Preliminary assessment of AVHRR/NOAA data to detect deforestation in the Pantanal

Abstract

This paper presents preliminary results of deforestation detection in the Pantanal, using AVHRR/NOAA image data. This initial analysis is part of the MULPAN Project, which investigates the usefulness of data gathered by different sensors for surveying thematic information of the Pantanal physical environment. This Project results from cooperation between INPE (Instituto Nacional de Pesquisas Espaciais) and Embrapa (Empresa Brasileira de Pesquisa Agropecuária). A full resolution AVHRR/NOAA image, channel 2 (0.72-1.1 µm) and channel 3 (3.5-3.9 µm) from Sept. 15, 1990, was used to detect deforestation. The image was geometrically corrected, remapped to conform to the cylindrical equidistant projection, and classified using a maximum likelihood algorithm. SPRING-1.1, a geographical information and image processing system, was used for both geometric correction and classification. Classification assessment was based on previous maps derived from Landsat-5/TM from 1990/91. Results indicated that the methodology is not efficient to detect deforestation, mainly because Pantanal is a region of a high occurence of savannas. The AVHRR/NOAA image discriminated only 20% of the total amount of 112 points of deforestation identified with Landsat-5/TM data. The remaining points of deforestation were wrongly classified as savanna or dense savanna.