Landslide susceptibility analysis along the coasts of Liguria region (northern Italy)

Abstract Coastal landslides pose significant hazards to populated areas and infrastructure, necessitating accurate assessment and mitigation strategies. In this study, landslide susceptibility maps for rockfalls/topples, rotational/translational slides, complex phenomena and rapid flows were developed in the Liguria region (Italy) from the coast to 2 km inland using the maximum entropy model (MaxEnt) and the inventory of Italian landslides, IFFI (Inventario dei Fenomeni Franosi Italiani - https://idrogeo.isprambiente.it/app/iffi). 16 environmental variabilities derived from the DTM, geological map, CORINE land cover and topographic map of the region were used in the models. After an initial general screening to eliminate variables with high collinearity, only the variables found to be most significant for each model were used for each landslide type. The landslide occurrence points were divided randomly into training (80%) and test set (20%). The accuracy of the processed models was evaluated by ROC (receiver operating characteristic) curves and AUC (area under the curve). The results showed good performance for all models, although this was dependent on the type of landslide considered. Specifically, the AUC values obtained from testing data were 0.927 for the rockfall/topple model, 0.797 for the rotational/translational slides model, and 0.819 for the complex landslides model. Finally, the AUC value of the rapid flow model is 0.903, but only for the training data, as the number of landslides available for analysis was not sufficient for a split into test and training data. The variables contributing most significantly to the models are the slope, lithology, and land cover, followed by distance from the shoreline and elevation. Susceptibility maps were created for each type of landslide and combined into a final map providing a comprehensive overview of the landslide hazard at the regional level.