Comparação da acurácia de índices de vegetação aplicados a classificação de imagens do satélite Landsat 8

The classification of soil use and coverage, as well as analysis of its changes are among most common applications for remote sensing. One of the most basic steps of the classification is the distinction of the vegetal cover in the other terrestrial surfaces. Landsat images are relevant sources of data in this analysis; and although there are several vegetation classification indices using Landsat data described in the literature, applications are limited by low accuracy in various situations. In this sense, the purpose of this study was to compare the available vegetation indexes and to identify the one that best applies to the classification of Landsat 8 satellite images, investigating vegetation indexes by leaf water content and leaf pigments. The values of the Normalized Difference Infrared Index (NDII), Simple Ratio (SR) and Visible Atmospherically Resistant Index (VARI) were evaluated for a specific region located in the Cerrado Biome. The performance of the vegetation index was compared with the performance of the Maximum Likelihood Classifier (MAXVER). The accuracy of the MAXVER classification was significantly higher than that of the vegetation index (Kappa - 0.95). Among the vegetation indexes, the classification of images was best applied to SR, demonstrating good agreement with the spectral targets, being the confusion between exposed and urban classes, and between the sparse and agricultural ranks classes, important sources of classification error.