Towards automation of river water surface detection

European rivers are increasingly impacted by frequent and lasting dry periods, with consequences on jeopardized ecosystems and local economies. Tools for monitoring the evolution of such impacts may be profitable exploited by public administration to assess environmental conditions and draw safeguard policies. This work presents the evolution of a methodology which integrates optical and radar imagery, by Copernicus Sentinel constellations, to map river water surfaces. Despite the base methodology being developed as a man-supervised classification, with necessity for the user to manually define training polygons, the proposed advancements will allow the system to automate training sample extraction. The process is based on the realization of binary masks, originated by processing optical and radar imagery with a BMax Otsu algorithm for image segmentation. The masks are then furtherly refined to obtain a reliable set of classified pixels, from which the training samples are extracted. A sensitivity analysis is performed for assessing the optimal amount of pixels to be considered, with respect to the total area of interest. Furthermore, the performances of several Machine Learning supervised classification algorithms are compared, leading to the selection of the best algorithm to be considered for future developments of the methodology.