A Physically-based Bayesian Network Model for Landslide Susceptibility Updating

In recent years, global warming has been associated with an increasing frequency and intensity of extreme meteorological events causing severe damage to land and infrastructure, and potential harm to people. For this reason, it is crucial to develop predictive tools to support land management. In May 2023, two extreme meteorological events struck the Emilia-Romagna region of Italy in succession. The first rainfall event was associated with a limited number of landslides while the following one triggered widespread landslides of various types, leading to extensive damage and forcing the closure of hundreds of roads. The extent of the damage in the affected areas was recorded by satellite imagery captured by Sentinel constellations of the European Space Agency (ESA). By looking for cloud-free pre-event, between-events and post-events images, two study areas were chosen for the development and calibration of a physically-based Bayesian network model, incorporating prior rainfall data and soil spatial variability. The primary objective is to calibrate the model using these training images and subsequently validate it across the entire affected regions utilizing available open-source data on landslide event inventories. With a given weather forecast, the resulting output aims to pinpoint the most hazardous areas in a timely manner. This research stems from a collaboration funded through the MSCA-SE UPGRADE project (GA 101131146).