Coupling a stochastic rainfall generator and a physically based infiltration and slope-stability model to investigate landslide triggering

Analysis of rainfall conditions potentially leading to landslide triggering is an important task for risk assessment and mitigation. In this paper a Monte Carlo approach based on coupling a stochastic rainfall model with a infiltration and slope stability model is developed and applied to generate synthetic rainfall and associated landslide data. More specifically, generated rainfall series are used as input to a physically based infiltration model, which enables to compute the pressure head response of the soil to rainfall events, given initial conditions. The TRIGRS model is used to compute transient pressure head response, and the initial conditions are derived by a linear-reservoir water table recession model. The resulting long series of landslide triggering and non triggering rainfall events is then analysed to identify most significant rainfall characteristics for defining landslide-triggering thresholds. First, widely adopted power-law thresholds, which is the most commonly encountered in literature, has been investigated. Then other models have been tested, which account for antecedent rainfall. Nonlinear regression techniques based on neural networks have also been exploited to choose the most suited functional form to link landslide triggering to variables. Validation with observed rainfall-landslide data shows the potential of this modelling technique.