IKONOS imagery for the Large Scale Biosphere–Atmosphere Experiment in Amazonia (LBA)

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.

Full text

Page 1

IKONOS imagery for the Large Scale Biosphere–Atmosphere Experiment
in Amazonia (LBA)
George Hurtta,b,*, Xiangming Xiaoa, Michael Kellera,c, Michael Palacea, Gregory P. Asnerd,
Rob Braswella, Eduardo S. Brondı ´zioe, Manoel Cardosoa, Claudio J.R. Carvalhof,
Matthew G. Fearona, Liane Guildg, Steve Hagena, Scott Hetrickh, Berrien Moore IIIa,
Carlos Nobrei, Jane M. Readj, Tatiana Sa ´f, Annette Schlossa,
George Vourlitisk, Albertus J. Wickelf,l
aInstitute for the Study of Earth Oceans and Space, University of New Hampshire, Durham, NH 03824, USA
bDepartment of Natural Resources, University of New Hampshire, Durham, NH 03824, USA
cUSDA Forest Service, International Institute of Tropical Forestry, Rio Piedras, PR, USA
dDepartment of Global Ecology, Carnegie Institution of Washington, Stanford University, Stanford, CA, USA
eDepartment of Anthropology, Indiana University, Student Building 130 Bloomington, IN 47405, USA
fEMBRAPA Amazo ˆnia Oriental, Bele ´m, PA 66095-100, Brazil
gEcosystem Science and Technology, NASA Ames Research Center, Moffett Field, CA 94035, USA
hAnthropological Center for Training and Research on Global Environmental Change Indiana University, Bloomington, IN 47405, USA
iCPTEC/INPE Instituto Nacional de Pesquisas Espaciais, Cachoeira Paulista, SP 12630-000 Brazil
jDepartment of Geography, Maxwell School, Syracuse University, Syracuse, NY 13244, USA
kBiological Sciences Department, California State University, San Marcos, CA 92096, USA
lCenter 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 CO2with 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.
D 2003 Elsevier Inc. All rights reserved.
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
Experiment in Amazonia (LBA) was initiated as an inter-
national research initiative led by Brazil with strong United
States and European Union participation. LBA is designed
to understand the climatological, ecological, biogeochem-
ical, 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
0034-4257/$ - see front matter D 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2003.04.004
* Corresponding author. Institute for the Study of Earth Oceans and
Space, University of New Hampshire, Durham, NH 03824, USA. Tel.: +1-
603-862-4185; fax: +1-603-862-0188.
E-mail address: george.hurtt@unh.edu (G. Hurtt).
www.elsevier.com/locate/rse
Remote Sensing of Environment 88 (2003) 111–127

Page 2

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 IKO-
NOS satellite.
Launched from Vandenberg Air Force Base in September
1999, the IKONOS satellite is the world’s 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 panchro-
matic 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 Pur-
chase (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. High-
resolution 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 field-
study sites. Eddy-flux tower sites are foci for local estimates
of CO2flux between the atmosphere and land surface and
other important ecophysiological measurements. In addi-
tion, they are typically the focus of many supporting
Table 1
Summary statistics on IKONOS holdings for LBA
fArea (km2)
N
Total Final
Total Beta
Total Complete
Total Final Unique
Total Beta Unique
Total Unique
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.
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.
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
112

Page 3

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. Fine-
scale 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-
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 km2have been acquired, consisting of
approximately 5268 km2at 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 Santare ´m.
Fig. 2. Examples of IKONOS imagery for two important LBA study regions near the cities of Manaus and Santare ´m. The images displayed are false color
representations of multispectral (4 m) imagery and, with the exception of Santare ´m 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.’’
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
113

Page 4

All IKONOS holdings collected for LBA are available to
LBA researchers following data-sharing guidelines estab-
lished 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 im-
agery in the LBA Experiment is for enhancing studies of
fine-scale heterogeneity in vegetation, land use, and land-
use 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 vege-
tation 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;
? Mappingagroforestrysystemsandsmall-scaleagriculture.
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 differ-
ent land cover types is important to carbon cycle and land
surface studies. Medium- to coarse-resolution optical sen-
sors (e.g., AVHRR, Terra-1 MODIS, SPOT-4 VEGETA-
TION) 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 rela-
tionships 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 end-
member 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:
Ri¼
X
k¼1
N
rkifkþ ei
ð1Þ
where Riis the reflectance of spectral band i in a multi-
spectral image (i=1, 2, ..., M), fkis the fractional cover of
end-member k (k=1, 2, ..., N) within a pixel, rkiis the
reflectance of end-member k at spectral band i, and eiis 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.
Table 2
CumulativestatisticsfromEOS-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.
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
114

Page 5

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 eddy-
flux Tower in Ro ˆndonia 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 re-
flectance 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
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 observa-
tional factors driving variations in apparent shadow are
known, little quantitative information exists on the magni-
tude 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., Land-
sat). The IKONOS satellite, with 1 m panchromatic and 4 m
multispectral capabilities, provides an opportunity to ob-
serve 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 panel—MODIS image (8-day composite of surface reflectance, July 20–26, 2000), false color
composite (band 2-6-1). Lower right panel—Fractional cover of forest within 500 m pixels, derived from spectral mixture analysis of MODIS data. This figure
‘‘includes materialn Space Imaging L.P.’’
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
115

Page 6

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 (FS.D.) shadow fraction
was 0.25F0.12, and within-scene (e.g., forest stand) vari-
ability 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 0.0F0.01 to 0.12F0.04 to 0.16F0.05 for areas with
woody vegetation at low (<25% cover), medium (25–75%)
and high (>75%) density, respectively. Landsat-like obser-
vations 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 f30–
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 measure-
ments 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 veg-
etation 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).TheIKONOSsatelliteaddsanewresource
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).
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
116

Page 7

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 pan-
sharpened 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 ground-
based 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 IKO-
NOS-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 esti-
mates derived manually from the 1 m panchromatic IKO-
NOS satellite imagery. The statistical distribution of crown
diameters from IKONOS was biased toward larger trees,
probably due to merging of smaller tree crowns, under-
estimation of understory trees, and over-estimation of
individual crown dimensions in the manual estimates
(Fig. 5a).
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 image’s 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 Tapajo ´s 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 clear-
cutting, selective logging significantly changes the rain-
forest by creating canopy gaps, disturbing understory veg-
etation, 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 (Nep-
stad 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 selec-
tively 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 higher-
resolution satellite data. Read (in press) compared perform-
ances of selected spatial methods using 15-m panchromatic
and 30-m multispectral Landsat-ETM+ data, and 1-m pan-
chromatic and 4-m multispectral IKONOS data, for distin-
guishing differing degrees of canopy disturbances resulting
from reduced-impact logging at the site near Manaus. The
study reports the calculated fractal dimension, spatial auto-
correlation 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-
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
117

Page 8

bance gradient comprised three treatments: areas of old-
growth (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 Tapajo ´s National Forest,
south of Santare ´m, 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, semi-
variograms 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
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 Tapajo ´s km 83 Tower area. This figure ‘‘includes materialnSpace Imaging L.P.’’
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
118

Page 9

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 cor-
related; Cressie, 1991) for the logged areas were detect-
able. 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 as-
ymptote (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 identifi-
able (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 log-
ging practices.
3.4. Effects of fine-scale heterogeneity in vegetation on
carbon balance
Fine-scale heterogeneity in land cover, such as is intro-
duced 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 CO2exchange, evapotranspiration,
and energy balance of intact tropical transitional forest to
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 frac-
tion of photosynthetically active radiation (fPAR) absorbed
by the intact and selectively logged forest canopies to
provide a way to link the canopy physiological measure-
ments obtained from eddy covariance towers (net CO2
exchange and evapotranspiration) to satellite data. Data
collected indicate that net CO2exchange and evapotranspi-
ration 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 fPARand 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.
3.5. Detection of biochemical and biophysical changes on
small land holdings
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 Arau ´jo, & Brienza Ju ´nior, 1997). Guild, Sa ´
and others are studying the influence of seasonal and inter-
annual 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 deter-
mine 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 for-
ests and agriculture more resilient to the effects of agricul-
tural 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)
LoggedNot logged
4
5
7
3
0
6
4
1
7
0
0
0
1
0
0
0
0
0
1
0.20
0.42
Mean
S.D.
4.11
2.47
The logged plots have more clusters than the not logged plots.
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
119

Page 10

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 enrich-
ment) 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 Ju ´nior, 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
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 materialnSpace Imaging L.P.’’
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
120

Page 11

resolution satellite imagery from the IKONOS system is a
superior surrogate to address biochemical and biophysical
effects introduced by these small land-holding practices.
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 Ac ¸aı ´ palm agroforestry system and floodplain forests in the Amazon estuary. Classified Landsat TM over IKONOS multi-
spectral image (top), aerial view of agroforestry canopy (middle), ground view of intensively managed agroforestry (bottom). This figure ‘‘includes materialn
Space Imaging L.P.’’
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
121

Page 12

vicinity of Igarape ´-Ac ?u, 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 mois-
ture, 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 can-
opy 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 bio-
geochemical cycles and the human aspect of the program.
Low data availability (e.g., cloud cover during the produc-
tion 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 un-
derstanding of the regional agriculture systems and land-
use 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 ac ?aı ´ 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 ?aı ´ palm forest
cover between 10,000 and 20,000 km2in 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 ac ?aı ´
palm agroforestry to a level of 81% accuracy (Brondı ´zio,
Moran, Mausel, & Wu, 1994, 1996). Estimates of manage-
ment intensity based on ac ?aı ´ palm’s importance value (a
ranking index based on density, frequency, dominance) have
been successfully correlated to Landsat TM’s NIR, MIR, and
red bands. IKONOS offers new opportunities to the under-
standing 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 pro-
duction and economic return of agroforestry stands. In
addition, IKONOS images may help to refine the accuracy
of Landsat-based agroforestry mapping, particularly by con-
tributing to the delineation between areas under different
management intensities as well as to quantify changes in the
structure offorest 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 in-
tensities, 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 improve-
ments in the definition of intensive and intermediary man-
aged agroforestry areas and in the definition of vegetation
regrowth stages.
4. Discussion and conclusions
IKONOS tasking for the Large Scale Biosphere–Atmo-
sphere Experiment in Amazonia has focused on the acqui-
sition 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.
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
122

Page 13

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 distri-
butions tend to over-represent relatively common climatic
conditions and under sample rare climatic conditions.
Adding factors to this analysis such as soil type, topogra-
phy, biogeography, and land-use history is likely to dilute
the representativeness of the current set of holdings even
further. With a current unique sample of approximately
3300 km2in an area of nearly 5,000,000 km2, current
IKONOS holdings represent a small and incomplete sam-
ple 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
104GB 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 sam-
pling methods have been used in the context of ground-
based sample measurements (e.g. USDA Forest Service,
1992). They are also necessary in high-resolution remote
sensing.
Understanding the importance of fine-scale heterogene-
ity over large geographical areas is a central challenge in
earth sciences that arises in many sub-disciplines (Ehler-
inger & 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 IKO-
NOS, are being developed and deployed. A critical chal-
lenge ahead will be in managing model complexity and
efficiently organizing high-resolution information. Remote-
sensing 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.
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 Head-
quarters in 2001 and 2002. Support for this research was
provided by NASA through the Terrestrial Ecology, Land-
Use Land-Cover Change, New Millennium, Interdiscipli-
nary Science, and Scientific Data Purchase Programs. D.
Clark and three anonymous reviewers provided sugges-
tions 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.eos-
webster.sr.unh.edu). This paper ‘‘includes materialn Space
Imaging L.P.’’
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).
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
123

Page 14

Map
ID
NameCountry lat nwlat swlat nelat selon nwlon sw lon nelon searea
(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
A´guas Emendadas
A´guas Emendadas
A´guas Emendadas
Brası ´lia East
Brası ´lia 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
Santare ´m Kilo 67
Santare ´m Kilo 67
Santare ´m Kilo 83
Santare ´m Kilo 83
Santare ´m Kilo 83
Santare ´m Kilo 83
Santare ´m Kilo 83
Santare ´m Pasture
Santare ´m 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 A. Database of IKONOS Tower Site images for LBA (http://www.eos-webster.sr.unh.edu)
Appendix B. Database of IKONOS Field Site images for LBA (http://www.eos-webster.sr.unh.edu)
Map
ID
NameCountrylat nwlat sw lat nelat selon nw lon swlon nelon seArea
(km2)
Date
acquired
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 Paraı ´so 1
Alto Paraı ´so 2
Alto Paraı ´so 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
?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
?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
?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
?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
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
124

Page 15

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 de-
tailed 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. Re-
mote 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 Re-
mote Sensing Symposium, Vancouver, BC (pp. 2069–2072).
Boardman, J. W. (1992). Sedimentary facies analysis using imaging spec-
trometry: 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 land-
cover distributions using global reflectance data. American Geophysical
Union Fall Meeting, San Francisco, CA.
Bradshaw, G. A., & Spies, T. A. (1992). Characterizing canopy gap struc-
ture in forests using wavelet analysis. Journal of Ecology, 80, 205–215.
Brienza Ju ´nior, 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.
Brondı ´zio, 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.
Brondı ´zio, E. S., Moran, E. F., Mausel, P., & Wu, Y. (1996). Changes in
land cover in the Amazon estuary: Integration of thematic mapper with
Appendix B (continued)
Map
ID
NameCountrylat nwlat sw lat nelat selon nwlon sw lon nelon 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 Vito ´ria
Igarape ´-Ac ¸u 2
Igarape ´-Ac ¸u 2
Igarape ´-Ac ¸u 2
Igarape ´-Ac ¸u 2
Igarape ´-Ac ¸u 2
Ituqui, Brazil scene A
Ituqui, Brazil scene B
Ituqui, Brazil scene C
Ituqui, Brazil scene D
Ituqui, Brazil scene E
Machadinho
Mato Grosso
Medicila ˆndia
Mil Madeiras
Pastaza
Pilche
Ponta de Pedras East
Ponta de Pedras West
Rohden East
Rohden Northeast
Rohden West
Ruro ´polis East
Ruro ´polis West
Santare ´m FLONA
Tapajo ´s
Sa ˜o Francisco do
Para ´ East
Sa ˜o Francisco do
Para ´ East
Sa ˜o Francisco do
Para ´ West
Sewaya
Tiguano
Tome ´-Ac ¸u East
Tome ´-Ac ¸u 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
83Brazil
?1.07
?1.20
?1.07
?1.20
?47.82
?47.82
?47.72
?47.72 175.557/18/2001
84Brazil
?1.07
?1.20
?1.07
?1.20
?47.83
?47.83
?47.72
?47.72 191.1012/30/2001
85Brazil
?1.07
?1.20
?1.07
?1.20
?47.83
?47.83
?47.80
?47.80 53.28 7/18/2001
86
87
88
89
90
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
G. Hurtt et al. / Remote Sensing of Environment 88 (2003) 111–127
125