Artificial Intelligence for a Multi-temporal Classification of Fluvial Geomorphic Units of the River Isonzo: A Comparison of Different Techniques

The pressure of human activities is particularly relevant on fluvial ecosystems, where activity such hydroelectric energy production can change natural dynamics. For this reason it is important to monitor, with a systematic approach, river geomorphic units distribution and their evolution over time. In particular this work consists of an application of different AI techniques to process Sentinel-2 optical data to acquire a multitemporal classification of fluvial geomorphic units (Channels, Pools, Bars, Island, Vegetation) on a study area of the river Isonzo in Friuli Venezia Giulia (Italy). Results showed that all the AI methods tested allow to perform accurate classification, with best results obtained by Random Forest, that reach an overall accuracy of 0.986, and the most confusion between Bars and Island classes with F-measure of 0.931 and 0.961 respectively.