Flood events rank among the most destructive natural hazards, necessitating comprehensive risk management strategies to mitigate their impact on society and the environment. Various approaches have been developed to map flood susceptibility. However, current methods still lack in accounting for dynamic changes in landscapes and infrastructure. This study leverages the potential of the Random Forest (RF) model to assess the flood susceptibility of the Italian territory. Hydrologic, geomorphic, and categorical information were used, evaluating a total of 25 conditioning factors. Through a novel approach called Average Merit of Information (AMI), particular attention was given to maximizing the information in the predictor factors. A rich database, including the Copernicus Emergency Management Service maps and regional records of historical flood events, was used for model calibration and validation. The RF model trained with mean maximum daily precipitation (MMDP), the Geomorphic Flood Index (GFI), distance from the nearest river (DNR), elevation, lithology, soil properties, NDVI, and land cover demonstrated superior generalization capacity compared to other tested sets, as evidenced by a quantitative comparison with official flood hazard maps (AUC greater than 0.9). Including the GFI significantly improved prediction accuracy in unexplored areas, though challenges persist in flat regions where geomorphic indicators are less distinct. Therefore, integrating satellite-derived information and complementary datasets facilitates the accurate identification of flood-prone areas, streamlining computational processes and enabling preliminary analyses for decision-makers. These findings underscore the importance of leveraging advanced modelling techniques and continuously updating data to inform flood risk management policies and practices.
How to cite: Saavedra Navarro, J., R. Zhuang, C. Albertini, and S. Manfreda. Mapping Flood Susceptibility Using Random Forest Exploiting Satellite Observations and Geomorphic Features. (under review), 2025.
