Sequential probabilistic back analyses of spatially varying soil parameters and slope reliability prediction under rainfall

Abstract

Accurately predicting slope reliability under a rainfall/rainstorm event is an important prerequisite for preventing rainfall-induced landslide hazards. However, the predicted probability of slope failure under the rainfall/rainstorm event is often larger than the observed frequency of slope instability. The spatial variability of multiple soil parameters was rarely accounted for. To address this issue, this paper proposes an efficient sequential probabilistic back analyses approach for learning multiple spatially varying soil parameters using Bayesian Updating with Subset simulation (BUS) method. Two survival records of a real slope in India (i.e., the slope stays stable before the rainfall and the slope keeps stable after a 57-day weak rainfall) are successively used in the sequential probabilistic back analyses of soil parameters. The results indicate that the proposed sequential probabilistic back analyses approach can effectively update the distributions of multiple spatially variable soil parameters by the fusion of slope survival records. More accurate statistics of soil parameters can be obtained when additional slope survival records are used in the probabilistic back analyses. Furthermore, two slope failure records under a 3-day heavy rainfall event and a rainfall event ranging from May 1, 2016 to June 30, 2016 in Chibo, India are, respectively, used to predict the slope reliability and further validate the effectiveness of the proposed approach. The predicted probabilities of slope failure under the target rainfall events are well consistent with the actual observation frequency. The proposed approach can provide a powerful and versatile tool for determining the statistics of soil parameters and early warning of landslide hazards under the future rainfall events.