A Full Probabilistic Approach to Landslide Forecast

Landslides are among the most destructive natural disasters that occur frequently worldwide, claiming lives and causing severe economic losses. The most common approaches for managing the short-term landslide risk is based on the definition of deterministic thresholds of a triggering event (a seismic quantity, or an amount of rain) above which the landslide is expected to occur. However, landslides, as well as most of natural events, is hardly predictable deterministically, owing to the unavoidable and ubiquitous presence of uncertainties of different kind. In this study, we present the first steps towards the development of a full probabilistic landslide forecasting model that accounts for the probabilistic forecasts of triggering events (such as earthquakes and/or rainfalls), and it includes a full appraisal of different kinds of uncertainty. Within a Bayesian mathematical framework, the model combines the probabilistic distribution of the mechanical parameters of the soil with the probability of observing a certain natural triggering event; the output is a space-time dependent probability of occurrence of landslides as a function of the probability of occurrence of their triggering event. In addition, we describe the landslide forecasts as a distribution of probability instead of one single value, to give a complete description of what we know and what we do not know. This approach provides a suitable scientific output that can be used by land use managers and decision-makers. Indeed, a formal probabilistic assessment fits more adequately the intrinsic non-deterministic nature of landslide occurrence. Moreover, it provides a more suitable framework that help defining roles and responsibilities of all actors involved in the full risk reduction process.