Modeling rainfall-induced landslide using the concept of local factor of safety: Uncertainty propagation and sensitivity analysis

Rainfall-induced landslides (RILS) are commonly investigated using hydro-mechanical models with the concept of local factor of safety. The inputs of these models are usually prone to uncertainties. An uncertainty analysis is required to investigate how uncertainties propagate through the model and impact the predictions. An appropriate strategy for uncertainty propagation analysis is suggested in this work to deal with nonlinearity and high dimensionality of RILS problems. It proceeds by performing a sensitivity analysis in two steps. A screening technique is first applied to eliminate insignificant parameters. Then, a global sensitivity analysis is performed to rank the parameters by order of importance. The Sobol indices are used as sensitivity metrics. The polynomial chaos expansion is used to compute the Sobol indices. The proposed strategy is first applied to a hypothetical benchmark and then to a more realistic configuration where prior knowledge of parameters and type of soil are considered. The results show that, when prior knowledge of soil is available, the most important parameters are the coefficient of cohesion, friction angle and air entry pressure head, respectively. The results also show that 10% uncertainty on these parameters leads to about 20% uncertainty on the prediction of slope stability.