APPLICATION OF DIFFERENT METHODOLOGIES USING LANDSAT 8 OLI AND TIRS IMAGES FOR LAND COVER CLASSIFICATION IN REGIONS OF THE BIOME PAMPA – BRAZIL

Abstract

Land cover maps and their dynamics within the Pampa fill historical gaps in initiatives and researches focused on grassland or non-forest vegetation in Brazil. For rural ecosystems, natural and managed, countless successful experiences with remote sensing are known. However, the operationalization of processing techniques, like classifications, still requires studies and tests to estimate its accuracies in face of the different kinds of images available (optical, thermal, and/or SAR – synthetic aperture radar). For vegetation mapping in the Pampa, challenges still demand the implementation of studies with different classifiers. This work had the objective of assessing the performance of parametric and nonparametric classifiers for land cover classification, with emphasis on vegetation covers, in regions of the biome Pampa – Brazil, including the comparison of the use of the optical and thermal bands of the OLI and TIRS/Landsat 8 sensors. For this aim, we tested the classifiers Maxver, Mahalanobis, Spectral Angle Mapper (SAM), and Support Vector Machine (SVM) with optical bands only, and combination of optical and thermal bands. The best performances were found with the SVM (optical bands only) and Maxver (optical and thermal bands) classifications. The thermal images showed potential for increasing the accuracy of the generated maps, arriving at a 3% increase in the results obtained with the combination of thermal and other bands.