Efficient stochastic analysis of unsaturated slopes subjected to various rainfall intensities and patterns

Rainfall infiltration poses a disastrous threat to the slope stability in many regions around the world. This paper proposes an extreme gradient boosting (XGBoost)-based stochastic analysis framework to estimate the rainfall-induced slope failure probability. An unsaturated slope under rainfall infiltration in spatially varying soils is selected in this study to investigate the influences of the spatial variability of soil properties (including effective cohesion c', effective friction angle φ' and saturated hydraulic conductivity k_s), as well as rainfall intensity and rainfall pattern on the slope failure probability. Results show that the proposed framework in this study is capable of computing the failure probability with accuracy and high efficiency. The spatial variability of k_s cannot be overlooked in the reliability analysis. Otherwise, the rainfall-induced slope failure probability will be underestimated. It is found that the rainfall intensity and rainfall pattern have significant effect on the probability of failure. Moreover, the failure probabilities under various rainfall intensities and patterns can be easily obtained with the aid of the proposed framework, which can provide timely guidance for the landslide emergency management departments.