LAND-COVER AND LAND-USE CLASSIFICATION BASED ON MULTITEMPORAL SENTINEL-2 DATA

In this paper, we focus on the analysis of multitemporal Sentinel-2 data for land-cover and land-use classification. Given a set of representative training areas, we use a Random Forest classifier for the semantic labeling of the considered scene with respect to seven classes. As the classifier allows assessing the relevance of involved features for the classification task, we also estimate the relevance of the spectral channels for different dates. The derived results clearly reveal the benefit of a multitemporal analysis of Sentinel-2 data, since it also addresses seasonal changes in the acquired data.