Hybrid stochastic-mechanical modeling of precipitation thresholds of shallow landslide initiation

Abstract

Numerous early warning systems based on rainfall measurements have been designed over the last decades to forecast the onset of rainfall-induced shallow landslides. However, their use over large areas poses challenges due to uncertainties related with the interaction among various controlling factors. We propose a hybrid stochastic-mechanical approach to quantify the role of the hydro-mechanical factors influencing slope stability and rank their importance. The proposed methodology relies on a physically-based model of landslide triggering, and a stochastic approach treating selected model parameters as correlated aleatory variables. The features of the methodology are illustrated by referencing data for Campania, an Italian region characterized by landslide-prone volcanic deposits. Synthetic intensity-duration (ID) thresholds are computed through Monte Carlo simulations. Several key variables are treated as aleatoric, constraining their statistical properties through available measurements. The variabilities of topographic features (e.g., slope angle), physical and hydrological properties (e.g., porosity, dry unit weight $γ_d$, and saturated hydraulic conductivity, $K_s$), and pre-rainstorm suction is evaluated to inspect its role on the resulting scatter of ID thresholds. We find that: i) $K_s$ is most significant for high-intensity storms; ii) in steep slopes, changes in pressure head greatly reduce the timescale of landslide triggering, making the system heavily reliant on initial conditions; iii) for events occurring at long failure times (gentle slopes and/or low intensity storms), the significance of the evolving stress level (through $γ_d$) is highest. The proposed approach can be translated to other regions, expanded to encompass new aleatory variables, and combined with other hydro-mechanical triggering models.