Multitemporal analysis of land cover and land use in the municipality of Itabirito (MG) for territorial management purposes

Abstract

The multiple applications of Remote Sensing for development are evident in actions promoted through public and/or private partnerships focused on technological innovation linked to research progress. These applications play a crucial role in decision-making and geospatial analyses related to planning and development. Therefore, this study aims to classify orbital imagery using machine learning in an open-source tool to support multitemporal geospatial analyses for the municipality of Itabirito (MG). The methodology applied in this study involved the use of the Dzetsaka plugin in QGIS version 3.22 for classifying satellite images, using the random forest algorithm as the classifier. For validation, the AcATAMa plugin, also integrated into QGIS version 3.22, was used along with the stratified random sampling method to evaluate the accuracy of the maps. A total of 3,925 validation samples were used, divided by year. The Kappa index calculation was employed to assess the reliability of the results, categorizing them as 'substantial' and 'almost perfect' for the years analyzed (2019-2022), with Kappa values ranging between 77% and 91%. The overall accuracy exceeded 86%, indicating the precision of land use and land cover classification in the municipality of Itabirito (MG).