Evaluation of sentinel-2, ndvi and mlme for mapping land use and land cover

Abstract

The Amazon Rainforest is considered one of the largest carbon reservoirs on Earth. However, indiscriminate anthropogenic land use and land cover changes, such as the conversion of forest to agricultural areas and pasture, generate large environmental impacts. The use of remote sensing techniques which helps mapping land use and land cover (LULC) becomes increasingly necessary. Vegetation indices such as NDVI (Normalized Difference Vegetation Index) and the Spectral Linear Mixing Model (MLME) are widely used for mapping and vegetation studies, because they allow analyzing and highlighting vegetation parameters and features in remotely sensed imagery. Thus, the aim of this paper was to evaluate the performance of land use and land cover (LULC) mapping using Sentinel-2B satellite data, added with the NDVI vegetation index and MLME using the Random Forest (RF) classifier. For this study, we used Sentinel-2B/ MSI images and both NDVI and MLME were calculated from Sentinel-2B bands. From image segmentation, attribute extraction was performed for each segment. The classification was performed by the RF method and validated using a Monte Carlo Simulation, observing Kappa and Global Accuracy (GA) values. To evaluate the gain obtained with the addition of NDVI and MLME variables, four classification scenarios were performed. We noticed that these scenarios presented similar results of Kappa index and GA, with no significant difference between them. The use of the Sentinel-2B/ MSI spectral bands showed a good alternative to mapping LULC, facilitating processing steps. However, the inclusion of the MLME for separation of the Degraded Forest (DF) class showed to be significant. In addition, it was shown that the use of the Random Forest classifier presents good results for the LULC mapping.