Territory occupation and consolidation in the Amazon region have some specific characteristics related to the dynamics of land use and land cover conversions, which can be analyzed using orbital remote sensing images. The aim of this study was to evaluate change detection products generated by change vector analysis (AVM) and image subtraction techniques derived from linear spectral mixing modeling (MLME), applied to Thematic Mapper/Landsat optical images, to study land use and land cover conversions occurring in agricultural settlement areas in the southeastern region of Roraima, Brazil. We analyzed change images derived from application of AVM (magnitude, alpha and beta) and subtraction of fraction images (soil, vegetation and shade), for their ability to identify and discriminate the existing conversions. An extensive field work was used as a guide to define the classes. Exploratory analyses of class behaviors were made and two supervised algorithms for image classification - Bhattacharyya and Support Vector Machine - were tested. By grouping (clumping classes), we sought to optimize conversion identification in the classification products. The results indicated better Bhattacharyya region classifier performance of conversion discrimination. The use of MLME fractions difference images as input into the classifier resulted a very good or excellent classification quality, which was better in comparison with products using AVM images, either in isolation or in conjunction with MLME difference images.
Land use; change detection; fraction images; remote sensing; Amazon
