Validação do uso e cobertura do solo utilizando imagens do satélite sentinel - 2 no estudo de desastres ambientais: o caso do rompimento da barragem de Brumadinho - MG

Abstract

The rupture of the B1 ore tailings dam in Brumadinho-MG triggered one of the worst environmental disasters in Brazil, affecting hundreds of lives and devastating an area of 297 hectares. With the aim of assessing changes in land use and occupation after the disaster, this study aimed to delimit and evaluate the extent of the impact of tailings mud, as well as to perform supervised classification of the municipality before and after the dam rupture, through a combination of techniques, including remote sensing imagery, geoprocessing tools, and RStudio software. Analysis of spectral indices NDVI, EVI, and NDWI revealed a significant and adverse impact on the affected region, highlighting vegetation loss. When conducting supervised classification, the RF and SVM algorithms showed similar performances, with overall accuracy and kappa above 90%. However, RF stood out as more effective in identifying different thematic areas of the surface when using Sentinel-2 satellite images. These results showed that machine learning algorithms, when combined with high-resolution spatial images, provide precise and reliable analysis for future environmental studies. This approach enables rapid identification and assessment of changes in land cover following catastrophic events such as dam ruptures, facilitating decision-making and proper planning for environmental recovery.