Land use and land cover classification in sao paulo, brazil, using landsat-8 oli images and derived spectral indices

This article presents a land use and land cover (LULC) classification map based on Random Forest (RF) classifier algorithm in the Sao Paulo State (Brazil), using Landsat-8 OLI data. The method consists in using time series images from January to December of 2020 based on the spectral and temporal characteristics of the LULC classes. We performed the classification class by class considering: water, urban area, forest, agriculture, forest plantation and pasture. Then, we pre-processed the selected images based on the spectral characteristics of the targets to highlight each LULC class. After that, the classification was performed using RF for each class individually and then we composed the final map with all LULC classes. The results showed a global accuracy of 89.10%, kappa value of 0.8692, producer accuracies greater than 79.80% and user accuracies greater than 76.82% for the classes mapped. Therefore, the method is consistent allowing to minimize the classification errors facilitating the pos-classification edition of individual classes mapped.