Katsalis, ParaskevasBagkis, EvangelosKaratzas, KostasWohlgemuth, VolkerNaumann, StefanArndt, Hans-KnudBehrens, GritHöb, Maximilian2022-09-192022-09-192022978-3-88579-722-7https://dl.gi.de/handle/20.500.12116/39405Remote sensing data have been employed for monitoring the differences in land use over time. This information serves as the basis of any further land-related analysis, modelling and decision making. It requires satellite coverage of an area of interest, in various bands, and intense analysis of the data to correctly identify the different land types and associate them to the geographical reality precisely. In this paper, we collect Sentinel 2, level 1C satellite data to extract spectral indices and utilise them as features for land cover classification. The method is based on the use of machine learning for properly mapping the Greater Thessaloniki Area, engaging the random forest algorithm. Two different classification configurations in terms of target labels are tested for their accuracy. The main goal of the study is to present a pipeline for researchers and practitioners that need to define non-generic classes and classify geographical areas accordingly. Results, evaluated with the confusion matrix, suggest excellent performance on the test set and bring to surface limitations of the approach concerning the lack of proper high-quality data for algorithm training.enremote sensingsatellite dataland usemachine learningNormalized Vegetation IndexText/Conference Paper1617-5468