Natural Resources, University of New Hampshire, Durham, NH 03824, USA
c
USDA Forest Service, International Institute of Tropical Forestry, Rio Piedras, PR, USA
d
Department of Global Ecology, Carnegie Institution of Washington, Stanford University, Stanford, CA, USA
e
Department of Anthropology, Indiana University, Student Building 130 Bloomington, IN 47405, USA
f
EMBRAPA Amazonia Oriental, Belem, PA 66095-100, Brazil
g
Ecosystem Science and Technology, NASA Ames Research Center, Moffett Field, CA 94035, USA
h
Anthropological Center for Training and Research on Global Environmental Change Indiana University, Bloomington, IN 47405, USA
i
CPTEC/INPE Instituto Nacional de Pesquisas Espaciais, Cachoeira Paulista, SP 12630-000 Brazil
j
Department of Geography, Maxwell School, Syracuse University, Syracuse, NY 13244, USA
k
Biological Sciences Department, California State University, San Marcos, CA 92096, USA
l
Center for Development Research (ZEF), Department of Ecology and Resource Management, University of Bonn, D-53113 Bonn, Germany
Received 19 July 2002; received in revised form 12 September 2002; accepted 22 April 2003
Abstract
The LBA-ECO program is one of several international research components under the Brazilian-led Large Scale Biosphere Atmosphere
Experiment in Amazonia (LBA). The field-oriented research activities of this study are organized along transects and include a set of primary
field sites, where the major objective is to study land-use change and ecosystem dynamics, and a smaller set of 15 operational eddy flux tower
sites, where the major objective is to quantify net exchange of CO2 with the atmosphere. To supplement these studies and help to address
issues of fine-scale spatial heterogeneity and scaling, high-resolution satellite imagery (IKONOS, 1 4 m) have been acquired over some of
these study sites. This paper begins with a description of the acquisition strategy and IKONOS holdings for LBA. This section is followed
with a review of some of the most promising new applications of these data in LBA.
Keywords: IKONOS; Remote sensing; Spatial heterogeneity; Land use; Land cover; LBA
1. Introduction
Tropical deforestation in Amazonia has been a growing
ecological and climatological concern for several years
(Fearnside, 1990; Nepstad et al., 1999; Skole & Tucker,
1993). In 1997, the Large Scale Biosphere Atmosphere
* Corresponding author. Institute for the Study of Earth Oceans and
Space, University of New Hampshire, Durham, NH 03824, USA. Tel.: +1603-862-4185; fax: +1-603-862-0188.
E-mail address: george.hurtt@unh.edu (G. Hurtt).
doi:10.1016/j.rse.2003.04.004
Experiment in Amazonia (LBA) was initiated as an international research initiative led by Brazil with strong United
States and European Union participation. LBA is designed
to understand the climatological, ecological, biogeochemical, and hydrological functioning of Amazonia, the impact
of land-use change on these functions, and the interactions
between Amazonia and the earth system. The scientific
questions of LBA are: (1) How does Amazonia currently
function as a regional entity? (2) How will change in land
use and climate affect the biological, chemical and physical
functions of Amazonia, including the sustainability of
112
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111127
Table 1
Summary statistics on IKONOS holdings for LBA
Total
Total
Total
Total
Total
Total
Final
Beta
Complete
Final Unique
Beta Unique
Unique
f Area (km2)
5268
1540
6808
3761
1491
5252
71
19
90
46
18
64
Beta and Final are the two data versions. Beta is a preliminary data product.
development in the region and the influence of Amazonia

on global climate (The LBA Science Planning Group,
1996)?
In order to address these and related scientific questions,
multidisciplinary investigations at a variety of spatial and
temporal scales are underway. Knowledge of phenomena
such as organic-matter decomposition, photosynthesis,
plant-community succession, and large-scale patterns of fire,
land use, and land abandonment must be understood and
integrated. To this end, data are being collected across a
range of scales, from short-term leaf-level measurements, to
forest-stand statistics, to remote sensing over the entire
region. New models are also being developed to integrate
observations and concepts across biological, temporal, and
spatial scales. In addition, new remote sensing instruments
including MODIS, MISR, and EO-1 Hyperion are being
brought to bear on the research topics of the LBA program.
One new instrument being employed in LBA is the IKONOS satellite.
Launched from Vandenberg Air Force Base in September

1999, the IKONOS satellite is the worlds first commercial
instrument to collect imagery at 1-m resolution. IKONOS is
polar orbiting, sun-synchronous, and has a 98.1j orbital
inclination relative to the earth. Images have one panchromatic band at 1-m resolution and four spectral bands (blue,
green, red, near-infrared, similar to Landsat TM spectral
bands 1 4) at 4-m resolution. With the support of the
NASA Terrestrial Ecology and NASA Scientific Data Purchase (SDP) programs, an ongoing effort to obtain an
IKONOS database of Amazonia has become an important
part of LBA. This paper begins with a description of the
acquisition strategy and IKONOS holdings for LBA. This
discussion is followed with a summary of some of the most
promising new applications of these data in LBA. Highresolution remote sensing data such as from IKONOS are
powerful tools for helping to meet LBA objectives and
should be included in future project planning.
2. Initial acquisition strategy and database

The initial strategy for acquiring IKONOS data for the
LBA project has focused on obtaining imagery at key LBA
study sites. For the purposes of this study, sites can be
categorized into two groups: eddy-flux tower sites and fieldstudy sites. Eddy-flux tower sites are foci for local estimates
of CO2 flux between the atmosphere and land surface and
other important ecophysiological measurements. In addition, they are typically the focus of many supporting
Fig. 1. Locations of IKONOS images for LBA-Ecology study sites acquired. Open stars represent locations of eddy-flux tower sites, and solid stars represent
field study sites. Map-ID labels for each star are used to cross reference site names listed in Appendices A and B.
ground-based studies. Field-study sites are mainly used for

ground-based studies of ecosystem structure and dynamics,
land-use change, and recovery following disturbance. Finescale disturbances such as selective logging are also studied
at some of these sites.
Input from the LBA research community was solicited
and used to obtain specific coordinates for sites suitable for
IKONOS tasking. To date, imagery requested by 19 differ-
113
ent investigation teams has been tasked and acquired. In

some cases, repeat images of the same location have been
obtained for studies of land-cover change. In all, 90 images
covering over 6808 km2 have been acquired, consisting of
approximately 5268 km2 at the final product level (Table 1,
Fig. 1, Appendices A and B). Fig. 2 provides examples of
IKONOS imagery collected for two key study regions
outside the cities of Manaus and Santarem.
Fig. 2. Examples of IKONOS imagery for two important LBA study regions near the cities of Manaus and Santarem. The images displayed are false color
representations of multispectral (4 m) imagery and, with the exception of Santarem Kilo83 11.14.01, cover a domain size of 7 7 km. Greater detail is
available in larger-scale (i.e. smaller domain) subsets of these images. White patches, which are clouds, can be seen in several images along with cloud
shadows. Standard tasking criteria require images to have less than 10% cloud coverage. This figure includes materialn Space Imaging L.P.
114
Table 2
Cumulative statistics from EOS-WEBSTER (http://www.eos-webster.sr.unh.
edu) pertaining to IKONOS holdings for LBA
Number of investigators/team requests
Number of registered IKONOS users
Number of visitors to IKONOS subsystem
Number of IKONOS products ordered
IKONOS data available to users (GB)
IKONOS data distributed (GB)
19
47
1248
671
16.6
87.6
Data first became available on EOS-WEBSTER in 2001.
All IKONOS holdings collected for LBA are available to

LBA researchers following data-sharing guidelines established between NASA and Space Imaging (Thorton, CO,
USA). The data are listed in the LBA Data Information
System Beija-Flor and are featured holdings at the NASA
Earth Science Information Partner EOS-WEBSTER (http://
www.eos-webster.sr.unh.edu). To date, EOS-WEBSTER has
nearly 50 registered IKONOS users and has had more than
1200 visitors to the IKONOS data collection during the past
year (Table 2).
3. Promising applications of IKONOS imagery in LBA

The primary motivation for incorporating IKONOS imagery in the LBA Experiment is for enhancing studies of
fine-scale heterogeneity in vegetation, land use, and landuse change. Multispectral imagery with a resolution of 4 m
and panchromatic with a resolution of 1 m can potentially
provide valuable information on the composition of vegetation and specific details of land-use changes. In LBA,
these data are being used to address a variety of different
research topics including:

Improvement of land-cover remote sensing;

Estimates of forest crown diameter, basal area, and
biomass;
Detection and quantification of selective logging;
Effects of fine-scale heterogeneity in vegetation on
carbon balance;
Detection of biochemical and biophysical changes on
small land holdings;
Mapping agroforestry systems and small-scale agriculture.
This paper reviews some of the most promising new

research in LBA using IKONOS data on these topics. This
research is at an early stage of development, as IKONOS
has been introduced only recently to LBA. The review is
intended to summarize and synthesize ongoing research at
the early stage of development when feedback and sharing
of ideas can be most valuable.
3.1. Improvement of land-cover remote sensing
Having precise estimates of the area coverage of different land cover types is important to carbon cycle and land
surface studies. Medium- to coarse-resolution optical sensors (e.g., AVHRR, Terra-1 MODIS, SPOT-4 VEGETATION) provide daily observations of land cover for the
globe. However, the land surface in the Amazon is often a
mixture of land-cover types at the spatial resolution of these
sensors (0.5 1 km). Quantifying the fractional coverage of
broad land-cover types within large pixels by extrapolating
knowledge that is obtainable by visual or automated
interpretation of IKONOS is a remote sensing problem of
current interest to LBA research. All of the approaches
discussed below rely on the development of scaling relationships between the spectral information in regional-scale
data and the spatial information in high resolution data
(IKONOS in this case), a process known as spectral
unmixing.
A number of spectral unmixing approaches have been
developed to estimate subgrid-scale land cover fractions. In
general, the methods are either based either on linear
mixture modeling (Boardman, 1989; Cross, Settle, Drake,
& Paivinen, 1991; Smith, 1990), or on nonlinear regression
using artificial neural networks (e.g. Foody, Lucas, Curran,
& Honzak, 1997). In traditional spectral mixture modeling,
the surface reflectance at each pixel of the image is assumed
to be a linear combination of the reflectance of each endmember present within the pixel. For an image pixel that has
M spectral bands and N end-members, the spectral linear
unmixing model is described in the following equation:
Ri
N
X
rki fk ei
k1
where Ri is the reflectance of spectral band i in a multispectral image (i = 1, 2, . . ., M), fk is the fractional cover of
end-member k (k = 1, 2, . . ., N) within a pixel, rki is the
reflectance of end-member k at spectral band i, and ei is the
residual term. Nonlinear regression takes the converse
approach, directly modeling the fractions as a function of
reflectance fk = Gk(R), where G is normally defined by a
simple back-propagation neural network (Foody et al.,
1997). To avoid overfitting, a Bayesian modification can
be made to these algorithms (Bishop, 1995; Braswell et al.,
2000) so that all the available high-resolution data can be
used in training the model.
Some studies have explored the use of linear spectral
unmixing of AVHRR data for sub-pixel characterization of
cropland (Quarmby, Townshend, Settle, & White, 1992) and
tropical forests (Cross et al., 1991). These studies have had
some success, but have been limited by sensor resolution
and the lack of relevant spectral bands in AVHRR data (only
red and near infrared bands for vegetation). The MODIS
sensor has higher resolution and more spectral bands related
to land cover and vegetation (2 bands at 250 m resolution
plus 5 bands at 500 m resolution). Because of the increased
resolution and added bands, research using MODIS data has
the potential to provide improved estimates of land cover
and land-cover changes.
Recently, Xiao et al. have begun using IKONOS

images to improve the spectral definition of end-members
(Fig. 3). In an exploratory study, a 7 7 km IKONOS
image acquired on April 6, 2000 covering the Jaru eddyflux Tower in Rondonia was used to select end-members
of six land cover types (primary forest, secondary forest,
crop, pasture, river, soil) through visual interpretation and
delineation of region of interest (ROI) for each land cover
type. The resulting end-members of individual land cover
types were co-registered and reconciled with an MODIS
image (standard 8-day composite product of surface reflectance at 500 m spatial resolution, July 20 26, 2000).
Spectral unmixing analysis of the MODIS image was then
conducted, using a spectral linear unmixing algorithm
implemented in the commercial software (ENVI version
3.2), which is based on earlier works of Boardman (1989,
1992). The resulting MODIS-based fractional coverage
maps of land-cover types compared favorably to Landsat
7 ETM+ data acquired on August 6, 1999. By using
IKONOS data at select sites to improve region-wide land
cover algorithms that rely on MODIS data, this research
115
may lead to improved land-cover mapping across the

entire Amazon basin benefiting both carbon cycle science
and land surface studies. Additional advances may come
from greater use of non-linear or Bayseian statistical
methods.
A second application addresses the issue of shadows in
remotely sensed images. The biological and structural
complexity of tropical forests and savannas results in
marked spatial variation in shadows inherent to remotely
sensed measurements. While the biophysical and observational factors driving variations in apparent shadow are
known, little quantitative information exists on the magnitude and variability of shadow in remotely sensed data
acquired over tropical regions. Even less is known about
shadow effects on multispectral observations from satellite
instruments often employed in tropical studies (e.g., Landsat). The IKONOS satellite, with 1 m panchromatic and 4 m
multispectral capabilities, provides an opportunity to observe tropical canopies and their shadows at spatial scales
approaching the size of individual crowns and vegetation
clusters.
Fig. 3. A comparison of IKONOS, Landsat 7 ETM+ and MODIS images. Upper left panel-IKONOS image acquired on April 6, 2000, false color composite
(band 4-3-2). Lower left panel-Landsat 7 ETM+ image acquired on September 6, 1999, false color composite (band 4-5-3). The white box illustrates the
domain of the IKONOS image in the upper left. Upper right panelMODIS image (8-day composite of surface reflectance, July 20 26, 2000), false color
composite (band 2-6-1). Lower right panelFractional cover of forest within 500 m pixels, derived from spectral mixture analysis of MODIS data. This figure
includes materialn Space Imaging L.P.
116
Asner and Warner (in review) used 44 IKONOS images

from the LBA data archive to quantify the spatial variation of
canopy shadow fraction across a broad range of forests in the
Amazon and savannas in the Cerrado. Forests had substantial
apparent shadow fractions as viewed from the satellite
vantage point. The global mean ( F S.D.) shadow fraction
was 0.25 F 0.12, and within-scene (e.g., forest stand) variability was similar to inter-scene (e.g., regional) variation.
The distribution of shadow fractions for forest stands was
skewed, with 30% of pixels having fractional shadow values
above 0.30 (Fig. 4). Shadow fractions in savannas increased
from 0.0 F 0.01 to 0.12 F 0.04 to 0.16 F 0.05 for areas with
woody vegetation at low ( < 25% cover), medium (25 75%)
and high (>75%) density, respectively. Landsat-like observations using both red (0.63 0.70 Am) and NIR (0.76 0.85
Am) wavelength regions were highly sensitive to sub-pixel
shadow fractions in tropical forests, accounting for f 30
50% of the variance in red and NIR responses. Asner and
Warner also found that a 10% increase in shadow fraction
resulted in a 3% and 10% decrease in red and NIR channel
response, respectively. However, the NDVI of the Amazon
forests was weakly sensitive to changes in shadow fraction.
For low-, medium- and high-density savannas, a 10%
increase in shadow fraction resulted in a 5 7% decrease in
red-channel response. Shadows accounted for f15 50% of
the overall variance in red-wavelength responses in the
savanna image archive. Weak or no relationships occurred
between shadow fraction and either NIR reflectance or the
NDVI of Brazilian savannas. Quantitative information on
shadowing is needed to validate or constrain radiative

transfer, spectral mixture and land-surface models of the
Amazonian biosphere and atmosphere.
3.2. Estimates of forest crown diameter, basal area, and
biomass
In forests, allometric relationships exist between different
measurements of tree size such as tree diameter at breast
height (DBH), tree height, crown diameter, and tree biomass.
Foresters and ecologists often use these relationships as a
basis for estimating important quantities such as biomass or
carbon stock from relatively simple ground-based measurements of DBH and/or height (Araujo, Higuchi, & Carvalho,
1999; Brown, 1997; Keller, Palace, & Hurtt, 2001). Remotely
sensed estimates of vegetation height using lidar have been
used to estimate important fine scale characteristics of vegetation including vegetation biomass (Drake, Dubayah,
Clark, Knox, & Blair, 2002; Drake, Dubayah, Clark, Knox,
Blair, Hofton et al., 2002; Harding, Lefsky, Parker, & Blair,
2001; Lefsky, Harding, Cohen, Parker, & Shugart, 1999;
Means et al., 1999). In addition, high-resolution imagery
based on aerial photographs and videography, declassified
historical reconnaissance imagery, and other satellite sensors
have been used in analyses to estimate attributes of forests
including properties of gaps, stand age, height, and biomass
(e.g. Bradshaw & Spies, 1992; Cohen & Spies, 1992; Cohen,
Spies, & Bradshaw, 1990; Shugart, Bourgeau-Chavez, &
Kasischke, 2000). The IKONOS satellite adds a new resource
Fig. 4. Histogram of shadow fractions from 29 IKONOS pan-chromatic images of Amazon tropical forests. Descriptive statistics are provided in upper right.
From Asner and Warner (submitted for publication).
with which to advance these studies using contemporary

high-resolution imagery.
Read, Clark, Venticinque, and Moreira (submitted for
publication) conducted a pilot study at a reduced-impact
logging operation in tropical moist forest near Manaus,
Brazil to assess the potential application of IKONOS
satellite data to research and management of tropical forests.
As part of this study the authors assessed the feasibility of
using pan-sharpened IKONOS data for estimating forest
crown diameter. The pan-sharpening transformation exploits
the spatial resolution of the 1-m panchromatic data and the
spectral resolution of the four 4-m multispectral data by
merging the two to create a 1-m multispectral dataset.
Crown areas for nine clearly distinguishable trees were
derived by manually digitizing tree crowns from the pansharpened IKONOS image. Measured trunk diameters and
an index of crown area calculated from ground-based
measurements for the same nine trees were correlated with
the digitized crown area measurements. Trunk diameter was
found to be significantly correlated with digitized crown
area, suggesting that estimates of trunk diameter, and thus
biomass, may be possible with IKONOS data. The groundbased index of crown area although correlated yielded a
weaker relationship. In a similar study of tropical wet forest
in Costa Rica, Clark et al. (submitted for publication) found
highly significant correlations between crown area digitized
from IKONOS data and the same ground-based index of
crown area. In both studies, the trees studied were selected
for having clearly defined crowns on the IKONOS image,
and these correlations are likely to weaken for trees with less
defined crown edges or partially obscured crown sections,
or where shadows from neighboring trees obscure sections
of the crown. These observations highlight the importance
of further research on shadow effects and crown edge
detection methods.
In another set of studies, Asner et al. (submitted for
publication) have made ground-based measurements of
important tree parameters and compared these to IKONOS-based estimates with the goal of advancing and
applying this technology in LBA. In order to develop a
calibration data set, ground-based measurements of tree
height, crown diameter and crown depth in a lowland
tropical forest in the eastern Amazon, Brazil were first
obtained using a handheld laser rangefinder. The sample
included 300 trees stratified by diameter at breast height
(DBH). Significant relationships between DBH and both
tree height and crown diameter were then derived from the
ground-based measurements. Ground-based measurements
of crown diameter and area were then compared to estimates derived manually from the 1 m panchromatic IKONOS satellite imagery. The statistical distribution of crown
diameters from IKONOS was biased toward larger trees,
probably due to merging of smaller tree crowns, underestimation of understory trees, and over-estimation of
individual crown dimensions in the manual estimates
(Fig. 5a).
117
Building on this work, the team is developing new

automated techniques to identify properties of trees from
IKONOS imagery to replace the manual interpretations (Fig.
5b e). The approach scans through an images digital
numbers and uses a derivative calculation to test for the
edge of a crown in multiple directions from a seeded point.
The threshold, number of directions scanned, and method
for estimating crown shape are tunable in the program. The
method allows for gaps to be present in the analysis, and for
crowns to overlap, both of which are typically not done in
manual interpretation methods. Current work using the
program has proven promising in analyses of tropical forests
in the Cauaxi and Tapajos regions and will be expanded in
future work.
3.3. Detection and quantification of selective logging
The Amazonian tropical moist forest is experiencing
intensive logging in many regions (INPE, 2000). As roads
are built through the forest, large areas of previously
unlogged forest become attractive to the timber industry.
While causing less damage to the ecosystem than clearcutting, selective logging significantly changes the rainforest by creating canopy gaps, disturbing understory vegetation, and altering the litter layer on the forest floor.
Reduced impact logging is practiced in some areas, but
the relative damage caused by both types of logging has
only begun to be documented (Asner, Keller, Pereira, &
Zweede, 2002; Pereira, Zweede, Asner, & Keller, 2002). In
general, maps of logging areas are only preliminary (Nepstad et al., 1999). Read (in press) visually compared 15-m
pan-sharpened Landsat Enhanced Thematic Mapper
(ETM+) and 1-m pan-sharpened IKONOS images of selectively logged areas in a reduced-impact logging operation
near Manaus, Brazil. The coarser resolution ETM+ data
were capable of discriminating major logging roads and
logging decks (cleared areas in the forest for temporary
storage of logs) but the finer spatial resolution of the
IKONOS data was necessary for distinguishing smaller
logging features, including some minor logging roads and
treefall gaps.
Methods that analyze the spatial patterns of surface
features in remotely sensed data have been found to be
useful for characterizing a variety of landscape features, and
there is potential with such methods for detecting fine-scale
differences in forest canopy structure using the new higherresolution satellite data. Read (in press) compared performances of selected spatial methods using 15-m panchromatic
and 30-m multispectral Landsat-ETM+ data, and 1-m panchromatic and 4-m multispectral IKONOS data, for distinguishing differing degrees of canopy disturbances resulting
from reduced-impact logging at the site near Manaus. The
study reports the calculated fractal dimension, spatial autocorrelation and texture measures using visible, near-infrared
and NDVI data for each dataset at two plot sizes (10 and
335 ha) over a gradient of canopy disturbance. The distur-
118
Fig. 5. (a) Cumulative frequency distributions of crown area detected by various methods. IKONOS_current_paper refers to manual interpretation. VB
Logic Routine refers to automated method. (b d) Automated crown detection algorithm of an IKONOS image. (e) Number of trees vs. DBH class. Analyses
of Tapajos km 83 Tower area. This figure includes materialn Space Imaging L.P.
bance gradient comprised three treatments: areas of oldgrowth (undisturbed) forest, logged forest excluding major
roads and logging decks, and logged forest including major
roads and decks. In all cases the spatial methods were
successful at discriminating between the three treatments.
These results indicate that it may be possible to combine
spatial analyses of higher-resolution datasets over small
areas with coarser resolution datasets over large areas, for
assessing basin-wide fine-scale canopy disturbances such as
those resulting from selective logging.
Using 4 m multispectral IKONOS data, Hagen et al. have
been examining a section of the Tapajos National Forest,
south of Santarem, a Brazilian city in central Amazonia.
This field research area has been separated into 1 km plots,
some of which were selectively logged at various times in
the last f5 years. A spatial analysis of IKONOS data was
conducted in an attempt to automate the discrimination of
logged and intact forests. As an alternative to analyzing the
raw IKONOS panchromatic brightness data, spectral bands
were combined to calculate the normalized vegetation index
(NDVI). This vegetation index is related to the fraction of

absorbed photosynthetically active radiation and the leaf
area index of the forest (Myneni, Hall, Sellers, & Marshak,
1995). Logged areas can have reduced NDVI levels due to
an increase in shadow fraction and reduction in the amount
of green vegetation relative to soil and non-photosynthetic
vegetation. Several test areas were analyzed in an attempt to
quantify differences in harvested and unlogged areas. Each
area chosen was 50 50 pixels (200 200 m). Nine areas
were selected with no logging history and ten areas were
selected that have been logged.
The analysis involves two separate metrics. First, semivariograms were created for each block. The variogram
captures the spatial variation and length scales of the
vegetation in the block. A curve was then fit to the
empirical variogram and the parameters of this fitted
function were examined. As expected, the variogram
parameters for the two types of forest are similar, probably
because the basic spatial structure in the two forests is
essentially the same. Canopy diameters will dominate the
119
length scales identifiable in the variograms because most

trees in a block remain intact even after logging. However,
slightly higher sill (value of semivariance as distance
reaches its asymptotic value; Cressie, 1991) and longer
range (distance beyond which observations are least correlated; Cressie, 1991) for the logged areas were detectable. The higher sill is a direct result of the higher variance
in the block (low NDVI areas of shadow and soil signal,
mixed with the high NDVI of remaining canopies). The
longer range is likely due to disturbances in the regular
pattern of canopies. An irregular pattern reaches an asymptote (sill) at a longer range.
To create the second metric, binary images were created
for each block by setting an NDVI threshold, above which
pixels are assigned a value of one and below which pixels
are assigned a value of zero. The numbers of pixel
clusters, consisting of three or more pixels with values
of zero, were tallied. By setting an NDVI threshold and
creating a binary image for each block, relatively clear
separation between logged and unlogged plots is identifiable (Table 3). Using these two metrics in a combined
decision algorithm, suggests that recently logged forests
can be separated from non-logged forests imaged with
IKONOS, and that further investigation might also provide
insight into reduced impact effects versus traditional logging practices.
include selectively logged systems and cattle pasture. These

land forms are readily visible from multi-temporal, high
resolution IKONOS imagery, making this imagery useful
for locating potential study sites and quantifying annual
variations in forest re-growth and/or land-cover change (Fig.
6). The group has already used IKONOS data for estimating
the spatial and temporal spread of selective logging over the
last 2 years because the development of logging roads is
readily visible in the high-resolution imagery (Fig. 6). They
are currently trying to calibrate IKONOS and other imagery
(i.e., AVHRR) to ground-based measurements of the fraction of photosynthetically active radiation ( fPAR) absorbed
by the intact and selectively logged forest canopies to
provide a way to link the canopy physiological measurements obtained from eddy covariance towers (net CO2
exchange and evapotranspiration) to satellite data. Data
collected indicate that net CO2 exchange and evapotranspiration are closely linked to seasonal variations in leaf area
index (LAI), and thus, fPAR (Vourlitis et al., 2002, in press).
Because satellite reflectance data appear to be sensitive to
spatial and/or temporal variations of surface biophysical
features such as fPAR and LAI (Sellers et al., 1997), the
multi-temporal nature and high spatial resolution make
IKONOS a potentially powerful tool for estimating the
effects of land cover change on net CO2 exchange and
evapotranspiration of tropical transitional regions.
3.4. Effects of fine-scale heterogeneity in vegetation on

carbon balance
3.5. Detection of biochemical and biophysical changes on

small land holdings
Fine-scale heterogeneity in land cover, such as is introduced by fine-scale disturbance and selective logging, is
hypothesized to be important to the carbon balance of
ecosystems (Moorcroft, Hurtt, & Pacala, 2001; Nepstad et
al., 1999). Work by Vourlitis and Priante-Filho is expanding
measurements of the net CO2 exchange, evapotranspiration,
and energy balance of intact tropical transitional forest to
Secondary forests in Amazonia serve as a potentially

important carbon sink offsetting some of the carbon released
to the atmosphere from land-use conversion of primary
forest, and slash and burn agriculture. As the area affected
by land uses grows, the area and importance of secondary
forests is likely to grow. The functioning of secondary
forests, however, may be limited due to intensified land
use, shortened fallow periods, and increased vulnerability to
climate fluctuations (de Sa, de Oliveira et al., 1998; de Sa,
de Oliveira, de Araujo, & Brienza Junior, 1997). Guild, Sa
and others are studying the influence of seasonal and interannual climate fluctuations on secondary forest dynamics
and agricultural water, nutrient and productivity status under
contrasting land-use/conversion practices. The goal of the
research is to determine the extent to which secondary
forests and agricultural systems are differentially susceptible
to extreme climate events (e.g. droughts) under contrasting
land-use/clearing practices. A secondary goal is to determine the extent to which remote sensing can quantify
productivity, nutrient, and water relations in these secondary
forests and agricultural fields.
Alternative land-use/clearing practices (mulching and
fallow vegetation improvement) may make secondary forests and agriculture more resilient to the effects of agricultural pressures and drought through (1) increased biomass,
soil organic matter and associated increase in soil water
Table 3
The number of clusters identified in each of the 19 plots (9 logged and 10
not logged) using the Threshold-Binary Analysis with an optimal threshold
value (T ) of 0.41 and group size (GS) of three pixels
Threshold-binary analysis (T = 0.41, GS> = 3)
Mean
S.D.
Logged
Not logged
4
5
7
3
0
6
4
1
7
0
0
0
1
0
0
0
0
0
1
0.20
0.42
4.11
2.47
The logged plots have more clusters than the not logged plots.
120
Fig. 6. The ratio between NIR/red for April 2000 (top panel) and May 2001 (bottom panel) for the transitional forest site near Sinop, Mato Grosso. The lighter
surfaces (NIR/red = 0 1) show the recently cleared areas while darker areas (NIR/red > 3) show forest. This figure includes materialn Space Imaging L.P.
storage, and nutrient retention and (2) greater rooting depth

of trees planted for fallow improvement. New alternatives to
burning (mulching and fallow vegetation improvement) are
under study by the Brazilian agricultural research agency
(EMBRAPA) in eastern Amazonia. These practices include
cutting, chopping, and mulching secondary vegetation, and
fallow improvement by planting with fast-growing legumes.
This experimental practice (chop-and-mulch with enrichment) has resulted in increased soil moisture during the
cropping phase, reduced loss of nutrients and organic
matter, and higher rates of secondary forest biomass accu-
mulation (Brienza Junior, 1999; de Sa & Alegre, 2001; de

Sa, de Oliveira et al., 1998; de Sa et al., 1997; de Sa,
Vielhauer, Denich, Kanashiro, & Vlek, 1998; Sommer,
2000; Vielhauer et al., 2001). All of these practices improve
the sustainability of the land-use regime.
Due to the large area involved, operational monitoring in
Amazonia generally requires the use of Landsat, MODIS,
ASTER, AVHRR or other regional monitoring system.
However, the spatial resolution of these systems (30 1000
m) is insufficient to resolve several of the study plots
associated with small land holdings. Therefore high spatial
resolution satellite imagery from the IKONOS system is a

superior surrogate to address biochemical and biophysical
effects introduced by these small land-holding practices.
121
Radiometrically calibrated data from the IKONOS system

offers spatial resolution with spectral bands approximating
those of Landsat Thematic Mapper bands 1 4. In the
Fig. 7. Multi-scale analysis of Aca palm agroforestry system and floodplain forests in the Amazon estuary. Classified Landsat TM over IKONOS multispectral image (top), aerial view of agroforestry canopy (middle), ground view of intensively managed agroforestry (bottom). This figure includes materialn
Space Imaging L.P.
122
vicinity of Igarape-Acu, Para, Guild and Sa are converting

IKONOS imagery to NDVI or similar vegetation indices,
and comparing them to ground measurements of leaf area
index (LAI) and fPAR, which is expected to be influenced by
plot treatment effects. Ground measurements for this study
focus on the key factors of leaves and canopies of secondary
forest and crops that can indicate canopy condition for
remote detection of changes in biomass and drought effects.
These measurements include leaf water potential, stomatal
conductance, LAI, and leaf and canopy radiative properties.
Additional data includes vapor pressure deficit, soil moisture, temperature, and precipitation as these measurements
indicate the status of the mechanisms influencing drought
stress. Mechanisms by which alternative practices of chop/
mulch decrease vulnerability to drought include increased
soil organic matter, increased soil water holding capacity
during the cropping phase and increased rooting depth to
exploit soil water at depth during the fallow phase. It is
expected that these differences will be manifested through
LAI, canopy architecture, canopy water relations, and canopy biochemistry.
3.6. Mapping agroforestry systems and small-scale
agriculture
Mapping and characterizing Amazonian agriculture and
land-use systems are among the most challenging tasks
required to fulfill the mission of the LBA program. These
tasks relate to questions of fundamental importance and
provide research opportunities for scientists studying biogeochemical cycles and the human aspect of the program.
Low data availability (e.g., cloud cover during the production season), the spatial and temporal scales of production
systems vis-a`-vis the resolution of remotely sensed data,
similarities among land cover classes (e.g., agroforestry
systems and secondary succession) have limited our understanding of the regional agriculture systems and landuse cycles. This limitation has been particularly pervasive
for studies of managed agroforestry systems where
changes in vegetation structure are usually subtle, such
as changes in species composition and canopy architecture.
This is the case of the aca palm fruit production (Euterpe
oleracea Mart.), arguably the economically most important
land-use system of the Amazon estuary. In most studies,
this type of agroforestry system was not distinguished from
surrounding floodplain forests. Managed ac a palm forest
cover between 10,000 and 20,000 km2 in the Amazon
estuary. This is probably the most significant Amazonian
land-use system maintaining forests of high economic
output without deforestation and displacement of local
communities.
Previous work integrating Landsat TM, field inventories,
and land-use interviews has allowed discrimination between
unmanaged floodplain forests and intensively managed aca
palm agroforestry to a level of 81% accuracy (Brondzio,
Moran, Mausel, & Wu, 1994, 1996). Estimates of manage-
ment intensity based on ac a palms importance value (a

ranking index based on density, frequency, dominance) have
been successfully correlated to Landsat TMs NIR, MIR, and
red bands. IKONOS offers new opportunities to the understanding of this type of land-use system as well as associated
small-scale upland agricultural systems. IKONOS images
help to reveal the level of management intensity across
agroforestry areas, thus, providing means to estimate production and economic return of agroforestry stands. In
addition, IKONOS images may help to refine the accuracy
of Landsat-based agroforestry mapping, particularly by contributing to the delineation between areas under different
management intensities as well as to quantify changes in the
structure of forest canopies. Another important application of
IKONOS in this area refers to the analysis of swidden
agricultural systems (shifting cultivation) and associated
stages of secondary vegetation regrowth. IKONOS images
can also be used in the delineation of riverine house-gardens
important to the understanding of resource management
strategies in the Amazon estuary. These are key features
for the analysis of land-use intensification and settlement
patterns in the region.
Building upon previous research, data from IKONOS
images are being used to correlate to field inventories of
floodplain agroforestry areas managed under different intensities, as well as secondary succession in different stages
of regrowth. Measures of vegetation structure and species
composition are compared to textural and spectral indices in
IKONOS and ETM+ data re-sampled at different spatial
resolution. Also, IKONOS data are being applied to the
definition of shifting cultivation cycles and to improve the
delineation of small-scale agriculture not visible in Landsat
TM and ETM+ data. Preliminary results using IKONOS
images in multiscale analyses (Fig. 7) indicate improvements in the definition of intensive and intermediary managed agroforestry areas and in the definition of vegetation
regrowth stages.
4. Discussion and conclusions

IKONOS tasking for the Large Scale Biosphere Atmosphere Experiment in Amazonia has focused on the acquisition of imagery at a set of key study sites. The data
acquired to date are currently being used by a number of
researchers to address a set of earth science applications that
can benefit from high-resolution. These applications are of
high priority to the international scientific community and
the LBA project in particular and include strategies for
improving estimates of terrestrial carbon stocks and fluxes
as well as estimates of land-use and land-cover changes
(Cerri et al., 1995; Prentice et al., 2001; The LBA Science
Planning Group, 1996).
As LBA moves forward, it is necessary to consider the
adequacy of the current set of holdings and potential
strategies for future tasking of high-resolution instruments.
123
further. With a current unique sample of approximately

3300 km2 in an area of nearly 5,000,000 km2, current
IKONOS holdings represent a small and incomplete sample of this study area (< 0.1%).
Given the usefulness of IKONOS data and the limited
size of the current set of holdings, how should future
tasking be planned? Collecting imagery over the entire
domain of Amazonia is prohibitive. For example, at current
prices a single coverage of the area of the size of Amazonia
would cost nearly US$1 billion and contain on the order of
104 GB of data. To make matters worse, repeat imagery are
needed for change detection, inflating these potential costs
and data volumes even more. The current strategy for
tasking IKONOS data has been to focus on key study
sites. To qualitatively improve the tasking strategy, an effort
is needed to design and implement an objective and
efficient sampling strategy. Objective and efficient sampling methods have been used in the context of groundbased sample measurements (e.g. USDA Forest Service,
1992). They are also necessary in high-resolution remote
sensing.
Understanding the importance of fine-scale heterogeneity over large geographical areas is a central challenge in
earth sciences that arises in many sub-disciplines (Ehleringer & Field, 1993; Hurtt, Moorcroft, Pacala, & Levin,
1998; Moorcroft et al., 2001). Toward this goal, new
models that formally address issues of scaling are being
developed (Moorcroft et al., 2001), and new technologies
for obtaining high-resolution data remotely, such as IKONOS, are being developed and deployed. A critical challenge ahead will be in managing model complexity and
efficiently organizing high-resolution information. Remotesensing technologies that provide needed information are
perhaps nowhere more important than in remote regions
such as Amazonia, where ground-based infrastructure and
access are limited.
Fig. 8. (a) Brazilian Legal Amazon (Grey), and IKONOS sample locations
(dots). (b) IKONOS sample distributions (dotted curves) and region-wide
distributions (solid curves) of average annual conditions for several
ecologically important environmental conditions. Conditions displayed
include temperature (TEMP), dew point (DEWP), wind speed (WSPD),
photosynthetically active radiation (PAR), and precipitation (PRECIP).
Environmental conditions are the average of statistics for 1987 and 1988
complied in ISLSCP I (Meeson et al., 1995; Sellers et al., 1995).
One way to begin to assess the adequacy of current

holdings is to calculate statistics on the representativeness
of current holdings. For example, Fig. 8 presents sample
distributions of several ecologically important climate
variables defined by locations with IKONOS imagery,
compared to those of the entire Brazilian Legal Amazon
domain. As expected from a limited sample, these distributions tend to over-represent relatively common climatic
conditions and under sample rare climatic conditions.
Adding factors to this analysis such as soil type, topography, biogeography, and land-use history is likely to dilute
the representativeness of the current set of holdings even
Acknowledgements
Much of the material for this paper was organized for
a series of lectures given at the High-Spatial Resolution
Commercial Imagery Workshops held at USGS Headquarters in 2001 and 2002. Support for this research was
provided by NASA through the Terrestrial Ecology, LandUse Land-Cover Change, New Millennium, Interdisciplinary Science, and Scientific Data Purchase Programs. D.
Clark and three anonymous reviewers provided suggestions that greatly improved this manuscript. D. Blaha, M.
Routhier, and S. Spencer provided assistance with
graphics and data management. The IKONOS data
referenced in this paper are hosted and made available
to the scientific community via the NASA Earth Science
Information Partner EOS-WEBSTER. (http://www.eoswebster.sr.unh.edu). This paper includes materialn Space
Imaging L.P.
124
Appendix A. Database of IKONOS Tower Site images for LBA (http://www.eos-webster.sr.unh.edu)
Map
ID
Name
Country
lat nw
lat sw
lat ne
lat se
lon nw
lon sw
lon ne
lon se
area
(km2)
Date
acquired
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
guas Emendadas
A
guas Emendadas
A
guas Emendadas
A
Braslia East
Braslia West
Caxiuana
Caxiuana
Cerrado
Cerrado
Fazenda Nossa Senhora
Fazenda Nossa Senhora
IBGE Campo Sujo
IBGE Campo Sujo
Jaru Tower
Jaru Tower
Jaru Tower
Manaus 1
Manaus 1
Manaus 2
Manaus 2
Santarem Kilo 67
Santarem Kilo 67
Santarem Kilo 83
Santarem Kilo 83
Santarem Kilo 83
Santarem Kilo 83
Santarem Kilo 83
Santarem Pasture
Santarem Pasture
Sinop Mato Grosso
Sinop Mato Grosso
Sinop Mato Grosso
Sugarcane
Sugarcane
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
15.52
15.52
15.50
15.62
15.58
1.72
1.72
21.63
21.63
10.73
10.73
15.92
15.92
10.03
10.05
10.05
2.56
2.56
2.58
2.58
2.83
2.83
2.99
3.01
2.99
3.04
3.05
2.99
2.99
11.38
11.38
11.38
21.07
21.07
15.58
15.58
15.60
15.72
15.68
1.78
1.78
21.70
21.70
10.79
10.79
15.98
15.98
10.13
10.11
10.11
2.62
2.62
2.64
2.64
2.89
2.89
3.05
3.05
3.05
3.07
3.08
3.05
3.05
11.44
11.44
11.44
21.13
21.13
15.52
15.52
15.50
15.62
15.58
1.72
1.72
21.64
21.63
10.73
10.73
15.92
15.92
10.03
10.05
10.05
2.56
2.56
2.58
2.58
2.83
2.83
2.99
3.01
2.99
3.04
3.05
2.99
2.99
11.38
11.38
11.38
21.07
21.07
15.58
15.58
15.60
15.73
15.68
1.78
1.78
21.70
21.70
10.79
10.79
15.98
15.99
10.13
10.11
10.11
2.62
2.62
2.64
2.64
2.89
2.89
3.05
3.05
3.05
3.07
3.08
3.05
3.05
11.44
11.44
11.44
21.13
21.13
47.63
47.63
47.65
47.99
48.08
51.49
51.49
47.65
47.65
62.39
62.39
47.90
47.90
61.98
61.97
61.96
60.15
60.15
60.24
60.24
54.99
54.99
55.00
55.00
55.00
55.01
55.00
54.92
54.92
55.36
55.36
55.36
48.10
48.10
47.63
47.63
47.65
47.99
48.07
51.49
51.49
47.65
47.65
62.39
62.39
47.90
47.90
61.98
61.96
61.96
60.15
60.15
60.24
60.24
54.99
54.99
55.00
55.00
55.00
55.01
55.00
54.92
54.92
55.36
55.36
55.36
48.10
48.10
47.57
47.57
47.54
47.89
47.97
51.42
51.42
47.58
47.58
62.33
62.33
47.84
47.84
61.88
61.90
61.90
60.08
60.08
60.18
60.18
54.93
54.93
54.94
54.94
54.94
54.94
54.94
54.86
54.86
55.29
55.29
55.29
48.03
48.03
47.57
47.57
47.55
47.89
47.97
51.42
51.42
47.58
47.58
62.33
62.33
47.84
47.84
61.88
61.90
61.90
60.08
60.08
60.18
60.18
54.93
54.93
54.94
54.94
54.94
54.94
54.94
54.86
54.86
55.29
55.29
55.29
48.03
48.03
49.06
49.57
123.71
124.10
124.17
49.06
48.83
49.03
50.43
49.03
49.43
49.06
49.64
110.29
49.06
48.43
49.06
49.13
49.03
49.11
49.06
49.03
49.03
33.06
49.03
25.86
23.54
49.00
49.02
49.03
49.73
49.42
49.03
50.80
6/23/2000
6/23/2001
6/1/2001
6/1/2001
6/1/2001
6/7/2000
9/3/2001
5/29/2000
5/29/2001
8/2/2000
8/2/2001
5/29/2000
5/29/2001
4/6/2000
5/17/2000
5/28/2001
7/22/2000
8/2/2001
8/24/2000
10/7/2001
6/21/2000
11/3/2001
8/29/2000
11/14/2001
11/19/2001
8/29/2000
7/5/2002
6/13/2000
7/27/2001
4/30/2000
5/19/2001
7/5/2002
6/12/2000
6/12/2001
Appendix B. Database of IKONOS Field Site images for LBA (http://www.eos-webster.sr.unh.edu)

Map
ID
Name
Country
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Altamira East
Altamira West
Alto Paraso 1
Alto Paraso 2
Alto Paraso 3
Apeu
Cangussu
Cauaxi
Cauaxi
Cauaxi C
Cauaxi R (Scene A)
Cauaxi R (Scene B)
Cauaxi R (Scene C)
Ecuador 1 East
Ecuador 1 West
Ecuador 2 East
Ecuador 2 West
Ecuador 3 East
Ecuador 3 West
Ecuador 4
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Ecuador
Ecuador
Ecuador
Ecuador
Ecuador
Ecuador
Ecuador
lat nw
3.17
3.17
9.65
9.42
9.61
1.27
9.95
3.69
3.69
3.61
3.71
3.71
3.71
0.21
0.21
0.19
0.19
0.19
0.19
0.21
lat sw
3.31
3.31
9.73
9.50
9.70
1.34
10.01
3.82
3.82
3.65
3.75
3.75
3.75
0.09
0.09
0.07
0.07
0.31
0.31
0.33
lat ne
3.17
3.17
9.65
9.42
9.61
1.27
9.95
3.69
3.69
3.61
3.71
3.71
3.71
0.21
0.21
0.19
0.19
0.19
0.19
0.21
lat se
lon nw
lon sw
lon ne
lon se
Area
(km2)
Date
acquired
3.31
3.31
9.73
9.50
9.70
1.34
10.01
3.82
3.82
3.65
3.75
3.75
3.75
0.09
0.09
0.07
0.07
0.31
0.31
0.33
52.26
52.35
63.19
63.35
63.29
48.00
50.04
48.33
48.33
48.50
48.43
48.43
48.43
76.84
76.94
76.28
76.30
76.92
76.94
76.60
52.26
52.35
63.19
63.35
63.29
48.01
50.04
48.33
48.33
48.50
48.43
48.43
48.43
76.84
76.94
76.28
76.30
76.92
76.94
76.60
52.17
52.26
63.15
63.30
63.25
47.94
49.98
48.27
48.27
48.39
48.37
48.37
48.37
76.82
76.82
76.18
76.20
76.82
76.91
76.49
52.17
52.26
63.15
63.30
63.25
47.94
49.97
48.27
48.27
48.39
48.37
48.37
48.37
76.82
76.82
76.18
76.20
76.82
76.91
76.49
171.32
162.14
45.03
50.63
50.19
49.06
49.00
91.51
91.51
48.76
28.10
28.10
28.10
23.74
76.97
154.42
152.19
146.04
34.84
160.59
10/14/2000
10/14/2000
7/25/2002
7/14/2002
7/17/2002
7/10/2001
7/7/2002
11/2/2000
6/18/2002
7/7/2002
8/9/2002
8/20/2002
8/23/2002
1/13/2002
1/5/2002
1/10/2002
2/15/2002
10/25/2000
10/25/2000
8/26/2001
125
Appendix B (continued)
Map
ID
Name
Country
lat nw
lat sw
lat ne
lat se
lon nw
lon sw
lon ne
lon se
Area
(km2)
Date
acquired
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
Ecuador 4 East
Ecuador 4 West
Fazenda Nova Vida
Fazenda Vitoria
Igarape-Acu 2
Igarape-Acu 2
Igarape-Acu 2
Igarape-Acu 2
Igarape-Acu 2
Ituqui, Brazil scene A
Ituqui, Brazil scene B
Ituqui, Brazil scene C
Ituqui, Brazil scene D
Ituqui, Brazil scene E
Machadinho
Mato Grosso
Medicilandia
Mil Madeiras
Pastaza
Pilche
Ponta de Pedras East
Ponta de Pedras West
Rohden East
Rohden Northeast
Rohden West
Ruropolis East
Ruropolis West
Santarem FLONA
Tapajos
Sao Francisco do
Para East
Sao Francisco do
Para East
Sao Francisco do
Para West
Sewaya
Tiguano
Tome-Acu East
Tome-Acu West
Zabalo
Ecuador
Ecuador
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Ecuador
Ecuador
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
Brazil
0.21
0.21
10.11
2.93
1.15
1.15
1.15
1.15
1.12
2.51
2.51
2.51
2.51
2.51
9.43
9.57
3.42
2.79
0.07
0.26
0.19
1.34
10.53
10.44
10.53
4.06
4.06
2.99
0.33
0.33
10.17
3.00
1.22
1.22
1.22
1.22
1.21
2.68
2.68
2.68
2.68
2.68
9.54
9.63
3.47
2.89
0.05
0.32
0.31
1.45
10.56
10.52
10.56
4.09
4.09
3.08
0.21
0.21
10.11
2.93
1.15
1.15
1.15
1.15
1.12
2.51
2.51
2.51
2.51
2.51
9.43
9.57
3.42
2.79
0.07
0.26
0.19
1.34
10.53
10.45
10.53
4.06
4.06
2.99
0.33
0.33
10.17
3.00
1.22
1.22
1.22
1.22
1.21
2.68
2.68
2.68
2.68
2.68
9.54
9.63
3.47
2.89
0.05
0.32
0.31
1.45
10.56
10.52
10.56
4.09
4.09
3.08
76.51
76.61
62.84
47.45
47.61
47.61
47.61
47.61
47.62
54.33
54.24
54.39
54.50
54.50
62.11
55.97
52.92
58.84
77.14
76.27
76.92
48.92
58.52
58.47
58.59
54.83
54.85
55.00
76.51
76.61
62.84
47.45
47.61
47.61
47.61
47.61
47.63
54.32
54.24
54.39
54.50
54.50
62.11
55.97
52.92
58.84
77.14
76.27
76.92
48.92
58.52
58.48
58.59
54.83
54.85
55.00
76.49
76.51
62.75
47.37
47.54
47.54
47.54
47.54
47.54
54.22
54.19
54.28
54.37
54.49
62.00
55.90
52.86
58.76
77.07
76.22
76.82
48.82
58.42
58.40
58.50
54.72
54.74
54.94
76.49
76.51
62.75
47.37
47.54
47.54
47.54
47.54
47.54
54.22
54.19
54.28
54.37
54.48
62.00
55.90
52.86
58.76
77.07
76.22
76.82
48.82
58.42
58.40
58.50
54.72
54.74
54.94
34.35
154.75
60.64
74.22
55.00
55.00
55.00
55.00
92.92
211.80
100.94
227.15
261.12
35.40
154.47
49.00
36.02
96.99
105.22
40.63
146.04
138.07
41.63
68.79
35.60
39.90
40.95
76.55
8/26/2001
8/26/2001
8/16/2001
7/29/2001
9/19/2001
10/3/2001
8/31/2002
9/19/2002
10/22/2002
7/27/2001
8/7/2001
8/7/2001
10/6/2001
10/9/2001
5/28/2001
7/5/2002
10/17/2002
7/25/2002
1/16/2002
1/10/2002
12/19/2000
12/19/2000
7/8/2002
10/1/2002
7/8/2002
7/8/2002
7/5/2002
7/5/2002
Brazil
1.07
1.20
1.07
1.20
47.82
47.82
47.72
47.72
175.55
7/18/2001
Brazil
1.07
1.20
1.07
1.20
47.83
47.83
47.72
47.72
191.10
12/30/2001
Brazil
1.07
1.20
1.07
1.20
47.83
47.83
47.80
47.80
53.28
7/18/2001
Ecuador
Ecuador
Brazil
Brazil
Ecuador
0.15
0.42
2.34
2.34
0.19
0.20
0.48
2.46
2.46
0.24
0.15
0.42
2.34
2.34
0.19
0.20
0.48
2.46
2.46
0.24
76.20
76.51
54.21
54.29
75.43
76.20
76.51
54.20
54.29
75.43
76.14
76.45
54.17
54.18
75.37
76.14
76.45
54.17
54.18
75.37
35.86
35.86
49.60
151.69
35.86
2/4/2002
1/13/2002
7/27/2001
7/27/2001
2/9/2002
83
84
85
86
87
88
89
90
References
Araujo, T. M., Higuchi, N., & Carvalho, J. A. (1999). Comparison of
formulae for biomass content determination in a tropical rainforest site
in the state of Para, Brazil. Forest Ecology and Management, 117,
43 52.
Asner, G. P., Keller, M., Pereira, R., & Zweede, J. (2002). Remote sensing
of selective logging in Amazonia: Assessing limitations based on detailed field measurements, Landsat ETM+ and textural analysis. Remote
Sensing of Environment, 80(3), 483 496.
Asner, G. P., Palace, M., Keller, M., Pereira Jr., R., Silva, J. N. M., &
Zweede, J. C. (submitted for publication). Estimating canopy structure
in an Amazon forest from laser rangefinder and IKONOS satellite
observations. Biotropica.
Asner, G. P., & Warner, A. S. (submitted for publication). Canopy shadow
in IKONOS satellite observations of tropical forests and savannas. Remote Sensing of Environment.
Bishop, C. (1995). Neural networks for pattern recognition. New York:
Oxford University Press (504 pp.).
Boardman, J. W. (1989). Inversion of imaging spectrometry data using
singular value decomposition. IEEE International Geoscience and Remote Sensing Symposium, Vancouver, BC ( pp. 2069 2072).
Boardman, J. W. (1992). Sedimentary facies analysis using imaging spectrometry: A geophysical inverse problem. PhD Thesis, University of
Colorado, Boulder, CO, 212 pp.
Braswell, B. H., Linder, E., Hagen, S., Xiao, X., Frolking, S., Moore III, B.,
& Liu, J. (2000). A Bayesian unmixing algorithm for retrieving landcover distributions using global reflectance data. American Geophysical
Union Fall Meeting, San Francisco, CA.
Bradshaw, G. A., & Spies, T. A. (1992). Characterizing canopy gap structure in forests using wavelet analysis. Journal of Ecology, 80, 205 215.
Brienza Junior, S. (1999). Biomass dynamic of fallow vegetation enriched
with leguminous trees in the Eastern Amazon of Brazil. PhD Thesis,
Goettingen Goettinger Beitrage zur Land- und Forstwirtschaft in den
Tropen und Subtropen, 133 pp.
Brondzio, E. S., Moran, E. F., Mausel, P., & Wu, Y. (1994). Land use
change in the Amazon estuary: Patterns of Caboclo settlement and
landscape management. Human Ecology, 22(3), 249 278.
Brondzio, E. S., Moran, E. F., Mausel, P., & Wu, Y. (1996). Changes in
land cover in the Amazon estuary: Integration of thematic mapper with
126
botanical and historical data. Photogrammetric Engineering and Remote Sensing, 62(8), 921 929.
Brown, S. (1997). Estimating biomass and biomass change of tropical
forests: A primer. FAO Forestry Paper 134, United Nations Food and
Agriculture Organization, Rome, Italy.
Cerri, C., Niro, H., Melillo, J., Krug, T., Fernandes, E., Forsberg, B.,
Houghton, R., Keller, M., Martinelli, L., Nepstad, D., Nobre, A., Richey, J., Reynaldo, V., Crill, P., Davidson, E., De Mello, W., Melack, J.,
Mozeto, A., Skole, D., Soares, J. V., Sterberg, L., & Trumbore, S.
(1995). The ecological component of an integrated Amazon study (also
known as LBA): The effects of forest conversion. Workshop Report,
Manaus, Brazil, 55 pp.
Clark, D. B., Read, J. M., Clark, M., Cruz, A. M., Dotti, M. F., & Clark,
D. A. (submitted for publication). Application of 1-m and 4-m resolution satellite data to studies of tree demography, stand structure and
land-use classification in tropical rain forest landscapes. Ecological
Applications.
Cohen, W. B., & Spies, T. A. (1992). Estimating structural attributes of
Douglas-fir/western hemlock forests stands from Landsat and SPOT
imagery. Remote Sensing of the Environment, 41, 1 17.
Cohen, W. B., Spies, T. A., & Bradshaw, G. A. (1990). Semivariograms of
digital imagery for analysis of conifer forest structure. Remote Sensing
of the Environment, 34, 167 178.
Cressie, N. (1991). Statistics for spatial data. USA: Wiley-Interscience
(900 pp.).
Cross, A. M., Settle, J. J., Drake, N. A., & Paivinen, R. T. M. (1991). Subpixel measurement of tropical forest cover using AVHRR data. International Journal of Remote Sensing, 5, 1119 1129.
de Sa, T. D. A. & Alegre, J. (2001). Praticas agroflorestais visando o
manejo de vegetac oes secundarias: Uma abordagem com enfase em
experiencias amazonicas. In CONGRESSO BRASILEIRO DE SISTEMAS AGROFLORESTAIS, 3., 2000, Manaus, AM. Embrapa Amazonia Ocidental. Documentos, 17, 102 115.
de Sa, T. D. A., de Oliveira, V. C., Coimbra, H. M., de Carvalho, C. J. R.,
Dias-Filho, M., Sommer, R., & Brienza Junior, S. (1998). Diurnal and
seasonal patterns of leaf water relations in spontaneous and enriched
secondary vegetation components: Tools to understand water exchange
and water stress behavior, Abstracts of the Third SHIFT Workshop,
ManausAM. March 15 19, p. A14.
de Sa, T. D. A., de Oliveira, V. C., de Araujo, A. C., & Brienza Junior, S.
(1997). Biophysical aspects of enriched fallow with leguminous fast
growing trees in Northeastern Para, Brazil: Spectral composition of
light and water vapour. In International Symposium on the Science
and Practice of Short-term Improved Fallows, Lilongwe, Malawi,
1997, Abstracts, Lilongwe, IUFRO/ISSS-AISS-IBG/ICRAF.
de Sa, T. D. A., Vielhauer, K., Denich, M., Kanashiro, M., & Vlek, P. L. G.
(1998). Towards improving natural resources use in Eastern Amazonia
through a modified sequential agroforestry system. In Resumos Expandidos: II Congresso Brasileiro em Sistemas Agroflorestais, 25.11.1998,
pp. 95 100.
Drake, J. B., Dubayah, R. O., Clark, D. B., Knox, R. G., & Blair, J. B.
(2002). Sensitivity of large-footprint lidar to canopy structure and biomass in a neotropical rainforest. Remote Sensing of Environment, 81,
378 392.
Drake, J. B., Dubayah, R. O., Clark, D. B., Knox, R. G., Blair, J. B.,
Hofton, M. A., Chazdon, R. L., Weishampel, J. F., & Prince, S.
(2002). Estimation of tropical forest structural characteristics using
large-footprint lidar. Remote Sensing of Environment, 79, 305 319.
Ehleringer, J. R., & Field, C. B. (Eds.) (1993). Scaling physiological processes leaf to globe. San Diego, CA: Academic Press, 388 pp.
Fearnside, P. M. (1990). The rate and extent of deforestation in Brazilian
Amazon. Environmental Conservation, 17, 213 226.
Foody, G. M., Lucas, R. M., Curran, P. J., & Honzak, M. (1997). Nonlinear mixture modeling without end-members using an artificial neural
network. International Journal of Remote Sensing, 18, 937 953.
Harding, D. J., Lefsky, M. A., Parker, G. G., & Blair, J. B. (2001).
Laser altimeter canopy height profiles: Methods and validation for
closed-canopy, broadleaf forests. Remote Sensing of Environment, 76,

283 297.
Hurtt, G. C., Moorcroft, P. R., Pacala, S. W., & Levin, S. A. (1998).
Terrestrial models and global change: Challenges for the future. Global
Change Biology, 4, 581 590.
INPE (2000). Monitoring the Brazilian Amazon Forest by Satellite 1999
2000. Amazon Institutional Program, General Coordination of Earth
Observation, National Institute of Space Research, Brazil.
Keller, M., Palace, M., & Hurtt, G. C. (2001). Biomass estimation in the
Tapajos National Forest, Brazil: Examination of sampling and allometric uncertainties. Forest Ecology and Management, 154, 371 382.
Lefsky, M. A., Harding, D., Cohen, W. B., Parker, G., & Shugart, H. H.
(1999). Surface lidar remote sensing of basal area and biomass in deciduous forests of Eastern Maryland, USA. Remote Sensing of Environment, 67, 83 98.
Means, J. E., Acker, S. A., Harding, D. J., Blair, J. B., Lefsky, M. A.,
Cohen, W. B., Harmon, M. E., & McKee, W. A. (1999). Use of largefootprint scanning airborne lidar to estimate forest stand characteristics
in the Western Cascades of Oregon. Remote Sensing of Environment,
67, 298 308.
Meeson, B. W., Corprew, F. E., McManus, J. M. P., Myers, D. M., Closs,
J. W., Sun, K. J., & Sunday, D. J. (1995). ISLSCP Initiative I-global
datasets for land-atmosphere models, 1987 1988. Washington, DC:
American Geophysical Union, Available on CD-ROM.
Moorcroft, P., Hurtt, G., & Pacala, S. (2001). A method for scaling vegetation dynamics: The ecosystem demography model (ED). Ecological
Monographs, 71(4), 557 586.
Myneni, R. B., Hall, F. G., Sellers, P. J., & Marshak, A. L. (1995). The
interpretation of spectral vegetation indices. IEEE Transactions on Geoscience and Remote Sensing, 33, 481 486.
Nepstad, D. C., Verissimo, A., Alencar, A., Nobre, C., Lima, E., Lefebvre,
P., Schlesinger, P., Potter, C., Moutinho, P., Mendoza, E., Cochrane, M.,
& Brooks, V. (1999). Large-scale impoverishment of Amazonian forests
by logging and fire. Nature, 398, 505 508.
Pereira, R., Zweede, J., Asner, G. P., & Keller, M. (2002). Forest canopy
damage from conventional and reduced impact selective logging in
Eastern Amazon, Brazil. Forest Ecology and Management (In press).
Prentice, I. C., Farquhar, G. D., Fasham, M. J. R., Goulden, M. L., Heimann,
M., Jaramillo, V. J., Kheshgi, H. S., Le Quere, C., Scholes, R. J.,
Wallace, D. W. R., Archer, D., Ashmore, M. R., Aumont, O., Baker,
D., Battle, M., Bender, M., Bopp, L. P., Bousquet, P., Caldeira, K., Ciais,
P., Cox, P. M., Cramer, W., Dentener, F., Enting, I. G., Field, C. B.,
Friedlingstein, P., Holland, E. A., Houghton, R. A., House, J. I., Ishida,
A., Jain, A. K., Janssens, I. A., Joos, F., Kaminski, T., Keeling, C. D.,
Keeling, R. F., Kicklighter, D. W., Kohfeld, K. E., Knorr, W., Law, R.,
Lenton, T., Lindsay, K., Maier-Reimer, E., Manning, A. C., Matear, R. J.,
McGuire, A. D., Melillo, J. M., Meyer, R., Mund, M., Orr, J. C., Piper, S.,
Plattner, K., Rayner, P. J., Sitch, S., Slater, R., Taguchi, S., Tans, P. P.,
Tian, H. Q., Weirig, M. F., Whorf, T., & Yool, A. (2001). The carbon
cycle and atmospheric carbon dioxide. In J. T. Hougton, Y. Ding, D. J.
Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, & C. A.
Johnson (Eds.), Climate Change 2001: The Scientific Basis. Contribution of the Working Group I to the Intergovernmental Panel on
Climate Change ( pp. 183 238). Cambridge, UK: Cambridge University Press.
Quarmby, N. A., Townshend, J. R. G., Settle, J. J., & White, K. H. (1992).
Linear mixture modeling applied to AVHRR data for crop area estimation. International Journal of Remote Sensing, 3, 415 425.
Read, J. M. (in press). Spatial analyses of logging impacts in Amazonia
using remotely-sensed data. Photogrammetric Engineering and Remote
Sensing.
Read, J. M., Clark, D. B., Venticinque, E. M., & Moreira, M. P. (submitted
for publication). Application of merged 1-m and 4-m resolution satellite
data to research and management in tropical forests. Journal of Applied
Ecology.
Sellers, P. J., Dickinson, R. E., Randall, D. A., Betts, A. K., Hall, F. G.,
Berry, J. A., Collatz, G. J., Denning, A. S., Mooney, H. A., Nobre,

C. A., Sato, N., Field, C. B., & Henderson-Sellers, A. (1997).
Modeling the exchanges of energy, water, and carbon between continents and the atmosphere. Science, 275, 502 509.
Sellers, P. J., Meeson, B. W., Closs, J., Collatz, J., Corprew, F., Dazlich, D.,
Hall, F. G., Kerr, Y., Koster, R., Los, S., Mitchell, K., McManus, J.,
Myers, D., Sun, K. J., & Try, P. (1995). An overview of the ISLSCP-I
Global datasets. On: ISLSCP Initiative I- Global datasets for landatmosphere models, 1987-1988, vol. 1 5. Washington, DC: NASA
(Published on CD).
Shugart, H. H., Bourgeau-Chavez, L., & Kasischke, E. S. (2000). Determination of stand properties in boreal and temperate forests using highresolution imagery. Forest Science, 46(4), 478 486.
Skole, D. L., & Tucker, C. J. (1993). Tropical deforestation and habitat
fragmentation in the Amazon: Satellite data from 1978 to 1988. Science,
260, 1905 1910.
Smith, M. O. (1990). Vegetation in deserts: A regional measurement of
abundance from multi-spectral images. Remote Sensing of Environment,
31, 1 26.
Sommer, R. (2000). Water and nutrient balance in deep soils under shifting
cultivation with and without burning in the Eastern Amazon, Gottingen,
Cuvillier, 240 pp.
127
The LBA Science Planning Group (1996). The Large Scale Biosphere
Atmosphere Experiment in Amazonia (LBA): Concise Experimental
Plan. LBA Project Office, CPTEC, Brazil, http://yabae.cptec.inpe.br/lba.
USDA Forest Service (1992). Forest service resource inventories: An overview, Washington, DC.
Vielhauer, K., Denich, M., de Sa, T. D. A., Kato, O. R., Kato, M. d. S. A.,
Brienza Jr., S., & Vlek, P. L. G. (2001). Land-use in a mulch-based
farming system of small holders in the Eastern Amazon. Proceedings of
the Deutscher Tropentag (Conference on International Agricultural Research for Development) One World: Research for a better quality of
life, Bonn, Germany, October 9 11 2001, pp. 1 9.
Vourlitis, G. L., Priante-Filho, N., Hayashi, M. M. S., de Souza Nogueira,
J., Caseiro, F. T., & Campelo Jr., J. H. (2002). Seasonal variations in
the evapotranspiration of a transitional tropical forest of Mato Grosso,
Brazil. Water Resources Research, 38(6), 30 31.
Vourlitis, G. L., Priante-Filho, N., Hayashi, M. M. S., de Souza Nogueira,
J., Caseiro, F. T., Raiter, F., & Campelo Jr., J. H. (in press). The role of
seasonal variations in meteorology on the net CO2 exchange of a Brazilian transitional tropical forest. Ecological Applications.

Multi-site evaluation of IKONOS data for classification of

tropical coral reef environments
Serge Andrefouet a,*, Philip Kramer b, Damaris Torres-Pulliza c, Karen E. Joyce d,
Eric J. Hochberg e, Rodrigo Garza-Perez f, Peter J. Mumby g, Bernhard Riegl h,
Hiroya Yamano i, William H. White j, Mayalen Zubia k, John C. Brock c,
Stuart R. Phinn d, Abdulla Naseer l, Bruce G. Hatcher l, Frank E. Muller-Karger a
a
Institute for Marine Remote Sensing, College of Marine Science, University of South Florida, 140 7th Avenue S., St. Petersburg, FL 33701, USA
b
Rosenstiel School of Marine and Atmospheric Science, University of Miami, Miami, FL, USA
c
Center for Coastal and Regional Marine Studies, United States Geological Survey, St. Petersburg, FL, USA
d
Biophysical Remote Sensing Group, Department of Geographical Sciences and Planning, University of Queensland, St. Lucia, Australia
e
Hawaii Institute of Marine Biology, University of Hawaii, Honolulu, Kaneohe, USA
f
Coral Reef Ecosystems Ecology Laboratory, Marine Resources Department, CINVESTAV-I.P.N. Unidad Merida, Merida, Mexico
g
Marine Spatial Ecology Laboratory, University of Exeter, Exeter, UK
h
Oceanographic Center, National Coral Reef Institute, Nova Southeastern University, Miami, FL, USA
i
Social and Environmental Systems Division, National Institute for Environmental Studies, Onogawa, Tsukuba, Ibaraki, Japan
j
Department of Marine Science and Coastal Management, The University of Newcastle, Newcastle upon Tyne, UK
k
Laboratoire Terre-Oceans, Universite de la Polynesie Francaise, Tahiti, French Polynesia
l
Department of Biology, Dalhousie University, Halifax, Nova Scotia, Canada
Received 3 June 2002; received in revised form 5 December 2002; accepted 22 April 2003
Abstract
Ten IKONOS images of different coral reef sites distributed around the world were processed to assess the potential of 4-m resolution
multispectral data for coral reef habitat mapping. Complexity of reef environments, established by field observation, ranged from 3 to 15
classes of benthic habitats containing various combinations of sediments, carbonate pavement, seagrass, algae, and corals in different
geomorphologic zones (forereef, lagoon, patch reef, reef flats). Processing included corrections for sea surface roughness and bathymetry,
unsupervised or supervised classification, and accuracy assessment based on ground-truth data. IKONOS classification results were
compared with classified Landsat 7 imagery for simple to moderate complexity of reef habitats (5 11 classes). For both sensors, overall
accuracies of the classifications show a general linear trend of decreasing accuracy with increasing habitat complexity. The IKONOS sensor
performed better, with a 15 20% improvement in accuracy compared to Landsat. For IKONOS, overall accuracy was 77% for 4 5 classes,
71% for 7 8 classes, 65% in 9 11 classes, and 53% for more than 13 classes. The Landsat classification accuracy was systematically lower,
with an average of 56% for 5 10 classes. Within this general trend, inter-site comparisons and specificities demonstrate the benefits of
different approaches. Pre-segmentation of the different geomorphologic zones and depth correction provided different advantages in different
environments. Our results help guide scientists and managers in applying IKONOS-class data for coral reef mapping applications.
Keywords: Landsat; Bathymetric correction; Glint; Accuracy; Habitat mapping; Seagrass
1. Introduction
Remote sensing provides an effective way to observe
and monitor shallow coral reefs worldwide, to characterize
* Corresponding author. Tel.: +1-727-553-3987; fax: +1-727-553-1103.

E-mail address: serge@seas.marine.usf.edu (S. Andrefouet).
doi:10.1016/j.rse.2003.04.005
inter-reef structural differences, and to map intra-reef

habitat diversity and zonations, assess bathymetric variations, design survey protocols, conduct biogeochemical
budgets, and map beta-diversity (Andrefouet, Claereboudt,
Matsakis, Page`s, & Dufour, 2001; Andrefouet, MullerKarger, Hochberg, Hu, & Carder, 2001; Andrefouet &
Payri, 2000; Capolsini, Andrefouet, Rion, & Payri, 2003;
Hochberg & Atkinson, 2000; Jupp et al., 1985; Liceaga-
S. Andrefouet et al. / Remote Sensing of Environment 88 (2003) 128143
Correa & Euan-Avila, 2002; Mumby, 2001; Mumby &

Edwards, 2002; Mumby, Green, Clark, & Edwards, 1998;
Palandro, Andrefouet, Dustan, & Muller-Karger, 2003;
Purkis, Kenter, Oikonomou, & Robinson, 2002; Roelfsma,
Phinn, & Dennisson, 2002). The recent increase of remote
sensing applications targeting reef environments (Andrefouet, Muller-Karger, et al., 2001) reflects the growing
concern about drastic and negative changes occurring on
reefs over the past three decades due to anthropogenic (e.g.
pollution, fishing, and coastal development) or natural (e.g.
global warming) stresses.
The satellite data most commonly used since the mid1980s for direct observation of coral reefs have been
medium spatial resolution digital images, i.e. a spatial
resolution of 10 30 m. This includes data delivered by
the Indian Remote Sensing Satellite C (IRS-C), Satellite
pour lObservation de la Terre (SPOT) 1 4 High Resolution Visible (HRV), Landsat 5 Thematic Mapper (TM), and
more recently by SPOT 4 5, Landsat 7 Enhanced Thematic Mapper Plus (ETM+), and Advanced Spaceborne
Thermal Emission and Reflection Radiometer (ASTER)
sensors. Conversely, high resolution images are those
with a spatial resolution greater than 10 m such as those
provided by IKONOS or Quickbird (1 4 m). This study
aims to clarify the potential of high spatial resolution
IKONOS data for reef mapping worldwide.
In their review presenting the use of remote sensing for
coastal tropical assessment, Green, Mumby, Edwards, and
Clark (1996) pointed out the difficulty of comparing
different reef assessments due to lack of consistency in
classification schemes, in in situ data collection and image
processing methods, and in accuracy assessment protocols.
SPOT HRV and Landsat TM data have been used most
frequently because of their availability starting in the mid1980s. The work of various independent investigators
worldwide has now helped define the potential of medium
spatial resolution data for reef applications (e.g. Ahmad &
Neil, 1994; Andrefouet, Muller-Karger, et al., 2001; Matsunaga & Kayanne, 1997; Purkis et al., 2002; Yamano &
Tamura, in press). It is now clear that for geomorphology
and habitat-scale applications, SPOT and Landsat data are
adequate for simple complexity mapping (3 6 classes), but
for more complex targets (7 13 classes) they are limited
by their spatial and spectral resolution and likely by their
digitization rate (8 bits) (Capolsini et al., 2003; Hochberg
& Atkinson, 2003; Mumby, Green, et al., 1998; Mumby &
Edwards, 2002).
The 1999 launch of the commercial IKONOS satellite,
operated by Space Imaging (SI), provides for the first
commercial space sensor with 11 bit, high-spatial resolution (4 m), calibrated data in four wide spectral bands that
are potentially useful for coral reef studies. The spectral
bands closely match the first four bands of the ETM+
sensor (Thome, 2001). Despite the quick attenuation of red
radiance in water, the near-infrared (NIR) band is potentially useful for very shallow water targets (Menges, Hill,
129
& Ahmad, 1998) and low tide conditions when benthos is

exposed. The IKONOS satellite data has generated great
interest among coral reef researchers that were previously
limited to the use of aerial color photographs (Andrefouet
et al., 2002; Palandro et al., 2003) or costly digital airborne
multispectral or hyperspectral data (Mumby, Green,
Edwards, & Clark, 1999) for very high resolution work.
To the best of our knowledge, four independent peerreviewed studies have already addressed the potential of
IKONOS for reef habitat mapping and there are probably
many more investigations in progress, judging by the
number of coral scenes acquired and archived by SI.
Mumby and Edwards (2002) compared IKONOS classification results of Caribbean (Turks and Caicos) coastal
areas between different spaceborne and airborne sensor
data. They concluded that IKONOS data were unable to
discriminate habitats in a complex (13 classes) classification scheme, but that they were adequate for moderate to
simple complexity mapping (9 5 classes). They found
IKONOS provides an acceptable accuracy (64 74% overall accuracy), although similar to that obtained with a
medium resolution sensor like the Landsat TM. Capolsini
et al. (2003) applied a similar multi-sensor comparative
approach for South Pacific (Tahiti) reefs and reached
similar conclusions (66 86% overall accuracy for 7 3
classes). Maeder et al. (2002) also mapped a Caribbean
reef (Roatan Island, Honduras) with good results (>85%
accuracy) using a simple (five classes, including a deep
water class) classification scheme. Finally, assuming linear
mixing and using in situ spectral reflectance measurements, Hochberg and Atkinson (2003) simulated IKONOS
classification for various sea floors made of different
proportions of algae, coral, and sand. Without actually
processing any IKONOS data, they suggested that coral
dominated habitat could not be accurately separated from
algae dominated habitat using IKONOS data. Using simulated IKONOS data, coral cover was overestimated and
algae cover underestimated on their test-site of Kanehoe
Bay, HI. Capolsini et al. confirmed this prediction using
real images of Tahiti reefs, where true coral habitats
(coral>60%) were poorly assessed (9.54% users accuracy)
in a simple classification scheme (ruble, sand, algae,
coral).
The number of IKONOS evaluations for reefs is already
quite impressive and, more importantly, each seems to
provide results consistent with the others. However, because of the variety of sites and methods considered, we
felt that some of the Green et al. (1996) remarks would
still be valid if many independent studies were conducted
without some coordination to prevent too much heterogeneity in terms of classification scheme (habitat description), classification algorithms, and accuracy assessment
protocols. Therefore, beginning in 2000, the University of
South Florida (USF) worked with the NASA Scientific
Data Purchase (SDP) program (Stennis Space Center) to
task IKONOS for acquisition of scenes of representative
130
reefs around the world. We worked with international

remote sensing and coral biology/ecology/geology scientists to process and evaluate the images on their most
intensively studied research sites where ground-truth data,
local reef expertise, and other satellite or airborne remote
sensing data were available. Most of these investigators
agreed to contribute to the present study, making this effort
certainly the largest international cooperation for remote
sensing of reefs.
This paper presents a synthesis of the results in terms of
classification and mapping of coral reef habitats using
IKONOS data, as well as a comparison with ETM+
performances for selected sites. Each investigator that
contributed to this study will likely publish their own
detailed results in the future on different subjects since
each of them has a specific interest in the IKONOS data
(e.g. change detection). Despite the initial wish to have
consistent methodology rigorously applied throughout the
mapping exercise, local specificities and expertise, and cost
of field work lead to slightly different means of data
processing, ground-truthing strategy, or evaluation of
results. Nevertheless, the bulk of work provides clear
trends and lessons discussed hereafter.
2. Material and methods

2.1. Reef-sites and image data-set
Ten different sites were considered for this comparative
study (Fig. 1). The sites represent the primary biogeographic coral regions of the world according to Veron (1995),
and includes bank reefs, fringing reefs, barrier reefs, and
atolls.
Table 1 summarizes the IKONOS and Landsat 7 data
used for this study. We avoided images with obvious
water quality issues (e.g. suspended sediments). Only
scenes with clear waters were used, thus minimizing
the effects of variable water optical properties for this
comparative study. Until early 2001, the IKONOS images
delivered to NASA SDP were systematically resampled
with cubic convolution (CC), but analysis of the same
image delivered with CC and nearest neighbor (NN)
resampling showed that the textural information was
significantly degraded using CC in different coral reef
shallow floors (SA, unpublished data). Thus, we systematically requested NN resampling after April 2001. IKONOS data were geocorrected in UTM WGS-84, as Master
Standard (MS) or Original Standard (OS) products (Dial,
Bowen, Gerlach, Grodecki, & Oleszczuk, in press), but
never as Precision products since no ground control
points were provided to Space Imaging for our study
sites. Landsat 7 ETM+ data (Table 1) were ordered
through the Eros Data Center in either Geotiff or HDF
format, with NN resampling and UTM WGS-84. The
Landsat image of Shiraho Reef, Japan was provided by
the National Space Development Agency of Japan, in

Fast-L7 format, NN resampling.
2.2. Habitats: similarities and differences between sites
One of the challenges of this comparative study is to
reconcile different habitat classification schemes. Indeed, the
10 sites are characterized by different bottom features, and
the methods used by investigators in each region addressed
their problems in particular ways. The various investigators
provided classification schemes ranging from simple (4 5
classes for Shiraho and Glovers), to moderately complex (7
8 classes for Arue, Addu, Biscayne Dubai, and Andros), and
then to very complex (13 15 classes for Mayotte, Heron,
and Boca Paila) (Table 2). The very complex sites were also
described in less detail by hierarchical simplification. Heron
Reef has been described using 13, 7, and 5 habitats classes.
Boca Paila has been described using 15 and 8 habitats.
Mayotte has been split from 1 general site with 14 general
classes into 2 sub-sites with 10 specific classes each, providing better thematic description for each sub-site.
To compare classification results throughout this range of
habitat complexity, it is desirable to put the various schemes
into the same hierarchical framework. For the four Atlantic/
Caribbean sites (Biscayne in USA, Andros in Bahamas,
Glovers in Belize, and Boca Paila in Mexico), we can refer
to the scheme provided by Mumby and Harborne (1999)
(Fig. 2). This is a multi-level hierarchical model with geomorphologic and benthic components. This scheme was
applied to each site but it was necessary to adjust the
thresholds in benthic cover and depth (Fig. 3). For instance,
Andros coral cover was high (5 15%) in most of the
geomorphological strata. Thus, most of the Andros reef
should be classified as coral since according to Mumby
and Harborne (1999) coral classes are characterized by a
cover >1%. However, a coral label for the entire Andros reef
would be misleading. Similarly, the notion of deep or
shallow lagoon floors differs between sites (Figs. 2 and 3).
Extrapolation of the Mumby and Harborne (1999)
geomorphology/benthic hierarchical framework to non-Caribbean sites is possible in theory. For Shiraho (Japan) and
Dubai, the geomorphology of the site is simple: shallow
(0 3 m) lagoon for Shiraho and gentle slope for Dubai.
Therefore, since there is only one geomorphologic unit,
there is no specific reference to geomorphology in the
classification scheme and only benthic features were used
(Table 2). Conversely, for Addu, Arue, Heron, and especially Mayotte, reference to geomorphological strata is
required (Table 2). Mayotte is by far the most complex
site. Quod, Bigot, Dutrieux, Maggiorani, and Savelli.
(1995) pointed out the richness of this site, characterized
by two contrasted barrier-reefs separated by the deep
( f 70 m) Longogori pass. The northern barrier reef
(Pamandzi Reef) is characterized by extensive seagrass
and algal beds and internal spur and grooves systems,
while the southern reef (Ajangoua Reef) is free of seagrass
131
Fig. 1. Location of the study sites. The map indicates the main coral biogeographic areas according to Veron (1995). Our sites represent the Arabian Gulf,
Indian Ocean, Indo Pacific, Pacific, and Caribbean biogeographic zones. Each site is presented using a RGB color composite based on the red, green, and blue
bands of the IKONOS sensor. Size (in km) of the image showing the most characteristic zone is indicated, though the actual processed area may be wider.
Includes material Space Imagingn.
and richer in small coral heads, with presence of enclosed

lagoons. Our own in situ survey of Mayotte Reef in
December 2000 highlighted nearly 30 different habitats,
the main difference with Quod et al. was the extensive
areas of dead corals, consequences of a 1998 coral

bleaching event. For mapping purposes, we considered as
a starting point only 14 broader classes for Mayotte (Table
2). Arue reef in Tahiti Island (French Polynesia) is repre-
132
Table 1
Site information and characteristics of the IKONOS and Landsat images
Latitudeb
Longitudeb
IKONOS
Landsat 7 ETM +
Acquisition date
Product
Resampling
Path/row
Acquisition date
Format
Ground-truth
Depth (m)
References
Stoddart (1964)
Kramer, Kramer,
and Ginsburg (1998)
Frouin and Hutchings (2001)
Jaap (1984)
Kramer and Kramer (2000)c
Riegl (1999)
McClanahan and
Muthiga (1998)
Rogers (1997),
Smith et al. (1998)
Addu AN, SA, BGH

Andros PK
0.6149
24.5833
73.1201
77.7833
14 March 2000
12 March 2001
MS
OS
CC
NN
145/60
13/43
20 December 2000
26 March 2000
HDF
EDC-Geotif
March 2002
2000/2001
0 15
0 30
Arue SA, MZ
Biscayne DTP, JB, SA
Boca Paila RG
Dubai BR
Glovers PJM, WHW
17.5750
25.3500
19.9166
24.9400
16.8166
149.6000
80.2166
87.5000
54.8900
87.8000
11 March 2000
18 March 2001
20 July 2000
2 May 2001
12 April 2001
OS
OS
MS
MS
OS
CC
CC
CC
NN
NN
53/72
15/42
19/46
160/42
18/48
6 June 2000
5 February 2000
24 July 1999
N/A
8 November 2000
Earthsat-Fast
EDC-HDF
EDC-Geotif
N/A
EDC-Geotif
June 2000
2001/2002
1999/2000
Fall 1995
1999 2001
0 12
2 12
0 30
09
0 18
Heron SA, KEJ, SRP
23.5258
151.8900
7 May 2001
MS
NN
90/77
14 November 1999
EDC-HDF
2001/2002
0 15
Mayotte SA
Shiraho HY
12.8933
24.3166
45.2180
124.2333
31 August 2000
28 March 2002
OS
OS
NN
NN
91/77
161/69
115/43
18 September 1999
30 August 2000
23 February 2002
EDC-HDF
EDC-Geotif
Fast-L7A EROSd
Dec. 2000
1999
0 15
03
OS: Original Standard, MS: Master Standard, CC: cubic convolution, NN: nearest neighbor convolution, N/A: not available.
a
Investigators: initials from author list.
b
Lower left corner of the IKONOS scene. Negative latitude for South, negative longitude for West.
c
Describe Yucatan reefs.
d
Provided by National Space Development Agency of Japan.
Quod et al. (1995)

Kayanne et al. (2002),
Harii and Kayanne
(submitted for publication)
Site and
investigatora
133
Table 2
Classification scheme for the 10 sites (Boca Paila 15 classes and Mayotte 14 classes are not shown)
Site
Class label
Site
Class label
Addu
Classes: 8
Sand/rubble, backreef
Sand, backreef
Sand, lagoon floor
Coral/algae, patch reef
Algae, crest
Coral, forereef
Seagrass, backreef
Coral, backreef
Dense brown algae (>80%), crest
High density coral heads on heterogeneous pavement
Low density coral heads on sandy floor
Dense brown algae (>50%) on reef flat, heterogeneous floor
Moderate brown algae (15 25%) on reef flat,
heterogeneous floor
Sparse brown algae ( < 5%) on reef flat, heterogenous floor
Sand and ruble (>90%)
Deep (>10 m) lagoon floor
Dense coral (>25%)
Moderate coral ( < 25%)
Seagrass/algae
Heterogeneous lagoon floor
Pavement
Shallow sand, lagoon floor
Deep sand, forereef
Andros
Classes: 8
Coral (Acropora sp.) crest

Dense coral (Montastraea sp.) on forereef
Gorgonian plain on forereef
Dense gorgonian on escarpment
Algae on low relief spur and grooves
Pavement on forereef
Deep (>12 m) sand on forereef
Shallow sand/ruble on forereef
Dense patches
Diffuse patches
Deep (>3 m) dense seagrass on lagoon floor
Deep (>3 m) sparse seagrass on lagoon floor
Shallow ( < 3 m) dense seagrass on lagoon floor
Arue
Classes: 8
Boca Paila
Classes: 7
Glovers
Classes: 5
Heron
Classes: 13
Forereef
Brown algae
Seagrass/Lobophora sp.
Seagrass
Sand
Branching corals (>75%) on forereef and crests
Multi-growth forms, dense corals (>50%) on reef flat
Multi-growth forms, moderate corals (25 50%) on reef flat
Coral pavement (>75%) on crest and reef flats
Coral head, forereef
Heterogeneous reef flat (coral < 10%, sand, rocks,
fleshy algae, coralline)
Dead coral (>80%) coated by encrusting coralline
Dead coral (>80%) covered by fleshy algae
Sand with rocks ( < 15%) covered by fleshy algae
Sand with scattered dead coral heads ( < 15%),
coated by encrusting coralline
Sand with dense dead coral heads (>50%),
covered by fleshy algae
Coral sand and mud (>90%)
Pavement and ruble (>90%)
Shiraho
Classes: 4
Biscayne
Classes: 8
Dubai
Classes: 8
Heron
Classes: 5
Mayotte
Classes: 14
Shallow ( < 3 m) moderate seagrass on lagoon floor

Shallow ( < 3 m) sparse seagrass on lagoon floor
Deep (>3 m) sand on lagoon floor
Dense coral
Sparse coral
Seagrass
Shallow algae
Deep algae
Pavement
Shallow sand
Deep sand
Branching and massive corals on forereef
Multi-growth forms corals (>25%) on reef flat
Heterogeneous (coral < 25%) reef flat
Heterogeneous reef flat, sand bottom
(10% < rock algae coral < 25%)
Sand (>90%)
Forereef
Dense coral margins on spur-and-grooves,
enclosed lagoon and pass
Dense coral heads (>25%) on heterogeneous floor, reef flat
Sparse coral heads ( < 10%) on sand floor, reef flat
Dense Thalassodendron sp. seagrass
and Padina sp. algae (>80%)
Dense Thalassodendron sp. seagrass (>80%)
Diffuse seagrass ( < 25%) on sand
Mixed seagrass and algae
Mixed seagrass beds, with corals, rocks, ruble, algae
Brown algae on reef flat and crests
Coralline mounts
Pavement and ruble
Deep sand channels on forereef
Shallow sand on reef flat
Coral
Seagrass/algae
Pavement
Sand
sentative of the Tahitian reefs under moderate influence of

the harbor and industrial activities (Frouin & Hutchings,
2001). This is a morphologically complex reef, with a
fringing reef, deep channels, a large barrier reef and three
large patch reefs (Fig. 2). Two wide passes connect the
channels with the ocean on each side of the reef complex.
Heron Reef is a large reef platform of the southern Great

Barrier Reef (Australia) and one of the most studied reef
sites in the world, including using remote sensing. Here,
we considered only the western side (Fig. 2) where most
of the ground-truth data were collected in 2001 and 2002.
It is also the most heterogeneous area since the eastern
134
Fig. 2. Top: Habitat classification scheme as proposed by Mumby and Harborne (1999) for mapping Caribbean coral reefs. Both geomorphologic and benthic
keys may define a habitat. For instance, three habitats encountered on dense patch reefs are highlighted: a coral zone with presence of massive colonies of
Montastraea sp., an algal zone dominated by the brown algae Lobophora sp., and a bare substrate zone densely covered by gorgonians. Depending on the
precision of the classification scheme, a practitioner can provide details only at geomorphological level, or at benthic levels, or a combination of both. Bottom:
for Glovers Reef, the final classification scheme is a simple five-class scheme, with a forereef (geomorphology) class and four generic benthic classes (brown
algae, seagrass, mixed seagrass, and sand) not related to a specific geomorphologic zone (e.g. sand includes sandy areas in any geomorphological zones).
side of the bank is dominated by a large shallow sandy

lagoon (Smith, Frankel, & Jell, 1998).
Table 2 summarizes the different classes, without details
on the main benthic species, which are available in the
references provided for most of the sites (Table 1). These
references describe in detail the benthic community structure
present on each site. Species-level description is the final
level of the hierarchy of habitats. Clearly, dominant coral,
algae, or seagrass species are generally different from one
region (or one site) to another. For instance, Mayotte has
extensive Thalassodendron sp. seagrass beds, while Thalassia sp. and Syringodium sp. dominate Biscayne and Halophila sp. and Halodule sp. are frequent in Dubai. Andros and
Boca Paila coral crests are dominated by Acropora palmata,
but behind Shirahos crest Montipora spp. and Heliopora
coerulea are dominant. Brown algae Sargassum sp. and
Turbinaria sp. are dominant on Arue patch reef, while

Lobophora sp. extensively colonized Glovers patch reefs.
2.3. IKONOS image processing
IKONOS data were processed differently depending on
image quality and characteristics of the site. Main stages of
processing included surface roughness correction, depth
correction, classification, and accuracy assessment.
2.3.1. Surface roughness correction
A large fraction of the images acquired via NASA SDP
suffer from sea surface effects due to wind-generated wave
patterns and associated sun glint. As of May 2002, nearly
50% of the 40 images (not all of them processed here)
delivered to USF suffered significantly from this problem
135
Fig. 3. Example of contrasted habitat classification schemes for two Caribbean sites: Biscayne Bay, FL (USA) and Andros island (Bahamas). Coloured sections
highlight the zones present on each reef. For Biscayne, the study area includes lagoon floor and patch reefs. For Andros, the area of interest includes the oceanic
side of the coral reef system with the forereef, crests, and spur-and-grooves. In both sites, benthic classes are related to a geomorphologic zone, highlighted with
similar color. For instance, for Biscayne, a class of dense seagrass has been defined both for the shallow lagoon floor (blue) and the deep lagoon (green). Sand
is related on the lagoon floor (yellow), without separation between deep and shallow. Diffuse gorgonians are related to diffuse patch reefs (purple). For Andros,
the same bi-component geomorphology-benthos hierarchy occur, e.g. dense gorgonians occur only on high relief escarpment (purple); forereef (green) is split
into four classes: sand (yellow), diffuse gorgonian (light purple), coral Montastraea (dark green), and bare bedrock (grey). Andros and Biscayne both comprise
eight classes of habitats, but with minimum overlap between the two schemes. This illustrates the variety of habitats classification scheme that can occur in the
same region (see also Fig. 2 for Glovers).
(Fig. 4). For this study, a sea surface roughness correction

was required for Heron and Biscayne. Dubai and Glovers
images were also corrected, but the wave patterns were not
that significant and the original images were used. Details
of the algorithm for sea surface correction are provided in
Hochberg, Andrefouet, and Tyler (2003). Briefly, the
process consists of analyzing the sea surface assuming
that the water is virtually opaque in the NIR band (Siegel,
Wang, Maritorena, & Robinson, 2000) and that the relative
amount of radiance reflected upward at the sea surface is
solely a function of geometry, independent of wavelength.
This means that pixels with glint contribution in NIR

bands also have similar glint contribution in total upward
radiance in visible bands. Identifying the pixels with
maximum and minimum radiances in the NIR enables
estimation of the percentage of glint contribution in each
pixel, which is then corrected to absolute radiance in the
visible bands.
2.3.2. Depth correction or depth-invariant indices
Lyzenga (1981) proposed a method to eliminate the
effects of water column attenuation on bottom radiances
136
Fig. 4. Pre-processing of IKONOS images. Top: Example of surface correction aimed at removing wave and glint effects that occur on almost 50% of the 40
images provided by NASA SDP to USF. The level of details available in deep areas is drastically improved after correction. Bottom: An example of false color
composite made with three depth-invariant bottom indices, one from each pair of IKONOS bands 1 3. The backreef appears in pink dividing the outer forereef
from the lagoon. This technique limits misclassification due to depth between shallow dark objects and deep bright objects. Surface correction and depthinvariant indices were not systematically applied, depending on image quality and site characteristics (see Table 3).
(or reflectances). Basically, Lyzenga showed that pixels of

the same bottom-type located at various unknown depths
appear along a line in the bidimensional histogram of two
log-transformed visible bands. The slope of this line is the
ratio of diffuse attenuation of the two bands. Repeating
this for different bottom types at variable depth results in a
series of parallel lines, one for each bottom type. Projection of these lines onto an axis perpendicular to their
common direction results in an unitless depth-invariant
bottom-index where all pixels from a given bottom-type
receive the same index-value regardless of its depth. Two
visible bands (or one 2D histogram) provide one index.
Three bands can provide three depth-invariant indices by
permutation. The main drawback of this method is that
index values cannot be related to radiance or reflectance
measurements. Also, in some cases, application of Lyzengas method is problematic because the same bottom type
may not occur over a wide range of depths, thus biasing
the accurate estimation of the ratio of diffuse attenuation
(Maritorena, 1996). This is the case for Biscayne and Arue
in this project. Nevertheless, it has been proven that
applying similar empirical techniques and classifying the
resulting index images instead of the initial images could

significantly increase the accuracies of the maps (Mumby,
Clark, Green, & Edwards, 1998). Here, the technique has
been applied to Glovers (Fig. 4), Heron, Andros, and Boca
Paila for two reasons: (1) sites presented a significant
depth range and (2) deep sand channels and shallow
lagoon sand pools were adequate to train and apply the
method.
2.3.3. Unsupervised and supervised classification
The regional investigators applied two strategies. For
Glovers and Shiraho, they first conducted an unsupervised
classification and then assigned the different segments to
a given benthic category according to expert knowledge
and ground-truth data. For Andros, Heron, Mayotte,
Dubai, Arue, Biscayne, Boca Paila, ground-truthed polygons in each class were used to train a supervised
maximum likelihood classifier. For Mayotte, the two
contrasted barrier-reefs were processed simultaneously
and then separately. For Glovers only, a contextual
decision rule was applied. It re-classified any lagoonal
pixel classified spectrally into forereef to the correct
137
bleaching event. Dubai reef areas that may have changed

after 1998 and that could bias the accuracy assessment
were removed. The assessment was systematic, along a
grid placed over the image. Observed and computed
classes at each grid-point were used to build the confusion matrix, for a total of 1086 points. The accuracy
metric that we have considered in this study is the overall
accuracy (Stehman, 1997), i.e. the proportion of control
points correctly classified.
category. Depending on the depth of the site or the tide

conditions, three or four IKONOS spectral bands were
used (Table 3).
2.3.4. Accuracy assessment
It was not possible to survey any of the reefs simultaneously with the acquisition of the images (Table 1)
even though some sites were visited several times in
consecutive years (Table 1). The time gap between
ground observations and image acquisition is generally
between a few months and a year. The exceptions were
Dubai, where large-scale video tracks (Riegl, Korrubel, &
Martin, 2001) were collected in 1995 and the image
acquired in 2001, and Shiraho, with a 3-year gap
(Kayanne, Harii, Ide, & Akimoto, 2002). Mumby and
Edwards (2002) also reported a gap of 5 years. They
stated that habitat delineation was unlikely to evolve
significantly during this time frame. This may actually
be questionable depending on the history of the site and
the type of perturbations that may have occurred (hurricanes, coral bleaching, etc.), but for our data set we have
considered this statement valid, except for Dubai. Formal
accuracy assessments were summarized in confusion matrices for each site with generally more than 100 independent control points for the total partition, with the
exception of Mayotte (1230 points along 25 transects of
variable length) and Heron (5 transects provided 93
points, in addition to 72 isolated independent points
reef-wide) (Fig. 5). The Dubai classification has been
compared with continuous shipborne video surveys that
highlight the rate of change on the reefs after the 1998
2.4. Landsat 7 image processing

Where both IKONOS and Landsat 7/ETM+ data were
available (Table 1), ETM+ data were processed similarly
to the IKONOS data, with the exception of sea surface
correction that was never applied. We did not try to
classify ETM+ data for high complexity classification
(>10 classes), but only for moderate complexity (5 10
classes). The same ground-truth point/transects were used
for accuracy assessments for both IKONOS and Landsat
classifications. Of particular interest is the tandem IKONOS/ETM+ for Mayotte, acquired only 1 day apart in
August 2000 in similar low-tide conditions. For Heron,
we considered two ETM+ images, one each in high and
low tide condition, and we used three and four bands,
respectively. For Andros, we do not provide an overall
accuracy because the area considered for Landsat mapping included the shallow lagoon and forereef, whereas
for IKONOS, only the forereef was processed. Moreover,
Landsat accuracy assessment was not done independently
but was based on original training polygons. Further, we
Table 3
Image processing parameters and classification results
Site
Surface
correction
Depth
correction
IKONOS
bands
Nb control
points
Classes
Overall accuracy
IKONOS (%)
Overall accuracy
Landsat 7 (%)
Addu
Andros
Arue
Biscayne
Boca
Paila
N
N
N
Y
N
N
Y
N
N
Y
4
3
4
3
3
400
150
200
123
150
8
8
8
8
15
66
74
70
84
45
56
N/A
52
56
N/A
Dubai
Glovers
N
N
N
Y
3
3
N/A
150
Heron
165a
Mayotte
1230c
Shiraho
104
7
8
11
5
13
7
5
14
10d
10e
4
74
71
51
77
42
61
78
61
73
68
81
53
N/A
42
71
N/A
N/A
66/61b
N/A
56
50
63
N/A: Not computed.

a
Partly independent point, partly clusters (see text for details).
b
Using image from 18/09/99 (low tide) and 14/11/99 (high tide), respectively.
c
Clusters only (see text for details).
d
Ajangoua barrier reef (no seagrass).
e
Pamandzi barrier reef (seagrass).
138
Fig. 5. Example of validation of IKONOS classification using in situ large scale transects on Heron Reef. The classifications of three sites (Dubai, Mayotte, and
Heron Reef) have been controlled using this technique. Here, on the south of Heron, the transition between five benthic classes can be compared with in situ
observation where the percent cover of different substrates has been estimated visually in 20 20 m units. Despite the difference in resolution (4 vs. 20 m), we
note the good agreement between in situ data and classification, with accurate definition of transition zones, homogeneous zones, and heterogeneous patchy
zones (data from Joyce, Phinn, Roelfsema, Neil, & Dennison, 2002).
compare IKONOS and Landsat only in qualitative terms

for this site.

Examples of classification accuracy and final maps are
provided in Figs. 5 and 6. Classification accuracies for
each site obtained with IKONOS and Landsat are provided
in Table 3. Fig. 7 presents the pooled overall accuracies vs.
habitat complexity achieved with IKONOS. We also added
the results available for Roatan (Maeder et al., 2002),
Turks and Caicos (Mumby & Edwards, 2002), and
Punaauia (Capolsini et al., 2003). For Dubai, image
classification in eight classes (Table 2) identified as algae
several areas previously dominated by corals in 1995.
Generally, the four different classes of dense corals identified by video were accurately recognized but only as a
broad coral class (Table 2). A posteriori ground-truthing
showed that the coral areas apparently misclassified as
algae really changed after the 1998 bleaching event (Riegl,
1999), suggesting that IKONOS could be used for monitoring changes in benthic communities for similar sites if a
reference had been available. This is confirmed by Palandro et al. (2003) who combined aerial photographs and one
IKONOS image of Carysfort Reef (Florida) to quantify the
rate of coral loss in the last 20 years.
The general trend in Fig. 7 is a linear decrease of
accuracy with increasing complexity. Accuracies range
from an average of 77% for 4 5 classes to 71% for 7
8 classes, 65% in 9 11 classes, and 53% for more than 13
classes. There is no obvious bias that could be explained
by the skills of the investigators. For instance, in 13 14
classes, both best and worst results are provided by the
same investigator (SA). At 8 10 classes, all results are
very consistent. The variations are the natural consequences of the nature of the site and the way that the images
have been processed (Table 3). We did not compute the
variance of each accuracy because the sampling schemes
for accuracy assessment were not the same and in some
cases the rigorous conditions of application would likely
have been violated (Foody, 2002; Stehman, 1997; Steh-
139
Fig. 6. Examples of classification maps for Arue, Andros, Biscayne, and Glovers. For the three first sites overall accuracy was >70%, while for Glovers the
11-class scheme led to a poor 50% accuracy, prompting the implementation of a more simple five-class scheme (Fig. 2).
man, 1999). For reference, Mumby and Edwards (2002) or

Capolsini et al. (2003) computed the 95% confidence
interval for different accuracy metrics (Tau and overall
accuracy). Confidence intervals were in the range F 5
10% for several hundred independent control points,
depending on the classification schemes (3 13 classes)
and the number of control points per class. Our initial
intention was to apply more systematic processing using
the same classification schemes, but this proved to be quite
difficult or impossible because of the many local specificities and constraints.
The linear decreasing trend in Fig. 7 could be used to
estimate a priori the accuracy to be expected for a given site
using the methods described here. Fig. 7 compares the
results achieved by IKONOS and Landsat. Landsat also
provides a linear trend, but at a level 15 20% lower than
IKONOS throughout the range of habitat complexity.
Depending on the range of depth, type of geomorphology

and habitats, and water clarity at the time of acquisition, the
final accuracy will be modulated in the range highlighted
here. For instance, for IKONOS, for simple four-class
mapping, 68 81% overall accuracy can be expected according to Turks and Caicos (T&C), Shiraho, and Punaauia.
Highest accuracies are obviously obtained for the shallow
fringing and barrier reefs (Shiraho and Punaauia), while
T&C comprised habitats between 0 and 25 m depth. Fig. 7
also can aid in deciding the level to which habitat complexity can be addressed for a given level of accuracy for both
sensors. Using the data presented in Fig. 7, two relations
may be derived: overall accuracy ( Y) vs. number of classes
(X) was Y = 3.90X + 86.38 (r2 = 0.63) for Landsat and
Y = 2.78X + 91.69 for IKONOS (r2 = 0.82). If an 80%
accuracy is required for scientific or management applications, only four to five classes can be used using IKONOS.
140
90
Overall accuracy (%)
80
70
60
50
40
30
20
2
10
11
12
13
14
15
Roatan
Punaauia
Turks and Caicos
Heron
Andros
Arue
Biscayne
Glovers
Mayotte
Shiraho
Boca Paila
Addu
Dubai
Number of habitat classes

90
Arue
Biscayne
Mayotte (Ajangoua)
Mayotte (Pamandzi)
Shiraho
Boca Paila
Addu
Heron (3 bandes)
Heron (4 bandes)
Punaauia
Turks and Caicos
Glovers
Overall accuracy (%)
80
70
60
50
40
30
20
2
10
11
12
13
14
15
Number of habitat classes

Fig. 7. Relation between habitat classification scheme complexity (number of classes) and overall accuracy for the IKONOS sensor (top) and Landsat ETM+
sensor (bottom). Roatan data are from Maeder et al. (2002). Punaauia data are from Capolsini et al. (2003). Turks and Caicos data are from Mumby and
Edwards (2002), using textural information for IKONOS processing and Landsat 5 TM data.
Around the 70% threshold, most of sites with less than 10

classes are included, but there are exceptions.
The accuracies reported here are for the overall partition. Individual habitats that may be of importance for
particular applications do not necessarily have the same
accuracy, for better or worse. For most of the sites, we do
not have enough control points to discuss accuracy within
each individual class (Congalton & Green, 1999), but
some observations are worth mentioning. For instance,
sand is always very well classified (e.g. 90 92% users
accuracy for deep and shallow sand in Dubai). Conversely,
with the exception of Dubai (59% users accuracy), true
coral areas are generally poorly classified when algaedominated habitats also occur on the same site at the same
depth in a patchy fashion, confirming the predictions of
Hochberg and Atkinson (2003) with 43.8% users accuracy. High accuracy for coral zones seems possible on
several sites (e.g. Andros and Heron) but this is because
the image processing has been stratified by geomorphologic zone (e.g. Andros with 97% users accuracy for
forereef-high relief Montastraea annularis zones) or be-
cause there is no real competition with other classes (e.g.

Heron does not have shallow dense seagrass beds spectrally similar to deep corals). High accuracy also occurs for
forereef geomorphologic zones, which are implicitly coralrich (95% users accuracy in Glovers).
Comparison of the accuracy of IKONOS and Landsat
assessments confirms previously published results obtained
with moderate resolution sensors (Landsat TM, SPOTHRV, ASTER). An exception was Chauvaud, Bouchon,
and Manie`re (1998) and Chauvaud, Bouchon, and Manie`re
(2001), whose accuracies are certainly optimistically biased
(>90% for more than 25 classes using SPOT-HRV). Landsat accuracies also decreased linearly with increasing
habitat complexity. It is adequate for simple habitat complexity mapping, and the results for seven to eight classes
are consistent, within a low range of accuracy (50 56%)
(Table 3). For Heron, lower complexity (five classes)
provided acceptable accuracy (>60%), especially with the
image acquired at low tide since this resolved the shallow
heterogeneous hard-bottom and coral zones along the
reef rim (Table 3). For the Andros forereef where ETM+
accuracy could not be computed without optimistic bias,

IKONOS better discriminated shallow water ( < 5 m)
habitats which are often small and patchy. The texture of
the habitat seems exploitable using the IKONOS sensor at
4-m resolution for this site (see below further discussion
on texture). In deeper (>5 m) forereef zones, habitats are
more spatially continuous (e.g. high relief Montastraea
zone visible in Fig. 6) and both sensors provide similar
patterns. Narrow geomorphological features such as terraces and steps are more evident on IKONOS. The decision
to use IKONOS or Landsat (or any other medium resolution sensor) is constrained generally by consideration
of cost-effectiveness (Mumby et al., 1999). IKONOS
is clearly not the best option for covering large areas
(Mumby & Edwards, 2002). For small areas of high
interest (research sites or marine protected areas), our
results show the level of improvement in mapping accuracy that can be expected for science or management
applications. This also demonstrates the potential of conducting multi-sensor reef mapping by coupling Landsat
broad (but accurate) 4/5-class mapping for large reef
stretch (e.g. Florida Keys, Bahamas, Great Barrier Reef,
etc.) with more precise (and still accurate) IKONOS
mapping for specific (patchy) areas.
Higher accuracies could be potentially achieved using
textural measurements as suggested by Mumby and
Edwards (2002). For coarse habitat mapping (four classes)
on Turks and Caicos, overall accuracy increased from 68%
to 75% when textural neo-channels (3 3 window variance) were combined with depth-invariant indices in the
classification process. However, this is likely to be site
(patchiness, depth) and image quality dependent. Indeed,
tests on Mayotte show that despite contrasted textural
signatures for different bottom-types, the texture did not
improve the overall accuracy. For Mayotte, the fact that we
could also use the NIR bands throughout most of the reef
system likely explains this lack of improvement. Mumby
and Edwards also noted that texture did not improve
classification results when using 10-band multispectral
airborne data. Thus, the extra spectral information seems
more important than texture. We suggest that more systematic tests are required to fully quantify and qualify the
benefits of using textural signatures in various conditions.
Another potential way to improve accuracy is to use
contextual knowledge to modify the classifications a posteriori (Mumby, Clark, et al., 1998). The benefit of contextual
knowledge can also be inferred from our results. Indeed,
higher accuracies were achieved when the reefs were a
priori segmented into main geomorphologic or contrasted
zones. Glovers was processed as a whole (including forereef, lagoon, and rim) and reached 77% accuracy in five
very broad classes (Table 2), even after some contextual
editing. Conversely, Andros reached 74% in eight very
specific classes (Table 2), but was initially pre-segmented
to process only the forereef, thus avoiding confusion with
classes present on the backreef (e.g. seagrass, small patch
141
reefs). The backreef processed alone would likely provide

results similar to Biscayne (lagoon floor with seagrass/sand,
patch reefs), where eight classes were mapped with 84%
accuracy. Finally, splitting the Mayotte reef system in 2
barrier reefs with 10 very specific habitats each yielded
better results than a global classification in 14 thematically
broader classes. This strongly suggests that the practitioner
should carefully pre-segment the image and then process by
zones to optimize results. This can be performed empirically, by visual interpretation, or in a formal way. Andrefouet,
Roux, Chancerelle, and Bonneville (2000) formalized with a
fuzzy membership function the distance to the shore
factor for the classification of SPOT images to avoid
misclassification between fringing mud and coral crests.
Suzuki, Matsakis, Andrefouet, and Desachy (2001) integrated directions (perpendicular, parallel) when classifying reef flats in atolls using Landsat imagery. However,
these are complex processes and for simplicity, in most of
the cases, pre-segmentation is made relatively easy by the
high resolution of the IKONOS images where fronts and
boundaries can be accurately delineated by simple visual
interpretation (e.g. Andros). Eventually, similar habitats in
different zones can be merged depending on the desired
classification scheme (e.g. algae on forereef merged with
algae on patch reefs).
It has been suggested that the depth correction technique should be systematically applied (Green, Mumby,
Edwards, & Clark, 2000) because of its simplicity and
potential benefit. Here, we had difficulties to do so at each
site. Biscayne, for instance, does not have the same
habitats at different depths and the Lyzenga (1981) method
cannot be applied. In this case, slightly different techniques
(e.g. ratios, Maritorena, 1996; Paredes and Spero, 1983;
Polcyn, Brown, & Sattinger, 1970) or ancillary field data
(bathymetry) are required to overcome the bathymetric
challenge. The patterns of misclassifications (patches classified as dense seagrass) on Biscayne suggest that a depth
correction could be useful, even though without bathymetric correction, the overall accuracy appears quite good
(84%). For Andros, tests showed that the maximum
distance between habitats was achieved using two depthinvariant indices (using bands 1 2 and bands 2 3) and the
original band 3. A similar conclusion arose for Boca Paila
where only the depth-invariant index based on bands 1 and
2 was useful, in conjunction with other unprocessed bands.
In general, there is no doubt that bathymetric correction at
the scale of habitats should enhance overall accuracy and
avoid misclassifications. However, many of these misclassifications may happen simply because very different
zones are considered simultaneously in the classification
process. In a Caribbean reef, considering lagoon and back
reef communities (e.g. seagrass) with forereef coral communities in the same process typically leads to problems.
Therefore, we suggest that future work should examine the
relative benefits of contextual knowledge and depth compensation to achieve high accuracy. Indeed, this is proba-
142
bly site-dependent. Mumby, Clark, et al. (1998) showed

that depth correction is more critical than contextual
editing for a Turks and Caicos site, but in this case, the
morphology of the site was simple.
Hooten, for their support for travel and field work in

Heron Isl. Finally, thanks to Adam Lewis (Great Barrier
Reef Marine Park Authority) who provided the Landsat
images of Heron Reef. This is IMaRS contribution 052.
4. Conclusion
References
This is the first international coordinated effort to assess

the potential of a high resolution spaceborne sensor for coral
reef studies. This is also the first consistent compilation of
coral reef applications from commercially available data,
made available via the NASA SDP program. This work
informs scientists and managers on the type of accuracy
they can expect using IKONOS and Landsat 7 ETM+ data
for a coral reef mapping applications. However, our objectives were partially met. We failed to adequately conduct
similar processing on a variety of sites as we initially
planned. We quickly realized that this is inherent to the
diversity of sites we have selected and to the different image
qualities dependent on environmental conditions (wind,
tide, water quality). Nevertheless, the processing of 10 sites,
completed by 3 independent studies, clearly highlight the
potential of IKONOS for coral reef habitat mapping in
general, using standard methods for reef habitat mapping.
Results can be improved for each site since more sophisticated techniques may work better for a given site and for a
given image. For instance, to achieve high accuracy, we
recommend to pre-segment a reef into different zones if
possible. The comparative approach also suggests which
algorithms need to be improved or developed in the future
(bathymetric correction, contextual edition, and texture) for
a given type of reefs.
Acknowledgements
Obviously, this study would not have been possible
without the spirit and crew of the NASA Scientific Data
Purchase program at Stennis Space Center under the
successive responsibility of Fritz Pollicelli and Troy
Frisbie. We greatly acknowledge their help in the tasking
requests and their support of our research since 1999.
Andrew Mattee and Michael Satter were our contacts at
Space Imaging. This research was supported by NASA
grants NAG5-10908 to SA and NAG-3446 to FMK and
Kendall Carder. Many individuals helped gather field data
on the various sites. We are indebted to Chris Roelfsema,
Bill Dennison, Fabienne Bourdelin, Claude Payri, Bernard
Thomassin, Michel Pichon, Hajime Kayanne, Saki Harii,
Yoshiyuki Tanaka, Ernesto Arias-Gonzales, Don Hickey,
Nancy Dewitt, Tonya Clayton, David Palandro, Chuanmin
Hu, and the staffs of Heron Island Research Station,
Biscayne National Park, and Service des Peches et de
lEnvironment Marin de Mayotte. We are grateful to GISLAGMAY (Mayotte) and the World Bank, via Andy
Ahmad, W., & Neil, D. T. (1994). An evaluation of Landsat Thematic

Mapper (TM) digital data for discriminating coral reef zonation:
Heron Reef (GBR). International Journal of Remote Sensing, 15,
2583 2597.
Andrefouet, S., Berkelmans, R., Odriozola, L., Done, T., Oliver, J., &
Muller-Karger, F. E. (2002). Choosing the appropriate spatial resolution
for monitoring coral bleaching events using remote sensing. Coral
Reefs, 21, 147 154.
Andrefouet, S., Claereboudt, M., Matsakis, P., Page`s, J., & Dufour, P.
(2001). Typology of atolls rims in Tuamotu archipelago (French Polynesia) at landscape scale using SPOT-HRV images. International Journal of Remote Sensing, 22, 987 1004.
Andrefouet, S., Muller-Karger, F., Hochberg, E., Hu, C., & Carder, K.
(2001). Change detection in shallow coral reef environments using Landsat 7 ETM+ data. Remote Sensing of Environment, 79, 150 162.
Andrefouet, S., & Payri, C. (2001). Scaling-up carbon and carbonate metabolism in coral reefs using in situ and remote sensing data. Coral
Reefs, 19, 259 269.
Andrefouet, S., Roux, L., Chancerelle, Y., & Bonneville, A. (2000). A
fuzzy possibilistic scheme of study for objects with indeterminate boundaries: Application to French Polynesian reefscapes. IEEE Transactions
on Geoscience and Remote Sensing, 38, 257 270.
Capolsini, P., Andrefouet, S., Rion, C., & Payri, C. (2003). A comparison of
Landsat ETM+, SPOT HRV, IKONOS, ASTER and airborne MASTER
data for coral reef habitat mapping in South Pacific islands. Canadian
Journal of Remote Sensing, 29, 187 200.
Chauvaud, S., Bouchon, C., & Manie`re, R. (1998). Remote sensing techniques adapted to high resolution mapping of tropical coastal marine
ecosystems (coral reefs, seagrass beds and mangrove). International
Chauvaud, S., Bouchon, C., & Manie`re, R. (2001). Cartographie des biocenoses marines de Guadeloupe a` partir des donnees SPOT (recifs coralliens, phanerogammes marines, mangroves). Oceanologica Acta, 24,
S3 S16.
Congalton, R., & Green, K. (1999). Assessing the accuracy of remotely
sensed data: Principles and practices. New York: Lewis Publishers.
IKONOS satellites, imagery, and products. Remote Sensing of Environment, 88, 23 36.
Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote Sensing of Environment, 80, 185 201.
Frouin, P., & Hutchings, P. A. (2001). Macrobenthic communities in a
tropical lagoon (Tahiti, French Polynesia, central Pacific). Coral Reefs,
19, 277 286.
Green, E. P., Mumby, P. J., Edwards, A. J., & Clark, C. D. (1996). A review
of remote sensing for the assessment and management of tropical coastal resources. Coastal Management, 24, 1 40.
Green, E. P., Mumby, P. J., Edwards, A. J., & Clark, C. D. (2000). Remote
sensing handbook for tropical coastal management. Paris: UNESCO.
Harii, S., & Kayanne, H. (submitted for publication). Larval dispersal,
recruitment, and adult distribution of the brooding stony octocoral Heliopora coerulea on Ishigaki island, southwest Japan. Coral Reefs.
Hochberg, E., & Atkinson, M. (2000). Spectral discrimination of coral reef
benthic communities. Coral Reefs, 19, 164 171.
Hochberg, E., & Atkinson, M. (2003). Capabilities of remote sensors to
classify coral, algae and sand as pure and mixed spectra. Remote
Hochberg, E. J., Andrefouet, S., & Tyler, M. R. (2003). Sea surface cor-

rection of high spatial resolution IKONOS images to improve bottom
mapping in near-shore environments. IEEE Transactions on Geociences
and Remote Sensing, 41, 1724 1729.
Jaap, W. C. (1984). The ecology of the South Florida coral reefs: A community profile. Report Fish Wildlife Service OBS-82/08.
Joyce, K. E., Phinn, S. R., Roelfsema, C., Neil, D. T., & Dennison, W. C.
(2002). Mapping the southern Great Barrier Reef using Landsat ETM
and the Reef Check classification scheme. Proceedings of 11th Australasian Remote Sensing and Photogrammetry Conference. Brisbane:
Causal Publications, CDROM.
Jupp, D. L. B., Mayo, K., Kuchler, D. A., VanClaasen, D., Kenchington,
R. A., & Guerin, P. R. (1985). Remote sensing for planning and managing the Great Barrier Reef of Australia. Photogrammetria, 40, 21 42.
Kayanne, H., Harii, S., Ide, Y., & Akimoto, F. (2002). Recovery of coral reef
populations after the 1998 bleaching on Shiraho Reef in the southern
Ryukyus, NW Pacific. Marine Ecology Progress Series, 239, 93 103.
Kramer, P. A., & Kramer, P. R. (2000). Ecological status of the MesoAmerican Barrier Reef System. Impacts of Hurricane Mitch and 1998
coral bleaching. Report to the World Bank, RSMAS, University of
Miami, 73 pp.
Kramer, P. A., Kramer, P. R., & Ginsburg, R. N. (1998). Assessment of
coral reef health, Andros Barrier Reef, Bahamas. Progress Report,
RSMAS-University of Miami.
Liceaga-Correa, M. A., & Euan-Avila, J. I. (2002). Assessment of coral reef
bathymetric mapping using visible LANDSAT TM data. International
Lyzenga, D. R. (1981). Remote sensing of bottom reflectance and water
attenuation parameters in shallow water using aircraft and Landsat data.
Maeder, J., Narumalani, S., Rundquist, D., Perl, R., Schalles, J., Hutchins,
K., & Keck, J. (2002). Classifying and mapping general coral-reef
structure using IKONOS data. Photogrammetric Engineering and Remote Sensing, 68, 1297 1305.
Maritorena, S. (1996). Remote sensing of the water attenuation in coral
reefs: A case study in French Polynesia. International Journal of Remote Sensing, 17, 155 166.
Matsunaga, T., & Kayanne, H. (1997). Observation of coral reefs on Ishigaki island, Japan, using Landsat TM images and aerial photographs.
Proceedings of 4th Int. Conf. on Remote Sensing for Marine and Coastal Environments ( pp. 657 666), Orlando, FL.
McClanahan, T., & Muthiga, N. (1998). An ecological shift in a remote coral
atoll of Belize over 25 years. Environmental Conservation, 25, 122 130.
Menges, C. H., Hill, G. J. E., & Ahmad, W. (1998). Landsat TM data
and potential feeding grounds for threatened marine species turtle in
northern Australia. International Journal of Remote Sensing, 19,
1207 1221.
Mumby, P. J. (2001). Beta and habitat diversity in marine systems: A new
approach to measurement, scaling and interpretation. Oecologia, 128,
274 280.
Mumby, P., Clark, C. D., Green, E. P., & Edwards, A. J. (1998). Benefits of
water column correction and contextual editing for mapping coral reefs.
Mumby, P. J., & Edwards, A. J. (2002). Mapping marine environments
with IKONOS imagery: Enhanced spatial resolution can deliver greater
thematic accuracy. Remote Sensing Environment, 82, 248 257.
Mumby, P., Green, E. P., Clark, C. D., & Edwards, A. J. (1998). Digital
analysis of multispectral airborne imagery of coral reefs. Coral Reefs,
17, 59 69.
Mumby, P. J., Green, E. P., Edwards, A. J., & Clark, C. D. (1999). The costeffectiveness of remote sensing for tropical coastal resources assessment
and management. Journal of Environmental Management, 3, 157 166.
Mumby, P. J., & Harborne, A. R. (1999). Development of a systematic
classification scheme of marine habitats to facilitate regional manage-
143
ment and mapping of Caribbean coral reefs. Biological Conservation,

88, 155 163.
Palandro, D., Andrefouet, S., Dustan, P., & Muller-Karger, F. E. (2003).
Change detection in coral reef communities using the IKONOS sensor and historic aerial photographs. International Journal of Remote
Sensing, 24, 873 878.
Paredes, J. M., & Spero, R. E. (1983). Water depth mapping from passive
remote sensing data under a generalized ratio assumption. Applied Optics, 22, 1134 1135.
Polcyn, F. C., Brown, W. L., & Sattinger, I. J. (1970). The measurement of
water depth by remote sensing techniques. Report 8973-26-F, Willow
Run Lab., University of Michigan, Ann Arbor.
Purkis, S., Kenter, J. A. M., Oikonomou, E. K., & Robinson, I. S. (2002).
High resolution ground verification, cluster analysis and optical model
of reef substrate coverage on LANDSAT TM imagery (Red Sea, Egypt).
Quod, J. P., Bigot, L., Dutrieux, E., Maggiorani, J. M., & Savelli, A.
(1995). La reserve de la Passe en S (Ile de Mayotte). Expertise
biologique et cartographie des peuplements benthiques. Report ARVAM/IARE/SCE DES PECHES, Sainte Clotilde, La Re union,
31 pp. + appendices.
Riegl, B. (1999). Corals in a non-reef setting in the Arabian Gulf (Dubai,
UAE): Fauna and community structure in response to recurring mass
mortality. Coral Reefs, 18, 63 73.
Riegl, B., Korrubel, J. L., & Martin, C. (2001). Mapping and monitoring of
coral communities and their spatial patterns using a surface based video
method from a vessel. Bulletin of Marine Science, 69, 869 880.
Roelfsma, C., Phinn, S., & Dennisson, C. W. (2002). Spatial distribution of
benthic microalgae on coral reefs determined by remote sensing. Coral
Reefs, 21, 264 274.
Rogers, R. W. (1997). Brown algae on Heron Reef flat, Great Barrier Reef,
Australia: Spatial, seasonal and secular variation in cover. Botanica
Marina, 40, 113 117.
Siegel, D. A., Wang, M., Maritorena, S., & Robinson, W. (2000). Atmospheric correction of satellite ocean color imagery: The black pixel assumption. Applied Optics, 39(21), 3582 3591.
Smith, B. T., Frankel, E., & Jell, J. S. (1998). Lagoonal sedimentation and
reef development on Heron Reef, southern Great Barrier Reef province.
In G. Camoin, & P. Davies (Eds.), Reefs and carbonate platforms in
the Pacific and Indian Oceans ( pp. 281 294). Oxford: Blackwell
Science.
Stehman, S. V. (1997). Selecting and interpreting measures of thematic
classification accuracy. Remote Sensing of Environment, 62, 77 89.
Stehman, S. V. (1999). Basic probability sampling designs for thematic
map accuracy assessment. International Journal of Remote Sensing,
20, 2423 2441.
Stoddart, D. R. (1966). Reef studies at Addu atoll, Maldive Islands: Preliminary results of an expedition to Addu atoll in 1964. Atoll Research
Bulletin, 116, 1 122.
Suzuki, H., Matsakis, P., Andrefouet, S., & Desachy, J. (2001). Satellite
image classification using expert structural knowledge: A method based
on fuzzy partition computation and simulated annealing. Proceedings of
Annual Conference of the International Association for Mathematical
Geology (CDROM) Cancun, Mexico.
Thome, K. (2001). Absolute radiometric calibration of Landsat 7 ETM+
using the reflectance-based method. Remote Sensing of Environment,
78, 27 38.
Veron, J. E. N. (1995). Corals in space and time: The biogeography and
evolution of the Scleractinia. Comstock/Cornell.
Yamano, H., & Tamura, M. (in press). Can satellite sensors detect coral
reef bleaching? A feasibility study using radiative transfer models in
air and water. Proceedings of 9th Int. Coral reef Symposium, Bali,
Indonesia.

Extending satellite remote sensing to local scales: land and water resource
monitoring using high-resolution imagery
Kali E. Sawaya*, Leif G. Olmanson, Nathan J. Heinert, Patrick L. Brezonik, Marvin E. Bauer
Department of Forest Resources, University of Minnesota, 1530 N. Cleveland Avenue St. Paul, MN 55108-6112, USA
Received 22 May 2002; received in revised form 19 March 2003; accepted 24 April 2003
Abstract
The potential of high-resolution IKONOS and QuickBird satellite imagery for mapping and analysis of land and water resources at local
scales in Minnesota is assessed in a series of three applications. The applications and accuracies evaluated include: (1) classification of lake
water clarity (r2 = 0.89), (2) mapping of urban impervious surface area (r2 = 0.98), and (3) aquatic vegetation surveys of emergent and
submergent plant groups (80% accuracy). There were several notable findings from these applications. For example, modeling and estimation
approaches developed for Landsat TM data for continuous variables such as lake water clarity and impervious surface area can be applied to
high-resolution satellite data. The rapid delivery of spatial data can be coupled with current GPS and field computer technologies to bring the
imagery into the field for cover type validation. We also found several limitations in working with this data type. For example, shadows can
influence feature classification and their effects need to be evaluated. Nevertheless, high-resolution satellite data has excellent potential to
extend satellite remote sensing beyond what has been possible with aerial photography and Landsat data, and should be of interest to resource
managers as a way to create timely and reliable assessments of land and water resources at a local scale.
Keywords: IKONOS; Remote sensing; High resolution imagery; Lake clarity; Aquatic vegetation; Impervious surface
1. Introduction
Although high-resolution imagery in the form of aerial
photography has been available for many years, the launch of
the IKONOS-2 by Space Imaging in September 1999 has
signaled a new era in satellite remote sensing. With multispectral digital imagery approaching that of small to medium
scale photography, we are in a new period of applications
development. Our objective has been to evaluate high-resolution satellite imagery in a variety of applications involving
monitoring of land and water resources. We approach each of
three applications at a local scale and address a pertinent
aspect of water quality in Minnesota. Each application
utilizes a single-date scene of IKONOS or QuickBird imagery covering approximately 11 11 km areas.
To demonstrate the potential of the imagery, these
applications each take a closer look at digitally classifying
a particular aspect of monitoring and mapping land and
E-mail address: ksawaya@gis.umn.edu (K. Sawaya).
doi:10.1016/j.rse.2003.04.0006
water resources. We include: (1) the classification of lake

water clarity, (2) the determination of urban impervious
surface area, and (3) aquatic vegetation surveys of emergent and submergent plant groups. Each application
attempts to take advantage of the increased spatial resolution and the multispectral properties of the data to digitally
classify aspects of the environment that have not been
possible to this degree of detail with satellite imagery in
the past. The first two applications take advantage of
existing methodologies for moderate resolution Landsat
data that have been modified for use with high-resolution
satellite data. These applications seem to make a smooth
transition to the new imagery type and should provide an
out of the box solution to resource managers.
Throughout the paper we address several considerations
common to applications development with high-resolution
satellite imagery including:
1. Imagery acquisition parameters. Appropriate acquisition
windows for each type of application are identified and
the challenges inherent in working with large geographic
extents of high-resolution satellite data are discussed.
K. Sawaya et al. / Remote Sensing of Environment 88 (2003) 144156
2. Field data sampling. Considerations for working with

point and polygon field data, preferred characteristics of
field reference data for use with high-resolution data, and
alternative approaches when ideal characteristics are not
available are discussed.
3. Spectral confusion. Instances of spectral confusion
associated with cover type classifications and how we
have amended our methods to account for them are
described.
4. Shadows. The effects of shadows on image classification
and ways these errors can be accounted for or avoided are
discussed.
5. Accuracy assessments. Approaches to modeling and
assessing the accuracy of output maps using various field
data sets, Landsat data of a similar date, and aerial photo
interpretation are discussed.
The paper is organized into three individual sections,
each reporting on a separate application and approach. A
concluding section revisits and synthesizes the above
considerations.
2. Lake water clarity classification

2.1. Background
Lakes are important recreational and aesthetic resources
that add to economic stability and quality of life. Protecting
and monitoring lake water quality is a major concern for
many local and state agencies. However, because of expense
and time requirements for ground-based monitoring, it is
impractical to monitor more than a small fraction of lakes by
conventional field methods. High-resolution satellite remote
sensing is another tool that can potentially be applied to
gather information needed for water clarity assessments in
lake-rich areas like Minnesota.
Results from several studies (e.g., Kloiber, Brezonik, &
Bauer, 2002; Kloiber, Brezonik, Olmanson, & Bauer, 2002;
Lillesand, Johnson, Deuell, Lindstrom, & Meisner, 1983)
have demonstrated a strong relationship between Landsat
Multispectral Scanner (MSS) or Thematic Mapper (TM)
data and ground observations of water clarity and chlorophyll a. In Minnesota, Olmanson, Bauer, and Brezonik
(2002) have developed a water clarity image processing
methodology and completed statewide assessments of over
10,500 lakes for the f 1990 and f 2000 time periods using
Landsat imagery. In this statewide assessment, water clarity
was evaluated for lakes over 8 ha (20 acres) in size, while
the smaller lakes and ponds were excluded due to the
relatively low spatial resolution of the Landsat imagery.
IKONOS imagery has four multispectral bands similar to
Landsat TM bands 1 4 and high spatial resolution, making
it a good candidate for applying previous methods to the
assessment of smaller lakes and ponds. In this study, we
used a September 4, 2001 IKONOS high-resolution satellite
145
image to assess the water clarity of smaller lakes and ponds

for a city scale analysis of water quality. These results are
compared to a lake water clarity classification using an
August 30, 2001 Landsat TM image. Assessment of smaller
lakes and ponds is important since they tend to be more
susceptible to impacts than larger lakes.
The overall objective of our research was to estimate
variables related to key management indicators, such as the
trophic state indices of Carlson (1977). The three common
water quality variables that indicate lake trophic state are
total phosphorus (TP), chlorophyll a (chla), and Secchi disk
transparency (SDT). Lake management agencies and organizations use these variables for measurements, along with
various transformations such as the trophic state indices
(TSI). SDT is the most consistently collected trophic state
indicator, and it is strongly correlated with the responses in
the blue and red bands of Landsat TM/ETM+ data (Kloiber,
Brezonik, & Bauer, 2002; Kloiber, Brezonik, Olmanson et
al., 2002). Therefore, most of our research to date has
involved calibrating Landsat TM data with ground-based
SDT measurements and estimating SDT for all lakes in an
image from the regression equation developed in the calibration step. The results then can be mapped directly as
distributions of SDT in the lakes, or the estimated SDT can
be converted to Carlsons trophic state index based on
transparency: TSI(SDT) = 60 14.41 ln(SDT).
It is important to recognize that other factors besides
algal turbidity (as indicated by chlorophyll levels) may
affect SDT in lakes. Most important of these (non-trophicstate) factors are humic color and non-algal turbidity (including soil-derived clays and suspended sediment). For this
reason, we report our results based on SDT calibrations as
satellite-estimated SDT or TSI(SDT), which clearly identifies the value as an index based on transparency, rather than
the generic term, TSI.
The specific objective for this study was to perform an
assessment of TSI(SDT) for the City of Eagan, MN. This
area was particularly well suited for this study since it has
375 small lakes, ponds, and wetlands and a well-established
lake monitoring program.
2.2. Methods
We used methods developed by Olmanson (1997) and
continued in subsequent studies (e.g., Kloiber, Brezonik, &
Bauer, 2002; Kloiber, Brezonik, Olmanson et al., 2002;
Olmanson et al., 2002) to apply Landsat imagery to regional
scale assessments of lake water clarity. We made minor
modifications including the addition of a lake polygon layer
to minimize spectral confusion between open water, shadows and asphalt features to make these methods compatible
with high-resolution satellite imagery.
2.2.1. Satellite imagery and lake reference data
Images for water clarity assessment were selected from a
late summer index period (July 15 September 15, with a
146
preference for August). This period was found to be the best

index period for remote sensing of water clarity in Minnesota (Kloiber, Brezonik, Olmanson et al., 2002). There are at
least two major advantages to using images from this
period: (1) short-term variability in lake water clarity is at
a seasonal minimum, and (2) most lakes have their minimum water clarity during this period. In addition, it is
preferable to have images from near anniversary dates for
change detection.
We acquired two satellite images, August 30, 2001
Landsat TM (path 27, rows 29 and 30) and September 4,
2001 IKONOS of Eagan, MN for this assessment. Both
images were of high quality and free of haze, clouds and sun
glint. It is critical to avoid IKONOS images displaying
specular reflection, or sun glint effects from lakes for this
application. Certain combinations of IKONOS view azimuths and zenith angles can result in bidirectional reflectance that saturates the sensor, making the data unusable for
this and other water related applications. Although the
imagery used in this analysis was suitable, specular reflectance effects have created problems for some of the other
images we acquired for lake water clarity assessments
between 1999 and 2001. We did not perform atmospheric
correction or normalization of the imagery for the regression
method used.
The availability of lake reference data was excellent due
to the City of Eagans Water Resources program and
volunteer participation in the Citizen Lake Monitoring
Program. The programs provided 94 SDT lake reference
points for the Landsat image taken within three days of the
image acquisition and 13 SDT lake reference points for the
IKONOS image taken within seven days of the image
acquisition. The SDT data were distributed over a wide
range of water quality.
2.2.2. Classification procedures
This section summarizes our image classification procedures; more detail is provided by Olmanson et al. (2002),
and the rationale for the procedures is described by Kloiber,
Brezonik, Olmanson et al. (2002). We used ERDAS Imagine image processing software, and ArcView geographical
information system (GIS) software, for the image processing
steps. Acquiring a representative image sample from each
lake used for calibration or accuracy assessment was our
primary objective. Ideally, the sample should represent the
center portion of the lake in at least five meters of water (or
twice the SDT measurement) where reflectance from vegetation, the shoreline, or the lake bottom do not affect the
spectral response. It was also critical to avoid shadowed
portions of lakes, which would lead to unnaturally clear lake
estimations.
We produced a water-only image by performing an
unsupervised classification in ERDAS Imagine. Because
water features tend to have very different spectral characteristics from terrestrial features, water was put into one or
more distinct classes that we could easily identify. We then
masked out terrestrial features creating a water-only image.

This method worked well for the Landsat imagery; however, with IKONOS imagery this clustering also included
pixels from shadows, asphalt, and other dark features.
Therefore, we created a lake polygon layer to help mask
the non-water dark features on the IKONOS image. Then
we performed a second unsupervised classification on the
water-only image and generated spectral signatures of each
class. We used these signatures, along with the location
where the pixels occur, to differentiate classes containing
clear water, turbid water, and shallow water (where sediment and/or macrophytes affect spectral response). Based
on this analysis, we recoded classes to avoid shadow,
vegetation, bottom, and terrestrial effects when selecting
lake sample locations. Digital number values from the
imagery were obtained to develop relationships with measured SDT. For this assessment, we used a polygon layer,
described in Olmanson et al. (2002), to help automate the
process. We used the signature editor in ERDAS Imagine
to extract the spectral data from the image for each sample
location.
We calibrated the Landsat image using the 94 SDT
measurements collected within three days of the image
acquisition date. Using log-transformed SDT data as the
dependent variable and Landsat Thematic Mapper band 1
(TM1) and the TM1/TM3 ratio as independent variables, we
performed a multiple regression. The regression model, with
r2 = 0.76 and SEE = 0.39, for prediction of SDT from the
Landsat TM data was:
lnSDT 1:493*TM1:TM3 0:035TM1 1:956 1
Once we developed the model for the Landsat image, we
used another polygon layer with all of Eagans lakes, ponds,
and wetlands 0.08 ha and larger to extract data from all
possible water-bodies with enough unaffected pixels to
predict water clarity. Forty-eight of Eagans lakes and ponds
had a sample of at least eight unaffected pixels that we used
to predict water clarity. We used the model above, developed from the entire Landsat image, to predict water clarity
for 48 of Eagans lakes and ponds. For this assessment, the
smallest pond that we were able to predict water clarity for
using the Landsat TM image was 1.5 ha.
We then calibrated the IKONOS image using two different datasets. The first dataset was 13 SDT measurements
collected within seven days of the image acquisition date.
The other dataset was 48 water clarity measurements
extracted from the Landsat image. We used this dataset to
compare the compatibility of the IKONOS imagery with
Landsat TM imagery and to explore an optional calibration
method if sufficient ground observations were not available
for a given scene. We performed a multiple regression using
log-transformed SDT data as the dependent variable and
IKONOS band 1 (IK1) and the IK1:IK3 ratio as independent
variables.
Lake water clarity maps can then be created from the

regression model by two methods. The first method
applies the model to each individual water pixel. This
method creates a pixel-level lake map. With this map, all
water pixels are classified and intra-lake variability can be
evaluated. The second method uses the digital numbers
collected from the sample of each lake to calculate an
average water clarity estimate. The data can then be
linked to a lake polygon layer to create a lake-level water
clarity map. This latter method has the advantage of
generating a single water clarity number for each lake
that can be used in other analyses or included in a water
clarity database.
2.3. Results and discussion
This assessment showed strong relationships between
both water clarity datasets (TM-derived and SDT-derived)
and the spectral-radiometric response of the IKONOS
data. The regression model (n = 48, r2 = 0.94, SEE = 0.15)
for prediction of SDT from the IKONOS data using the
water clarity data derived from the Landsat TM image
was:
lnSDT 1:958*IK1:IK3 0:004IK1 2:957
The regression model (n = 13, r2 = 0.89, SEE = 0.22) for

prediction of SDT from the IKONOS data using the
available SDT data (Fig. 1) was:
lnSDT 1:832*IK1:IK3 0:004IK1 2:813
The very strong relationship of r2 = 0.94 using the

Landsat-derived SDT to calibrate the IKONOS image
indicates that the two types of imagery are compatible
147
and have similar spectral-radiometric responses. The strong

relationship between SDT and spectral-radiometric response of the IKONOS imagery that are similar to results
we have seen with Landsat imagery indicates that IKONOS imagery can be used to assess the water clarity of
smaller lakes and ponds. The similarity in the resulting
models also indicates that Landsat imagery can be used to
calibrate IKONOS images that do not have sufficient field
reference data. A comparison of the resulting TSI(SDT)
values calculated from the Landsat image and from the
IKONOS image using each model for 48 lakes and ponds
both indicated a very strong agreement with an r2 = 0.95
(Fig. 2).
The 4-m resolution of the IKONOS imagery allowed us
to assess smaller water bodies than is possible with
Landsat 30-m resolution imagery. We easily assessed
water bodies with a minimum size of 0.08 ha included
in the lake, pond, and wetland polygon layer when open
water conditions and unaffected pixels existed. We used
347 lake, pond, and wetland polygons 0.2 ha and larger
on the Eagan lake polygon layer to extract the spectralradiometric data from the imagery. Of those lakes and
ponds, 236 had a sample of at least eight unaffected pixels
used to predict water clarity. In contrast, we could only
estimate and map the water clarity of 48 lakes and ponds
in Eagan when using Landsat imagery, as the minimum
lake size assessed with Landsat 30-m resolution imagery
was 1.5 ha.
High-resolution satellite data were also useful for general analysis of city land use and land cover features. Visual
assessment of how land use/cover affects water clarity can
be investigated by overlaying the classified lakes on the
original IKONOS imagery. Fig. 3 shows some of the
different land use/cover features and corresponding effects
on water clarity. For example, in the southeastern corner of
Eagan, the Lebanon Hills Regional Park is an area with an
Fig. 1. Comparison of IKONOS and field measurements of lake water clarity, TSI(SDT).
148
Fig. 2. Comparison of Landsat and IKONOS estimates based on Landsat modeling of lake water clarity, TSI(SDT).
abundance of forest and wetland areas and limited development in the form of parking lots and pavilions for park
visitors. This area has relatively high lake water clarity with
TSI(SDT) of around 50 and SDT of f 2 m. In many of the
residential and commercially developed areas, stormwater
is directed into lakes using them as convenient reservoirs.
The increase in impervious surface area and direct connection to the stormwater system has dramatically changed the
hydrology of many water bodies in Eagan. These changes
of increased watershed size, amount of runoff and quality

of runoff water have impacted many of Eagans water
bodies. These impacts can be seen in Fig. 3 where water
clarity of lakes and ponds in many of the developed areas is
generally TSI(SDT) 65 70 and SDT of f 0.5 m due to
eutrophication.
This study showed the usefulness of high-resolution
satellite imagery for water clarity assessments at a city
scale. This type of information can be very useful in city
Fig. 3. Lake water clarity classification of IKONOS multispectral data overlaid on panchromatic land image. (Imageryn Space Imaging L.P.).
planning and lake management to help monitor and protect

water clarity at a local scale.
3. Impervious surface mapping

3.1. Background
Impervious surfaces, or areas impenetrable by water,
negatively affect the natural environment. These built environmentsincluding roads, rooftops, sidewalks and parking
lotsincrease the rate of stormwater runoff, which transport
surface pollutants to receiving lakes and ponds. Thus,
impervious surfaces threaten water quality (Barnes, Morgan,
& Roberge, 2000; Sleavin, Civco, Prisloe, & Giannatti,
2000). In addition, Estes, Gorsevski, Russel, Quattrochi,
and Luvall (1999) suggest that impervious surfaces are
related to energy balances, urban heat island effects, habitat
fragmentation, and poor landscape aesthetics. To monitor
these impacts, resource managers have quantified the degree, extent, and spatial distribution of impervious surface
areas using various methods including ground surveys,
aerial photography interpretation, and satellite remote sensing. Satellite imagery, particularly Landsat TM and ETM+,
has recently emerged as an approach with the capability to
effectively estimate the percentage of impervious cover in
urban areas (Bauer, Doyle, & Heinert, 2002; Civco & Hurd,
1997; Ridd, 1995).
Previous research using Landsat TM data acquired in
1986, 1991, 1998, and 2000 across the Twin Cities Metropolitan Area suggests that a strong relationship (correlation
f 0.9) exists between the Landsat spectral-radiometric
response and percent impervious surface area calculated
from measurements of digital orthophoto quadrangle
(DOQs) imagery (Bauer et al., 2002). In this study, we
149
investigate whether a similar method is suitable for highresolution IKONOS imagery at a local scale for the City of
Eagan, MN. Eagan is located southwest of St. Paul, MN and
covers 89 km2. In the 1990s, Eagans population growth
increased 33%, reaching 63,557 in 2000. The landscape
consists mostly of single/multi-family residential, commercial/industrial, parks and recreational areas, vacant and
agricultural lands, and numerous small lakes and ponds.
3.2. Methods
We performed several background tests to determine
which spectral transformation provided the strongest relationship to percent impervious surface area. For IKONOS
data, a correlation of 0.90 0.95 suggested that the normalized difference vegetation index (NDVI) provided the best
relationship. Other transformations considered were IR/red
ratio and principal components. The ratio had a correlation
of 0.93 and the second principal component had a correlation of 0.86 with percent impervious. The ratio and the
NDVI had similar results, but we selected NDVI due to its
familiarity. In past research with Landsat TM imagery, we
used tasseled cap greenness (Bauer et al., 2002), but this
was not possible since the coefficients for a tasseled cap
transformation of IKONOS data have not been determined.
IKONOS digital imagery of the study area was acquired
on September 4, 2001. The imagery was clear and cloudfree; therefore, no atmospheric correction was necessary.
Using 270 ground control points, we rectified and registered
the imagery to DOQs covering the same area, with an RMS
error of 1.24 m.
We began our image processing by performing a general
land cover classification using a supervised, maximum
likelihood approach in ERDAS Imagine. Due to spectral
confusion among some water features and asphalt, we hand
Fig. 4. Sample areas delineated over Digital Orthophoto Quadrangles (DOQs) used to calculate varying degrees of impervious surface area for model
calibration and evaluation.
150
digitized open water and wetland areas using heads up

digitization from the IKONOS imagery. Due to spectral
confusion between bare areas and concrete, we also hand
digitized bare fields and extraction areas. Using these
techniques, we were able to achieve an overall classification
accuracy of 93% for urban, vegetation, bare areas, and
water. Pixels from the urban class were retained for the
determination of percent imperviousness. Vegetation and
bare classes were recoded as 0% impervious.
Using DOQs, we then selected well distributed samples
across the study area to represent varying degrees of
imperviousness. The objective was to obtain a complete
range of percent imperviousness with at least two samples
for each 10% interval. Sample areas ranged in size from 300
to 2400 pixels. We used ArcView 3.2 to digitize the
boundaries around each impervious feature and calculate
the impervious surface percentage for that individual sample
area (Fig. 4).
After calculating and attributing the percent imperviousness for each sample area, we determined the mean NDVI
value for each sample area by overlaying the sample onto
the NDVI transformed IKONOS image. We then created a
second order polynomial regression equation with the mean
NDVI value as the independent variable and percent impervious surface area as the dependent variable. With this
regression equation, we estimated impervious surface percent for each image pixel and rescaled the output values
from 0% to 100%. To evaluate classification accuracy, we
compared 25 additional independent samples of impervious
surface area determined from the DOQs to the IKONOS
map of imperviousness.

The results of the regression analysis showed a very
strong relationship (r2 = 0.95) between the NDVI and per-
cent imperviousness (Fig. 5). The resulting polynomial

equation used to estimate percent imperviousness across
the image was:
Y 44:339X 2 196:39X 95:645
The agreement between IKONOS estimates and measurements from the additional sample areas of DOQs was
very high with a linear correlation of 0.98 (Fig. 6). Fig. 7
displays a map of all impervious surfaces in the City of
Eagan as a continuous gray scale of 0 100% impervious
surface area. Note the degree of impervious detail that is
possible to discern in the built environment including bike
trails, vegetation along road medians, and individual
driveways. Resource managers can now recode the 0
100% scale to appropriate indices for their individual
analysis.
The land cover classification step helped to stratify and
improve the accuracy of the final map. Separating water,
vegetation, and bare areas helped prevent mapping several
pervious features as impervious. Without masking, open
water features are erroneously classified as 80 100%
impervious due to their low NDVI. The presence of tree
canopy shadows and bare patches in lawns can also
introduce a small amount of error into the impervious
surface estimations. For future land cover classifications
of high-resolution imagery, we anticipate using objectbased classification methods to minimize some of these
effects.
Shadows from buildings and other tall features affected
the overall classification of impervious surface in highresolution satellite imagery. They tended to be most problematic around the perimeter of forested and urban areas.
Shadows have low NDVI values, thus giving a false
measure of impervious features in certain areas. It is difficult
to ascertain what features are within a shadowed pixel using
digital classification. Therefore, manual image interpretation
Fig. 5. Relationship of percent impervious surface area determined from DOQs and IKONOS NDVI.
151
Fig. 6. Comparison of measured and IKONOS estimates of percent impervious surface area.
may be necessary if it is crucial to know the composition of

shadowed pixels. The high radiometric resolution option
(11-bit for IKONOS) may also provide a means to resolve
and characterize ground features within shadowed areas and
will be evaluated in future studies. Though the presence of
shadows can affect accuracy in this application, they tend to
affect individual pixels or small groups of pixels. Most
environmental models would analyze numerous pixels together (watershed, municipality, etc.), minimizing this error
across the landscape.
The September 4, 2001 imagery represented a time of
leaf-on conditions where tree crowns and other vegetation
could potentially obscure some impervious features. Our
work with Landsat data indicates that these conditions

introduce a statistically insignificant amount of error. We
have recently acquired leaf-off IKONOS imagery (November 2002) to determine differences or advantages between
using leaf-on, leaf-off, or a combination of the two with
high resolution satellite imagery.
We believe this approach to mapping impervious surface
is superior to assigning an assumed impervious value to a
land cover type of an existing map. We suggest that a pixelby-pixel approach is more accurate and better represents the
gradation of impervious values across features in neighborhoods and industrial areas. This application takes advantage
of the increased spatial resolution of the high-resolution
Fig. 7. Maps of percent impervious surface area derived from classification of IKONOS NDVI data.
152
satellite imagery and should be of interest to natural

resource and city managers who need to evaluate stormwater runoff, heat island effects, the protection of green
corridors, and other types of assessments quantifying built
environments.
5. Aquatic vegetation surveys

5.1. Background
While lakes are well known for their recreational and
aesthetic value, traditionally, society has considered wetlands as nuisances and problems that need to be cleaned
up. Statewide, Minnesota has drained over half of the
original wetlands (around 4.5 million hectares) for agricultural and development purposes, and many of the remaining wetlands are degraded. A frequent cause of wetland
degradation is increased storm water discharge resulting
from changes such as increases in impervious surface area
or installation of storm water systems in urban and suburban areas, as well as tiling and ditching systems in
agricultural areas. Changes in hydrology affect the water
quality and quantity, and may severely impact the function
of wetlands (Daily, 1997). When too many wetland plants
are removed or impacted, water quality, wildlife, and fish
populations can suffer (Mitsch & Gosselink, 1993). These
plants are important because they help protect water
quality, provide habitat for fish and wildlife, and provide
economic and aesthetic opportunities (Barbier, Burgess, &
Folke, 1994).
Aquatic plants in lakes and wetlands are beginning to
be recognized as important ecosystem features in need of
protection. As a result of this greater appreciation for
aquatic plants in wetland and lake environments, aquatic
plant surveys and assessments are becoming part of routine
monitoring efforts conducted by consultants, citizen
groups, and state and local agencies. Aquatic plant diversity and abundance are important indicators of lake or
wetland health, but accurate maps and data are difficult to
acquire. Because ground-based mapping requires much
time and human resources, only a small fraction of this
large resource has been mapped by natural resource
agencies.
The principal objective of this study was to evaluate the
capability of high-resolution satellite imagery to map and
classify aquatic plant groups for use by resource management agencies. To do the evaluation using IKONOS imagery, we conducted an aquatic plant survey on Swan Lake in
Nicollet County, MN. Swan Lake is a large (>3600 ha),
type-4 wetland meaning it is classed as a deep fresh water
marsh with standing water and abundant aquatic vegetation.
To evaluate the use of QuickBird imagery for assessment of
aquatic plants in open water lakes, we conducted aquatic
plant surveys of three lakes south of Lake Minnetonka. The
lakes include Christmas Lake (104 ha) mostly in Hennepin
County, and West Auburn (55 ha) and Shutz (38 ha) Lakes
in Scott County, MN.
5.2. Methods
We adapted methods typically used for land cover classification (Lillesand et al., 1998) and developed for water
clarity assessments (Olmanson et al., 2002) to do the
evaluation. The aquatic plant classification methods consisted of two procedures: separation of image features into
discrete units and classification of the pixels in each unit.
5.2.1. Satellite image data
For this study, we used an IKONOS image acquired on
September 1, 2001 of the Swan Lake area and a QuickBird
image acquired on July 28, 2002 of the Lake Minnetonka
area. An acquisition window of July 15 through September
15 captures the presence of aquatic vegetation in Minnesota.
The images were of high quality with only minor cloud
cover over the southern portion of the IKONOS image. We
applied a resolution enhancement of the multispectral bands
using the panchromatic band. Atmospheric correction or
normalization of the imagery was not necessary for the
methods used in this study.
5.2.2. Aquatic vegetation reference data
Due to the size of Swan Lake and the abundance of
aquatic plants in the lake, the collection of reference data
would be very difficult without the aid of modern technology. We used Global Positioning System (GPS) technology
and the advanced GPS tracking software in ERDAS Imagine 8.5. We collected field reference data shortly after
acquiring the imagery using a Fujitsu pen computer. While
in the field, we identified different types of aquatic vegetation and located them directly on the IKONOS image using
the field computer. Being able to accurately identify specific
locations on the image while in the field was especially
useful on this large wetland. Having the image available
quickly after its acquisition for use in reference data
collection was also a significant advantage in field sampling
because we could identify unique areas with different
spectral-radiometric responses on the image and target them
for field identification. We targeted emergent vegetation for
the evaluation, but also noted the location of submerged
vegetation appearing at the surface.
For the three lakes south of Lake Minnetonka the
Minnesota Department of Natural Resources collected the
field data. For this study the field data specifically targeted
submerged plants using an echo sounder equipped with a
GPS unit. The hydroacoustic survey collected data pertaining to lake depth and plant depth at thousands of points in
the shallower portions of the lakes where aquatic plants
were present but not at the surface. General point survey
data were collected in areas where the submerged plants
were topped at the surface or where there were floating or
emergent plants.
5.2.3. Classification procedures

The first step for the Swan Lake IKONOS image was to
separate wetland features from terrestrial features by digitizing the aquatic terrestrial boundary around the entire
wetland and all islands. We identified this boundary using
spectral-radiometric differences and spatial patterns visible
on the image. We then subset the image with the wetland
polygon to mask out all terrestrial features and create a
wetland-only image.
Swan Lake has a maximum depth of 2 m, clear water
throughout, and an abundance of aquatic vegetation. Consequently, we assumed that aquatic vegetation was present
throughout the wetland. An aquatic vegetation survey conducted by Tyler and Madsen (personal communication), in
which the presence or absence of 27 species of aquatic
plants at 118 evenly distributed sample points on the lake
was recorded, verified this assumption. The next step was to
stratify the wetland into emergent and submergent vegetation by performing an unsupervised classification specifying
10 classes. Emergent vegetation had very different spectral
characteristics from most submerged vegetation and were
put into distinct classes easily identified from the locations
where they occurred. We masked out submerged vegetation
features to create an emergent vegetation image, and conducted a second unsupervised classification on the emergent
image using 100 classes. Because of the difficulty in
separating some emergent plant types and our finding that
some areas with very thick submergent plant mats floating
on the surface were included in the image, we conducted
further cluster busting by stratifying the image further
153
into two emergent images and a thick submergent image.

We performed a third level of unsupervised classifications
using 100 classes for the emergent vegetation images and 10
classes for the thick submergent vegetation image. Using the
field reference data, we identified five different emergent
classes on the emergent vegetation images and recoded the
images to create an emergent vegetation map. We repeated
this procedure for the thick submerged vegetation image and
identified two submerged classes.
Next, we created a submerged vegetation image by
masking the emergent vegetation features. We clustered
the submergent vegetation image into 10 classes to identify
different types of submergent aquatic vegetation. Using a
graph of the spectral-radiometric signatures of the different
classes and the reference data, we identified classes of
different aquatic plant densities. Assuming that water clarity
is similar throughout the wetland and that aquatic plants are
located throughout the wetland, we attributed the differences
in spectral response to differences in submerged aquatic
plant depth and densities. We identified classes with higher
radiometric response as areas where the submerged vegetation was highly dense and appeared at the surface, and areas
with lower radiometric response as areas where submerged
vegetation was deeper or thinner. We created the submerged
vegetation map by combining the two submerged plant
images and recoding the map into four different submerged
plant density classes. Finally, we created an aquatic plant
classification map by overlaying the submerged aquatic
plant map and the emergent aquatic plant map over the
panchromatic image (Fig. 8).
Fig. 8. IKONOS image classification of aquatic vegetation of Swan Lake in Nicollet County, MN (Imageryn Space Imaging L.P.).
154
We followed similar procedures for the three lakes south

of Lake Minnetonka with the exception of also stratifying
the open water areas before classification of the submerged
and emergent/floating aquatic plants.
Stratification and unsupervised classification of the resolution-enhanced multispectral high-resolution imagery provided us with high-resolution maps that identify different
aquatic vegetation groups throughout the Swan Lake and
three lakes. The accuracy evaluation of the lake assessments
is ongoing and will be included in subsequent papers. To
evaluate the accuracy of the Swan Lake aquatic vegetation
map, we had two datasets available. The first is the field
data we collected and used to classify the imagery. Unfortunately, these locations were not random and are biased by
our prior knowledge of the aquatic vegetation types.
The other dataset, collected by Tyler and Madsen (unpublished), has 118 evenly distributed data points for which
they identified the presence or absence of 27 species of
aquatic plants. In reviewing the locations of these data, it
became clear that some points were not accurately positioned on the map. For example, the GPS coordinates of
some aquatic plant points were in forested areas at least 50
m away from the nearest aquatic plants. Therefore, instead
of using this survey for an accuracy assessment, we used it
to compare the statistical distribution of plant species in
Swan Lake.
5.3. Results and discussion
The final classification map for Swan Lake with five
classes of emergent and four classes of submerged aquatic
vegetation is shown in Fig. 8. Qualitative comparison to the
field observations that were initially acquired with the
imagery indicate that the classes of vegetation have been
mapped quite accurately. There was strong agreement between the field survey of plant distribution by Tyler and
Madsen and the IKONOS classification (Table 1). We also
have quantified the accuracy in an error matrix for the
emergent classes and the submerged class as a whole (Table
2). The overall accuracy was 79.5%, with producers and
users accuracies from 36% to 100% for the individual
classes. The highest accuracies occurred with plants that
grow in homogeneous dense beds such as cattail and water
lily/floating leaf pondweed. Plants that tend to have sparse
growing characteristics and areas with more heterogeneous
Table 1
Comparison of IKONOS and field survey estimates of the distribution of
aquatic vegetation diversity in Swan Lake, MN
Submerged plants
Cattail
Sedge and dead sedge
Bulrush
Water lily and FLPW
a
IKONOS (%)
Field survey a (%)
63.6
28.1
2.5
1.6
4.2
58.5
23.7
5.1
6.8
5.9
Tyler and Madsen, unpublished data.
Table 2
Accuracy of IKONOS image classification of aquatic vegetation for Swan
Lake, MN
Classified
data
Reference data
Cattail Sedge Sedge, Bulrush Lily Submerged Users
dead
accuracy
(%)
Cattail
24
3
Sedge
4
Sedge,
dead
Bulrush
Lily
Submerged
4
Producers
100.0 36.4
Accuracy (%)
88.9
66.7
100.0
2
5
7
1
13
100.0 92.9
13
3
62.5
4
53.8
100.0
92.9
54.2
79.5
plant communities were more difficult to classify accurately.

Ideally, the image would be taken back into the field after
classification, shortly after image acquisition, to verify the
classification using the GPS tracking software. Being able to
accurately identify specific locations on the classified image
while in the field would be especially useful for accuracy
assessments and for improving the classification.
Preliminary review of the aquatic plant assessments of
the three lakes south of Lake Minnetonka indicates that
submerged plants can be separated from open water areas
and classified to a plant top depth of around 2 m. It also
appears that submerged plants with more dense growing
characteristics like Eurasian watermilfoil (an exotic to
Minnesota lakes) can be separated from other submerged
plants that tend to have less dense growing characteristics.
Ozesmi and Bauer (2002), in their review of the methods
and results of satellite remote sensing of wetlands, primarily
with Landsat and SPOT data, found that accurately mapping
wetlands is a challenging task. However, this study indicates
that the use of IKONOS and QuickBird imagery for aquatic
plant surveys is promising. Stratification and unsupervised
classification of pan-sharpened multispectral high-resolution
imagery provided a map identifying different aquatic vegetation groups throughout a large type-4 wetland and three
lakes. The high spatial resolution of the imagery enables the
assessment of aquatic vegetation variation within lakes and
wetlands that cannot be obtained from Landsat data and the
multispectral data enable classification beyond what is
commonly done with aerial photography. Future work will
include object-based approaches to classification.
6. Conclusions
Through our experiences with these three projects, we
have identified a number of considerations for applications
development using high resolution satellite imagery. First,
we have had the most success applying high-resolution
imagery to local or city scale analyses. When imagery is
acquired over large geographic areas, scene differences can
and do exist due to different acquisition dates, view angles,

sun angles, and atmospheric conditions. Because of this
reality, we believe that this form of satellite data is best
suited to analyses of relatively small geographic areas (i.e.,
single images), or potentially north south oriented areas
that could fall along a single path. In addition, the cost of
high-resolution imagery may be prohibitively expensive for
large area assessments, but it is likely affordable for many
cities since the imagery cost can be shared between departments and used for several applications.
It has become clear that point data is often insufficient
field reference to calibrate high-resolution imagery. Due to
differences in GPS coordinates and rectification accuracy,
assigning a single field reference point to a pixel is difficult.
However, improved global positioning technology has been
a timely and valuable asset in the acquisition of field
reference data for use with high-resolution imagery. It is
now possible using field computers with GPS capability and
GIS and image processing software, to take high-resolution
satellite imagery into the field and capture points and
polygon reference areas directly correlated with the imagery.
High levels of precision for both the GPS and the imagery
are imperative for this work. The clarity of the imagery also
provides for more visual interpretation than has been previously possible, allowing the analyst greater certainty in
some field referencing.
While the spatial detail of high-resolution imagery is
impressive, the problem of spectral-radiometric similarity
between certain classes is, if anything, compounded. Mixed
pixels are still present and the variability within classes may
be greater. We have identified a number of incidences of
spectral confusion along the range of digital number values
as a result of these three applications. Dark feature confusions include open water, asphalt and shadows, mid-range
confusions include wetlands, shadows and forest damage
and bright feature confusions include concrete, bare fields
and recreational fields.
The presence of shadows that are readily resolved has
been a problem in virtually every application. Shadowed
portions of lakes in clarity assessments can lead to unnaturally clear lake estimations. Shadows from tall buildings in
an impervious surface analysis can skew estimates to greater
amounts of imperviousness than actually exist. The high
radiometric resolution of the data may provide a means to
extract information from areas that would otherwise be
obscured by shadows, but we have not yet had an opportunity to investigate this possibility. In some cases it may be
possible to obtain imagery at dates close to the summer
solstice to minimize shadows.
One of the challenges of accuracy assessment of highresolution classification is to match the resolution of the
reference data and the classification. As suggested above,
this can be especially challenging at the pixel level and is
one of the reasons we are interested in the use of segmentation and object-based classification approaches. Reference
data that might be appropriate for evaluating moderate
155
resolution classifications (e.g., Landsat TM, ETM+) may

be inadequate for high-resolution imagery when individual
features within a class are being resolved.
Similar spectral relationships exist between Landsat TM
and high-resolution IKONOS and QuickBird data for applications involving continuous variables such as lake water
clarity and impervious surface area. This is a great benefit to
the development of high-resolution satellite applications as
it can build on past research. This similarity also allows an
analyst to create and apply a modeled relationship from a
similar date of Landsat data and field data surrounding or
just outside of the IKONOS or QuickBird image itself. This
ability opens new possibilities for resource assessments at
different scales and for places where it is difficult to gather
field data.
In addition to the considerations and limitations previously mentioned, we believe that applications utilizing highresolution satellite imagery have several advantages over
other mapping methods. While traditional methods of land
cover mapping can produce high-resolution maps, their
production is expensive and time consuming. The highresolution imagery classification methods provide uniform
results over a local scale at high resolution with minimal
time and effort. By mapping variables, such as lake clarity
and impervious surface area, as continuous variables, the
high-resolution satellite imagery approach provides a highly
customizable set of classes. In addition, new software
options can easily make the conversion between raster and
vector formats. High-resolution satellite imagery is GISready for additional analysis. We are also encouraged by the
potential for repeat image acquisition and processing, making change detection possible for all three of these applications. These advantages make a high-resolution satellite
imagery-based approach to mapping and monitoring natural
resources affordable, time efficient, repeatable, and reliable.
In addition, several characteristics of satellite acquired highresolution data, including similarity in many aspects to
aerial photography with the added advantages of digital
format, multitemporal sequences, and multispectral imagery,
should make it attractive to natural resource managers.
The initial U.S. high-resolution commercial observation
satellites are now providing a steady source of high-resolution panchromatic and multispectral imagery data for a
broad range of commercial and government customers.
Resource managers are developing interest in utilizing
high-resolution satellite imagery to aid in their management
activities. The City of Eagan, the site of our lake water
clarity and impervious surface area research, is one example
of a promising, early participant. Obviously not every local
government unit will have the capabilities to process its own
satellite imagery, but these products will no doubt provide
impetus for commercial data distributors or value-added
companies to supply the environmental information services
such as these and others for local scale mapping and
monitoring. With increasing availability of high-resolution
satellite imagery, these approaches are feasible for local land
156
and water resource managers who need detailed and accurate monitoring information.
Acknowledgements
We gratefully acknowledge the support of the NASA
Science Data Purchase for providing IKONOS images, the
Upper Great Lakes Regional Earth Science Applications
Center (NASA grant NAG-13-99002), and the University of
Minnesota Agricultural Experiment Station. We also
appreciate the interest and cooperation of the City of Eagan,
Metropolitan Council, Minnesota Pollution Control Agency
and Minnesota Department of Natural Resources. Melanie
Tyler and John D. Madsen, Biological Sciences Department,
Minnesota State University, Mankato, provided the aquatic
vegetation survey field data. We also thank the reviewers
whose comments helped us improve this paper. Finally, we
thank Sarah Finley for her help in compiling and editing this
manuscript.
References
Barbier, E. B., Burgess, J. C., & Folke, C. (1994). Paradise lost? The
ecological economics of biodiversity. London: Earthscan Publications,
267 pp.
Barnes, K. B., Morgan, J. M., & Roberge, M. C. (2000). Impervious
surfaces and the quality of natural and built environments. Baltimore,
MD: Department of Geography and Environmental Planning, Towson
University.
Bauer, M. E., Doyle, J. K., & Heinert, N. J. (2002). Impervious surface
mapping using satellite remote sensing. Proceedings, IEEE Geoscience
and Remote Sensing Symposium, Toronto, Canada, June 24 28, Unpaginated CD ROM, 3 pp.
Carlson, R. E. (1977). A trophic state index for lakes. Limnology and
Oceanography, 22, 361 369.
Civco, D. L., & Hurd, J. D. (1997). Impervious surface mapping for the
state of Connecticut. Proceedings, 1997 ASPRS/ACSM Annual Convention. Seattle, WA, April 7 10 ( pp. 124 135).
Daily, G. C. (Ed.). (1997). Natures services: Societal dependence on natural ecosystems. Washington, DC: Island Press, 392 pp.
Estes, M. G., Gorsevski, V., Russel, C., Quattrochi, D., & Luvall, J. (1999).
The urban heat island phenomenon and potential mitigation strategies.
Proceedings, 1999 National Planning Conference: Approaching the
Millennium, Seattle, WA.
Kloiber, S. M., Brezonik, P. L., & Bauer, M. E. (2002). Application of
Landsat imagery to regional-scale assessments of lake clarity. Water
Research, 36, 4330 4340.
Kloiber, S. M., Brezonik, P. L., Olmanson, L. G., & Bauer, M. E. (2002). A
procedure for regional lake water clarity assessment using Landsat
multispectral data. Remote Sensing of Environment, 82(1), 38 47.
Lillesand, T., Chipman, J., Nagel, D., Reese, H., Bobo, M., & Goldman, R.
(1998). Upper Midwest gap analysis program image processing protocol. Report prepared for the U.S. Geological Survey, Environmental
Management Technical Center, Onalaska, Wisconsin, EMTC 98G001. 25 pp.+Appendixes A C.
Lillesand, T. M., Johnson, W. L., Deuell, R. L., Lindstrom, O. M., &
Meisner, D. E. (1983). Use of Landsat data to predict the trophic state
of Minnesota lakes. Photogrammetric Engineering and Remote Sensing, 49, 219 229.
Mitsch, W. J., & Gosselink, J. G. (1993). Wetlands (2nd ed.). New York:
Van Nostrand Reinhold, 722 pp.
Olmanson, L. G. (1997). Satellite remote sensing of the trophic state conditions of the lakes in the Twin Cities Metropolitan Area. M.S. Paper,
Water Resources Science, University of Minnesota, St. Paul, MN. 28 pp.
Olmanson, L. G., Bauer, M. E., & Brezonik, P. L. (2002). Water quality
monitoring of 10,000 Minnesota Lakes: statewide classification of lake
water clarity using Landsat imagery. Proceedings, 15th William T. Pecora Memorial Remote Sensing Symposium. November 8 15, Denver,
Colorado, unpaginated CD ROM, 5 pp.
Ozesmi, S. L., & Bauer, M. E. (2002). Satellite remote sensing of wetlands.
Wetlands Ecology and Management, 10(5), 381 402.
Ridd, M. K. (1995). Exploring a V-I-S (vegetation-impervious surface-soil)
model for urban ecosystem analysis through remote sensing: comparative anatomy for cities. International Journal of Remote Sensing,
16(12), 2165 2185.
Sleavin, W. J., Civco, D. L., Prisloe, S., & Giannattti, L. (2000). Measuring
impervious surfaces for non-point source pollution modeling. Proceedings, 2000 ASPRS Annual Conference. May 22 26, Washington, DC,
11 pp.

High spatial resolution spectral mixture analysis of urban reflectance

Christopher Small *
Lamont Doherty Earth Observatory, Columbia University, Palisades, NY 10964 USA
Received 9 May 2002; received in revised form 17 January 2003; accepted 22 April 2003
Abstract
This study uses IKONOS imagery to quantify the combined spatial and spectral characteristics of urban reflectance in 14 urban areas
worldwide. IKONOS 1-m panchromatic imagery provides a detailed measure of spatial variations in albedo while IKONOS 4-m
multispectral imagery allows the relative contributions of different materials to the spectrally heterogeneous radiance field to be determined
and their abundance to be mapped. Spatial autocorrelation analyses indicate that the characteristic scale of urban reflectance is consistently
between 10 and 20 m for the cities in this study. Spectral mixture analysis quantifies the relative contributions of the dominant spectral
endmembers to the overall reflectance of the urban mosaic. Spectral mixing spaces defined by the two low-order principal components
account for 96% to 99% of image variance and have a consistent triangular structure spanned by high albedo, low albedo and vegetation
endmembers. Spectral mixing among these endmembers is predominantly linear although some nonlinear mixing is observed along the gray
axis spanning the high and low albedo endmembers. Inversion of a constrained three-component linear mixing model produces stable,
consistent estimates of endmember abundance. RMS errors based on the misfit between observed radiance vectors and modeled radiance
vectors (derived from fraction estimates and image endmembers) are generally less than 3% of the mean of the observed radiance. Agreement
between observed radiance and fraction estimates does not guarantee the accuracy of the areal fraction estimates, but it does indicate that the
three-component linear model provides a consistent and widely applicable physical characterization of urban reflectance. Field validated
fraction estimates have applications in urban vegetation monitoring and pervious surface mapping.
Keywords: IKONOS imagery; Urban reflectance; Spectral mixture analysis
1. Introduction
Human settlements occupy a relatively small fraction of
Earths surface area, but their extent, distribution and evolution have enormous impact on environmental and socioeconomic dynamics worldwide. Despite their fundamental
importance, urban areas have not been mapped and characterized with remote sensing to the same extent that other land
cover types have. Optical sensors on operational satellites
provide an efficient means for quantifying past and present
distributions of human settlements as well as their physical
reflectance properties. This is important at two scales. At a
global scale, it is important to understand the physical
characteristics that distinguish developed urban areas from
other types of human modified and undeveloped land surfaces in order to map and monitor the extent and evolution of
urban areas with moderate resolution (20 30 m) optical
* Tel.: +1-845-365-8354; fax: +1-845-365-8179.
E-mail address: small@LDEO.columbia.edu (C. Small).
doi:10.1016/j.rse.2003.04.008
sensors. The reflectance properties of the urban mosaic are

central to both the discrimination of urban land cover and the
understanding of its role in the urban environment. At the
intraurban scale, it is important to constrain the effect that
urban land cover and building materials have on the immediate physical environment. Optical reflectance characteristics have a direct impact on urban microclimate because
they modulate the solar energy flux through the built
environment. The spatial scale of the reflectance determines
the spatial scale of surface temperature variations which, in
turn, have a strong influence on heat flux, convection and
urban microclimate. Numerical simulation of atmospheric
circulation has now advanced to the point where the effects
of the built environment on microclimate and air quality can
be modeledgiven adequate constraints on physical properties like albedo and fractional vegetation. Detailed characterization of urban land cover also has application to
pervious surface mapping and urban vegetation monitoring.
Previous urban remote sensing studies have generally
focused on identification of specific materials or land cover
C. Small / Remote Sensing of Environment 88 (2003) 170186
classes. Ridd (1995) proposed an urban landcover classification scheme based on the distribution of vegetation,
impervious surface and soil (the VIS model) but acknowledged the difficulty of distinguishing between soil and
impervious surfaces with optical sensors. Subsequent studies (e.g., Flanagan & Civco, 2001; Liu & Lathrop, 2002;
Madhavan et al., 2001) have employed a variety of classification methods (Maximum Likelihood, Unsupervised,
Decision Tree) with moderate resolution (30 m) imagery,
but traditional hard classification algorithms are impeded by
the abundance of spectrally mixed pixels. Mixed pixels are
problematic for statistical classification methods because
most algorithms are based on the assumption of spectral
homogeneity at pixel scale within a particular class of land
cover. The classification task is often further complicated by
inconsistencies between the thematic classes sought and the
reflectance properties that can be discriminated with moderate resolution broadband sensors. Urban areas provide
examples of spectrally diverse, scale-dependent thematic
classes containing large numbers of pixels that are spectrally
indistinguishable from other land cover classes. The diversity of land cover types and scales in the urban mosaic
therefore results in relatively high rates of misclassification
between urban and other land cover classes. Combining
spectral, textural and ancillary information can improve
classification accuracy (e.g., Stefanov et al., 2001), but a
physical characterization of reflectance is still necessary to
accommodate the physical processes that influence the
upwelling radiance measured by optical sensors. Several
recent studies have used physical rather than statistical
classifications of urban land cover in individual cities with
some degree of success (e.g., Kressler & Steinnocher, 1996,
2000; Phinn et al., 2002; Rashed, Weeks, Stow, & Fugate,
2002; Small, 2001a; Wu and Murray, 2002). High-resolution aerial photographs reduce the abundance of mixed
pixels (e.g., Akbari, Rose, & Taha, 1999), but film-based
images are less amenable to multispectral image analysis
and classification. Hyperspectral imagery can provide sufficient spatial and spectral resolution to map a wide variety
of urban surfaces (e.g., Herold, Gardner, Hadley, & Roberts,
2002), but relatively few cities have yet been mapped with
imaging spectrometers.
Characterization of urban reflectance is constrained by
the spatial resolution of the sensor. The 30-m resolution of
the Landsat ETM+ sensor and the 20-m resolution of the
Spot HRV sensor are generally not sufficient to discriminate
individual features (e.g., buildings, streets, trees) within the
urban mosaic. As a result, almost all the urban pixels
imaged by these sensors represent a composite radiance
field emanating from several distinct features with different
reflectances within the sensors field of view. This spectral
heterogeneity at scales comparable to the ground instantaneous field of view (GIFOV) of the sensor results in a
preponderance of spectrally mixed pixels. Mixed pixels
violate the cardinal assumption of statistical clustering
algorithms commonly used to classify land cover types.
171
The increased spatial resolution of IKONOS imagery provides an opportunity to image urban areas at scales sufficient to resolve many (but not all) of the individual features
in the urban mosaic. IKONOS orbital platform also makes
it possible to image a wide variety of urban areas worldwide
for a self-consistent analysis and comparison of the reflectance properties of urban land cover. A consistent, physically based description of urban reflectance properties could
help to advance our understanding and simulation of urban
microclimate by providing spatially explicit constraints on
albedo, evapotranspiration and spatial distribution of pervious surface. Characterization of urban reflectance at 4 m
scale would also provide constraints on the spectral endmembers and mixing processes responsible for the mixed
pixels imaged by moderate resolution sensors. If the scaledependent reflectance properties of urban mixed pixels
could be distinguished from the reflectance properties of
other types of land cover it would provide a basis for
mapping urban areas in moderate resolution imagery collected over the past 30 years.
The objective of this study is to explore a physical
characterization of urban reflectance properties in a variety
of urban settings. The characterization will incorporate the
spatial scale and optical reflectance properties of the variety
of land covers contained in the urban mosaics. Spatial
autocorrelation is used to quantify the characteristic scale
lengths of urban reflectance within and among different
cities. Spectral mixture analysis is used to quantify and
compare the reflectance properties of these urban areas.
Spectral mixture analysis provides a physically based representation of composite radiance measurements that allows
surface reflectance to be described as combinations of
spectral endmembers. Representing reflectance as continuous gradations within a spectral mixing space provides a
more flexible, and accurate, description than that resulting
from classification algorithms that assign each pixel membership in one (and only one) of a small number of classes.
It is important to distinguish between identification of
specific target materials and the physical representation of
the radiance image discussed here. Because many different
materials can have indistinguishable reflectance signatures
as measured by IKONOS, it is not generally possible to
identify specific target materials in these unvalidated spectral mixture analyses. The objective here is to determine the
consistency of urban reflectances and to assess the feasibility of using spectral mixture analysis to provide a general
and widely applicable representation of these reflectances.
The key questions addressed by this study are related to the
consistency of urban reflectance properties both within and
among a variety of urban areas worldwide. This study uses a
collection of 14 IKONOS images provided by the NASA
Scientific Data Purchase Program. The acquisition parameters for the NASA-funded IKONOS acquisitions are described in greater detail elsewhere in this volume.
Throughout the analysis it is assumed that the spatial
variations in radiance measured by IKONOS are primarily
172
a result of spatial variations in surface reflectance and that

spatial variations in atmospheric turbidity are not significant
at the scale of hundreds of meters or less. Hence radiance
will be used when referring to the IKONOS measurements
and reflectance will be used when referring to the surfaces
from which the radiance emanates.
2. Scales of urban reflectance

The spatial scale of urban reflectance influences physical
processes like surface heating and convection as well as the
spectral mixing that produces the mixed pixels observed in
moderate resolution imagery. It is important to quantify this
scale to better understand both of these processes. The
degree to which different cities (and areas within cities)
exhibit different scales of spatial variability determines the
feasibility of a general characterization of urban reflectance.
Urban mosaics are composed of a wide variety of features
with distinct sizes and reflectance properties, but the size
distribution is generally dominated by features at the scale

of streets and buildings. The reflectance of an individual
feature will generally have some degree of internal variability that is small compared to the reflectance contrast with
other features at a scale of 1 to 4 m. This is why individual
features like streets and buildings can be usually discriminated in IKONOS imagery. The spatial scale of reflectance
variability is therefore determined by the size distribution of
individual features in the urban mosaic. In the context of this
study, the characteristic scale length refers to the linear
dimension of the features of greatest areal abundance. For
a typical urban mosaic it is more informative to refer to a
range of lengths encompassing the most areally abundant
features.
Spatial autocorrelation of IKONOS panchromatic imagery can provide a quantitative measure of the characteristic
spatial scales of urban reflectance. The properties of the
autocorrelation function as applied to remotely sensed
images are described in detail by Getis (1994) and Wulder
and Boots (1998). At small lag distances the internal
Fig. 1. Characteristic scale length of urban reflectance. IKONOS panchromatic imagery of Caraz Peru (A) shows spatial variations in visible/near infrared
(VNIR) radiance resulting from the mosaic of roofs, streets, and open spaces. The superimposed grid spacing is 10 m. (B) The normalized two-dimensional
(2D) spatial autocorrelation function (ACF) of a 100 100-m subscene shows the azimuth dependence and quasi-periodic structure resulting from the
dominant scale of building size and street layout. (C) Asymptotic behavior of orthogonal one-dimensional autocorrelation functions extracted from the 2D ACF
shows the consistency of the slope of the central peak. The scale lengths estimated from the 0.1-slope threshold of these two functions are 10 and 12 m and are
consistent with the size of most features in (A). Includes materials n Space Imaging.
173
Fig. 2. Distribution of scale lengths for a variety of urban areas. Each of the 6357 sites corresponds to a 100 100-m subscene of IKONOS panchromatic
imagery. The distribution of scale lengths for the 14 cities indicates that characteristic scales are consistently between 10 and 20 m. Inset histograms show
distributions for different cities.
consistency in the reflectance of individual features results

in high correlations, but as the lag distance approaches the
characteristic spatial scale of the mosaic the reflectance
contrast between adjacent features causes the autocorrelation to decrease. The width of the central peak of the
autocorrelation therefore provides an indication of the
spatial scale of the individual features that contribute significant variance to the observed radiance field. Secondary
peaks in the function indicate repetitive patterns in the
reflectance at scales and azimuths where bright and dark
features are aligned in phase. In this study, two-dimensional
(2D) autocorrelation functions (ACFs) are used to estimate
the characteristic spatial scale of urban reflectance and to
assess the consistency of these scales within and among a
variety of urban areas worldwide. For each city, 2D ACFs
are calculated for a number of 100 100-m subscenes of
IKONOS panchromatic imagery, and characteristic scale
length estimates are derived from each ACF. The scale

length is estimated as the lag distance at which the radial
average of the 2D ACF attained a radial slope of 0.1 times
the maximum slope of the central peak. The choice of 0.1 is
rather ad hoc but the slope of the ACF diminishes asymptotically with distance between the central peak and the first
trough so the scale length estimates are not very sensitive to
the precise value used for the slope threshold below ~0.1.
The slope threshold returned estimates that were consistent
with random measurements of feature size from several
images. In most of the cities investigated here the 2D ACF
indicated the direction and scale of the dominant street grid
pattern as radial ridges of higher correlation, but the central
peaks are consistently axisymmetric (Fig. 1).
A wide variety of urban areas have similar distributions
of scale lengths. Fig. 2 shows scale length distributions for
some of the cities investigated and the total distribution of
Fig. 3. False color composites of IKONOS MSI imagery (left) and corresponding principal components (right) for 14 urban areas. Natural color composites of
visible bands (R/G/B = 3/2/1) do not discriminate between vegetation and low albedo surfaces as effectively as false color composites (R/G/B = 3/4/1)
incorporating the NIR band 4. False color composites of the three low-order principal components (R/G/B = PC1/PC2/PC3) minimize visible band correlations
thereby enhancing contrast between surfaces with subtle differences in reflectance. The first principal component corresponds to albedo with bright areas
appearing red in all but one image (blue in She Xian). The second principal component corresponds to high NIR reflectance so vegetation appears green. The
third principal component is orthogonal to the primary mixing plane so pixels with relatively low proportions of PC1 and PC2 and more pronounced nonlinear
mixing outside the primary mixing plane span the range from black to blue. Each image is 1 km2. Includes materials n Space Imaging.
174
all 14 cities. The total number of 100 100-m estimates

calculated for each city was determined by the size of the
IKONOS image and the total built-up area within each
image. All of the large cities had similar skewed distributions like those in Fig. 2. Some of the smaller cities did not
provide a large enough number of subscenes to attain a
strongly skewed distribution. All the cities investigated,
except Alta Mira and She Xian, had similar distributions
with modal scale lengths between 10 and 20 m. These
smaller cities had smaller scale lengths. These scale length
distributions explain why moderate resolution sensors with
GIFOVs greater than 10 m rarely resolve individual features
in urban areas. This also suggests that a significant percentage of the 4-m IKONOS MSI pixels should image spectrally
homogeneous features within urban mosaics with scale
lengths greater than 10 m. Urban areas with scale lengths
of 10 to 20 m will still have a significant percentage of
mixed pixels because the 4-m GIFOV is similar enough to
the scale length to result in a significant number of boundary
pixels spanning two or more features. The larger the
characteristic scale length of features in the urban mosaic,
the higher the percentage of spectrally homogeneous pixels
and the lower the percentage of mixed pixels. The panchromatic autocorrelation provides a conservative estimate of
spectral heterogeneity because spectrally distinct features
can have similar VNIR albedos as imaged by IKONOS
panchromatic band.
3. Spectral mixing
Spectral mixture analysis (SMA) provides a systematic
way to quantify spectrally heterogeneous urban reflectance.
SMA is based on the observation that, in many situations,
radiances from surfaces with different endmember reflectances mix linearly within the IFOV (Nash & Conel, 1974;
Singer, 1981; Johnson, Smith, Taylor-George, & Adams,
1983; Singer & McCord, 1979). This observation has made
possible the development of a systematic methodology for
SMA (Adams et al., 1986, 1989, 1993; Gillespie et al.,
1990; Smith, Ustin, Adams, & Gillespie, 1990; Roberts,
Smith, & Adams, 1993) that has proven successful for a
variety of quantitative applications with multispectral imagery (e.g., Adams et al., 1995; Elmore, Mustard, Manning, &
Lobell, 2000; Pech, Davies, Lamacraft, & Graetz, 1986;
Roberts, Batista, Pereira, Waller, & Nelson, 1998; Smith et
al., 1990). If a limited number of distinct spectral endmem-
175
bers are known, it is possible to define a mixing space

within which mixed pixels can be described as linear
mixtures of the endmembers. Given sufficient spectral
resolution, a system of linear mixing equations may be
defined and the best fitting combination of endmember
fractions can be estimated for the observed reflectance
spectra. The strength of the SMA approach lies in the fact
that it explicitly takes into account the physical processes
responsible for the observed radiances and therefore accommodates the existence of mixed pixels.
The diversity of land covers in the urban mosaic influences the topology of the mixing space while the spectral
dimensionality of the image is constrained by the number of
bands and their spectral resolution. The limited spatial and
spectral resolution of the IKONOS sensor results in a
projection of a high-dimensional mixing space onto a
lower-dimensional representation that is constrained by the
ability of the sensor to discriminate different surface reflectances at GIFOV scales. Analyses of AVIRIS hyperspectral
imagery suggest that some urban areas have as many as 30
to 50 spectral dimensions (Green & Boardman, 2000; Small,
2001c), but the IKONOS sensor can resolve only four of
these dimensions at most. A central question of this analysis
is whether these four dimensions provide an adequate basis
for a systematic characterization of urban reflectance. Is the
information content provided by the IKONOS sensor sufficient to characterize the differences in scale and reflectance
between urban areas and other land cover types in a
consistent manner?
In this study, the mixing space characterization is provided
by a principal component transformation of the multispectral
imagery. The principal component rotation minimizes the
correlations among bands and enhances the contrast between
different surface reflectances. The accompanying eigenvalue
distribution provides a quantitative estimate of the variance
partition between the signal- and noise-dominated principal
components of the image. With hyperspectral sensors this
partition and the number of signal-dominated components
can form the basis of a dimensionality estimate of the image
(Green & Boardman, 2000). With broadband sensors like
IKONOS, the four-band limit to the number of distinguishable dimensions is generally less than the number of spectrally distinct endmembers (and hence the inherent
dimensionality). The multidimensional feature space of the
low-order principal components represents the spectral mixing space that can be used to define spectral mixtures as
combinations of spectral endmembers (Boardman, 1993;
Fig. 4. Mixing space representation of IKONOS MSI imagery. Normalized eigenvalues (left column) give the variance partition among the principal
components. The two-dimensional scatterplots of the principal components are projections of the three low-order dimensions of a four-dimensional spectral
mixing space. The side view shows the two dimensions associated with the majority (>96%) of the variance while the top and end views incorporate the third
dimension. Variance in the third dimension is primarily associated with nonlinear mixing along the gray axis between the high and low albedo endmembers
and generally accounts for less than 3% of the total variance. The fourth dimension (not shown) contributes less than 1% of the variance but still contains
coherent spectral information. Orthogonal projections of spectral mixing spaces in a variety of urban areas show a similar structure with a narrow triangular
mixing space spanned by high albedo (H), low albedo (L) and vegetation (V) endmembers. Straight edges correspond to linear mixing among endmembers
while convex edges indicate nonlinear mixing. Tapering of the mixing space approaching the Vegetation endmember suggests that a three-component mixing
model is very well constrained for vegetation fractions. The apex seen in the third dimension generally corresponds to a soil (S) endmember.
176
Jackson, 1983; Johnson, Smith, & Adams, 1985). The mixing

space could be represented with scatter plots of the unrotated
bands but using scatter plots of the principal components
(PCs) provides an optimal projection of the mixing space
because the PC rotation orders the projections with respect to
the variance they contribute to the scene.
In this analysis, a minimum noise fraction (MNF)
principal component transformation is used to project the
mixing space onto a series of 2D scatter plots. These scatter
plots are referred to geometrically with the side view
corresponding to the projection of the two lowest order PCs
containing the largest amount of image variance. The end
view and top view incorporate the third dimension of
the mixing space and help to represent the cloud of pixels
occupying the three low-order dimensions of the mixing
space as a 3D object. The MNF transformation implemented in ENVI is analogous to the maximum noise
transformation described by Green, Berman, Switzer, and
Craig (1988) but differs in the ordering of the principal
components from high to low signal variance (RSI, 2000).
With IKONOS imagery, the MNF transformation usually
produces principal components similar to those resulting
from a traditional covariance-based PC rotation but offers
the added benefit of normalizing the eigenvalues relative to
the variance of the sensor noise estimate. For this analysis,
all MNF transformations were applied using noise covariance statistics derived from a July 2001 acquisition over the
Central Park Reservoir in New York City under relatively
clear atmospheric conditions. The Reservoir is an enclosed,
noncirculating body of water ~600 m in diameter with
negligible reflectance from suspended sediment or biological productivity and therefore provides a reasonable approximation of a dark target. Normalized eigenvalue
distributions quantify the partition of variance among the
principal components indicating how many spectral dimensions are required to represent the information content in
the image. The larger eigenvalues are associated with the
low-order principal components representing the dominant
reflectance patterns while the smaller eigenvalues are
associated with the higher-order principal components
associated with the pixel scale variance commonly assumed
to be noise. The signal-to-noise ratio of the IKONOS
sensor is sufficiently high that all of the principal components in all scenes investigated showed significant information content well above the noise level. The eigenvalue
distributions of IKONOS imagery therefore do not indicate
the dimensionality of the image but rather the variance
partition among the dimensions.
177
Subscenes of the 14 images used in this analysis are

shown in Fig. 3. The full-size images vary in area, cloud
cover and location, so 1-km2 subscenes are used for most of
the comparative analyses. These subscenes were chosen to
represent the spectral diversity typically observed within the
built-up core of each city. The results are generally consistent with those of the larger subscenes but are more easily
compared because of their uniform size. The false color
composites incorporating the NIR band (R/G/B = 3/4/1)
emphasize the distinction between vegetation and low
albedo surfaces. False color composites of the three loworder principal components maximize the spectral contrast
between objects of different reflectance, but they also reveal
a consistency in the reflectance of the urban areas used in
this study. In each case, the low-order principal component
(PC1) corresponds to the overall albedo of the image.
Hence, bright features usually appear red in Fig. 3. The
second principal component distinguishes vegetation from
nonvegetated surfaces resulting in a correspondence of the
green areas in the composites in Fig. 3. The third principal
component accounts for only a few percent of the total
image variance so areas that appear blue are characterized
by lower albedo unvegetated areas with other types of
reflectance such as dark soils.
Spectral mixing spaces provide a self-consistent basis for
comparison of urban reflectance characteristics (Small,
2002a). IKONOS MSI imagery shows a simple, but consistent, spectral mixing space structure for the urban areas
investigated in this study. The similarity of the triangular
mixing spaces shown in Fig. 4, as well as those not shown,
indicates that all 14 of the urban areas in this study have a
consistent mixing space topology in the form of a narrow
triangular cloud of pixels. The distribution is referred to as
narrow because the variance of the two primary dimensions is
considerably greater than the other dimensions. Each principal component is scaled by its full range of values so the
relative widths of the mixing space are exaggerated. The
actual variance is indicated by the corresponding eigenvalue
so the third and fourth dimensions are much thinner than
the first two dimensions. In each case, the three endmembers
defining the apexes of the triangular mixing space correspond
to high albedo (H), low albedo (L) and vegetation (V)
reflectances. Although the internal distributions of mixed
pixels within the mixing spaces vary, the overall form is
consistent. The apexes of the primary 2D mixing space
corresponding to the spectral endmembers are generally well
defined and the edges between the apexes are generally
straight or concave. This indicates that the mixing among
Fig. 5. Mixing spaces for nonurban land cover. At 1 km2 scales, other types of land cover do not generally produce the triangular mixing space characteristic of
urban land cover. While these mixing spaces do show cases of linear mixing among well-defined endmembers, nonlinear mixing is evident from the convex
edges of the mixing spaces. Multiple scattering within the open canopy vegetation of the Argentinian pampa produces strong nonlinear mixing. The mixing
space representing the sparsely vegetated Peruvian highland has a well-defined low albedo endmember, but mixing becomes increasingly nonlinear as albedo
increases along dimension 1. The densely vegetated mountains in central China produce a mixing space dominated by illumination differences in which an
exposed soil endmember resides at the bottom of the secondary dimension. The spectrally diverse riparian land cover of the Brazillian Amazon most closely
resembles the triangular mixing space seen in urban areas. Its primary mixing space is dominated by two linear mixing continua and complex topology at the
high albedo apex.
178
these three endmembers is primarily linear, while nonlinear

mixing is indicated by the convex edges seen in the third
dimension. The lower variance of the third dimension indicates that the magnitude of the nonlinear mixing is small
compared to the linear mixing represented in the two primary
dimensions. This suggests that a three-component linear
mixing model may provide a consistent and accurate way to
represent urban reflectances. A more detailed discussion of
the mixing space concept is provided by Boardman (1989a,b).
In contrast to the consistency in urban mixing spaces, the
land cover mosaics in nonurban environments produce a
diversity of mixing space topologies. Fig. 5 shows examples
of four IKONOS images selected from different environments and their mixing space representations. Even though
all of these environments contain several different types of
land cover of varying reflectance, none of their mixing
spaces resembles the triangular mixing space that is characteristic of the urban areas used in this study. Agricultural
areas do, however, often exhibit a triangular mixing space
similar to those of the urban areas. These triangular mixing
spaces bear a strong resemblance to the tasseled cap
feature space described by Kauth and Thomas (1976) and
can be considered representative of visible/near infrared
reflectance in certain instances where the scene contains
high albedo, low albedo and vegetation endmembers simultaneously. In fact, it would be possible to define a transformation for IKONOS imagery analogous to the tasseled cap
transform defined for Landsat TM and MSS. The spectral
mixture analysis described below is, however, preferable to
a predefined transformation because it can represent a wider
variety of spectral endmembers and can accommodate
different atmospheric effects.
The topology of the mixing spaces depends on the
combination of reflectance patterns contained within the
image. In general, the larger urban images used in this study
179
had mixing spaces similar to those of the 1-km images

shown in Fig. 4. Other 1-km images chosen from areas
surrounding the built-up part of each city sometimes had
different mixing space topologies, but inclusion of the builtup areas generally results in the triangular topology seen in
Fig. 4. A similar topology is seen in the low-order dimensions of urban mixing spaces generated from Landsat and
AVIRIS imagery (Small, 2001b). Spectral dimensionality
does, however, tend to diminish at smaller spatial scales
(Small, 2001c). In the case of IKONOS urban imagery, this
generally results in a different partition of variance among
the principal components rather than a change in mixing
space topology. Fig. 6 indicates that the larger urban scenes
generally have a greater fraction of the variance in the
second principal component as a result of a greater percentage of vegetated area. The 1-km images generally have a
greater fraction of variance associated with the low-order
dimension. This is not surprising as the smaller areas have a
lower areal percentage of vegetation and are dominated by
spectral variations along the gray axis between the high
albedo and low albedo endmembers.
The topologic consistency of the mixing spaces is
complemented by a consistency in endmember reflectances. Fig. 7 shows the radiance vectors corresponding to
pixels at the apexes of each mixing space. The amplitudes
of the endmembers are variable but the shape is remarkably consistent. The low albedo endmember generally
corresponds to deep shadow so its shape represents the
atmospheric path radiance component that is present in
every pixel. The most pronounced differences are related
to atmospheric conditions, but there is also a moderate
correlation (0.67) between the peak amplitude of the high
albedo endmember and the solar zenith angle at the times
the images were acquired. The collection geometry varied
between elevation angles of 61j and 87j. Given the
Fig. 6. Scale dependence of variance partition. Normalized eigenvalue distributions for larger urban areas (left) indicate that the second dimension of the mixing
space accounts for much of the variance (20 50%) in urban reflectance at spatial scales of several kilometers. Eigenvalue distributions of smaller (1 km2) city
center areas show a greater fraction of variance associated with the primary dimension spanned by high and low albedo endmembers. The difference is due to
the greater abundance of vegetated area within the larger images than within the 1-km2 city centers.
180
Fig. 7. Spectral endmembers. Radiance vectors corresponding to the apexes of the triangular mixing spaces show a remarkable consistency of form, although
amplitude is variable. Variations in amplitude result from actual differences in endmember reflectance as well as differences in illumination and atmospheric
conditions. The slope of the low albedo endmembers (thin curves) is consistent with wavelength-dependent atmospheric scattering effects. IKONOS DNs were
converted to radiance using the calibration coefficients provided by Space Imaging and NASA (Zanoni et al, Pagnutti et al, this volume).
differences in solar zenith, collection geometry, atmospheric turbidity, it is surprising that the amplitude differences
among the endmembers are not greater.
4. Linear mixture modeling
fH e 11 fV e12 fL e13 r1
The consistency of the spectral mixing space for a variety

of urban areas suggests that a simple three-component linear
mixture model may provide a consistent, general characterization of urban reflectance. Representing urban reflectance,
of individual pixels or aggregate distributions, as threecomponent mixtures makes it possible to compare reflectance patterns within and among cities. The high albedo,
low albedo and vegetation endmembers also provide a
useful physical description since albedo and vegetation
cover are two of the primary surface properties that control
urban microclimate. While the statistical moments of the
mixing space (mean, variance, skewness, etc.) provide a
description of the aggregate reflectance of an urban area,
each individual pixel within the aggregate can be described
by its relative areal abundance of each endmember. Spectral
mixture modeling provides a way to derive estimates of
endmember abundance for individual pixels.
Inversion of the urban three-component linear mixing
model for each pixel yields fraction estimates for each
endmember. The linear three-component mixing model is
given in continuous form by:
Rk fH EH k fV EV k fL EL k
corresponding to the high albedo (H), vegetation (V) and

low albedo (L) endmembers. The corresponding endmember fraction estimates we seek are fH, fV and fL. The discrete
implementation of the model, applicable to IKONOS MSI
radiances, is given by:
where R(k) is the observed radiance profile, a continuous

function of wavelength k. The E(k) are the spectra
fH e21 fV e22 fL e23 r2

where ri is the observed radiance vector corresponding to

discrete estimates of integrated radiance within the four
IKONOS MSI bands. The eij are the endmember radiance
vectors corresponding to the high albedo (H), vegetation (V)
and low albedo (L) endmembers. The indices i and j indicate
the spectral band and endmember of each element, respectively. An additional unity sum constraint equation can be
incorporated to urge the fractions to sum to 1. With four or
less endmembers, the system has more equations than
unknowns and can be solved for an optimal set of
endmember estimates chosen to minimize misfit to the
observed radiance vector.
The overdetermined linear mixing model, incorporating
measurement error, can be written in matrix notation as:
r Ef E
where E is an error vector which must be minimized to find

the fraction vector f which gives the best fit to the observed
radiance vector r. There are a number of ways to solve this
type of problem (e.g., Pech et al., 1986; Settle & Drake,
181
182
Fig. 9. Maximum endmember fractions and RMS error distributions. (A) The cumulative distributions of the largest endmember fraction of each pixel span the
full range of permissible maximum fractions. Cities with significant areas of unmodeled soil (Caraz, Juarez, Phoenix, Pune, and Salvador) have more than 10%
area with maximum fractions less than 0.33. The three-component model does not account for the soil endmember and cannot replicate the spectra of these
pixels. (B) The RMS of the differences between the modeled and observed radiance vectors is normalized by mean pixel radiance to account for the magnitude
of the misfit relative to the amplitude of the observed radiance vector. Distributions of radiance-normalized RMS error are consistently low with error
amplitudes less than 3% of the mean radiance of the corresponding pixel. The three urban areas with higher errors (Bangkok, Caraz, Salvador) correspond to
examples with an unmodeled fourth endmember in the third dimension.
1993; Smith, Johnson, & Adams, 1985; Smith et al., 1990).

The procedure used to invert the urban three-component
linear mixing model for endmember reflectances is described
in detail and the stability of the inversion is demonstrated in
Small (2001a). A unit sum constrained least squares inversion of the three-component model was performed on all 14
of the 1-km images. The endmember fraction images for
seven of the 1-km IKONOS images are shown in Fig. 8.
Most of the urban areas used in this study were characterized
by high fractions of the low albedo endmember. This
represents the abundance of shadow as well as the frequent
use of low albedo building materials in urban areas. Shadow
can mask true target reflectance thereby increasing the
apparent abundance of low albedo targets.
Inversion of the linear mixing model produces areal
abundance estimates, but these should not be interpreted in
the context of subpixel resolution. Areal estimation of an
individual feature within a single pixel is complicated by
the spatial variation in the sensors response function. The
nonlinear mixing effects of the sensors point spread
function can introduce significant error when the spatial
scale of the endmember fractions is comparable to the scale
of the GIFOV. The endmember fraction estimates are most
accurate when the spatial scale of the endmember components is small relative to the GIFOV. This nonlinear mixing
is further complicated by the application of modulation
transfer function compensation (MTFC) filters. The MTFC
procedure attempts to compensate for the spatial averaging
effects of the point spread function by increasing the

contrast of adjacent pixels (Pagnutti, Ryan, Kelly, & Holekamp, this issue; Ryan et al, this issue; Zanoni et al., this
issue). To quantify the effect (positive or negative) of the
MTFC on the fraction estimates, it would be necessary to
compare field validation measurements with estimates
derived from the same imagery with and without MTFC
applied. All but two of the IKONOS images (New York
and Hollywood) used in this study had MTFC applied as
they were acquired before the NASA investigators were
given the option of processing without MTFC.
Cumulative distributions of the maximum fractions for
each pixel in each image are shown in Fig. 9A. These
distributions indicate that most of images have fraction
distributions spanning the full range (0.33 to 1.0) of allowable abundances. Maximum fractions near 1.0 are associated
with pure pixels while maximum fractions near 0.33 are
associated with mixed pixels. Maximum fractions below
0.33 indicate that the three-endmember model did not
accurately represent some of the radiance vectors. In this
study, cities with large areas of unmodeled soil had significant numbers of pixels with fractions that did not sum to
unity. For some of these cities (Caraz, Juarez, Phoenix,
Pune, Salvador), a four-endmember model containing a soil
endmember would have been more appropriate.
Several of the urban areas in this study had mixing
spaces that could accommodate a fourth endmember. In
most of these cases, this fourth endmember would corre-
Fig. 8. Endmember fraction estimates and linear mixing model errors. Inversion of a constrained, three-component linear mixture model results in three
endmember fraction estimates for each pixel in the image. Endmember fraction images range from 0 (black) to 1 (white). The RMS error for each pixel is
calculated for the difference between the observed radiance and the forward implementation of the linear model using the endmember vectors and the
endmember fraction estimates. RMS error images range from 0% to 6% and are displayed with a 2% linear stretch to emphasize specific features not well fit by
the three-component linear model. Most urban areas have a strong low albedo component corresponding to shadow and the low surface reflectance of many
building materials.
spond to soil, but in some of the urban areas the fourth

endmember appears to represent a spectrally distinct high
albedo roofing material. Adding a fourth endmember will
generally reduce the RMS error so the three-component
linear mixing model provides a conservative indication of
how well linear mixture models can be expected to represent
urban reflectance in IKONOS imagery.
The suitability of the linear mixing model is indicated by
the magnitude and distribution of the RMS errorthe root
mean square of the difference between the observed radiance vectors and the modeled radiance vectors. The modeled radiance vector is generated by forward implementation
of the linear mixing model as a fraction-weighted sum of the
endmember radiance vectors using the estimated fractions.
Distributions of normalized errors are shown in Fig. 9B. The
RMS error of each pixel is normalized by the mean radiance
of the pixel to indicate the magnitude of the error relative to
the amplitude of the observed radiance vector. Fig. 9B
indicates that the majority of the pixels were fit to within
3% of the amplitude of the observed radiance vector in most
scenes. Three of the images used in the study had larger
errors. These images had mixing spaces indicating the
presence of a fourth endmember. It is encouraging, however,
that even in these cases the errors rarely exceed 6% of the
observed amplitude.
A low RMS error is necessary but not sufficient condition for accurate estimation of actual areal abundances of
specific endmember materials within the GIFOV. In other
words, a small error does not guarantee that the endmember
abundances will agree with field validation measurements,
but a large error does indicate that the endmembers and
fraction estimates do not accurately represent the composite
radiance measured by the sensor for a given pixel. Field
validation is necessary if the endmember abundances are to
be related to specific quantities of endmember materials on
the ground. The fraction estimates produced in this study
have not been validated with field measurements, but the
small errors to the observed radiances indicate that the threecomponent linear model is generally well posed and worthy
of validation and further investigation.
It is important to distinguish between spectral endmember abundances and specific target materials. Hyperspectral
imagery of urban areas reveals the presence of as many as
60 spectral dimensions (Green & Boardman, 2000; Small,
2001c) related to the wide variety of spectrally distinct
building materials (Herold et al., 2002). Many of these
materials are spectrally indistinguishable with broadband
sensors like IKONOS. High albedo roofing material is often
indistinguishable from high albedo soil while many low
albedo materials are indistinguishable from clear water and
deep shadow. Broadband sensors may not be able to discriminate spectrally similar materials but the continuum
representation provided by mixture models accommodates
a far wider range of targets and conditions than hard
classification because it represents the continuous gradations among spectrally distinct materials.
183
5. Potential applications
Representation of multispectral radiances, or surface
reflectances, as spectral endmember abundances provides
a simple way to describe reflectance properties imaged by
the IKONOS sensor. The strength of the spectral mixture
model is its ability to represent a wide variety of surface
reflectance types as simple combinations of endmember
abundances. The inversion of the linear mixing model is
computationally trivial compared to statistical classifications
like maximum likelihood. Spectral mixture analysis is
preferable to hard classification for many physical science applications because it accommodates the fundamental
physical process responsible for the preponderance of mixed
pixels observed in almost all multispectral imagery. Representing a high-dimensional urban mixing space with only
three endmembers is obviously a gross simplification, but it
does provide a more flexible representation than hard
classifications which attempt to represent a continuum of
reflectance characteristics with exclusive membership in one
of a limited number of idealized classes. Endmember
fraction distributions can also be hardened into a finite
number of thematic classes if necessary (Adams et al., 1995;
Roberts et al., 1998). Spectral mixture models have the
added advantage of producing output in physically intuitive
units (fractional area). Relative abundances of high and low
albedo surface and vegetation can be converted to physical
quantities like albedo, pervious surface area and leaf area
index more easily than radiance measurements can and
more accurately than thematic classes can.
Vegetation fraction estimates derived from IKONOS
imagery provide a tool to monitor urban vegetation health
and abundance. An example of intraurban variations in
vegetation fraction is shown for Chicago in Fig. 10. The
spatial distribution and abundance of vegetation has a direct
impact on the urban environment by modulating solar
energy flux and evapotranspiration. Urban vegetation may
also have an impact on urban air quality as leaves sequester
surface reactant pollutants and particulates (Abdollahi &
Ning, 2000). Mapping spatial distribution and abundance of
urban vegetation is therefore important for modeling urban
air quality. IKONOS 4-m GIFOV approaches the scale of
crown diameter for many tree species. Quantitative assessment of vegetation fraction at this scale could be useful for
monitoring defoliation and blight in large urban parks.
Vegetation distribution can also be used as a proxy for
pervious surface when modeling urban hydrology (Small,
2002b).
Examination of spectral mixing spaces may also prove
useful for feature extraction tasks. In spite of its limited
spectral resolution, IKONOS 11-bit radiometry is capable
of distinguishing subtle differences in surface reflectance
that would be indistinguishable with 8-bit imagery. IKONOS combination of high spatial resolution and bit depth
results in greater clustering within the mixing space. The
Phoenix mixing space shown in Fig. 4 provides an example
184
Fig. 10. Example applications of spectral mixture analysis of urban IKONOS imagery. Vegetation fraction estimates have sufficient spatial resolution to map
intraurban vegetation abundance at street scale in Chicago. Warmer colors show higher vegetation abundance with fractions greater than 50% saturated red.
Localization of features with consistent reflectance properties within the mixing space can facilitate extraction of features like road networks (green) in
Pasadena. Includes materials n Space Imaging.
of this clustering (Fig. 4). Targets with distinct and consistent reflectance will appear as distinct clusters in IKONOS
mixing space whereas their reflectance would be mixed
with, or indistinguishable from, adjacent features in the
mixing space of a sensor with lower spatial resolution or
fewer resolvable brightness levels. An example of a fine
scale feature extraction from mixing space clustering is
shown in Fig. 10. The paved road surfaces in this image
of Pasadena have sufficiently consistent and distinct reflec-
tance that they can be discriminated from other low albedo

features with relatively high accuracy.
6. Conclusions
Spectral mixture analysis provides a physically based
approach to quantify the optical reflectance properties of the
urban mosaic. The spectral mixing space concept accom-
modates the inevitable existence of mixed pixels and provides insight into the variety of distinct and gradational
reflectance patterns present in IKONOS imagery. The mixing space representation also reveals the presence of spectral
endmembers and the extent to which mixing among the
endmembers is linear. When mixing is predominantly linear,
it is possible to define linear mixture models that can be
inverted to yield endmember fraction estimates and measures of misfit to observed data.
Spatial autocorrelation of IKONOS panchromatic imagery provides statistical estimates of the spatial scale of urban
reflectance variations. Two-dimensional autocorrelation
functions consistently show a well-defined, axisymmetric
peak with a width corresponding to the spatial scale of the
most areally abundant features. The width of the peak
therefore provides an estimate of the characteristic spatial
scale of the high contrast reflectance patterns corresponding
to individual features within the urban mosaic (roads,
buildings, trees, etc.). The distribution of length scale
estimates from 6357 sites in 14 urban areas indicates that
the majority of sites have characteristic length scales between 10 and 20 m. This explains why urban areas are
characterized by spectral heterogeneity when imaged by
moderate resolution (20 30 m) sensors. This also suggests
that a significant fraction of IKONOS 4-m pixels will be
spectrally heterogeneous in urban imagery. This is supported by the maximum fraction distributions in Fig. 9.
Principal component transformation of urban IKONOS
imagery allows reflectance patterns to be interpreted in the
context of a spectral mixing space. Eigenvalue distributions
of the imagery used in this study indicate that the majority
(55% to 95%) of image variance corresponds to albedo
variations represented in the first principal component (PC).
A significant, but lesser, fraction of image variance is
associated with vegetation. In the urban areas investigated
here, almost all (>96%) image variance is associated with
the first two PCs. The remaining two PCs contain useful
information about distinct reflectances not discriminated in
the first two PCs and may be useful for isolating additional
endmembers. The spectral mixing space defined by the first
three principal components consistently takes the form of a
triangular scatterplot with linear or concave edges indicating
that mixing is predominantly linear. The spectral endmembers residing at the apexes of the mixing space correspond
to high albedo, low albedo and vegetation endmembers. The
mixing spaces of some of the urban areas used in this study
also revealed the presence of a fourth endmember, usually
corresponding to soil.
Inversion of a simple three-component linear mixture
model produces stable, consistent estimates of endmember
abundance fractions for each pixel in the image. RMS
misfits between observed radiance vectors and modeled
radiance vectors based on fraction estimates are generally
less than 3% of the mean of the observed radiance vector.
Most of the urban areas used in this investigation are
dominated by the low albedo endmember as a result of
185
shadowing and low reflectance building materials. Abundance of vegetation and high albedo features varies considerably within and among cities. Endmember abundance
maps derived from IKONOS can be used to constrain spatial
variations in solar energy flux and evapotranspiration as
well as to map spatial distributions of vegetation and
pervious surfaces.
Acknowledgements
The IKONOS data used in this study were provided by
Space Imaging through the NASA Scientific Data Purchase
program. Includes materials n Space Imaging. This research
would not have been possible without the support of the
Columbia Earth Institute, CIESIN and the NASA SocioEconomic Data and Applications Center (SEDAC).
References
Abdollahi, K. K., & Ning, Z. H. (2000). Urban vegetation and their relative
ability in intercepting particle pollution (PM2.5). Third Symposium on
the Urban Environment, Davis, CA, 1.15.
Adams, J. B. (1986). Spectral mixture modeling: A new analysis of rock
and soil types at the Viking Lander 1 site. Journal of Geophysical
Research, 91, 8089 8122.
Adams, J. B., Sabol, D. E., Kapos, V., Filho, R. A., Roberts, D. A., Smith,
M. O., & Gillespie, A. R. (1995). Classification of multispectral images
based on fractions of endmembers: Application to land cover change in
the Brazilian Amazon. Remote Sensing of Environment, 52, 137 154.
Adams, J. B., Smith, M. O., & Gillespie, A. R. (1993). Imaging spectroscopy: Interpretation based on spectral mixture analysis. In C. M. Englert, & P. Englert (Eds.), Remote Geochemical Analysis: Elemental and
Mineralogical Composition ( pp. 145 166). New York: Cambridge
Univ. Press.
Akbari, H., Rose, L. R., & Taha, H. (1999). Characterizing the Fabric of the
Urban Environment: A Case Study of Sacramento, CA. Report No.
LBNL-44688, Lawrence Berkeley National Laboratory, Berkeley, CA.
Boardman, J. W. (1993a). Automating spectral unmixing of AVIRIS data
using convex geometry concepts. Proceedings Third JPL Airborne
Earth Science Workshop, Pasadena, CA.
Boardman, J. W. (1989b). Inversion of imaging spectrometry data using
singular value decomposition. IGARSS89 12th Canadian Symposium
on Remote Sensing ( pp. 2069 2072).
Elmore, A. J., Mustard, J. F., Manning, S. J., & Lobell, D. B. (2000).
Quantifying vegetation change in semiarid environments: Precision
and accuracy of spectral mixture analysis and the normalized difference
vegetation index. Remote Sensing of Environment, 73, 87 102.
Flanagan, M., & Civco, D. L. (2001). Subpixel impervious surface mapping. Proc. 2001 ASPRS Annual Convention, St. Louis, MO (13 pp).
Getis, A. (1994). Spatial dependence and heterogeneity and proximal databases. In S. Fotheringham, & P. Rogerson (Eds.), Spatial Analysis and
GIS ( pp. 105 120). London: Taylor and Francis.
Gillespie, A. R., Smith, M. O., Adams, J. B., Willis, S. C., Fischer, A. F., &
Sabol, D. E. (1990). Interpretation of residual images: Spectral mixture
analysis of AVIRIS images, Owens Valley, California. Proceedings of
the 2nd Airborne Visible/Infrared Imaging Spectrometer (AVIRIS)
Workshop, Pasadena, CA ( pp. 243 270).
Green, A. A., Berman, M., Switzer, P., & Craig, M. D. (1988). A transformation for ordering mutispectral data in terms of image quality with
implications for noise removal. IEEE Transactions on Geoscience and
186
Green, R. O., & Boardman, J. (2000). Exploration of the relationship

between information content and signal/noise ratio and spatial resolution in AVIRIS data. In R. O. Green (Ed.), Proceedings of the Ninth JPL
Airborne Earth Science Workshop, Pasadena, CA ( pp. 195 206).
Herold, M., Gardner, M., Hadley, B., & Roberts, D. (2002). The spectral
dimension in urban land cover mapping from high-resolution optical
remote sensing data. Proceedings of the 3rd Symposium on Remote
Sensing of Urban Areas, June 2002, Istanbul, Turkey.
Jackson, R. D. (1983). Spectral indices in n-space. Remote Sensing of
Johnson, P. E., Smith, M. O., & Adams, J. B. (1985). Quantitative analysis
of planetary reflectance spectra with principal components analysis.
Journal of Geophysical Research, 90, C805 C810.
Johnson, P. E., Smith, M. O., Taylor-George, S., & Adams, J. B. (1983). A
semiempirical method for analysis of the reflectance spectra of binary
mineral mixtures. Journal of Geophysical Research, 88, 3557 3561.
Kauth, R. J., & Thomas, G. S. (1976). The tasseled capa graphic description of the spectral-temporal development of agricultural crops as
seen by Landsat. Proceedings of the Symposium on Machine Processing
of Remotely Sensed Data ( pp. 4041 4051).
Kressler, F., & Steinnocher, K. (1996). Change detection in urban areas using
satellite images and spectral mixture analysis. Proceedings of the ISPRS.
Kressler, F., & Steinnocher, K. (2000). Monitoring urban development
using satellite images. Proceedings of the Second International Symposium on Remote Sensing of Urban Areas, Regensburg, Germany.
Liu, X., & Lathrop, R. G. (2002). Urban change detection based on an
artificial neural network. International Journal of Remote Sensing,
23(12), 2513 2518.
Madhavan, B. B., & Kubo, S. (2001). Appraising the anatomy and spatial
growth of the Bangkok Metropolitan area using a vegetation impervious soil model through remote sensing. International Journal of
S0034-4257(03)00231-1)
Pech, R. P., Davies, A. W., Lamacraft, R. R., & Graetz, R. D. (1986).
Calibration of Landsat data for sparsely vegetated semi-arid rangelands.
Phinn, S., Stanford, M., Scarth, P., Murray, A. T., & Shyy, P. T. (2002).
Monitoring the composition of urban environments based on the vegetation impervious surface soil (VIS) model by subpixel analysis techniques. International Journal of Remote Sensing, 23(20), 4131 4153.
Rashed, T., Weeks, J. R., Stow, D., & Fugate, D. (2002). Measuring temporal compositions of urban morphology through spectral mixture analysis: Toward a soft approach to change analysis in crowded cities.
Proceedings of the Third International Symposium on Remote Sensing
of Urban Areas, Istanbul, Turkey, 11 13 June.
Ridd, M. K. (1995). Exploring a V I S (Vegetation Impervious Surface Soil) model for urban ecosystem analysis through remote-sensingComparative anatomy for cities. International Journal of Remote
Sensing, 16(12), 2165 2185.
Roberts, D. A., Batista, G., Pereira, J., Waller, E., & Nelson, B. (1998).
Change identification using multitemporal spectral mixture analysis: Applications in Eastern Amazonia. In C. Elvidge, & R. Lunetta (Eds.),
Remote Sensing Change Detection: Environmental Monitoring Applications and Methods ( pp. 137 161). Ann Arbor: Ann Arbor Press.
Roberts, D. A., Smith, M. O., & Adams, J. B. (1993). Green vegetation,
nonphotosynthetic vegetation and soils in AVIRIS data. Remote Sensing
of Environment, 44, 255 269.
S. (2003). IKONOS spatial resolution and image interpretability. Remote Sensing of Environment, 88, 37 51. (doi:10.1016/S00344257(03)00230-X)
Singer, R. B., & McCord, T. B. (1979). Large scale mixing of bright and
dark surface materials and implications for analysis of spectral reflectance. Journal of Geophysical Research, 1835 1848.
Settle, J. J., & Drake, N. A. (1993). Linear mixing and the estimation of
ground cover proportions. International Journal of Remote Sensing,
14(6), 1159 1177.
Small, C. (2001a). Estimation of urban vegetation abundance by spectral
mixture analysis. International Journal of Remote Sensing, 22(7),
1305 1334.
Small, C. (2001b). Multiresolution analysis of urban reflectance. Proceedings of the IEEE Workshop on Remote Sensing and Data Fusion of
Urban Areas, 8 9 Nov. 2001, Rome, Italy.
Small, C. (2001c). Spectral dimensionality of urban radiance. Proceedings
of 10th JPL Airborne Earth Science Workshop, NASA Jet Propulsion
Laboratory, Pasadena, CA.
Small, C. (2002a). A global analysis of urban reflectance. International
Journal of Remote Sensing (In press).
Small, C. (2002b). Reflectance properties of pervious and impervious surfaces. Proceedings of the ASPRS Pecora Land Remote Sensing Conference, Denver CO.
Smith, M. O., Johnson, P. E., & Adams, J. B. (1985). Quantitative determination of mineral types and abundances from reflectance spectra
using principal component analysis. Journal of Geophysical Research,
90, 792 804.
Smith, M. O., Ustin, S. L., Adams, J. B., & Gillespie, A. R. (1990).
Vegetation in deserts: I. A measure of abundance from multispectral
images. Remote Sensing of Environment, 31, 1 26.
Stefanov, W. L., Ramsey, M. S., & Christensen, P. R. (2001). Monitoring
urban land cover change: An expert system approach to land cover
classification of semiarid to arid urban centers. Remote Sensing of
Environment, 77(2), 173 185.
Wu, C. S., & Murray, A. T. (2003). Estimating impervious surface distribution by spectral mixture analysis. Remote Sensing of Environment,
84(4), 493 505.
Wulder, M., & Boots, B. (1998). Local spatial autocorrelation characteristics of remotely sensed imagery assessed with the Getis statistic.
International Journal of Remote Sensing, 19(11), 2223 2231.
Snyder, G. L., & Lee, G. (2003). The joint agency commercial imagery
approach. Remote Sensing of Environment, 88, 17 22. (doi:10.1016/
S0034-4257(03)00226-8)

Use of IKONOS and Landsat for malaria control in the Republic of Korea
Penny M. Masuoka a,*, David M. Claborn b, Richard G. Andre c, Joseph Nigro c,
Scott W. Gordon d, Terry A. Klein e, Hung-Chol Kim e
a
Uniformed Services University of the Health Sciences, Goddard Space Flight Center, Code 920, Greenbelt, MD 20771, USA
b
Navy Disease Vector Ecology and Control Center, Naval Air Station, Box 43, Jacksonville, FL 32212-0043, USA
c
Department of Preventive Medicine and Biometrics, Uniformed Services University of the Health Sciences, Bethesda, MD 20814-4799, USA
d
U.S. Army Center for Health Promotion and Preventive Medicine, 1312 Cobb Street SW, Fort McPherson, GA 30330, USA
e
Preventive Services Directorate, 18th Medical Command, Unit 15281, APO AP 96205-0054, USA
Received 26 April 2002; received in revised form 17 February 2003; accepted 24 April 2003
Abstract
Malaria reemerged in the Republic of Korea (ROK) in 1993. While limited numbers of U.S. soldiers in high-risk areas use chloroquine/
primaquine chemoprophylaxis to prevent malaria, control of mosquito larvae through larviciding also can be used to reduce the risk of
malaria transmission. In order to estimate the cost of larviciding, accurate estimates of the spatial extent of mosquito larval habitats are
necessary. The purpose of this study was to determine whether an accurate estimate of the area covered by mosquito larval habitats can be
obtained using Landsat 7 Enhanced Thematic Mapper+ (ETM+) and/or IKONOS data for the Korean test site.
To estimate the area covered by larval habitats near Camp Greaves [Paekyeon-Ri, near Tongil-Chon (village)] in the ROK, an IKONOS
and a Landsat 7 ETM+ image were classified using a parallelepiped classification. In a comparison with rice paddy field sites, 24 (92%) of
the sites were classified correctly on the IKONOS image and 17 (65%) were classified correctly on the Landsat image. Comparing the
classifications on a pixel-by-pixel basis, the agreement between the two classifications was 79%. Part of the disagreement was due to the
difference in resolution of the two images. In spite of local differences, the two classifications produced similar area estimates.
Although either Landsat or IKONOS could be used in Korea for a reasonable estimate of habitat area, only IKONOS can resolve small
irrigation ponds. While ponds represent a small portion of the total larval habitat, they are an important source for mosquito breeding during
the late rice-growing season in the ROK since they contain higher larval densities. High-resolution imagery, such as IKONOS, would be
necessary for planning and implementing treatment of these smaller habitats.
Keywords: IKONOS; Landsat; Malaria control
1. Introduction
After being absent from the Republic of Korea (ROK)
since the 1970s, Plasmodium vivax malaria reemerged in
1993 with the occurrence of two cases (Chai et al., 1994).
The number of cases has grown almost every year since,
resulting in 1642 cases in 1997 (Feighner, Pak, Novakoski,
Kesley, & Strickman, 1998) and peaking in 2000 with 4142
cases (Korea National Institute of Health). The focus of
malaria has been just south of the Demilitarized Zone
(DMZ) in Kyonggi and Kangwon Provinces. The primary

E-mail address: penny@ltpmail.gsfc.nasa.gov (P.M. Masuoka).
doi:10.1016/j.rse.2003.04.009
vectors of malaria in South Korea are Anopheles sinensis

Wiedemann and perhaps A. lesteri Baisas and Hu, species
that are associated with the rice paddy environment
(Tanaka, Misusawa, & Saugstad, 1979). During 2001, the
number of malaria cases declined to 2533, with 29 (1.1%)
of the cases occurring in U.S. military personnel who had
been stationed in the Republic of Korea (Preventive Services Directorate, 18th Medical Command, Seoul, ROK,
personal communication).
Since 1999, U.S. Army personnel stationed near the
DMZ have used chloroquine/primaquine chemoprophylaxis
and other preventive measures such as permethrin-impregnated bednets and uniforms and topical deet mosquito
repellents. The use of larvicides in rice paddies and ponds
surrounding the military bases also has been suggested
(Strickman et al., 1999). This study is part of a larger effort
188
P.M. Masuoka et al. / Remote Sensing of Environment 88 (2003) 187194
to determine the utility and cost of various methods to

reduce the risk of malaria transmission to U.S. military
personnel. The purpose of this study was to determine
whether an accurate estimate of the area covered by mosquito larval habitats can be obtained using Landsat 7
Enhanced Thematic Mapper+ (ETM+) and/or IKONOS data
for the Korean test site. The area estimates of mosquito
habitats then can be used to estimate the cost of larviciding
near U.S. military bases in Korea.
The use of remote sensing to detect mosquito breeding
habitats has been shown to be possible by several authors,
and a review of the techniques can be found in an article by
Hay, Snow, and Rogers (1998). Remote sensing has been
used to predict which California rice fields will have the
highest production of A. freeborni larvae nearly 2 months
before the peak larval density occurs (Roberts & Rodriguez,
1994; Wood et al., 1991). Beck et al. (1994) used images
from Landsat Thematic Mapper (TM) to estimate the risk of
malaria in 40 villages in Chiapas, Mexico, based on two
environmental factors: transitional swamp and unimproved
pasture. Rejmankova, Roberts, Pawley, Manguin, and
Polanco, (1995) used classified multispectral SPOT data to
identify marshes containing vegetation favorable for mosquito habitat in Belize. Thomas and Lindsay (2000) used
SPOT imagery to estimate the risk of exposure to malaria in
rural Gambian children. The current study utilized Landsat
ETM+ and IKONOS images to quantify the size of larval
habitats within the vectors flight range around two U.S.
Army bases in the ROK.
IKONOS imagery is a commercial product acquired by
the IKONOS satellite and sold by Space Imaging (http://
www.spaceimaging.com). ETM+ is acquired by the Landsat
7 satellite and sold by the United States Geological Survey
(http://earthexplorer.usgs.gov). Characteristics of IKONOS
and Landsat are listed in Table 1. One important difference
in the two sensors is that IKONOS lacks mid-infrared (IR)
channels, which are very useful in vegetation studies. Scene
size is different for the two sensors: 11 11 km for the
IKONOS versus 185 185 km for Landsat. Generally,
IKONOS acquisitions must be requested, whereas Landsat
is automatically acquired every 16 days over the United
States and approximately every 4 months elsewhere (Arvidson, Gasch, & Godward, 2001). Although costs have
historically varied for both sensors, IKONOS imagery has
always been more expensive per square kilometer than
Landsat. The resolution of the two sensors is 15 and 30 m
for the Landsat panchromatic and multispectral bands,
respectively, versus 1 and 4 m for the IKONOS panchromatic and multispectral bands, respectively. Using IKONOS
mosaics for large areas not only costs more than Landsat,
the large image size can be very difficult to store and
process. However, for certain projects, the increased resolution of IKONOS may be necessary. One of the questions
that this study attempts to answer is whether the highresolution imagery is more effective in delineating selected
types of small larval mosquito habitats, and, if so, how this
Table 1
Characteristics of IKONOS and Landsat images
Image
Band
Wavelength
(Am)
IKONOS
(1) Blue
(2) Green
(3) Red
(4) Near-IR
Panchromatic
0.45 0.52
0.52 0.60
0.63 0.69
0.76 0.90
0.45 0.90
4
4
4
4
1
Landsat ETM +
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.45 0.52
0.53 0.61
0.63 0.69
0.75 0.90
1.55 1.75
10.4 12.5
2.09 2.35
0.52 0.90
30
30
30
30
30
60
30
15
Image
Scene size (km)
IKONOS
Landsat ETM +
11 11
185 185
Blue-green
Green
Red
Near-IR
Mid-IR
Thermal
Mid-IR
Panchromatic
Resolution
(m)
The two sensors have significant differences in radiometric properties, and

the gains are different. Furthermore, the conversion from measured analog
signal to digital numbers results in 8-bit data for Landsat and 11-bit data for
IKONOS (Goward, Davis, Fleming, Miller, & Townshend, this issue;
Pagnutti et al., this issue).
affects the overall estimate of total habitat as predicted by

the low-resolution image.
2. Methods
2.1. Field work
Field work for this project was performed from June
through September 2000, and concentrated on two military
bases near the DMZ: Camp Greaves and Camp Casey.
Camp Greaves [Paekyeon-Ri, near Tongil-Chon (village)]
is located in a rural area just south of the DMZ. Camp Casey
(Tongducheon) is approximately 35 miles east of Camp
Greaves in a more populated area with less agriculture.
Standard larval survey techniques using a plastic dipper
in all types of standing fresh water were conducted at both
sites. Seven potential types of larval habitats were identified
and sampled: (1) rice fields, (2) streamside pools, (3)
irrigation ponds, (4) irrigation ditches, (5) drainage ditches,
(6) swamps, and (7) rivers. Only two small swamps occurred in the study area and were not found to be important
larval habitats. The Imjin River parallels the rear boundary
of Camp Greaves and could not be sampled due to military
security measures. However, a nearby upriver site was
sampled and was negative for mosquito larvae.
During larval sampling, water quality testing was conducted to determine the potential effect of minerals on
mosquito larvae populations. Samples were tested for nitrate
and phosphate concentrations, total dissolved solids, and pH.
Statistical analysis of these data showed that none of these
factors was predictive for the presence or absence of Anopheles larvae (Claborn, Hshieh et al., 2002). These results
suggest that most standing or slowly moving fresh water is
suitable as a larval habitat for A. sinensis in this area.
Each field site was located by using a Garmin III Global
Positioning System (GPS) unit (Garmin International,
Olathe, KS, http://www.garmin.com/). Since GPS readings
could not be taken in the center of the rice fields due to
potential crop damage, four readings were taken at the
corners of the field and then the points were averaged to
get an accurate estimate of the center of the field. GPS
points were plotted on topographic maps in the field using
ArcView version 3.2 (Environmental Research Systems
Institute, Redlands, CA, http://www.esri.com/). After the
completion of the fieldwork, points also were displayed
on the IKONOS image. Almost all of the 93 points collected
were located within 5 m of the actual ground points. Four
points had a greater locational error, possibly due to
electromagnetic interference from nearby transmission lines
(Earth Observation Magazine (EOM) Archives, 2002, http://
www.eomonline.com/Common/Archives/Oct95/gps.htm,
April 15, 2002) or from deliberate interference from local
189
military forces (Ward & Johannessen, 1996). The four offset

points were manually corrected using the topographic maps
or IKONOS image as a reference.
Because of the interest in estimating the amount of larval
habitat in areas that would affect the camps, buffer zones
were created around the two camps using Arc/Info software
(Environmental Research Systems Institute, Redlands, CA,
http://www.esri.com/). The boundary of Camp Casey was
digitized and a 1-km buffer zone was placed around the
camp based on an approximate 1-km flight range of the A.
sinensis mosquito. Due to the size and shape of Camp
Greaves, a 2-km radius circle was created around a point
marking the approximate center of Camp Greaves, resulting
in an approximate 1-km buffer zone around the perimeter of
the camp. Fig. 1 shows the larval sampling sites and the 1km buffer zone around Camp Greaves.
2.2. Image analysis
Two images were used for this study: a Landsat 7 ETM+
image acquired on April 29, 2000, and an IKONOS image
acquired on August 2, 2000. The Landsat image covers both
Fig. 1. IKONOS image with buffer zone around Camp Greaves and larval sampling sites. (Includes material from Space Imagingn.)
190
the Camp Greaves and the Camp Casey sites. The IKONOS
image lies entirely within the Landsat image but only covers
the Camp Greaves site. The images were georeferenced to a
UTM projection with a WGS-84 datum. A visual comparison of the images with each other and with georeferenced
topographic maps showed that the geometric correction of
the images was good.
PCI remote sensing software (PCI Geomatics, Richmond
Hill, Ontario, Canada, http://www.pcigeomatics.com) was
used to perform supervised classifications on the IKONOS
and Landsat images. Training sites for the classification
were selected at locations where researchers had sampled
standing water for anopheline larvae. Various classification

algorithms were investigated including minimum distance,
maximum likelihood, and parallelepiped programs. The
parallelepiped algorithm with a maximum likelihood as a
tiebreaker appeared to be the most accurate classification of
the images based on a visual comparison with a plot of the
sampling sites on the IKONOS image. For the IKONOS
image, training sites were collected for the rivers, ponds,
ditches, and rice fields. Because of the lower resolution,
training sites on the Landsat image included only rice fields
and the river; ponds and ditches were too small to be
resolved. The river was considered to be nonhabitat based
Fig. 2. Comparison of IKONOS false color composite and classification with Landsat. On classification images, the river class is shown in yellow, rice fields in
green, and ponds in red. Each image is 5.5 km in width. (Includes material from Space Imagingn.)
on sampling at upriver sites. Other nonhabitat areas of the

images were classified as urban and forest, but these were
later grouped together as a single, nonhabitat class for the
purpose of estimating area.
PCIs MODEL program was used to generate reports on
the area covered by each class in the classified images. To
compare the classification results between the two imagery
types, a subset of the Landsat image was made to include
only the area covered by the IKONOS image. A confusion
matrix was generated to compare the classification accuracy
of the Landsat and IKONOS with field sample sites common to both images. The MODEL program also was used to
compare the Landsat and IKONOS classifications on a
pixel-by-pixel basis and to create a new image that depicted
matching and nonmatching pixels.
3. Results
3.1. Classification
Figs. 2 and 3 show the result of the classifications of the
Landsat and IKONOS images. Based on a visual comparison between the classification and the original image,
191
rivers, ponds, and rice fields were successfully classified

on the IKONOS imagery. Ditches could not be successfully
classified on the IKONOS imagery, possibly due to trees,
shrubs, and other plants that grow along the ditches and
make them spectrally similar to other land cover classes. On
the Landsat imagery, rivers and rice fields could be classified, but ponds and ditches were too small and could not be
used for collecting training sites. A visual comparison of the
Landsat and IKONOS classifications shows that they are
quite similar in the classification of the river (Figs. 2 and 3).
Large areas of rice fields were classified fairly accurately on
the Landsat classification (Fig. 3). However, small rice
fields, as seen in the NW corner of the IKONOS image
(Fig. 3), were not correctly classified by the Landsat
classification.
A confusion matrix (or error matrix) was calculated for
each of the classifications using the larval collection sites
that fell within the IKONOS image as the reference data
(Table 2). Twenty-six rice fields were sampled; 24 (92%)
were classified correctly on the IKONOS image and 17
(65%) were classified correctly on the Landsat. Only one
swamp was sampled, and since this was not a land cover
type for which a training site was selected, it was accurately
classified as unknown on the IKONOS image, but as rice on
Fig. 3. An enlarged subset of the images from Fig. 2 comparing an IKONOS false color composite and classification with Landsat. On classification images,
the river class is shown in yellow, rice fields in green, and ponds in red. Image width is 1.9 km. (Includes material from Space Imagingn.)
192
Table 2
Confusion matrices of land cover at larval sample sites compared to land
cover determined by classification of the satellite imagery
Classified data
Field sample sites

Rice
Confusion matrixIKONOS
Rice
24
Pond
0
Ditch
0
Swamp
0
Unknown
2
Total
26
Pond
Ditch
Swamp
Total
0
0
0
0
3
3
4
0
0
0
9
13
0
0
0
0
1
1
28
0
0
0
15
43
8
0
0
0
5
13
1
0
0
0
0
1
28
0
0
0
15
43
Confusion MatrixLandsat 7 ETM +

Rice
17
2
Pond
0
0
Ditch
0
0
Swamp
0
0
Unknown
9
1
Total
26
3
The same sample site locations were used for both matrices.
the Landsat image. Ditches were classified as either rice or

unknown on both the IKONOS and the Landsat images.
Only three ponds were sampled in the field in the area
covered by the IKONOS image. Although a visual inspection of the classification with the IKONOS 1-m panchromatic image shows that many ponds were accurately
classified, none of the three ponds sampled in the field
was correctly identified in the classification. The misclassification of the ponds on the IKONOS image is probably due
to the small size of these sampled ponds; the largest sampled
pond was only 9 7 m.
A comparison of the Landsat and IKONOS classification
was done using the PCI MODEL program to compare the
classification on a pixel-by-pixel basis. White pixels in Fig.
4 represent the pixels that were classified the same on the
two images; black pixels were classified differently. A report
generated by PCIs MODEL program calculated a 79%
agreement in pixel classification on the Landsat and IKONOS images. Differences in the classification may in part
represent differences in resolution. Also, since the Landsat
image was acquired early in the growing season (April 29)
and the IKONOS was acquired late in the growing season
(August 2), the differences in classifications may be partially
due to the different stages of rice development.
More modeling was done to explore the differences in
the two classifications. In Fig. 5, black pixels represent areas
that were classified as habitat on Landsat but as nonhabitat
on IKONOS. One reason for the difference is that Landsat
classified the small roads and large dikes separating the rice
fields as rice paddies because of lower resolution, whereas
IKONOS classified them as nonhabitat. There also may be
some mixed pixels on the Landsat that were classified as
rice but were separated into rice and nonhabitat on the
IKONOS image.
In Fig. 6, black pixels represent areas that were classified
as habitat on IKONOS but as nonhabitat on Landsat. Larger
Fig. 4. Image showing the agreement of pixels in the classification of

Landsat and IKONOS images. Black pixels (21% of image) were classified
differently in the two images. White pixels were assigned to the same class.
Image covers the same area as Fig. 2 and is 5.5 km wide.
patches of black on this image represent fields that were

inaccurately classified as nonhabitat on the Landsat. A
comparison with the Landsat image shows that some of
these fields are obviously spectrally different from the other
rice fields on the same image and may represent a different
Fig. 5. Image showing pixels (in black) that were classified as habitat on
Landsat but as nonhabitat on IKONOS. Image covers the same area as
Fig. 2 and is 5.5 km wide.
193
Table 4
Cost comparison of chemoprophylaxis to larviciding for control of malaria
within two military bases and a surrounding 1-km buffer zone
Camp Greaves
Camp Casey
430.4 ha of habitat
Larvicide treatment = US$40,263.13
Chemoprophylaxis cost for 760
persons = US$28,522.80
122.5 ha of habitat
Larvicide treatment = US$11,450.37
Chemoprophylaxis cost for 8000
people = US$330,264.00
Habitat estimates are based on the Landsat classification.
Fig. 6. Image showing pixels (in black) that were classified as habitat on
IKONOS but as nonhabitat on Landsat. Image covers the same area as
Fig. 2 and is 5.5 km wide.
stage of rice planting (nonflooded versus flooded). In the

IKONOS image that was acquired later in the growing
season, rice crops were mature and no apparent differences
existed between fields. Some of the small rice fields that
were classified as nonhabitat on Landsat also are apparent
on Fig. 6. A comparison of Fig. 6 with Fig. 2 shows that a
scattering of pixels from forested areas was misclassified as
rice on the IKONOS image.
Area estimates for the buffer zone around Camp Greaves
are shown in Table 3. Although the difference in the
classification of Landsat and IKONOS images was approximately 20%, the area estimates were very close. The
Landsat 7 ETM+ estimate of rice paddies was only 2.4%
less than the IKONOS estimate.
3.2. Larvicide treatment cost estimates
The habitat land cover estimates obtained in this paper
have been used to estimate the cost of mosquito larviciding
around the camps as a control method. The method of
Table 3
Comparison of land cover area estimates (m2) for Camp Greaves and
surrounding area, Republic of Korea
Class
Image
IKONOS
Landsat
Rice fields
Ponds
River
Nonhabitat
4,198,151
48,709
1,465,431
6,789,022
4,304,250
None
1,604,925
6,502,500
Nonhabitat includes all areas other than rivers, rice fields, and ponds.
estimating cost for larvicidal treatments is described in detail

in Claborn, Masuoka, Andre, Hooper, & Klein, 2002.
Briefly, estimates were based on current costs of a growthregulating insecticide, wages for workers who will apply the
larvicide, fuel cost, and maintenance expense for groundbased dispersal equipment. These were compared to the cost
of providing antimalarial medications based on the price of
the drugs and required pretreatment testing.
Larviciding cost estimates were compared to the cost of
chemoprophylaxis for U.S. personnel stationed at Camp
Greaves and Camp Casey. Results of the estimates are
shown in Table 4. Because of the large habitat area and
the low number of soldiers at Camp Greaves, malaria is
most cost-effectively controlled with chemoprophylaxis. In
Camp Casey, it is cheaper to use larviciding than chemoprophylaxis due to the large number of soldiers and the
small mosquito habitat area.
4. Conclusions
We have found that similar land cover area estimates of
mosquito larval habitat can be obtained from IKONOS and
Landsat 7 ETM+ imagery. For estimating the costs of
larviciding and other types of planning, which need only
rough estimates of the major habitat areas defined, Landsat
would be adequate. On a local level, the IKONOS image
allowed a better classification of rice fields, accurately
identifying 92% of the field sites versus 65% for the Landsat. Although we were unable to classify the three very
small ponds sampled in the field, many larger ponds were
visible on the IKONOS image. The use of IKONOS has the
advantage of being able to portray and classify land cover
features such as ponds and rice fields that are less than
30 30 m in size. Although ponds represent a relatively
small portion of the total habitat area, they are an important
breeding habitat for mosquitoes in Korea since they contain
higher larval densities than the rice fields late in the growing
season (Claborn, Hshieh et al., 2002). For areas where small
features represent the majority of the habitat, high-resolution
imagery would be necessary. In addition, high-resolution
imagery would be much more useful in planning sampling
collections and larviciding tasks than Landsat imagery.
The two imagery types could be used together in planning and implementing a malaria control program. Landsat,
in conjunction with information on the location and number
194
of cases in a geographic information system (GIS), could be

used in the first stages of planning to estimate the costs and
to plan the location of the spraying. The size of the area to
be treated could be adjusted within the GIS depending on
the budget available for the control program. Once local
treatment areas are selected, IKONOS imagery could be
used to locate habitats and track the local spraying efforts.
Worldwide, mosquitoes breed in a wide variety of
habitats. Many mosquito habitats such as small marshes
and streams can be mapped only on high-resolution imagery
such as IKONOS. In many regions of the world, highresolution imagery would be very useful for studying and
controlling malaria and other mosquito-borne diseases.
Acknowledgements
The work described in this paper was partially supported
by NASA grant NAG5-8532. The authors would like to
thank the following people for their support: Locke Stuart,
Nancy Maynard, Donald Roberts, Rita Aissi-Wespi, McKinley Rainey, William Herman, Kenneth McPherson, and
Alex Ornstein. The soldiers of the 702 Preventive Medicine
Section, 2nd Infantry Division, 18th Medical Command,
provided assistance in the field work phase of this research.
Several anonymous reviewers and the editors of this issue
provided helpful comments and suggestions.
References
Arvidson, T., Gasch, J., & Godward, S. N. (2001). Landsat 7s long term
acquisition planan innovative approach to building a global archive.
Special issue on Landsat 7. Remote Sensing of Environment, 78(1 2),
13 26.
Beck, L. R., Rodriguez, M. H., Dister, S. W., Rodriguez, A. D., Rejmankova, E., Ulloa, A., Meza, R. A., Roberts, D. R., Paris, M. A., Spanner,
R. K., Washino, C., Hacker, L. J., & Letgers, J. F. (1994). Remote
sensing as a landscape epidemiological tool to identify villages at high
risk for malaria transmission. American Journal of Tropical Medicine
and Hygiene, 51(3), 271 280.
Chai, I. H., Lim, G. I., Yoon, S. N., Oh, W. I., Kim, S. J., & Chai, J. Y.
(1994). Occurrence of tertian malaria in a male patient who has never
been abroad. Korean Journal of Parasitology, 32, 195 200.
Claborn, D. M., Hshieh, P. B., Roberts, D. R., Klein, T. A., Zeichner, B. C.,
& Andre, R. G. (2002). Environmental factors associated with larval

habitats of malaria vectors in Northern Kyunggi Province, Republic of
Korea. Journal of the American Mosquito Control Association, 18(3),
178 185.
Claborn, D. M., Masuoka, P. M., Andre, R. G., Hooper, T., & Klein, T. A.
(2002). A cost comparison of two malaria control methods in Kyonggi
Province, Republic of Korea, using remote sensing and geographic
information systems. American Journal of Tropical Medicine and Hygiene, 66(6), 680 685.
Earth Observation Magazine (EOM) Archives, http://www.eomonline.com/
Common/Archives/Oct95/gps.htm, April 15, 2002.
Feighner, B. H., Pak, S. I., Novakoski, W. L., Kelsey, L. L., & Strickman,
D. (1998). Re-emergence of Plasmodium vivax malaria in the Republic
of Korea. Emerging Infectious Diseases, 4(2), 295 298.
Goward, S. N., Davis, P. E., Fleming, D., Miller, L., & Townshend, J. R. G.
sites. Remote Sensing of Environment. (doi:10.1016/j.rse.2003.06.007)
Hay, S. I., Snow, R. W., & Rogers, D. J. (1998). From predicting mosquito
habitat to malaria seasons using remotely sensed data: Practice, problems and perspectives. Parasitology Today, 14(8), 306 313.
Schiller, S. (2003). Radiometric characterization of IKONOS multispectral imagery. Remote Sensing of Environment. (doi:10.1016/S00344257(03)00231-1)
Rejmankova, E., Roberts, D. R., Pawley, A., Manguin, S., & Polanco, J.
(1995). Predictions of adult Anopheles albimanus densities in village
based on distances to remotely sensed larval habitats. American Journal
of Tropical Medicine and Hygiene, 53(5), 482 488.
Roberts, D. R., & Rodriguez, M. H. (1994). The environment, remote
sensing, and malaria control. Annals of the New York Academy of Science, 740, 396 402.
Strickman, D., Miller, M. E., Kelsey, L. L., Lee, W. J., Lee, H. W., Lee,
K. W., Kim, H. C., & Feighner, B. H. (1999). Evaluation of the
malaria threat at the Multipurpose Range Complex, Yongpyong, Republic of Korea. Military Medicine, 164, 626 629.
Tanaka, K., Misusawa, K., & Saugstad, E. S. (1979). A revision of the adult
and larval mosquitoes of Japan (including the Ryukyu Archipelago and
the Ogasawara Islands) and Korea (Diptera: Culicidae). Contributions
on Entomology, International, 16 (987 pp.).
Thomas, C. J., & Lindsay, S. W. (2000). Local-scale variation in malaria
infection amongst rural Gambian children estimated by satellite remote
sensing. Transactions of the Royal Society of Tropical Medicine and
Hygiene, 94, 159 163.
Ward, N., & Johannessen, R. (1996). Interference to GPS in the marine
environment. Oceanographic Literature Review, 43(12), 1286.
Wood, R., Washino, R., Beck, L., Hibbard, K., Pitcairn, M., Roberts, D.,
Rejmankova, E., Paris, J., Hacker, C., Salute, J., Sebesta, P., & Letgers,
L. (1991). Distinguishing high and low anopheline-producing rice fields
using remote sensing and GIS technologies. Preventive Veterinary Medicine, 11, 277 288.

IKONOS imagery for resource management: Tree cover, impervious

surfaces, and riparian buffer analyses in the mid-Atlantic region
Scott J. Goetz a,b,*, Robb K. Wright b, Andrew J. Smith b, Elizabeth Zinecker b, Erika Schaub b
a
The Woods Hole Research Center, PO Box 296, Woods Hole, MA 02543-0296, USA
Department of Geography, University of Maryland, College Park, MD 20742-8225, USA
Received 9 August 2002; received in revised form 4 June 2003; accepted 4 July 2003
Abstract
High-resolution imagery from the IKONOS satellite may be useful for many resource management applications. We assessed the utility of
IKONOS imagery for applications in the mid-Atlantic region, including mapping of tree cover, impervious surface areas, and riparian buffer
zone variables in relation to stream health ratings. We focused on a 1313-km2 area in central Maryland using precision-georeferenced
IKONOS products. We found the IKONOS imagery to be a valuable resource for these applications, and were able to achieve map accuracies
comparable to manual aerial photo interpretation. We were also able to use derived data sets for consistent assessments over areas that would
be difficult to accomplish with traditional photographic mapping methods. For example, we found that a stream health rating of excellent
required no more than 6% impervious cover in the watershed, and at least 65% tree cover in the riparian zone. A rating of good required less
than 10% impervious and 60% tree cover. A number of issues associated with application of the IKONOS data arose, however, including
logistics of image acquisition related to phenological and atmospheric conditions, shadowing within canopies and between scene elements,
and limited spectral discrimination of cover types. Cost per unit area was also a nontrivial consideration for the image data products we used,
but allowed us to provide valuable derived products to agencies in support of their planning and regulatory decision-making processes. We
report on both the capabilities and limitations of IKONOS imagery for these varied applications.
Keywords: IKONOS imagery; Resource management; Mid-Atlantic region
1. Introduction
Information on land cover has become an integral part of
the environmental and developmental planning process. It
has aided in the advancement of more effective land use
planning, habitat assessments, and hydrological applications. Historically, land cover and land use information
was obtained by a combination of field measurements and
aerial photo interpretation. This approach typically required
intensive interpretation by expert analysts, and cross validation methods to ensure that analyst interpretations were
consistent. Recently, satellite imagery has become available
at spatial resolution nearly comparable to aerial photographs, with the added advantage of digital multispectral
* Corresponding author. The Woods Hole Research Center, PO Box
296, Woods Hole, MA 02543-0296, USA. Tel.: +1-508-540-9900; fax: +1508-540-9700.
E-mail address: sgoetz@whrc.org (S.J. Goetz).
doi:10.1016/j.rse.2003.07.010
information more complete even than those provided by

digital orthophotographs (DOQs). The platform stability of
high-resolution (1 5 m) imagery acquired from Earthobserving satellites provides another advantage to aerial
photographs acquired from aircraft, which roll, pitch, and
yaw during flight and require corrections for those effects.
Because of these advances and advantages, as well as
commercial potential, several companies have launched or
plan to launch high-resolution satellites in the near future
(Stoney, 2001).
High-resolution multispectral imagery has many potential benefits to government organizations, nonprofit agencies, and a wide array of mapping and related commercial
ventures (Dial, Bowen, Gerlach, Grodecki, & Oleszczuk,
this issue; Sawaya et al., this issue; Tanaka & Toshiro,
2001). Applications of these data can aid and assist the
monitoring and management of resource lands, parks, wetlands, and other protected areas, as well assess the effects of
natural disasters or complement protective measures in areas
196
S.J. Goetz et al. / Remote Sensing of Environment 88 (2003) 195208
that have the potential to burn or flood, to name just a few.

High spatial resolution imagery is not a panacea for these
applications, however, owing to a number of issues that
arise with their use. For example, the increased textural
information available in fine-resolution imagery allows for
improved interpretation based on the shape and texture of
ground features, but techniques that have been developed to
process and analyze current satellite data, including vegetation indices or multitemporal classification techniques that
utilize mid-infrared or thermal channels (e.g., Varlyguin,
Wright, Goetz, & Prince, 2001), may not be applicable to
the additional information provided by high-resolution satellites. Other issues include difficulties in acquiring consistent sequential acquisition dates, intensive computer
processing time and disk space required to store larger
image data sets, and considerations of economic efficiencies
(Fisher & Goetz, 2001).
The objective of this paper was to assess the practicality
of using high-resolution imagery, in this case provided by the
IKONOS satellite operated by Space Imaging, for resource
management applications in Montgomery County in the
mid-Atlantic region of the United States. We sought to utilize
the IKONOS imagery as an alternative to air photo interpretation for updating the forested lands map, as well as to map
changes that had occurred in land use, particularly residential
development and intensification of impervious surface areas.
Additional specific goals were to assess the utility of the
IKONOS imagery for production of tree cover and impervious surface area maps, and to examine those map products
and derivatives, including riparian buffer zone land cover,
relative to stream water quality. Related applications include
the use of IKONOS to train subpixel algorithms of tree cover
and impervious surfaces using coarser resolution imagery
(e.g., Landsat). The tree and impervious area maps, when
combined, provide an improved ability to meet resource
management goals, particularly with respect to comprehensive planning, rural land protection, and goals for improved
water quality. We identify and address some of the benefits
and limitations of high-resolution imagery for these research
applications.
2. Study area and data sets

The study area encompasses all of Montgomery County,
MD (1313 km2) (Fig. 1), an area comprised of a mixture of
forest and farms interspersed with a range of residential
developments and industrial/commercial zones. The county
was selected for this research because it seeks to develop
improved geospatial information, capabilities, and technologies to assess impacts of environmental change, and a
range of resource management decisions. The Maryland
National Capital Parks and Planning Commission (MNCPPC) and the Montgomery Department of Environmental Protection (DEP) are well advanced in their utilization of
geographic information systems (GIS) and have been na-
tionally recognized and awarded for their countywide forest

preservation and stream protection strategies. The Planning
Board of Montgomery County has established master plans
for each region to plan future development, and has been at
the forefront of many of Marylands Smart Growth
programs. One of these includes the Legacy Open Space
program to fund the purchase of culturally and environmentally unique lands and to protect the countys surface water
supply (M-NCPPC, 2000). To date, more than 12,000 ha
have been protected as parks and open space, at a cost of
over US$137 million, and additional acquisitions are
planned within a 39,000-ha agricultural preserve. There is
a desire by the DEP, M-NCPPC, and others to have these
lands connected to other parcels reserved under the
counties parks system and related land conservation programs, providing the county a green infrastructure of
natural areas to promote recreation opportunities, protect
water resources, and maximize biodiversity (Weber & Wolf,
2000). In the 10 years since the last aerial photo-based forest
land cover classification, however, there had been considerable residential development and expanded transportation
projects, with associated predictions of rapid future rates of
land conversion (Jantz, Goetz, & Shelley, in press). These
factors provided an impetus for the applications research
presented here.
2.1. IKONOS imagery
IKONOS is the first commercially owned satellite
providing 1-m resolution panchromatic image data and 4m multispectral imagery (Dial et al., this issue). The
multispectral image data include three visible and one
infrared channel (Table 1). Data are collected in 11-bit
radiometric resolution and provided in a format compatible
with image analysis software. The tile size for each
individual scene is 11.3 11.3 km. Because of its pointable
off-nadir viewing capability, the satellite revisit interval is
as little as 3 4 days.
The IKONOS images of Montgomery County (Fig. 2)
were obtained through the NASA Scientific Data Purchase, a
program designed to make remote sensing data sets available
for research and practical applications (Birk, Stanley, &
Snyder, this issue). The acquired imagery was equivalent
to Space Imagings Carterra Precision product, which is a
precision-georeferenced image data product with an absolute horizontal geometric accuracy of 5 m. The NASA
Scientific Data Purchase contract specification allowed for
mislocation errors of 250 m in standard products (see Helder,
Coan, Patrick, & Gaska, this issue), thus a precision product
was required for the analyses we wished to conduct. The
orthocorrected product we acquired had an RMS error of
approximately 1.9 m. Eleven IKONOS image tiles were
acquired in six swaths to provide complete coverage of the
entire county (Table 2). The nominal cost of the imagery was
US$141 km 2. Although the IKONOS platform has a short
repeat schedule, a cloud-free image could not be acquired for
197
Fig. 1. Study area map of Montgomery County, MD, within the mid-Atlantic region and Chesapeake Bay watershed.
Panel 7 during a 12-month acquisition window. For this

panel, a cloud and cloud shadow mask were created manually, through which all processings were filtered. Note that
several of the panels were acquired in late spring during leafoff conditions, while others were acquired in early summer
leaf-on conditions (Fig. 2). We also note that the off-nadir
view of the county for all panels, in conjunction with the
collection azimuth (Table 2), affected the radiance detected
at the sensor from the combined atmosphere and surface
constituents.
2.2. Planimetric and natural resource data layers
The sole source of reference (i.e., training and validation)
data came from planimetric GIS data sets provided by the
M-NCPPC. The primary data set for tree cover mapping was
a natural features database that contained countywide
polygon data for forested lands, agricultural lands, and
water features. The database was visually interpreted by a
commercial vendor contracted by a multiagency consortium, based on use of 1992 aerial photography. The derived
forest cover map had a large minimum mapping unit to

represent forest areas, rather than individual or small clusters of trees. The agricultural areas contained a wide range
of cover types, including row crops, pasture land, fallow
fields, and grassy regions.
Additional reference data were provided by the planimetric data of the built environment, which included highly
detailed polygons of road and building footprints. This
provided the primary data set used for impervious area
mapping. Because the planimetric and natural features
coverages were developed using photography flown some
10 years ago (1992), changes had occurred in the county up
Table 1
IKONOS spectral band widths specified as full width at half maximum
Band
Spectral wavelength (nm)
1
2
3
4
Pan
445 516
506 595
632 698
757 853
450 900
Blue
Green
Red
Near-infrared
Panchromatic
198
Fig. 2. Location and acquisition dates of IKONOS image panels of the study area. Note the cloud cover remaining in Panel 7 despite a 14-month acquisition
window. (Includes material from Space Imagingn.)
to the time of the IKONOS acquisition. This required

evaluation of the accuracy of the impervious and natural
features coverages, and screening for data sets suitable for
algorithm training and map validation.
3. Methods
3.1. IKONOS image preprocessing
The four multispectral IKONOS bands were radiometrically corrected to at-sensor reflectance utilizing the methods
Table 2
IKONOS acquisition metadata
Panel
Date
Local
time
Sun
elevation
Sun
azimuth
Collection
elevation
Collection
azimuth
0
1
3
4
5
7
5/1/2001
5/1/2001
5/23/2001
4/6/2000
6/26/2000
4/20/2000
10:59
11:00
11:02
10:35
10:53
10:45
62.08
62.16
67.36
51.28
66.98
57.25
143.46
143.80
139.09
139.97
126.80
140.47
78.25
88.49
85.83
76.46
79.06
77.35
24.83
95.46
276.78
119.51
301.54
349.31
outlined by Goetz (1997) and the calibration parameters

provided by Space Imaging in the imagery metadata. These
corrections to top-of-atmosphere reflectance minimize the
errors associated with solar elevation and allow for betweenband comparisons of data values.
Preliminary supervised and unsupervised forest/nonforest classifications were done using the IKONOS imagery.
An unsupervised classification was run using an ISODATA algorithm, specifying 255 clusters, 10 iterations,
and a confidence interval of 0.96. The resultant clusters
were assigned to three categories (forest, nonforest, or
unknown) using manual interpretation of the imagery
itself. Clusters assigned to the unknown class were
located in both the forested and nonforested areas. These
were used as a mask to create a new image, which was
missing the areas that were earlier assigned to either
forest or nonforest. The unsupervised classification was
then run again on the revised image, consisting entirely of
unknown areas, using the same set of clustering parameters. This process was repeated five times, until it was
difficult to identify any new unknown areas. A supervised
maximum likelihood classification (MLC) was also developed using visually interpreted training areas for each
of four land cover types (tree/forest, agriculture, urban/

built, and water). A total of 100 training areas was
identified. Classifications produced using these two
approaches were assessed for accuracy using 1000 randomly selected samples that were visually interpreted to
the appropriate class, of which 100 were visited on the
ground to verify the interpretations.
Based on other work using Landsat Thematic Mapper
(TM) imagery, we suspect that the modest classification
accuracies we found (75 85%) using these classification
approaches with IKONOS imagery were due to a combination of leaf-on/leaf-off imagery, limited image spectral bands and spatial spectral variability within and
between image components, as well as limitations in
the use of hand-delineated training areas or statistically
defined clusters of similar objects. As a result, we
decided to take a different approach to the classification,
focused on the use of a decision tree classifier and a
large number of training samples (described below). We
also derived four additional image variables to aid the
spectral discrimination process including: (i) Normalized
Difference Vegetation Index (NDVI), (ii) Atmospherically
Resistant Vegetation Index (ARVI), (iii) NIR/red (simple
ratio), and (iv) NIR/blue. The NDVI was added to aid in
the discrimination of sediment-laden water from wet
fields. ARVI (Kaufman & Tanre, 1992) was included to
help remove artifacts within the canopy itself, such as
canopy shadow. The simple ratio and NIR/blue ratio were
included to aid the discrimination of agricultural fields
and shrubs from trees.
3.2. Reference data preprocessing
In the case of tree cover mapping and accuracy assessment, natural features reference data screening was done
by overlaying the IKONOS imagery with the reference
coverage and by deleting polygons whose cover type did
not match. The incorrectly attributed polygons included
forested and agricultural areas that had been converted to
residential or commercial (developed) land use, as well as
forested or agricultural areas that had some development
within the polygon to the extent that it could skew the
training and validation data. The water polygons were also
evaluated for accuracy with errors occurring mostly in
farm ponds that had been drained or developed, as well
as portions of water bodies occurring in rapids and over
rock that were subject to changes associated with flow
volume. These screening processes permitted the use of the
natural features coverage to develop independent training
and cross-validation data sets for the decision tree classifier, and subsequent development of a tree cover map
rather than forest cover per se (where forest would imply,
e.g., >60% tree cover).
In order to minimize edge pixels, the forest, grass, and
water polygons were internally buffered by 20 m (five
IKONOS pixels). Areas converted from a natural to a
199
built environment since the creation of the data set were

not added to the training data set used for image
classification since the existing data included enough
samples for the spectral responses of buildings and roads.
The planimetric data were rasterized to 4-m pixels
utilizing the same pixel footprint as the IKONOS imagery, so that there was exact pixel coregistration. The final
reference data set consisted of a coregistered raster file
containing forest, crops/grasses, water, buildings, and
roads.
Following data quality assessment, the reference data sets
were used as the basis for accuracy assessment. Tree cover
was assessed using the natural features data set and impervious cover with the planimetric data. A combination of
random selection and reserved sample sets was used, as well
as independent interpretation of imagery and field visits.
These were derived differently for tree and impervious
cover, as described below.
3.3. Tree cover mapping
Having the natural features data set available for training the classification algorithm allowed for 1.7 million
unique randomly generated samples to be used in a land
cover stratification sampling scheme from the training
data, and retention of a proportional number of samples
in relation to the training data. Of the 1.7 million points,
600,000 were randomly removed from the data set to be
used as mutually exclusive validation points. The number
of training samples allocated to each panel depended on
the respective area of the panel, with the smallest receiving
70,000 and the largest receiving 250,000 samples. The
sample size also varied between forests, agricultural lands,
water, and urban features. Virtually no urban or suburban
trees or grasses were used in the training data because
these managed areas have high vegetation indices that
create statistical confusion with, for example, deciduous
forest spectra.
We used a decision tree classifier, or more specifically
a classification tree algorithm, within the statistical software package S-PLUS, developed by Insightful. This
univariate decision tree algorithm, based on Breiman,
Freidman, Olshend, and Stone (1984), recursively thresholds the training data into increasingly homogeneous
partitions using nonparametric rules. Decision trees have
become popular for land cover mapping (Friedl & Brodley, 1997; Hansen, Dubayah, & DeFries, 1996; McCauley
& Goetz, in press) because the tree output is intuitive,
with each IKONOS band or variable threshold listed for
each successive partition. The decision tree classifier in SPLUS works by evaluating individual sample points, so
the values of the IKONOS pixels for the original bands,
and the derived indices and ratios were used.
Because temporal differences in the IKONOS scene
acquisition dates (Table 2) caused type confusion in early
classifications, leaf-on and leaf-off imagery required the
200
development of more scene-specific decision trees based

on sample points from each individual scene only. To aid
in the development of a robust classification, each of the
six IKONOS panels was classified individually, and the
resulting decision trees were used to classify its respective
IKONOS image by applying the threshold breakpoints
listed in the classification tree to the eight image variables
described above. The panels were then mosaicked back
together for the final map of the entire county. An
accuracy assessment was completed on the final classification using the data points reserved from the initial
sampling of the natural features coverage, without duplication of sample selections.
3.4. Impervious area mapping
The impervious surface mapping was also done using a
classification tree approach, but made use of the planimetric
map of buildings and roads rather than the natural features
database. Upon inspection, it was apparent that the building
features had few errors but the transportation features
contained numerous errors of omission, particularly parking
lots. In order to evaluate the completeness of the data, the
county was divided into 689 blocks, each of which consists
of many thousands of IKONOS pixels. Each block was
checked for errors against recent DOQs and the imagery
itself. Areas developed since the creation of the planimetric
data were not included in the training data set. Of the 689
original blocks, 197 were deemed sufficient for algorithm
development and validation-based tests of spectral discrimination with sample size. Approximately 600,000 data
samples (4-m pixels) resulted.
Samples within each block were coded either impervious or nonimpervious based on interpretation of the planimetric data. The sample was then divided into training and
validation data sets based on each swath of IKONOS
imagery, resulting in about one-fourth (140,366) being
reserved for validation. This site-based approach reduced
the influence of spatial autocorrelation and artificially high
validation statistics. Additional training data were derived
for nonimpervious features such as water bodies and wetlands based on the natural features coverage described
above.
Predictor variables were chosen based on an analysis of
impervious feature signatures and their vegetation indices.
As with the tree cover mapping, the set of IKONOS spectral
variables used for prediction consisted of the four optical
bands, the NDVI, and the simple ratio (NIR/red). Two
additional spectral ratios that showed promise in discriminating open soil plots from impervious features were also
used (NIR/blue and NIR/green). The spectral information
and impervious class for each of these data sets were used to
grow impervious surface classification trees. The resulting
tree-based algorithms were then applied to the corresponding
IKONOS image tile, and thematic accuracy statistics were
calculated.
3.5. Stream health analyses

Statistical analyses were performed on the impervious
and tree cover maps to determine whether there were
statistically significant differences between small watersheds (approximately equivalent to hydrological unit code
HUC 14), as rated by water quality experts into one of four
possible stream resource condition categories. These were
based on countywide baseline stream monitoring data,
including pH, dissolved oxygen, and temperature, among
others (Van Ness, Brown, Haddaway, Marshall, & Jordahl,
1997). This multiagency effort also assessed a number of
biological indicators based on fish and benthic macroinvertebrate surveys, and were used to define an index of
biological integrity (IBI) (Stribling, Jessup, White, Boward,
& Hurd, 1998). Of the 296 small watersheds in the county,
245 had been analyzed and represented by a categorical
stream resource quality rating of excellent, good, fair, or
poor. We refer to these as stream health ratings. About onethird of these (93) was indicated as preliminary assessments, which were analyzed as separate categories from the
completed assessments.
We examined the amount of tree cover within the
entire area of each watershed, as well as within 100-ft
riparian buffers (a size determined by independent buffer
monitoring and restoration activities) to explore whether
there were links to the stream health ratings. The stream
buffer was calculated using a hydrology vector layer
provided by the county. The amount of impervious
surface area within each watershed was also calculated
and expressed, as with tree cover, as a proportion of total
watershed area.
One-way analysis of variance (ANOVA) was used to
assess the relation of each variable to the stream health
ratings. Logistic regression was also used to explore the
relationship of stream health with the mapped variables
(i.e., impervious cover, watershed tree cover, and riparian
buffer tree cover). Logistic regression is a nonlinear and
nonparametric technique that uses maximum likelihood
estimation to predict the odds probability of categorical
dependent variables (in this case stream health rating).
This is unlike ordinary least squares regression estimation
of changes in a numerical dependent variable. The technique also has the advantage of not assuming normal
distributions or homoskedastic independent variables
(Menard, 1995). It is often used to test binary variables,
Table 3
Comparison of county area classified as tree or nontree using supervised
and unsupervised approaches
Classification type
Type
Percent of area
Unsupervised
Tree
Nontree
Tree
Nontree
44.6
55.4
49.7
50.3
Supervised

Table 4
Accuracy assessment of supervised and unsupervised classifications
Classification
Overall
accuracy
j statistic
Z statistic
Supervised
Unsupervised
0.86
0.83
0.72
0.67
15.7
13.3
but also can be applied to multinomial dependents with

more classes, as in our case with ordinal ratings of stream
health.
4. Results
4.1. Tree cover
The extent of the tree cover classifications derived using
the supervised and unsupervised approaches is summarized
in Table 3. The unsupervised classification provided an
estimate of f 5% (67 km2) more forest than the supervised
classification. Accuracy statistics generated for the initial
201
supervised and unsupervised classifications (Table 4) indicate that both classifications were significantly better than
random, but the differences between the classified maps,
while not statistically significant ( p = 0.07), were substantial enough to convince us to take a different approach (i.e.,
decision trees trained with a large sample of reference
data).
The tree cover map derived from the decision tree
approach (Fig. 3) was spatially detailed and boundaries
between scenes are virtually nonexistent, despite the range
of dates in the IKONOS acquisitions in both years and
seasons. This was due, in large part, to the accuracy of the
decision tree algorithm and the abundance of training data.
Trees occupied nearly 38% of the county (Table 5). Areas
obscured by cloud or cloud shadow, which could not be
correctly classified, comprised 7% of the total county area.
The original forest classification done ca. 1992, and
subsequently manipulated for use as training data for our
mapping, depicted just 28% of the county as forested
(36,849 ha). The primary reason for the difference was the
depiction of a forest/nonforest classification in the natural
Fig. 3. IKONOS tree cover map of Montgomery County, MD. Note the forest cover data (from natural features GIS coverage; Fig. 4) used in the areas obscured
by clouds.
202
Table 5
Classification of county using decision tree classifier
Classified type
Area (ha)
Percent of county
Tree
Nontree
Cloud
Total
49,522
72,268
9478
131,268
37.7
55.1
7.2
100
features layer, whereas use of IKONOS permitted us to

capture, in many cases, individual trees. A subset of the
IKONOS tree cover map, along with the reference data
(natural features coverage), shows differences between the
two mapping approaches (Fig. 4), where tree cover is
shown in lighter green compared to the reference forest

coverage (darker green). Note the missing trees in the
subdivisions and between the links of the golf course in
the lower left. It would be possible, if desired, to filter the
tree cover map in such a way as to approximate the forest
cover map (e.g., by retaining only those areas with at least
five of the surrounding eight pixels in a 3 3 window
classified as tree).
Accuracy statistics based on the independent 600,000
reference data samples reserved for validation show very
high map accuracies, with omission errors of 2.1% and
commission errors of 4.1% (Table 6). Overall classification
accuracy was 97.3% (j = 0.95).
Fig. 4. IKONOS image detail with classification results. (a) IKONOS imagery; (b) 1992 natural features coverage of forest areas; (c) tree cover derived from
imagery using a decision tree classifier; and (d) IKONOS image with natural features coverage and derived tree cover maps overlaid. Note the small patches of
trees in (c) compared to the forest cover in (b) and the substantial amount of tree cover missing from the forest area coverage (d). (Includes material from Space
Imagingn.)

Table 6
Accuracy assessment of tree cover map
Tree
Nontree
Validation
sample
(natural
features)
Classified
(IKONOS)
Number
correct
Producers
accuracy
Users
accuracy
212,703
287,297
217,224
282,577
208,351
278,236
98.0
96.9
95.9
98.5
4.2. Impervious surfaces

The impervious surface map showed a level of detail
sufficient to map individual houses within subdivisions,
as well as clearly depicting areas where residential
development had occurred since the original planimetric
coverage was produced (Fig. 5). This map of impervious
areas depicts a wide range of materials, some of which
have very different spectral properties (e.g., pavement,
concrete, roof tiles, etc.). Other mapping methods based
on the use of regression trees, rather than classification
trees, permit estimation of imperviousness as a continuous variable (i.e., as a percentage between 0 and 100).
We have used the planimetric reference data to train an
algorithm of continuous subpixel impervious maps based
203
on Landsat imagery (Smith et al., fourthcoming), but the

IKONOS impervious map could be used in place of
planimetric data.
Internal accuracy assessment checks are provided by
the decision tree software, including estimates of misclassification rates for each intermediate and terminal node,
as well as an overall algorithm assessment. Additional
checks of misclassification error were done by crossvalidating the algorithm using 10% of the data samples
in an iterative procedure, cycling through the full data set
(after Friedl, Brodley, & Strahler, 1999). Analysis of
these accuracy measurements for various tree-growing
scenarios strengthened the final algorithm selection and
the overall classification accuracies. Independent statistical validation using the 140,366 samples reserved from
the image sampling approach demonstrated the ability of
the algorithm to discriminate impervious from nonimpervious component materials in the imagery (Table 7).
Omission errors in the impervious map were 9.4% and
commission errors were 10.8%. Overall classification
accuracy was 84.2% (j = 0.36). Despite our efforts to
screen for mislabeled and omitted features (e.g., driveways and sidewalks) in the planimetric data used for
training, some remained present in both the training and
Fig. 5. Impervious surface areas (yellow) for a portion of Montgomery County derived from IKONOS imagery (right) compared to planimetric coverage
derived from aerial photographs (left), showing changes in residential development since the area was initially mapped (ca. 1993). IKONOS tree cover (green)
is also shown, with the IKONOS panchromatic image as a backdrop. (Includes material from Space Imagingn.)
204
Table 7
Accuracy assessment of impervious surfaces map
4.3. Riparian buffers and stream health
Validation
Classified Number Producers Users
sample
(IKONOS) correct accuracy accuracy
(planimetric)
Impervious
108,730
Nonimpervious 31,636
110,466
29,900
98,513
19,683
90.6
62.2
89.2
65.8
validation data samples, which produced somewhat lower

accuracy statistics than expected. We believe, based in
part on results such as those shown in Fig. 5, that the
accuracy of the map is actually higher than that suggested by the interpretation of planimetric data.
Average tree cover within riparian buffers ranged from

as little as 6% to over 98% by watershed, impervious cover
within the watersheds ranged from 0% to 43%, and total
tree cover (both inside and outside buffers) ranged between
6% and 94%. This range of observations was more than
sufficient to detect the sensitivity of the stream health to
buffer quality. Differences between watersheds were also
visually evident in map comparisons for these same
variables (Fig. 6), as well as in the statistics derived for
each rating category (Table 8 and Fig. 7). ANOVA showed
significant differences between watersheds with different
Fig. 6. Maps of small watershed aggregated values for (a) stream health rankings, (b) impervious surface cover (%), (c) tree cover (%), and (d) riparian buffer
tree cover (%). (b) (d) were calculated using the IKONOS-derived maps of these land cover variables.
amounts of impervious surface area, tree cover, and riparian tree buffers. Variance ratios ( F statistics) were 42.1,
16.3, and 15.0, respectively, all significant at p < 0.001
(N = 245). Streams with excellent ratings had significantly
lower impervious cover and higher tree cover, both within
the watershed and riparian buffers, than streams rated
good, fair, or poor. This trend continued progressively
through each rating category, with increasing average
impervious area and decreasing tree cover as stream health
decreased. All pairwise comparisons confirmed statistically
significant differences ( p < 0.001) between stream rating
categories. There were no substantial differences between
those streams with preliminary assessments and those with
completed assessments, except in the case of those rated
excellent (n = 32) versus preliminary excellent (n = 6), but
this was likely a result of the small sample size of the
latter.
The goodness of fit test of Hosmer and Lemeshow
(1989) failed to reject the null hypothesis of no difference
between the observed and model-predicted values of stream
health; thus, the multivariate model was an adequate predictor of stream health rating. There is no varianceexplained statistic in logistic regression when the dependent
variable is categorical, but reduction in residual deviance
from the null model was highly significant (v2 = 33.1,
p < 0.001). There was, however, multicollinearity among
variables in the regression. Across watersheds, the extent of
tree cover both within and outside of the 100-ft buffer was
positively correlated with each other (r = 0.71), so the
properties of the 100-ft buffer were not independent of
the surrounding land cover. Impervious area and total tree
cover were inversely correlated (r = 0.55), as were impervious area and tree buffers (r = 0.43), but these relationships were reduced in agricultural areas, which had both
low tree and impervious cover. As implied by the ANOVA,
the individual independent variables in the logistic regression were each statistically significant, and stepwise inclusion tests revealed that impervious cover was the primary
predictive variable, followed by tree cover within the
buffers, and then total tree cover. Residual deviance was
significantly reduced when riparian buffers were added to a
model based on impervious surfaces alone, but subsequent
addition of total watershed tree cover did not further
contribute to model significance. The results suggest that
guidelines for excellent stream health rating would be no
more than 6% impervious with at least 65% forested
Table 8
Small watershed sample size and average statistics by stream health rating
category
Stream health
rating
Area
(km2)
Impervious
(%)
Tree
cover (%)
Buffered
(%)
Excellent
Good
Fair
Poor
38
81
76
50
272
658
451
356
3.6
4.9
13.9
19.5
50.6
44.6
37.0
29.6
76.8
71.3
63.2
56.3
205
Fig. 7. Stream health rankings in relation to (a) impervious surface cover,

(b) watershed tree cover, and (c) riparian buffer tree cover, each derived
from the IKONOS image data.
buffers, and no more than 10% impervious with at least

60% buffered for a rating of good. These values would
clearly be affected by the landscape configuration of buffers
in relation to impervious areas, and by mitigation measures
like sediment retention ponds.
5. Discussion
5.1. Mapping applications
Land cover classification in the study region using
IKONOS imagery with traditional supervised and unsupervised approaches produced substantially different maps
and, for example, tree cover estimates. There were several
factors that impacted the classifications, including the
timing of data acquisition (thus phenological stage),
associated viewing and atmospheric conditions, the quality of training data (in the case of supervised classification), and insufficient discrimination resulting from
limited IKONOS spectral bands. Some of the more
pronounced classification errors were between wet agricultural fields, dense forests, and shadows. For example,
spring small grains (wheat, barley, and rye) and hay
(alfalfa and grasses) had emerged in agricultural fields,
which produced a similar spectral response as deciduous
trees in partial leaf flush. In other areas, the visibility of
the forest floor through the tree canopy produced spectral
responses similar to residential areas. Shadows were also
an especially important consideration with the use of this
high spatial resolution imagery, particularly shadowing
within the forest canopy from adjacent trees. Some of
these issues can be reduced through the use of multitemporal imagery, but this would be logistically difficult
206
to accomplish in many areas, and cost-prohibitive for

most applications using IKONOS.
In order to reduce the impact of some of these issues,
the decision tree classification approach we adopted made
use of a large number of reference (training) data for the
study area. The resulting classification reinforced the
considerations noted above regarding the use of IKONOS
imagery for vegetation mapping. Two classification issues
were especially problematic, each dependent on the timing
of scene acquisition. In the early spring images (Panels 4
and 7) (see Table 1 and Fig. 1), the primary source of
classification error was confusion between mature deciduous forests and bare ground areas such as untilled agricultural or partially vegetated abandoned areas. In these
scenes, mature forests had just begun to leaf out, and
undergrowth, bare ground, and litter beneath the trees
were spectrally similar to agricultural fields with crop
residue still in place from no-till farming practices. In
the spring and early summer images (Panels 1 3 and 5),
the primary confusion occurred between active agricultural
fields and deciduous trees. Despite these issues, we were
able to achieve accurate maps of tree and impervious
cover.
Although the above points are not unique to IKONOS
imagery, they are accentuated by the high spatial resolution relative to other imaging sensors such as Landsat
TM. In the case of the early season deciduous forest, the
4-m IKONOS resolution included data in the intercanopy
space, whereas with TM, the pixels include a mixture of
canopy and higher biomass woody stems, generally
producing higher NDVI values. In the case of the leafon scenes, the high vegetation index values of actively
growing agricultural fields appeared spectrally similar to
that of the dense deciduous tree canopy. This was partly
due to greater spatial variability in IKONOS scene
elements. For example, a single tree may be contained
within one TM pixel, whereas the same tree may be
represented by 5 or 10 IKONOS pixels, resulting in
greater variation with foliage density and canopy shape
(see also Blonski, 2001; Goward, Davis, Fleming, Miller,
& Townshend, this issue). Inclusion of the ARVI image
helped reduce this effect, but cannot completely resolve
this unique aspect of high spatial resolution imagery.
Clearly, this issue could be considered an advantage
for some applications (e.g., assessments of defoliation or
tree health). There have also been suggestions that highresolution imagery could provide information capable of
discriminating different habitat types and complex forest
associations (e.g., oak/hickory versus beech/maple), or
permit mapping of invasive species such as Kudzu. Based
on our results with IKONOS, we believe that these
applications would be difficult to accomplish within
reasonable confidence limits.
The capability of IKONOS to point in a desired
direction adds to the flexibility in acquisition strategies,
and also has implications for the spectral properties of
image components. The relationship between sun, surface,

and sensor for our image acquisitions (Table 1) shows that
Panels 0 and 1 were acquired on the same orbit, but the
sensor azimuth angles differed. As a result, many of the
water bodies in Panel 0 show the effect of sun glint,
whereas those in Panel 1 do not. Similar, albeit less
dramatic, conditions occur in areas that are mostly vegetation or impervious cover types. This type of information
can be useful for analyses of bidirection reflectance properties, but complicates image classification for many applications, particularly where many IKONOS tiles or
overlapping image panels are required to completely map
a study area.
The work reported here also permitted algorithm development for application to other IKONOS imagery in the
region, and significantly allowed for the development of
subpixel mapping techniques using Landsat imagery,
which we will report on in future publications (e.g., Smith
et al., fourthcoming).
5.2. Stream health application
Riparian buffers have been recognized as important
landscape features that provide unique habitat for many
wildlife species (Iverson, Szafoni, Baum, & Cook, 2001),
as well as filtering capabilities for removing nutrient
pollutants from agricultural runoff before they reach waterways (Cooper, Gilliam, Daniels, & Robarge, 1987; Correll, 1997; Lowrance et al., 1997). Traditional approaches
to mapping and monitoring riparian zone vegetation have
relied on photographic interpretation (e.g., Lonard, Judd,
& Desai, 2000), but this is not practical over large areas.
For example, the Chesapeake Bay Program has been
tasked with establishing 2000 miles of forested riparian
buffers by the year 2010, but does not have a practical
methodology to accurately assess current buffer statistics,
let alone monitor new plantings. Landsat data are simply
not of sufficient spatial resolution to adequately map
riparian buffer vegetation within the widely accepted
100-ft ( f 30 m) buffer width used as a common reference
for buffer effectiveness.
Our analyses of impervious surfaces and tree cover
within the small watersheds and riparian buffer zones of
Montgomery County provide an example of the capabilities conveyed by consistent mapping over large areas
using IKONOS imagery. A key advantage to the IKONOS-derived maps was the fine spatial resolution that
allowed very-local-scale analysis of riparian buffers, and
statistically meaningful sample populations within the
small watersheds. The results confirm that there are clear
linkages between the land cover within a watershed and
the stream water quality. Specifically, the amount of
impervious surface area within a watershed and tree cover
within riparian buffer zones provided robust indicators of
stream health rating. This suggests that the buffers were
functioning to reduce pollutants and sediments from built
areas before they entered the waterways. Conversely,

buffers alone were insufficient for protecting waterways
in highly impervious areas or where forest cover outside
buffer zones was low, including agricultural areas that
presumably affected stream health through chemical runoff. Additional analyses beyond the scope of this paper
are being completed to explore these relationships further,
and to assess the effectiveness of mitigation approaches in
controlled and uncontrolled paired watersheds.
The high-resolution land cover maps that can be derived
from IKONOS imagery and the associated landscape
variables that can be calculated from them (e.g., buffer
configuration) thus provide an important contribution to
these types of resource management applications. In the
case of Montgomery County, these statistics are actively
used to determine whether resources are focused on
watershed preservation (streams of excellent health), protection (good or excellent health), or restoration (fair or
poor health).
207
protection and restoration management decision making.

Other applications for which the maps are currently being
utilized include resource lands monitoring and acquisition
planning, analyses of habitat connectedness and bird habitat
suitability, and hydrodynamic modeling for mitigation of
stream bank erosion. IKONOS imagery can augment traditional mapping approaches and, in many cases, may provide
a cost-effective alternative.
Acknowledgements
This work was funded by NASAs Earth Science
Applications Division (grants NAG1302010 and NAG1399011). We acknowledge Montgomery County MNCPPC (Mary Dolan and Nazir Baig) and Montgomery
DEP (Cameron Wiegand) for data sets and collaborative
assistance. IKONOS image data were acquired by Space
Imaging and provided by the NASA Scientific Data
Purchase program.
6. Conclusions
Utilizing high spatial resolution imagery can be beneficial to many different resource management applications.
Imagery like IKONOS can aid in the development of a wide
range of mapping and spatial modeling applications, but a
number of issues must be considered with these relatively
new and unique data sets. The issues include (i) programmatic considerations of the timing of acquisitions to capture
features of interest (e.g., leaf-off for impervious versus leafon for tree cover), as well as timely data acquisition (ours
took 18 months to acquire); (ii) technical considerations,
such as compensation for the effects of object shadowing
resulting from improved resolution of individual scene
elements, and limited spectral resolution and range; and
(iii) economic factors, such as cost per square kilometer
relative to DOQs, and feasibility of consistent repeat
acquisitions. For example, we noted an inability of IKONOS to adequately spectrally discriminate some land cover
types due to high spatial variability within scene elements
resulting from variable illumination and viewing conditions.
Spectral variability within scene objects also contributed to
reduced class type discrimination between more generalized
land cover types (e.g., confusion between deciduous forest
and some agricultural crops).
New sets of interpretation strategies need to be developed to maximize the information obtained from IKONOS,
while minimizing the problematic issues specific to high
spatial resolution imagery. Our exploration of several such
approaches relevant to resource mapping applications suggests the great practical utility of IKONOS imagery, particularly for impervious surface, tree cover, and riparian
buffers, all of which are related to stream health. Our results
provided very specific guidelines for predicting stream
health ratings, which allows for targeted and adaptive
References
Birk, R.J., Stanley, T., & Snyder, G.L. (2003). U.S. Government commercial
data purchase programs for use in government-sponsored science and
applications activities. Remote Sensing of Environment, 88, 3 16.
(doi:10.1016/S0034-4257(03)00227-X)
Blonski, S. (2001). IKONOS-based simulations of Landsat 7 VNIR data:
Comparison with actual, coincident images. In Proceedings of the 2001
High Spatial Resolution Commercial Imagery Workshop, Greenbelt,
MD, USA, March 19 21 (sponsored by NASA/NIMA/USGS Joint
Agency Commercial Imagery Evaluation Team) (CD-ROM).
Breiman, L., Freidman, J., Olshend, R., & Stone, C. (1984). Classification
and regression trees. Monterey, CA: Wadsworth.
Cooper, J. R., Gilliam, J. W., Daniels, R. D., & Robarge, W. P. (1987).
Riparian areas as filters for agricultural sediment. Soil Science Society of
America Journal, 51, 416 420.
Correll, D. L. (1997). Buffer zones and water quality protection: General
principles. In N. E. Haycock, T. P. Burt, K. W. T. Goulding, & G. Pinay
(Eds.), Buffer zones: their processes and potential in water protection
( pp. 7 20). Harpenden, UK: Quest Environmental.
Dial, G.F., Bowen, H., Gerlach, B., Grodecki, J., & Oleszczuk, R. (2003).
IKONOS satellite, sensor, imagery, and products. Remote Sensing of
Environment, 88, 23 36. (doi:10.1016/S0034-4257(03)00229-3)
Fisher, J., & Goetz, S. J. (2001). Considerations in the use of high spatial
resolution imagery: An applications research assessment. In American
Society for Photogrammetry and Remote Sensing (ASPRS) Conference
Proceedings, St. Louis, MO (p. 8), Available from ASPRS http://
www.asprs.org and at http://www.geog.umd.edu/resac.
Friedl, M. A., & Brodley, C. E. (1997). Decision tree classification of land
cover from remotely sensed data. Remote Sensing of Environment, 61(3),
399 409.
Friedl, M., Brodley, C., & Strahler, A. (1999). Maximizing land cover
classification accuracies produced by decision trees at continental to
global scales. IEEE Transactions on Geoscience and Remote Sensing,
37(2), 969 981.
Goetz, S. J. (1997). Multi-sensor analysis of NDVI, surface temperature
and biophysical variables at a mixed grassland site. International Journal of Remote Sensing, 18(1), 71 94.
Goward, S.N, Davis, P.E., Fleming, D., Miller, L., & Townshend, J.R.G.
208
measurements for Selected Earth Observation System (EOS) Validation

Sites. Remote Sensing of Environment, 88, 79 98. (doi:10.1016/j.rse.
2003.06.007)
Hansen, M., Dubayah, R., & DeFries, R. S. (1996). Classification trees: An
alternative to traditional land cover classifiers. International Journal of
Remote Sensing, 17(5), 1075 1081.
Helder, D., Coan, M., Patrick, K., & Gaska, P. (2003). IKONOS geometric
characterization. Remote Sensing of Environment, 88, 68 78.
(doi:10.1016/S0034-4257(03)00232-3)
Hosmer, D. W., & Lemeshow, S. (1989). Applied logistic regression. New
York: Wiley.
Iverson, L. R., Szafoni, D. L., Baum, S. E., & Cook, E. A. (2001). A
riparian wildlife habitat evaluation scheme developed using GIS. Environmental Management, 28(5), 639 654.
Jantz, C. A., Goetz, S. J., & Shelley, M. A. (2003). Using the SLEUTH
urban growth model to simulate the land use impacts of policy scenarios
in the Baltimore Washington metropolitan region. Environment and
Planning, B (in press).
Kaufman, Y. J., & Tanre, D. (1992). Atmospherically Resistant Vegetation
Index (ARVI) for EOS-MODIS. IEEE Transactions on Geoscience and
Remote Sensing, 30, 260 271.
Lonard, R. I., Judd, F. W., & Desai, M. D. (2000). Evaluation of colorinfrared photography for distinguishing annual changes in riparian forest vegetation of the lower Rio Grande in Texas. Forest Ecology and
Management, 128(1/2), 75 82.
Lowrance, R., Altier, L. S., Newbold, J. D., Schnabel, R. R., Groffman,
P. M., Denver, J. M., et al. (1997). Water quality functions of riparian forest buffer systems in the Chesapeake Bay watershed. Environmental Management, 21, 687 712.
McCauley, S., & Goetz, S. J. (2003). Mapping residential density patterns
using multi-temporal Landsat imagery and a decision-tree classifier.
International Journal of Remote Sensing (in press).
Menard, S. (1995). Applied logistic regression analysis. Quantitative Applications in the Social Sciences, 106. Thousand Oaks, CA: Sage Publications.
M-NCPPC (2000). Legacy Open Space functional master plan, Montgomery County Department of Park and Planning, Silver Spring,
MD, 80 pp. plus appendices and maps (http://www.mc-mncppc.org/

legacy_open_space).
Sawaya, K., Olmanson, L., Holden, G., Sieracki, J., Heinert, N., & Bauer,
M. (2003). Extending satellite remote sensing to local scales: Land and
water resource management using high resolution imagery. Remote
Sensing of Environment, 88, 143 155. (doi:10.1016/j.rse.2003.04.006)
Smith, A. J., Goetz, S. J., Prince, S. D., Wright, R., Melchoir, B., Mazzacato, E. M., & Jantz, C. (fourthcoming). Estimation of sub-pixel impervious surface area using a decision tree approach, IKONOS and Landsat
imagery. Remote Sensing of Environment.
Stoney, W. (2001). The past, present, and future of the medium- and highresolution satellite world. In: J.C. Baker, K.M. OConnell, & R.A.
Williamson (Eds.), Commercial observation satellites: At the leading
edge of global transparency ( pp. 652 661). Santa Monica, California:
RAND Corporation and the American Society for Photogrammetry and
Remote Sensing. (Available online at http://www.rand.org/publications/
MR/MR1229/)
Stribling, J. B., Jessup, B. K., White, J. S., Boward, D., & Hurd, M. (1998).
Development of a benthic index of biotic integrity for Maryland
streams. Annapolis: Montgomery Department of Natural Resources,
62 pp.
Tanaka, S., & Toshiro, S. (2001). A new frontier of remote sensing from
IKONOS images. International Journal of Remote Sensing, 22(1),
1 5.
Van Ness, K., Brown, K., Haddaway, M., Marshall, D., & Jordahl, D.
(1997). Montgomery County water quality monitoring program: Stream
monitoring protocols. Watershed Management Division, Department of
Environmental Protection, Silver Spring, Maryland.
Varlyguin, D., Wright, R., Goetz, S. J., & Prince, S. D. (2001). Advances in land cover classification for applications research: A case
study from the mid-Atlantic RESAC. American Society for Photogrammetry and Remote Sensing (ASPRS) Conference Proceedings,
St. Louis, MO . (p. 7).
Weber, T., & Wolf, J. (2000). Marylands green infrastructureusing landscape assessment tools to identify a regional conservation strategy.
Environmental Monitoring and Assessment, 63, 265 277.

Acquisition of earth science remote sensing observations from commercial

sources: lessons learned from the Space Imaging IKONOS example
Samuel N. Goward a,*, John R.G. Townshend a, Vicki Zanoni b, Fritz Policelli b, Tom Stanley b,
Robert Ryan c, Kara Holekamp c, Lauren Underwood c, Mary Pagnutti c, Rose Fletcher c
a
Department of Geography, 2181 LeFrak Hall, University of Maryland, College Park, MD 20242, USA
Earth Science Applications Division, NASA Stennis Space Center, Stennis Space Center, MS 39529, USA
c
Remote Sensing Directorate, Lockheed Martin Space Operations-Stennis Programs, John C. Stennis Space Center, MS 39529 USA
b
Received 7 February 2003; received in revised form 28 April 2003; accepted 10 June 2003
Abstract
Over the last decade, NASA and other federal agencies have been increasingly encouraged to explore commercial sources of land remote
sensing data rather than pursuing government-funded sources of these measurements (prior to U.S. Commercial Executive, 2003. U.S.
Commercial Remote Sensing Policy. White House, Washington, DC). The science and applications users have been skeptical of this move to
commercial sources of observations both because of uncertainty over the capabilities of the private sector and the unsatisfactory outcomes of
previous efforts to commercialize satellite-acquired land remote sensing observations. In an effort to more fully explore the potential of
commercial remotely sensed land data sources, the NASA Earth Science Enterprise (ESE) implemented an experimental Scientific Data
Purchase (SDP) that solicited bids from the private sector to meet ESE-user data needs. The SDP activity spanned nearly 5 years and supplied
many US and international researchers with sources of land remote sensing observations that had not been previously available. The images
from the Space Imaging IKONOS system provided a particularly good match to the current ESE missions such as Terra and Landsat 7 and
therefore serve as focal point in this analysis.
Throughout the SDP process, there have been many lessons learned concerning interactions between US industry, government agencies
and the science user community. The specifics of the Space Imaging IKONOS experience under the NASA SDP are most valuable with
respect to possible future uses of commercial vendors to supply NASA ESE user needs for space-acquired land observations. Areas where
valuable lessons were learned included the technical, scientific, proprietary and management aspects of the interactions. As this activity has
evolved, user confidence in the technical and scientific qualities of the IKONOS measurements has increased substantially. There are still
areas where further progress could be achieved, with respect to proprietary and management aspects of scientific commercial data buys.
To date, the NASA scientific and applications users who have examined the IKONOS imagery have found the data to be of high quality,
providing substantial value to their specific pursuits. They have found that the novel attributes of IKONOS, particularly in the spatiotemporal
domain have introduced new analysis challenges not previously experienced with EOS sensors such as Landsat and MODIS. The technical
qualities of the observations have been substantially improved during the SDP activity as a result of independent validation and verification
by the Joint Agency Commercial Imagery Evaluation (JACIE) of the IKONOS observations.
The experience gained from the Space Imaging IKONOS SDP activity, suggests that US private sector is technically capable of meeting
the needs of NASA ESE science and application users. The future success of such interactions between industry, government and users
appears to be far more dependent on the organizational and legal aspects of such arrangements than technical capabilities of the data
providers.
1. Introduction
The NASA Scientific Data Purchase (SDP) experiment
to acquire scientific-quality satellite-acquired remotely
E-mail address: sgoward@umd.edu (S.N. Goward).
doi:10.1016/j.rse.2003.06.007
sensed earth observations from private industry suppliers,

conducted under the NASA Scientific Data Purchase (SDP),
has provided valuable and important lessons, which will
prove useful for the potential future success of commercial
remote sensing systems in satisfying the needs of US
terrestrial scientists.
Of the private industry vendors selected for the SDP
(Birk, Stanley, & Snyder, 2003), Space Imaging presents the
210
Fig. 1. The spectral band passes of IKONOS versus Landsat 7. Note the differences particularly in the near infrared. This leads to a divergence in spectral
measurements from the two systems (see Goward et al., this issue for further details).
best case study for describing how well US industry may be

able to fulfill the role, historically filled by US government
laboratories (particularly the NASA Goddard Space Flight
Center), of supplying space-acquired remotely sensed land
measurements. The experience with the Space Imaging
IKONOS system, launched successfully in September
1999, now encompasses over 4 years and provides a useful
foundation to consider what processes have worked, and
which might be improved.1
the Landsat Thematic Mapper (TM), the Earth Observing

System (EOS), Advanced Spaceborne Thermal Emission
and Reflection Radiometer (ASTER) and the French HRVSPOT systems (Fig. 1, Table 1). The red and near infrared
measurements are also comparable to red and near infrared
spectral measurements from the Advanced Very High Resolution Radiometer (AVHRR), the EOS Moderate resolution
Imaging Spectrometer (MODIS) as well as the French
VEGETATION instrument as well as several other sensors
(see Table 1 in (Townshend & Justice, 2002))
2. The space imaging IKONOS mission
Details concerning the specifics of the Space Imaging

IKONOS observatory are covered elsewhere in the issue
(Dial, Bowen, Gerlach, Grodecki, & Oleszczuk, 2003).
From the perspective of the NASA Earth Science Enterprise
(ESE) community, the following are considered interesting
characteristics of the IKONOS observatory:
The four multispectral bands closely approximate three
of the visible and the near infrared spectral measurements of
1
It is interesting to note that a portion of Space Imagings heritage
originates in EOSAT, the corporation that was created to commercially
operate Landsats 4 & 5, as well as build, launch and operate Landsat 6.
For many reasons, this earlier Landsat commercialization was not
successful, and the Landsat mission was returned to government management in 1992 (Sheffer, 1994).
The pixel area of the multispectral measurements is more

than a factor 50 improvement over Landsat 7 (16 m2
versus 900 m2), and the 1 m panchromatic band is over
200 times better than the 15-m pan band on Landsat 7.
This substantial increase in spatial resolution helps
provide the types of detailed spatial measurements
needed to assess scaling issues encountered when
Table 1
Spectral band passes for IKONOS and Lansat-7 visible and near-infrared
bands
Spectral band
IKONOS spectral range
Lansat-7 spectral range
Blue
Green
Red
NIR
445 516
506 595
632 698
757 853
450 515
525 605
630 690
750 900
nm
nm
nm
nm
nm
nm
nm
nm
moving from local to global-scale observatories (Woodcock & Strahler, 1987) (Fig. 2).
The radiometric precision of the IKONOS sensor, with its
11-bit analog to digital data conversion, is a factor of 8
improvement compared to the 8-bit radiometry for
Landsat 7s ETM+, providing a comparable improvement
in radiometric measurement precision that IKONOS
supplies spatially. Potentially, this combination of radiometric precision and spatial detail could provide the high
quality means needed to validate the lower resolution
measurements from Landsat and other systems.
The orbital equatorial crossing time is also nearly the
same as Landsat, Terra, EO-1 and SPOT, although the
IKONOS systems nadir, repeat frequency is much lower
at ~ 140 days, a necessary trade-off for the increased
spatial resolution of the IKONOS system. With platform
agility it can revisit most locations with 3 days within
F 30j from nadir (Dial et al., 2003).
As a result, IKONOS supplies an important step in
multiscale land remote sensing, one that is most closely
associated with field measurements and ground truth.
The similarities in spectral measurements and orbit that
exist between IKONOS and other systems, combined with
the increased radiometric precision and spatial resolution of
the IKONOS satellite, make it a valuable tool in comparing
and evaluating higher and lower resolution systems. Thus,
the IKONOS images offer significant potential in providing
the validation and local assessments needed to support
regional and global scale studies of the earth system, using
Landsat, SPOT, ASTER, MODIS, AVHRR and VEGETATION as the primary observatories.
211
3. Science community perspective on commercially

acquired scientific measurements
The significance of the IKONOS data purchase for the
U.S. science community can best be appreciated from the
context of their previous experiences with the U.S. (foreign)
private sector U.S. scientists have noted several aspects of
commercial observations that might limit their value for
scientific query (National Research Council, 2001, 2002).
These can be broadly categorized as follows:

Technical: The research community believes that the

commercial remote sensing sector is targeting markets
where the visual quality of the imagery is valued more
highly than the spectroradiometric precision of the pixel
level measurements. Experience with sensors such as the
Landsat Return Beam Vidicon (RBV) and the NOAA
Advanced Very High Resolution Radiometer have
demonstrated how limited observatory quality can impact
the scientific quality of remotely sensed measurements
(Teillet & Holben, submitted for publication). Because
scientists and engineers consider commercial systems
black boxes containing many unknowns and therefore
of uncertain value, they tend to be skeptical of using
measurements from such systems.
Scientific: Each new observatory brings new capabilities
and new challenges. How best to exploit such systems
requires a significant investment in time and resources.
However, researchers must continue to make successful
progress on their current research goals, if they are to
have any hope of winning renewed funding for their
research. Consequently, it is frequently difficult for
Fig. 2. Comparison of IKONOS (left); acquisition date: 04/02/2001, image upper left corner latitude/longitude: 45.45/12.38, and Landsat 7 (right); acquisition
date: 08/26/2001, image upper left corner latitude/longitude: 47.0/11.12 observations for the city of Venice, Italy. The differing spatial resolution of the two
sensors is clear. The trade off is that Landsat systematically monitors all land areas of the earth seasonally whereas the IKONOS system can only sample small
portions of the earths land areas each year.
212
research teams to dedicate the effort needed to develop

applications of new observatories without targeted
support for these efforts.
Proprietary: Commercial providers often reluctant or
even unable to reveal details concerning sensor design,
acquisition procedures or pre-processing of imagery. In
the commercial marketplace, proprietary advantages may
add significant revenue potential. Added to this profit
concern are the increasing governmental pressures not to
export advanced US technologies outside the US. On the
other hand, the science community, in fulfilling its
research goals, expects both full disclosure of observatory and data specifications as well as participation in
the discovery and resolution of identified observatory
problems. Science community experience has also shown
that as data sets age, much of the original knowledge
base used to create and operate an observatory is quickly
lost and not recoverable. Researchers worry that the
inherent long-term value of the original observations can
be lost through a failure to originally document the
mission specifications.
Organization and Management Processes: Access to and
shared use of commercial remote sensing data is a major
concern for scientific applications researchers that utilize
these images. The vendors approach to these issues is
reflected in their data licensing and mission operations
strategies.

Mission ManagementSatellite observatories that

exclusively serve science interests have mission
operation strategies that are specifically targeted to
science needs (Arvidson, Gasch, & Goward, 2001).
For example, when a research team has staff in the
field collecting ground measurements, mission operations can place highest priority on acquisition of
that site. However, with a commercial system, where
scientists are but one of several types of customer,
coordination of acquisition schedules to meet specific
scientific goals may either be quite costly or
impossible. This may suggest that more automated
ground measurement systems may be needed in the
future.
Data LicensingA difficult lesson learned from
Landsat commercialization resulted from the highly
restrictive licensing agreements associated with the
data imposed by EOSAT; they nearly eliminated
scientific exploration of the new Thematic Mapper
data (Marshall, 1989a, 1989b). Inability to share data
among researchers is a fundamental problem within
the context of the scientific method that depends, in
part, on repeatability and traceability of experiments
and analyses.
The scientific communitys previous concerns about
receiving remotely sensed data from the private sector
originated from both a lack of understanding of how the
commercial approach could be tailored to meet science

needs, as well as prior failed attempts to partner with private
industry. The 1980s Landsat privatization and the SEAWIFS
ocean color observation procurement are examples that the
science community considers when evaluating commercial
suppliers of remotely sensed scientific measurements (National Research Council, 1985, 1995). The NASA SDP
differ from the earlier Landsat privatization effort in that
under the SDP NASA is buying data from an existing
commercial provider rather attempting to commercialize
an existing NASA science mission (National Research
Council, 2002).
4. Lessons learned
From 1999 to present, the NASA SDP procured approximately 2400 IKONOS data products (550 GB), costing
over $11 million, for NASAs Earth Science Enterprise
researchers. These data sets have been used to support a
wide range of earth science applications, as noted by the
papers published in this special issue.
There is no question that NASA-funded researchers have
enjoyed the access to these novel observations that has been
provided under this experimental NASA Scientific Data
Purchase. Analysis of these observations has posed some
new challenges, but they have also provided new insights
about land cover properties. Many of the early concerns that
the NASA ESE community of users had about the measurements have been alleviated through working with the actual
data.
An observatory as novel as IKONOS takes some time to
adjust to both in terms of getting used to the new perspective provided as well as becoming comfortable with the
technical characteristics of the observations. As stated in the
JACIE paper (Zanoni et al., 2003), it took a large team of
engineers and scientists from US government laboratories
and universities over 3 years to complete a detailed assessment of metric properties of the imagery supplied by Space
Imaging. The extent of the IKONOS validation is consistent
with prior validation experiences with government-funded
ESE science missions. It takes time to develop full confidence in the output of new observatories. Having a team of
scientists and engineers working on such system validation
is vital to mission scientific success (Goward & Masek,
2001; Justice & Townshend, 1994, 2002).
4.1. Technical assessment
Overall, NASA and affiliated researchers, in conjunction
with Space Imaging engineers and technicians, believe that
the IKONOS system supplies high quality, high spatial
resolution multispectral imagery that is suitable to support
the research goals and requirements of NASA-supported
Earth scientists. Validation and independent scientific assessment of the IKONOS observations extended over 4
years. A scientifically comprehensive validation takes considerable time and effort. After 4 years, there are few
technical questions related to sensor performance and Space
Imaging-applied, post-acquisition processing that remain to
be addressed. Technical areas that have been evaluated and
required detailed attention include:

Sensor Radiometric Calibration: initial radiometric

calibration information supplied by Space Imaging for
the multispectral observations was found to have an error
that increased with increasing spectral wavelength, when
compared with the independent JACIE evaluation
(Goward, Davis, Fleming, Miller, & Townshend, 2003;
Pagnutti et al., 2003). Space Imaging is employing a
calibration procedure that is relatively novel to the earth
science community involving use of stellar observations
to periodically calibrate the measurements (Bowen,
2002). Space Imaging employed the JACIE results to
adjust their calibration information.
The detector-to-detector calibration is remarkably good.
The only artifact that was seen in early products, across
the nearly 3000 detectors, was the contrast noted between
adjacent arrays (Pagnutti et al., 2003). Space Imaging
later adjusted for this effect.
Sensor Geometric Properties: overall, the image geometry is exceptionally good with only minor problems
noted. This is especially noteworthy given that the sensor
often captures images at angles substantially different
from nadir by tilting the spacecraft (Helder, Coan,
Patrick, & Gaska, 2003). A minor error in the processing
algorithm was identified that when addressed, further
reduced geometric errors by 50% to nominal levels
(Helder et al., 2003). Interestingly, in the process of
discovering this latter problem, the current basis for one
of the Federal Geographic Data Committee standards are
called into question (Helder et al., 2003).
Image Modulation Transfer Function Compensation
(MTFC): modulation transfer function compensation is
an image enhancement technique that is rarely used in the
Earth sciences although there may be some real value in
doing so (Huang, Townshend, Liang, Kalluri, & Defries,
2002; Townshend et al., 2000). Initial examination of the
imagery metadata, at NASA Stennis revealed the phrase
MTFC applied: Yes(Ryan et al., 2003). This was
surprising, since such processing had not been requested
as part of the NASA SDP. Discussions with Space
Imaging revealed that MTFC is a standard part of Space
Imagings data production process, with the prime
objective of making the visual appearance of images
sharper to the user. JACIE analyses revealed that the
MTFC, as applied, introduces some noise as a result of
over-compensating the data, though this occurs for much
less than 1% of the observed pixels (e.g. (Goward et al.,
2003; Pagnutti et al., 2003; Ryan et al., 2003). Further
assessment at the Stennis Space Center showed that the
MTFC processing kernel had been rotated 90j, sharpen-
213
ing the across-track direction more than the along-track

direction. This resulted in the over-compensation
observed in the initial empirical analyses (Ryan et al.,
2003).
On-board Compression: to achieve adequate satellite to
ground telemetry of these high-volume observations,
Space Imaging uses a proprietary data compression
scheme developed by Kodak. Although this introduces
uncertainties in the ground-based calibrations methods
employed by the JACIE team since it violates the
linearity assumptions upon which such methods are
based, the JACIE team found little evidence of
detrimental effects resulting from this compression. The
11-bit data, even with compression, provides more
dynamic range than the 8-bit systems used on sensors
such as Landsat ETM+.
The JACIE technical assessment of the IKONOS data
revealed specific technical measurement uncertainties that,
if not discovered, could have reduced the value of IKONOS
as a source of scientific measurements. These uncertainties
were addressed in an open collaboration between industry,
government and the research community. This collaborative
activity converged on effective answers that resolved and
validated the qualitative and quantitative integrity of the
measurements, observatory and post-acquisition processing
procedures.
It is clear from the JACIE experience that successful
adoption of commercial remote sensing data by the land
science community necessitates open dialogue between the
system developers and the scientists and engineers who use
the data. An unanticipated outcome of this JACIE activity
was the increased trust that developed over time between
representatives from industry, research and government
involved in the technical assessment of the IKONOS data.
These analyses produced benefits for all involved and
demonstrated that all parties could work together in a
constructive framework. It also demonstrated that the development of such a successful working relationship
requires a substantial commitment of time and resources
from all sides.
4.2. Scientific assessment
The availability of IKONOS imagery to the terrestrial
science community has generally been greeted with considerable enthusiasm (Andrefouet et al., 2003; Goetz, Wright,
Smith, Zinecker, & Shaub, 2003; Hurtt et al., 2003;
Masuoka et al., 2003; Morisette et al., 2003; Sawaya,
Olmanson, Heinert, Brezonik, & Bauer, 2003; Seelan,
Laguette, Casady, & Seielstad, 2003; Small, 2003; Thenkabail et al., 2003). The IKONOS observatory provides a
valuable source of spatially detailed land visible and near
infrared measurements that may be used to validate and
evaluate coarser spatial resolution multispectral visible and
near infrared measurements acquired by NASA observato-
214
ries such as Landsat and MODIS. This is particularly true

for foreign investigations, where access to fine resolution
imaging is often difficult, if not impossible, to achieve (Figs.
2 and 3). Space-acquired high spatial resolution measurements overcome serious problems encountered with using
aircraft-based sensors relative to both restricted airspace and
unfavorable flying conditions. An IKONOS-type system
substantially improves global access to such high spatial
resolution multispectral measurements.
There are some novel aspects of an IKONOS-type
observatory that introduce new interpretation challenges
not previously encountered with systems such as Landsat
and MODIS. The IKONOS observatory basically follows
the same polar, sun-synchronous orbit as Landsat and Terra.
However, the basic observation image is approximately
11 11 km. In order for the observatory to provide reasonable repetitive coverage for most land areas, it is an
extremely agile observatory, having the capacity to point
off nadir up to at least F 60j and in any azimuthal direction
relative to the ground track (Dial et al., 2003). The IKONOS
near-nadir repeat cycle ( F 1j) is better than 140 days, but
with its agility, IKONOS can observe any land location
within 3 days and look angles between F 30j, assuming
that the particular land area is cloud-free at that time. This
operational configuration introduces several interpretation
issues.
The time delay between acquisitions of adjacent scenes

for a given land region can be as large as several months,
introducing significant variations across adjacent scenes,
as a result of sun angle and vegetation phenology and
other temporal changes, such as tide stage. Such delays
can make analysis of adjacent scenes exceptionally
difficult, if not impossible (Andrefouet et al., 2003;
Seelan et al., 2003) (Fig. 4). Similar timing problems are
experienced with moderate resolution systems such as
Landsat, Aster and SPOT again demonstrating the tradeoffs between spatial and temporal resolution with
contemporary satellite remote sensing systems.
The limited areal extent of a scene, combined with
persistent cloud cover and other demands for the
observatory, have made coordination between satellite
observations and ground measurements quite difficult
(Helder et al., 2003; Morisette et al., 2003; Pagnutti et al.,
2003).
Specular reflectance as a result of sun glint from local
water bodies can be a major problem, depending upon
the relation between solar zenith and azimuth angles,
relative to the view zenith and azimuth angles, preventing some applications of the measurements (Dial et al.,
2003; Sawaya et al., 2003).
Scene contamination with haze and cirrus-type clouds
(Fig. 5), when they had met the clear scene criteria for
Fig. 3. (a) IKONOS imagery from Antarctica; acquisition date: 10/05/2000; image upper left corner latitude/longitude: 65.25/ 60.89. (b) IKONOS image
of Kerguelen Island in the Indian Ocean; acquisition date: 09/24/2002; image upper left corner latitude/longitude:
49.09/70.54. These observations
demonstrate the value of a space-based high spatial resolution observatory. Acquisition of such detailed imagery from aircraft would have been quite difficult
and very expensive. A space-based system such as IKONOS provides ready access to all areas of the earths land areas, an important attribute for an
observation system that is used to study global changes.
215
Fig. 4. Seasonal changes recorded between adjacent IKONOS scenes acquired from an area in Maryland just north of Washington, DC; one acquired on 04/07/
2000, image upper left corner latitude/longitude: 39.36/ 77.16, and one acquired on 04/06/2000, image upper left corner latitude/longitude: 39.33/ 77.02.
Note that at this time of the year, vegetation foliage are rapidly growing so that the scene from the early date shows much less green foliage than the later date.
Fig. 5. Impact of haze and cirrus clouds on observation quality for two images acquired of Congo, Africa. Note the significant loss of contrast in the image on
the right, acquired on 03/26/2001, image upper left corner latitude/longitude: 1.19/16.01, due to haze and clouds (the image without the haze and cirrus clouds,
on the left, was acquired on 10/23/2000).
216
SDP was found to be a problem in several locations

around the globe (Goward et al., 2003; Sawaya et al.,
2003)
A few investigators (Hurtt et al., 2003) also note the need
for a short-wave infrared (e.g. 1.65 Am) measurement
particularly in forestry applications, based on their
experiences with Landsat, MODIS and SPOT-HRV. This
poses significant challenges to sensor design, since it
would likely lead to more costly high spatial resolution
array technology using non-silicon detectors. Hopefully,
in the near future, this will become more feasible.
Overall, the NASA ESE scientific experience with
IKONOS imagery has been positive. Access to these high
spatial resolution, digital multispectral measurements is yet
another major step forward in our capacity to observe the
earths land areas over a wide range of spatial and temporal
scales.
It is worth noting that most of the scientific concerns
with IKONOS observations result from the necessary
technical trade-offs that must be made to produce such a
high spatial resolution system. As with any new technology,
the learning curve for the science community for this new
high spatial resolution satellite has been steep and more
time will be needed to develop a full scientific understanding of this new observation resource. Certainly, this first
step under the NASA Scientific Data Purchase has provided
and excellent start.
4.3. Proprietary/ITAR factors
Over time, one of the more interesting changes that
occurred during the SDP IKONOS evaluation was that
Space Imaging became less and less proprietary about
their system configuration and operations.
In the development of commercial remote sensing, private industry is facing the challenge that small technical
differences can result in a significant competitive advantage.
Also this an area of technology, where the political system is
becoming increasingly concerned about the export of advanced US technologies overseas. Based both on concerns
about their own competitive advantages as well as government regulations placed on them during licensing under
current US ITAR (International Traffic In Arms Regulations), companies such as Space Imaging are either unable
or quite reluctant to reveal the details of their observatory
design to user communities, particularly those that are
technically and scientifically well informed.
In initial discussions of sensor radiometric calibration,
sensor performance characteristics and other mission characteristics were considered proprietary and were not made
available for distribution to the broad user community. As
more questions were posed concerning mission performance, information gradually began to flow. By the first
High Spatial Resolution Commercial Imagery Workshop in
2001, Space Imaging was publicly offering descriptions of
systems design and mission operations. This release of

information helped to develop substantial confidence within
the user community regarding the quality of the acquired
data.
A primary goal of partnerships between government and
commercial entities aimed at supporting the science community should be to keep the proprietary knowledge
needed to understand an observatory and its measurements
at an absolute minimum. The role of government scientists
and engineers, serving as moderators of the interactions
between private industry and the science community facilitated a dialogue that led to effective interactions. The
government staff served as a neutral agent assuring industry
that their proprietary and ITAR interests would not be
violated, while at the same time assuring users that best
engineering and science practices were being followed. This
brokering role may be critical to future efforts to pursue
such industry science community interactions.
4.4. Administrative considerations
The science community holds the view, based on years of
working with NASA earth observation missions, that the
practices restricting access to knowledge of mission operations and data licensing limit the value of such data sets for
scientific purposes. This might appear to raise significant
barriers preventing the use of commercial sources for
science measurements, but present experience with reference to Space Imaging IKONOS data suggests they can be
overcome.
4.4.1. Mission operations
Many of the science users were quite disappointed with
the length of time between image acquisition, when they
were notified of this event and finally when the data were
delivered. As a model, many of the users have become
accustomed to the ready access to such information supplied
by the USGS EROS Data Center for Landsat 7. The Landsat-7 acquisition schedule is placed on the EROS web site a
few hours prior to acquisition, and within 48 h, a JPEG of
each acquired image can be viewed. Such an approach
keeps the user well informed and significantly reduces
time-consuming communications between users and suppliers (Morisette et al., 2003; Sawaya et al., 2003). This
appears to be an easily resolvable administrative issue.
4.4.2. Data licensing
A more complex problem is encountered in data
licensing and costs. Under the NASA Scientific Data
Purchase, NASA Stennis acted as broker for individual
NASA ESE principal investigators for the acquisition of
approximately $11 million of Space Imaging IKONOS
data products. NASA Stennis negotiated a licensing
agreement with Space Imaging that permitted sharing
these data freely among NASA-affiliated researchers,
which includes virtually anyone currently funded through
the NASA Earth Science Enterprise. Users are very

positive about the fact that the data can be shared among
NASA investigators.
The US science community has become increasingly
aware that the single government agency license negotiated
for access of the IKONOS data did not address a fundamental activity that has applied to most previously acquired
NASA earth observations, archival preservation at the
USGS EROS Data Center (National Research Council,
2002). Most, if not all land observations originally acquired
by NASA, ultimately migrate to the archival repository at
the USGS Earth Resources Observation System (EROS)
Data Center. A fundamental criterion for data sets which are
archived at EROS is that they are publicly available to any
interested users. Based on the original licensing agreement
on IKONOS data between NASA and Space Imaging,
submission to the EROS data archive under these terms
will be impossible for data acquired under the NASA SDP.
In fact, this problem is not unique to the IKONOS data
purchase but all licensing agreements, with the exception of
the Earth Satellite Corp., (Birk et al., 2003), developed
under the NASA SDP. This archival issue simply was not
considered in these deliberations. This is a fundamental
lesson learned that extends well past that learned from
working with Space Imaging alone.
4.4.3. Future data costs
Under the NASA SDP, individual investigators procurement of these observations was through a modest proposal
process that was approved by NASA Headquarters. There
was no realized cost to individual investigators for IKONOS
data products other than writing the proposal. Thus, under
the NASA SDP individual investigators viewed these IKONOS observations as free to their funded research activities. This process is quite similar to access to other NASA
EOS observations, with the exception of Landsat data.
However, as the NASA investigators considered continued use of IKONOS data to support their scientific goals,
they recognized that they could probably not count on
continued free access to the NASA Scientific Data
Purchase. After reviewing NASA-negotiated data costs,
most of the investigators concluded that they could not
afford, within their grant budgets, to purchase many such
data sets. This suggests that if use of such commercial
resources are to be considered for scientific research in
the future that negotiated arrangements between NASA and
other federal agencies and individual companies will most
likely be needed to make the costs acceptable to individual
users.
These issues of data licensing and who should bear the
direct costs for access to commercially supplied remotely
sensed data are at the heart of the clear cultural differences
between the open-access expectations of the governmentfunded scientific community and the proprietary, commercial concerns of private industry. How these cultures are
ultimately integrated is well beyond the scope of this
217
current assessment but has been well exposed in NRC

evaluations of the NASA SDP of IKONOS data (National
Research Council, 1997, 2001, 2002). Areas that need
further consideration include legal factors related to intellectual property rights and data cost issues, including how
the fees should be paid and the specific costs that are
acceptable to both industry and users. These are complex
questions that are the crux of relations between the US
government, industry, and academia. Answering these questions may prove difficult, but addressing them will ultimately be critical to successfully meet the goals of all
partners involved.

The NASA Scientific Data Purchase activity was directed to evaluate whether the US private sector, in place of the
government, is capable of supplying remotely sensed land
imagery to science users that meet their needs. In this
activity, Space Imaging, space-based IKONOS mission,
achieved this goal, demonstrating that a private industry
financed satellite observatory is capable of meeting some of
the observation needs of the US science community.
The primary conclusion that can be reached from the
SDP IKONOS experience is that US private industry, in this
case Space Imaging, is technically able to supply useful
remotely sensed digital imagery to the US science community. This overcomes a major concern of the science community that typical commercial observatories are not of
sufficient technical quality to meet science needs. There
are several considerations that should be given further
attention in any future efforts by NASA or others to procure
scientific measurements from the commercial sector.
5.1. Some guidelines for future scientific data purchases
There are specific lessons learned from the SDP IKONOS experience that should be recognized when future
efforts to acquire scientific measurements from the US
private sector to support government-funded researchers
are to be considered. These include:

Independent assessment and validation of acquired

measurements is a critical part of the acquisition
process. This assessment could be performed either by
individual investigators or by an oversight team, such as
the NASA/USGS/NIMA JACIE team used in the SDP
activity, or even by an independent group such as the
National Institute of Standards and Technology (NIST) or
Underwriters Laboratories (UL) evaluators. There are
some real advantages in using the single organization
point of contact for these activities because critical
technical and intellectual resources are in short supply.
Focusing these resources produced reasonably quick (1 3
years) turnaround on the results.
218
An open dialogue between industry and ultimate data

users is essential to ensure that the best outcome is
achieved in the industry science community interaction.
Some of the difficulties that the science community
experienced with Space Imaging may have occurred as a
result of the organizational arrangements made by the
government entities. For example, the uncertainties concerning acquisition schedules could have been easily
resolved if Space Imaging and/or the responsible government agency simply had a web site where interested users
could check on the status of their requests.
Finding a balance between the perceived necessity of full
disclosure in the science community and the restrictions
in disclosure caused by competitive and ITAR concerns
of the private sector should be a major goal in
interactions between government and industry. The
Space Imaging experience has shown that as trust and
confidence develop, these issues have become substantially less problematic; each realizing that the overarching goal is to satisfy everyones best interests. There
also is a further need for appropriate government
agencies to review and evaluate the importance of ITAR
restrictions in protecting national security relative to the
substantial negative impact such restrictions have on
scientific and technical developments. With the advent of
foreign systems, the topic will need considerable further
attention. History has shown that to the degree possible,
open-access has the largest benefits for all involved.
Workable data licensing procedures are still difficult to
achieve in satellite land remote sensing. Industrys desire
to preserve the economic potential and of the observations though strict copyrighting, so they can benefit from
repetitive sales, tends to conflict with the user communitys interests in preserving the scientific value of the
observations through sharing and long-term repetitive
access. This is an area where industry, government
agencies and science community representatives need to
work together to seek effective compromises that satisfy
each community. For example, a possible alternate
licensing agreement could include some type of a timelimit clause, where after, for example, 5 years, the
observations acquired under a commercial data buy
would revert to the public archive and be made generally
available to the scientific community. It will probably
take more time for the commercial market place to
evolve before a rational long-term answer to this issue
will emerge. However, it is likely useful to continue
experiments such as the NASA SDP to encourage
emergence of new insights into the appropriate licensing
arrangements.
Much of what has been revealed in this effort by NASA
and Space Imaging jointly to meet NASA ESE scientists
data needs reinforces what had been previously discovered
in earlier efforts in this direction (National Research Council, 1985; Pace, Frelinger, Lachman Brooks, & Gabriele,
2000; Pace, Sponberg, & Maccauley, 1999). Moving away

from government sources to commercial sources of scientific land remote sensing observations is not simple and will
require further efforts to refine the models of this interaction
that have been employed to date.
Not withstanding remaining uncertainties, the NASA
Scientific Data Purchase of Space Imaging IKONOS high
spatial resolution, multispectral, satellite-acquired imagery,
has served as a successful step forward by NASA and Space
Imaging in moving towards commercial sources of valuable
scientific data. The particular lessons learned in this NASA
SDP activity should be of value to the Landsat Data Continuity Mission (LDCM) procurement that is now underway.
Acknowledgements
This assessment of the NASA Scientific Data Purchase
experience with the Space Imaging IKONOS observations
was carried out with support from the Scientific Data
Purchase administered by the civil servant and contractor
staff at the NASA Stennis Space Center. Drs. Goward and
Townshend received support from NASA Grant NAG
1398001 from NASA Stennis to pursue this study. All of
the participants in the JACIE program as well as contributors to the JACIE Workshops dedicated to IKONOS
assessment substantially contributed to the understanding
noted here. It is our hope that we have properly represented
all participants perspectives in this regard.
References
Andrefouet, S., Kramer, P., Torres-Pulliza, D., Joyce, K. E., Hochberg, E. J.,
Garza-Perez, R., Mumby, P. J., Riegl, B., Yamano, H., White, W. H.,
Zubia, M., Brock, J. C., Phinn, S. R., & Muller-Karger, F. E. (2003).
Multi-sites evaluation of IKONOS data for classification of tropical coral
reef environments. Remote Sensing of Environment, 88, 127 142.
(doi:10.1016/j.rse.2004.04.005)
Arvidson, T., Gasch, J., & Goward, S. N. (2001). Landsat 7s long term
acquisition planan innovative approach to building a global archive,
special issue on Landsat 7. Remote Sensing of Environment, 78(1 2),
13 26.
Birk, R. J., Stanley, T., & Snyder, G. L. (2003). U.S. government commercial
data purchase programs for use in government-sponsored science and
applications activities. Remote Sensing of Environment. (doi:10.1016/
j.rse.2003.07.007)
Bowen, H. S. (2002). Absolute radiometric calibration of the IKONOS
sensor using radiometrically characterized stellar sources. Pecora
15/Land Satellite Information IV/ISPRS Commission I/FIFOS 2002,
pp. CD-ROM. Denver, CO: American Society of Photogrammetry and
Remote Sensing.
IKONOS satellite, imagery, and products. Remote Sensing of Environment, 88, 23 36. (doi:10.1016/j.rse.2003.08.014)
Goetz, S. J., Wright, R., Smith, A. J., Zinecker, E., & Shaub, E. (2003).
IKONOS imagery for resource management: tree cover, impervious
surfaces and riparian buffer analyses in the Mid-Atlantic region. Remote
Sensing of Environment, 88, 194 207. (doi:10.1016/j.rse.2003.07.010)
Goward, S. N., Davis, P. E., Fleming, D., Miller, L., & Townshend, J. R. G.

sites. Remote Sensing of Environment. (doi:10.1016/j.rse.2003.07.009)
Goward, S. N., & Masek, J. (2001). Landsat 7 Special Issue. Remote
Sensing of Environment, 78(1 2), 1 325.
Helder, D., Coan, M., Patrick, K., & Gaska, P. (2003). IKONOS geometric
characterization. Remote Sensing of Environment, 88, 68 78.
(doi:10.1016/j.rse.2003.04.002)
Huang, C., Townshend, J. R. G., Liang, S., Kalluri, S. V. N., & Defries, R. D.
(2002). Impact of sensors point spread function on land cover characterization, assessment and deconvolution. Remote Sensing of Environment,
80(203 212).
Hurtt, G., Xiao, X., Keller, M., Palace, M., Asner, G. P., Braswell, R.,
Brondizio, E. S., Cardoso, M., Carvalho, C. J. R., Fearon, M. G., Guild,
L., Hagen, S., Sa, T., Schloss, A., Vourlitis, G., Wickel, A. J., Moore III,
C., & Nobre, C. (2003). IKONOS imagery for the large scale biosphereatmosphere experiment in amazonia (LBA). Remote Sensing of Environment, 88, 110 126. (doi:10.1016/j.rse.2003.04.004)
Justice, C. O., & Townshend, J. R. G. (1994). Data sets for global remote
sensing: Lessons learnt. International Journal of Remote Sensing, 15,
3621 3639.
Justice, C. O., & Townshend, J. R. G. (2002). Special issuethe moderate
resolution imaging spectrometer (MODIS): A new generation of land
surface monitoring. Remote Sensing of Environment, 83(1 2), 1 359.
Marshall, E. (1989a). News and comment. Landsats: Drifting toward oblivion? Science, 243, 999.
Marshall, E. (1989b). News and comment. Landsat wins a reprieve. Science, 243, 1429.
Masuoka, P. M., Clayborn, D. M., Andre, R. G., Nigro, J., Gordon, S. W.,
& Klein, T. A. (2003). Use of Ikonos and Landsat for malaria control in
the Republic of Korea. Remote Sensing of Environment, 88, 186 193.
(doi:10.1016/j.rse.2003.04.009)
Morisette, J. T., Nickeson, J. E., Davis, P., Wang, Y., Tian, Y., Woodcock,
C. E., Shabanov, N., Hansen, M., Schaub, D. L., Huete, A. R., Cohen,
W. B., Oetter, D. R., & Kennedy, R. E. (2003). The use of NASAs
commercial data purchase program in support of MODIS land validation. Remote Sensing of Environment. (doi:10.1016/j.rse.2003.04.003)
National Research Council (1985). Remote sensing of the earth from space:
A program in crisis. Washington, DC: National Academy Press.
National Research Council (1995). Earth observations from space: History,
promise and reality. Washington, DC: National Academy of Sciences.
National Research Council (1997). Bits of power: Issues in global access to
scientific data. Washington, DC: National Academy Press.
National Research Council (2001). Resolving conflicts arising from the
privatization of environmental data. Washington, DC: National Academy Press.
National Research Council (2002). Toward new partnerships in remote
sensing: Government, the private sector and earth science research.
219
Pace, S., Frelinger, D., Lachman, B., Brooks, A., & Gabriele, M.
(2000). The earth below: Purchasing science data and the role of
public private partnerships. Washington, DC: Rand Science and
Policy Institute.
Pace, S., Sponberg, B., & Maccauley, M. (1999). Data policy issues and
barriers to using commercial resources for mission to planet earth.
Washington, DC: Rand Science and Policy Institute.
j.rse.2003.07.008)
S. (2003). IKONOS spatial resolution and image interpretability. Remote Sensing of Environment, 88, 37 51. (doi:10.1016/j.rse.
2003.07.006)
Sawaya, K., Olmanson, L., Heinert, N., Brezonik, P., & Bauer, M. (2003).
Extending satellite remote sensing to local scales: Land and water resource monitoring using high-resolution imagery. Remote Sensing of
Environment, 88, 143 155. (doi:10.1016/j.rse.2003.04.006)
Remote sensing applications for precision agriculture: A learning community approach. Remote Sensing of Environment, 88, 156 168.
(doi:10.1016/j.rse.2003.04.007)
Small, C. (2003). High resolution spectral mixture analysis of urban reflectance. Remote Sensing of Environment. (doi:10.1016/j.rse.
2003.04.008)
Teillet, P. M., & Holben, B. N. (1993). Toward operational radiometric
calibration of NOAA AVHRR imagery in visible and near infrared channels. Canadian Journal of Remote Sensing (submitted for publication).
Thenkabail, P. S., Stucky, N., Griscom, B. W., S.Ashton, M., Diels, J.,
Meer, B. V. D., & Enclona, E. (2003). Biomass estimations and carbon stock calculations in the oil palm plantations of African derived
savannas using IKONOS data. Remote Sensing of Environment (in
review).
Townshend, J. R. G., Huang, C., Kalluri, S. N. V., Defries, R. S., Liang, S.,
& Yang, K. (2000). Beware of per-pixel characterization of land cover.
Townshend, J. R. G., & Justice, C. O. (2002). Towards operational monitoring of terrestrial systems by moderate resolution imaging. Remote
Woodcock, C. E., & Strahler, A. H. (1987). The factor of scale in remote
sensing. Remote Sensing of Environment, 21, 311 322.
Snyder, G. L., & Lee, G. (2003). The joint agency commercial imagery
approach. Remote Sensing of Environment, 88, 17 22. (doi:10.1016/
j.rse.2003.07.005)

(Geology) Remote Sensing of Environment - IKONOS Special Issue (Geologos)

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

(Geology) Remote Sensing of Environment - IKONOS Special Issue (Geologos)

Hochgeladen von

Copyright:

Verfügbare Formate

Remote Sensing of Environment 88 (2003) 1 2

A new direction in Earth observations from space: IKONOS

the government. The technical performance of IKONOS has

Remote Sensing of Environment 88 (2003) 3 16

Government programs for research and operational uses

* Corresponding author. Tel.: +1-202-358-1701; fax: +1-202-3583098.

istration (NASA), the National Oceanic and Atmospheric

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

mapping. Information capacity from U.S. civil government

operated by the public sector. Privatization policies of

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

2. NASA commercial data initiatives

In support of the NASA mission to understand and

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

data for Earth science research programs. Through this

NASA Scientific Data Purchase project. Fig. 3 indicates

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

Fig. 2. NASAs Scientific Data Purchase project user affiliation.

Karger, 2003), forestry (Ashton et al., 2002; Hansen,

higher (600-m) spatial resolution imagery of regions of

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

scientists to study phenomena related to elevation and

and in studying hydrological processes in support of

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

Platform and sensor

F 100 m (center pt)

Data provider: DigitalGlobe/Intermap

Data provider: Positive Systems

Model stereo pair

research and applications projects, including NASA Solid

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

Platform and sensor

Data provider: EarthSat

Circa 2000 orthorectified,

Circa 2000 orthorectified,

F 100 m (center pt)

Data provider: Positive Systems

Data provider: DigitalGlobe/Intermap

Data provider: DigitalGlobe

F 15.2 m (CE 90%)

Data provider: AeroVironment

x,y = 1.83 m (CE 90%)

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

have been delivered for Part II of the Scientific Data

to identify overlapping requirements in advance of data

3. USGS commercial imagery initiatives

3.2. Commercial imagery acquisition

Remotely sensed data and technology have been widely

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

Project4 and to research on forest structure and stand

4. NIMA Commercial Imagery Program

Additional emphasis early-on was on educating the

Fig. 4. NIMA TPED information cycle.

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

Commercial sources have collected and delivered more

NIMA has disseminated over 35,000 images electronically or on CD-ROMs.

4.4. Future of commercial imagery

5. Joint Agency Commercial Imagery Evaluation

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316

confidence of knowing that the systems are performing

NIMA, the USGS, and NASA have determined that the

NASA assessed the performance of the Scientific Data

R.J. Birk et al. / Remote Sensing of Environment 88 (2003) 316