GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 2.2 GNGTS 2024 Mapping recent flood covers using machine learning techniques. A. Mendicelli 1 , F. Mori 1 , C. Varone 1 , M. Simionato 1 , M. Moscatelli 1 1 CNR-IGAG, Is9tuto di Geologia Ambientale e Geoingegneria, Area della Ricerca di Roma 1, Via Salaria km 29.300, 00015 Monterotondo Stazione, Rome, Italy) Abstract At present, the most detailed geological map covering the enCre Italian naConal territory is the 1:100,000 scale geological map of Italy created by ISPRA. To esCmate the straCgraphic amplificaCon of seismic moCon at the surface over a large area, it is crucial to beber define the geological and lithotechnical characterisCcs of covering soils and geological bedrocks. This work is aimed at improving the definiCon of recent alluvial covers (Holocene and Upper Pleistocene deposits) compared to the 1:100,000 geological map of Italy. For this purpose, a methodology based on machine learning models has been developed. It considers both categorical and numerical variables to predict the presence/absence of recent flood coverage with good accuracy. To train the machine learning model, both geomorphometric parameters and geological databases at different scales were used. IniCally, the methodology was tested in the Calabria Region and in the Marche Region, for which promising results were obtained with good performances in the external test. The next step, sCll in the development phase, consists in the applicaCon of the methodology in a wider area which includes not only the Calabria and Marche regions but also Tuscany, Emilia-Romagna and Umbria. The model thus obtained will be tested across the enCre naConal territory. Acknowledgements This research was supported with funds from the PNRR, from the project: “NaConal Center for HPC, Big Data and Quantum CompuCng – HPC – SPOKE 5” – CN00000013. Corresponding author: amerigo.mendicelli @ cnr.it