A probabilistic approach to model spatio-temporal landslide susceptibility

Shallow landslides pose significant hazards globally, particularly in regions with steep topography and susceptible geological conditions. These landslides are often triggered by intense rainfall or rapid snowmelt, and the understanding of their spatial and temporal dynamics is essential for hazard assessment and risk mitigation, especially in the context of climate change.

This study develops a statistically-based spatiotemporal model using Generalized Additive Models (GAMs) to evaluate shallow landslide susceptibility in the Orba basin(595 km2), Northern Italy. The model integrates static predictors such as slope and lithology with dynamic rainfall descriptors, particularly maximum rainfall intensity and antecedent cumulative rainfall, with the aim of finding a failure probability associated with certain values of antecedent cumulative and maximum rainfall intensity. Values for the rainfall descriptors are derived from a copula analysis that allows to estimate these parameters for defined return periods. In this way, both the spatial and the temporal components are included within the analyses.

Results highlight the nonlinear influence of cumulative rainfall on slope stability, consistent with suction stress theory, and the irrelevant effect of extreme rainfall intensities beyond a threshold. Susceptibility matrices derived from the model enable time-dependent assessments at the slope unit scale, offering valuable tools for early warning systems and climate change scenario analyses. In particular, the probabilistic methods using copula modelling allowed for the quantification of landslide susceptibility associated with specific return periods. Also, the deterministic and probabilistic analyses of future climate scenarios under varying RCP pathways revealed complex temporal trends in landslide susceptibility, demonstrating the significant impact of climate change on slope stability.