Aplicação do algoritmo Random Forest para classificação do uso e cobertura da terra na Bacia Hidrográfica do Rio Sucuru utilizando o Google Earth Engine.

Abstract

Through the inadequate anthropogenic actions over the years in the Sucuru River Basin, where it has been undergoing changes in the ecosystem. Land use and land cover maps are important tools to provide accurate information for dynamic monitoring, planning and watershed management. With the advent of cloud computing platforms and machine learning classifiers, new opportunities are emerging for more accurate and large-scale land use and land cover classification. The objective was to obtain a classification of land use and land cover in the Sucuru River Basin, for the year 2022, through the Random Forest classifier, using the RGB and RGB bands combined with spectral indices (NDVI, NDWI, NDBI and SAVI), using as accuracy parameters the Confusion Matrix, Kappa Coefficient, General Accuracy, Producer Accuracy and User Accuracy. The use of the Google Earth Engine platform for the preparation and analysis of land use and land cover maps provided satisfactory results with speed and accuracy. It was observed that the best performance for the Random Forest classifier was the RGB-INDICES combination, obtaining a good performance in distinguishing the classes, as in the classification of land use and land cover, achieving a lower spectral confusion and an accuracy above 89.98%, the addition of spectral indices generated significantly satisfactory levels of accuracy.