Multisensor analysis for environmental targets identification in the region of Funil dam, state of Minas Gerais, Brazil

Abstract

The use of remote sensing to map land cover and changes in land use has proven to be a practical, reliable, and accessible approach. These images provide precise details about the landscape, utilizing image processing techniques, modeling, and classification algorithms. This study aimed to identify different areas, such as coffee plantations, water bodies, urban areas, forests, exposed soil, and pastures in the Funil reservoir region of Minas Gerais, Brazil. Image data from Landsat-8, Sentinel-1, and Sentinel-2 satellites for June 2021 were used. Different supervised classification algorithms such as rf, rpart1SE, and svmLinear2 were applied based on a large volume of remote sensing data. The analyses and maps were performed using the software RStudio, considering a significance level of 5%. The highest accuracy and kappa index values were found for the rf algorithm, followed by svmLinear2 and rpart1SE. The results showed that the rf algorithm achieved the highest accuracy and kappa index values, followed by svmLinear2 and rpart1SE. However, during the validation phase, the svmLinear2 algorithm outperformed based on the statistical results of the confusion matrix. Therefore, it was considered the most suitable for generating the thematic mapping of the landscape. This is because svmLinear2 identified a more significant number of coffee areas and better-distinguished vegetation areas.