| Resumo: | In past years, studies about Land Use and Land Cover (LULC) have been approached extensively in remote sensing for providing information on the environmental and global changes in the landscape. In the forest species mapping, one of the major challenges when using Sentinel-2 (S2A) multispectral data is to delineate and discriminate areas of heterogeneous forest components with spectral similarity at the canopy level. In this context, the main objective of this study was to evaluate the S2A data performance for LULC mapping, using a Random Forest classifier (RF). A set of 26 independent variables derived from the 2019 summer period S2A data, with a spatial resolution of 10 m, was used. A total of eight object-based LULC classes were created, four forest classes (Quercus suber, Quercus rotundifólia, Eucalyptus sp, and Pinus pinea) and four other uses. For this propose supervised classification method was applied using the RF classifier. The cartography accuracy assessment was performed using the statistics confusion matrix and Kappa coefficient (k). This study showed that the RF classifier achieved high overall accuracy (92%) and Kappa (91%) for the four forest classes defined using S2A data.
|
|---|