Forecasting of rainfall-induced landslides in pyroclastic soil deposits through hydrometeorological information.

Shallow landslides pose a major geohazard impacting mountainous regions all around the world, and wide slope areas in Campania (southern Italy) covered by loose granular deposits overlapping a karstic bedrock are known for hosting the most destructive landslides of the region in the last decades. The landslide triggering factor in this case is clearly the rainfall. Nonetheless, there are concurring causes linked to the hydrological conditions predisposing slopes to failure (Bogaard and Greco, 2016). In the present study, the landslide-inducing factors are divided in static and dynamic (Moreno et al., 2023). The static factors (e.g., topography, slope, forest ratio) are well investigated in numerous studies on landslide susceptibility assessment. However, the modelling of dynamics factors (e.g., rainfall, soil moisture) is a relatively new issue and has been addressed only in few studies. In this study, Generalized Additive Models (GAMs) were applied for spaciotemporal data-based modelling of landslide prediction for eleven years (2010-2020). The study area is located on the Sarno and Partenio mountains in Campania where pyroclastic soil deposits cover about 370 km2 of carbonate massifs. In a first step, the modelling of static components, controlling landslide susceptibility in the area, was carried out by utilizing the historical data of landslide events along with other factors (slope, forest ratio etc.,) significantly affecting the static probability of landslide occurrence. Afterwards, the dynamic component was modelled by considering the triggering rainfall and the antecedent soil moisture for landslide events. The soil moisture data was taken from ERA5-Land soil moisture product. Lastly, the static and dynamic components were integrated to model the dynamic probability of landslide occurrence. A cross-validation technique was used for model training. The novel integrated model approach showed trustworthy improvements in the assessment of the probability of landslide. The model was also successfully tested for different rainfall events reproducing the landslide triggering conditions in the study area.