coordinate information as UTM-labeled tic marks. These data
were archived at EDC, and distributed to the project
implementation teams at the University of New Hampshire,
Michigan State University, and University of Maryland on
8mm tape cartridges. A 1:250,000 scale photographic 3-band
color composite print and a 1:1,000,000 color transparency
were also produced and distributed.
LANDSAT digital data acquired for this project represent a
distillation and high-grading of the best available scenes for
the regions of interest for the three time periods. Most scenes
have 20% cloud cover or less. Without the highly
collaborative nature of this project, such a valuable data set
may never have been assembled in a single archive.
3.2 Overview of End-to-end Processing
Previous research on an International Space Year project to
map deforestation in the Brazilian Amazon (Skole and Tucker
1993) and methods development for this project suggested
that the use of digital image processing in conjunction with
editing, georeferencing and spatial analysis in a Geographic
Information System was an effective means for quantifying
deforestation, and that the use of high resolution LANDSAT
data may in fact yield much better precision than AVHRR-
based analyses. The use of digital pre-processing with visual
post-processing greatly reduces analysis time over that of
hand digitization of a photographic product. and greatly
reduces the confusion of classes associated with purely digital
processing techniques. Nearly 3000 scenes will be processed
and analyzed using these techniques to produce individual
classifications containing areas of forest, deforestation.
regrowth, non-forest vegetation, cloud, cloud shadow, and
water.
The dataset resulting from the digital classification was then
plotted as a polygon on clear vellum at 1:250,000 scale and
compared with the 1:250,000 scale color composite prints of
the Landsat images provided by EDC. Digitizers checked the
label on each and every polygon for accuracy, made any
necessary changes and added any missing polygons. This
process was iterative, with quality assurance checking repeated
until the coverage was accepted for archiving. Quality control
was carried out by trained supervisors with GIS training,
forestry backgrounds and field experience.
The final coverages were then stitched together scene by scene
to build regional coverages. Spatial analysis took place at this
regional level, where areas of deforestation, forest, and
secondary growth were calculated. These activities created the
products that will be distributed to the science user community
by a DAAC or international organization such as IGBP and
used by collaborators in global carbon balance modeling
project.
To obtain an estimate of the accuracy of the data set, the HTF
project developed a field based accuracy assessment program.
Our objectives were to quantify the thematic and areal variance
480
of our results. We did this at three levels of analysis. First, we
conferred with experts in each region, gaining insights from
their extensive knowledge of local conditions. To facilitate a
close working relationship with experts in the countries and
regions in which we were studying, we had a visiting
scientists program which provided support to colleagues from
tropical countries to spend time in residence at the HTF labs.
Second, we conducted preliminary and cursory field excursions
to various areas, where we developed a good on-the-ground
sense of conditions and established initial classification rules
and procedures. Third, we conducted systematic field validation
exercises, where points on the field were selected and
measurements were made using a Global Positioning System.
Results of these field exercises were used to develop a
statistical accuracy assessment using standard methods of
presentation in contingency tables.
3.3 Data Acquisition
The project obtained digital Landsat MSS and TM data from
three sources: (a) the U.S. national archives held at the EROS
Data Center, primarily for historical MSS data, (b) foreign
ground stations, and (c) directly from the EOSAT corporation.
Much of the data were purchased with funding from NASA and
EPA at rates set by the USGAU (U.S. Government Affiliated
User) agreement, although the project had been able to get
discounts as a result of the large number of scenes being
acquired from any given source. The EROS Data Center
provided the MSS data from the U.S. national archive as in
kind support to the project.
At the start of the project a detailed data plan was developed to
determine the best dates and sources for data. This data plan
was created using a computerized Information Management
System (IMS) which the project constructed to sort through
over 3 million metadata entries obtained from the EROS Data
Center and foreign ground stations. The EROS Data Center
provided the full digital metadata listing of its archives as well
as the most current listing of metadata for selected ground
stations through the Landsat Ground Station Operators
Working Group (LGSOWO). These listings were supplemented
with a significant number of additional entries provided
directly by ground stations that had not provided metadata
listings to LGSOWG (c.g. Thailand and Indonesia).
The project then used the IMS to determine how much data was
available with less than 20% cloud cover, and to select the
most appropriate dates for large-arca acquisitions in each of
the regions covered by this study (i.e.. the Amazon Basin,
Central Africa, and Southeast Asia).
The project established cooperative arrangements with ground
stations in Brazil, Ecuador, Kenya, Gabon, South Africa,
India, Thailand, Indonesia, and Australia. We worked closely
with the ground station operators to select and purchase data
from their historical archives as well as new acquisitions. The
University of New Hampshire had the primary responsibility
for coordinating all data orders from the ground stations. The
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
proj
estal
proc
Data
wher
inve!
proc
grou
purcl
that
(CCI
Once
by tl
distri
Univ
analy
The
Brazi
The à
grour
data |
probl
inclu
chron
lack
dimin
probl
data. '
availa
future
Lands
3.4
This :
used ¢
Hamp
the Bi
by the
GSFC
Pre-pr
analyz
contro
the IN
projec
Recor
and en
(http:/
Next,
defore
image
Steps t
