Application of different methodologies with the use of OLI and TIRS LANDSAT 8 images for classification of land cover in areas of the Pampa - Brazil biome

The operationalization of processing techniques, such as the classification of usage and coverage of land, still needs studies and tests in order to estimate its efficacy, especially due to the function of different types of images available with a focus on rural or non-forest vegetation in Brazil. This essay had as its objective to assess the performance of parametric and non-parametric classifiers for the classification of usage and coverage of land in areas of the Pampa - Brazil biome, including the comparison of the use of optical and thermal bands of the OLI and TIRS/Landsat 8 sensors. Thus, 8 classifications were generated from the Maxver, Mahalanobis, Spectral Angle Mapper (SAM), and Support Vector Machine (SVM) algorithms considering the following combination of data: 1 -optical bands only, and 2 - combination of optical and thermal bands. The classes considered for classification were: arboreal vegetation, cultivation area, field, water, and others. The parametric classifiers presented Global Precision of 72% to 83%, and the non-parametric of 66% to 80%. The class of cultivation area was that which most presented confusion with the field areas in all of the classifiers. The highest rates of accuracy were ascertained in the SVM (optical bands only) and Maxver (optical and thermal bands) classifiers. The inclusion of thermal images presented a rise of 1% to 3% in the accuracy of the generated maps, values that still must be superseded through new classification tests.