Bayesian Modeling of Rainfall-Induced Landslides

Landslides are a serious geologic hazard, posing substantial and increasing risks, and resulting in widespread disruptions. Correctly assessing the relationship between terrain conditions and landslides is fundamental to understand these instability-triggering factors, and to implement effective monitoring campaigns. In this work, we study rainfall-induced landslide events occurred from 2002 and 2020 in the Peri-Adriatic area of the Abruzzo region (Italy). We propose a zero-inflated Poisson (ZIP) model to estimate spatial landslide susceptibility while accounting for the presence of excess zeros in the data. Thanks to the probabilistic framework, it is possible not only to estimate the spatial susceptibility but also to locate the areas that are likely to have under-reported events. Results show that the most susceptible areas are in the vicinity of river basins and in high coastal areas. The probability of not having landslides generally increases as the distance from the Adriatic Sea increases.

Keywords: Bayesian, Landslides, Under-reporting, Zero-inflated Poisson, Zero-inflation.