Optimizing Landslide Susceptibility Models with Enhanced Parameters and Machine Learning

Landslides are one of the major disasters which is affecting the mountainous region of the world, and with the change in climate its impact is probably going to increase further, thus landslide susceptibility models play a key role for risk mitigation. The present work analyses landslides in Italy, focusing on rapid flow landslides and rainfall-triggered landslides. Data are derived from two national-scale inventories: the IFFI database, which provides spatially distributed landslide information, and the ITALICA catalogue, which records temporally referenced landslide events; the entire analysis was conducted using Google Earth Engine (GEE) with the Random Forest (RF) algorithm and Support Vector Machine (SVM), incorpo-rating 16 factors. Initial analyses using these parameters obtained a relatively fair accuracy, but by integrating two additional parameters, soil density and Normalized Difference Vegetation Index (NDVI) along with optimisation, the overall accuracy is further improved. Comparative analysis indicates that the RF algorithm outper-forms the SVM algorithm, making RF the superior method for landslide susceptibility modelling and prediction in this study, and confirming the choice of the variables using response surface method with reduced variables (Most influencing in RF). This highlights the validity of using advanced machine learning techniques and comprehensive parameter sets to improve landslide risk assessment in Italy.