MONITORING TROPICAL FOREST FROM SPACE: THE PRODES DIGITAL PROJECT 
D. M. Valeriano, E. M. K. Mello, J. C. Moreira, Y. E. Shimabukuro, V. Duarte, 
I. M. e Souza, J. R. dos Santos, C. C. F. Barbosa, R. C. M.de Souza 
National Institute for Space Research (INPE) 
C.P. 515-CEP 12210-970. Säo José dos Campos, SP. Brazil 
e-mail:(dalton, kalil, yosio, valdete, iris, jroberto)@dsr.inpe.br 
(moreira, claudio, cartaxo)@dpi.inpe.br 
Commission VII, WG VII/3 Integrated Monitoring Systems for Resource Management 
KEY WORDS: Terrestrial, Monitoring, Land Use, Change Detection, Segmentation, Classification, Forestry, GIS 
ABSTRACT: 
This paper presents the results of a new methodological approach to identify and to estimate deforested areas. The work refers to the 
PRODES DIGITAL PROJECT, which develops an automatic procedure to analyze satellite data, using the techniques of Linear 
Spectral Mixture Model, Image Segmentation and Classification by regions, that are available on software SPRING 
(GEOREFERENCED INFORMATION PROCESSING SYSTEM), a state-of-the-art GIS and remote sensing image processing 
system with an object-oriented data model, which integrates raster and vector data representations in a single environment. 
1. INTRODUCTION 
PRODES (Monitoring the Brazilian Amazon Gross 
Deforestation) is the largest forest monitor project in the world, 
based on orbital remote sensing. 
For many years, the NATIONAL INSTITUTE FOR SPACE 
RESEARCH (Ministry of Science and Technology) has been 
promoting the interpretation of images provided by the U.S. 
LANDSAT satellite to monitor the evolution of the extent and 
rate of gross deforestation in the Brazilian Amazon (Figure 1). 
Due to the geometric problems faced during the development of 
the manual interpretation of multi-temporal TM images (image 
hardcopies with different scales, overlays digitizing/scanning, 
complexity of deforestation pattern), the availability of these 
deforestation maps in a digital format has been restricted to 
some areas. 
Over the last years, an effort to find a methodology to overcome 
this problem has been successtul (Mello et al., 2003 and 
Moreira ct al., 2003). 
The digital image analysis is an alternative to perform this task. 
However, the conventional digital classification using a pixel by 
pixel approach has shown to be feasible only for small (local) 
areas, where is possible to have a good quality control of the 
image classification. 
For large areas, as the ones studied by PRODES, two factors 
had been restricting the digital analysis of TM images: lack of a 
reliable image classifier and the huge amount of disk space 
required to process several TM scenes. 
With the increasing capacity of computers in terms of 
processing and storage, the development of a methodology to 
study large areas using digital techniques is now possible 
(Shimabukuro at al., 1998). 
Data from PRODES DIGITAL PROJECT, properly 
georeferenced, form a database of multi-temporal layers with 
easy access and integrated with different information sources 
and or computer systems (www.obt.inpe.br/prodes). 
2772 
  
Figure 1 — Brazilian Amazon - WRS-TM/LANDSAT 
2. ANEW METHODOLOGICAL APPROACH 
2.1 Satellite Data and Image Processing System 
Amazon Brazilian WRS-TM/LANDSAT images of bands 3, 4, 
and 5 were used. The Landsat TM data were resampled to à 
pixel size of 60 m by 60 m resulting on an image size of 3326 
pixels by 3072 lines, in order to reduce computer processing 
time. 
Digital Charts at 1: 100 000, edited IBGE and DSG, were used 
for georeferencing the satellite images. 
The GIS software package SPRING (GEOREFERENCED 
INFORMATION PROCESSING SYSTEM) with basic 
functions of image processing (including ‘segmentation’ and 
'per-field clustering classifier’ — ISOSEG) was used 
(www .dpi.inpe.br/spring). Spring is a state-of-the-art GIS and 
remote sensing image processing system with an object-oriented 
  
Internati 
er 
data moc 
represen 
2.2 Imag 
2.2.1 Sh: 
There a 
proportic 
squares 1 
the Land 
soil, and 
of the m 
method 1 
members 
Landsat 
vegetatio 
was veri 
represent 
and there 
2.2.2 Im: 
The proc: 
region ¢ 
homogen 
(Zucker, 
procedure 
The segm 
image, ai 
attributes 
classificat 
Two parai 
(1) simila 
numbers | 
two regioi 
to be cons 
2.2.3 Clas 
An unsups 
was applic 
techniques 
algorithm 
1993) uses 
to estimate 
acceptance 
distance tl 
regions cai 
belonging 
First, the 
vegetation. 
Spectral M 
segmented 
algorithm 
edited. 
To estimat 
image, the 
difference 
increment « 
2001, 200]