. First, we
ights from
facilitate a
untries and
a visiting
agues from
HTF labs.
excursions
the-ground
ation rules
] validation
lected and
g System.
develop a
nethods of
data from
|. the EROS
b) foreign
rporation.
NASA and
Affiliated
ble to get
nes being
ata Center
hive as in
veloped to
data plan
anagement
rt through
EROS Data
ata Center
es as well
ed ground
Operators
plemented
provided
| metadata
h data was
select the
n each of
on Basin,
ith ground
th Africa,
>d closely
chase data
tions. The
onsibility
ions. The
project cooperated with ground station personnel and
established procedures for ordering, formatting and
processing, shipping, and scheduling of data acquisitions.
Data were sent directly to the University of New Hampshire
where shipping manifests were reconciled against orders,
inventoried and forwarded to the EROS Data Center for pre-
processing. The project coordinated between the foreign
ground stations and the EROS Data Center to insure that data
purchased from the ground stations were delivered in a format
that EDC could pre-process into the project specific format
(CCT-P).
Once at EDC the data were pre-processed to a format specified
by the project, they were shipped back to the project for re-
distribution to the appropriate laboratory for analysis. The
University of New Hampshire and Michigan State University
analyzed data for the Brazilian Amazon and Southeast Asia.
The University of Maryland analyzed data for the non-
Brazilian Amazon and Central Africa.
The archive acquired by the project is unique. For at least two
ground stations, the processing systems for historical MSS
data have failed since the project began acquiring data. This
problem is a trend observed worldwide, due to several factors,
including failures of MSS tape handling hardware and the
chronic degradation of older tapes. This problem along with
lack of money for maintaining the archive due to the
diminished interest in historical MSS data presents a major
problem in the use of Landsat data as a long term source of
data. The data acquired by this project may be the only copies
available to the community at the present time and in the
future. Preservation of an HTF archive containing high quality
Landsat data is an additional benefit of this project.
3.4 Data Analysis
This section describes the details of data processing methods
used at Michigan State University and the University of New
Hampshire for analyzing the data covering Southeast Asia and
the Brazilian Amazon. These methods are similar to those used
by the University of Maryland, University of Virginia, and
GSFC.
Pre-processed data were sent to UNH and MSU where it was
analyzed for deforestation. The first step was inventory
control, where each scene received from EDC was recorded into
the IMS. Scenes were then rectified to "north-up" in a UTM
projection using the corner points in the Data Descriptor
Record as reference points. A browse product was then made
and entered into the IMS and placed in the project web page
(http://pathfinder-www.sr.unh.edu/pathfinder).
Next, the data were analyzed to produce GIS based maps of
deforestation. We used a hybrid methodology based on digital
image processing combined with visual interpretation. The
Steps used in our analysis were as follows:
International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
Data loading. Landsat MSS and TM image data arrived on
8mm tapes in P format. P format corresponds to system
corrected, map projected in Universal Transverse Mercator
(UTM) projection (not precision geocoded) digital data with
embedded tics marks and their corresponding UTM
coordinates. Each individual tape contained either 5 files
(MSS) or 8 files (TM). The first file on each tape was the Data
Descriptor Record (DDR) containing information about the
acquisition date, scene size (number of rows and columns and
pixel size) and UTM coordinates of the four corners of the
scene matrix. The information on the DDR was used to
automate the image referencing step.
Image Referencing. Landsat scenes were georeferenced to
"north-up" using the four corner points provided in the DDR as
Ground Control Points (GCPs) to define the linear
transformation from files coordinates to UTM projection. The
rectification was performed using nearest neighbor resampling
technique to minimize convolution of the data.
Image Classification. Two different classification
methods were used in our analysis: Image Thresholding and
Iterative, Self-organizing, Unsupervised Classification
(ISODATA).
Image Thresholding. This classification technique was
used for our analysis of MSS images of the Amazon. In this
region we were interested in identifying seven thematic
features: forest, deforestation, secondary forests, water,
, clouds, cloud shadows, and non-forest. After an initial R&D
effort we decided that image thresholding was best suited for
the Amazon considering time efficiency and accurate initial
classification of the data.
Identification of Spectral Characteristics of Thematic Classes:
The analyst examined the statistics of the digital values for
representative areas for each of these surface features found
within a scene. This was accomplished by drawing polygons
around representative areas and calculating the mean, std. dev.,
maximum, and minimum for each band for all pixels within the
polygon.
Single Band Thresholding: Utilizing the spectral
characteristics of the data for each thematic class threshold
values were chosen such that individual classes would be
separated. Once a threshold value had been chosen a two class
classification was created by level slicing the image based on
the DNs of one band. The output classification was evaluated
by overlaying the classification on the imagery. The
threshold values would then be adjusted accordingly. These
steps of identifying the threshold value, creating a level slice,
and evaluation of the classification were repeated up to five
times depending on the features within the scene.
Combination of Individual Level Slices: The single band
thresholding step created up to six classifications, each of
which identified one of the thematic features. These
classifications were combined into a single classification
481
