Prediction of area and length of landslides: application to the events of May 2023 in Emilia Romagna region

This thesis aims to develop a forecast model for landslide area and length run-out in the Emilia-Romagna area, Italy, using a data-driven approach integrating machine learning techniques with GIS-based spatial analysis. Using geological, geomorphological, land use and hydrological parameters, the study forecasts these two parameters using the Random Forest approach. Particularly attention is paid to the effect of rainfall parameters and slope-related properties in influencing model performance. To find the impact of the predictors in the accuracy of the estimation, several model configurations were tested both with and without certain variables. Models including all the selected predictors have the highest expected accuracy; rainfall attribute considering field measures is particularly significant since it directly gauges localised rainfall occurrences. Models lacking crucial rainfall or slope conditions showed poor performance, thereby confirming their usefulness in landslide dynamics. The results of this study holds significant implications for practical applications in land-use planning and disaster mitigation strategies towards landslide phenomena. The predictive capability of the model provides policymakers and risk management authorities with a scientific basis for land planning, infrastructure resilience actions, and resource allocation for landslide-prone areas.