Application of Sentinel-1 and Sentinel-2 data via Machine Learning for Land Use and Land Cover Mapping in the Ibirapuitã Environmental Protection Area, Pampa Biome, using the Random Forest Classification Algorithm

Abstract

The joint approach of optical sensor images and synthetic aperture radar (SAR) has been effective in land cover mapping. In this study, conducted in the Ibirapuitã environmental protection area, machine learning techniques were employed to classify land use and cover. The Random Forest (RF) algorithm was applied using statistical attributes from Sentinel-2 optical image products, such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Soil-Adjusted Vegetation Index (SAVI), along addition to attributes from images Sentinel-1 SAR images, including backscattering coefficient, polarimetric parameters, and interferometric data. The results demonstrated the robustness of the RF classifier, with average values of Overall Accuracy, Kappa Coefficient, and F1-Score reaching 96.89%, 0.9495, and 0.8909, respectively. The combination of SAR attributes and optical data allowed for better discrimination in certain classes, such as urban areas, wetlands, and agriculture. The proposed methodology achieved high accuracy and precision in land use and cover classification, except when using isolated Sentinel-1 data. Notably, the inclusion of interferometric coherence resulted in the best performance among the proposed scenarios.