Efficient reliability analysis of unsaturated slope stability under rapid drawdown using XGBoost-based surrogate model

Reservoir slope stability during water level drawdown has drawn increasing concern in geotechnical engineering in recent years. In this study, an efficient reliability analysis framework based on the extreme gradient boosting (XGBoost) surrogate model is employed to evaluate the failure probability of unsaturated slopes subjected to the rapid drawdown considering the depth-dependent properties of spatially varying soils. A c-φ slope is selected as an illustrative example to investigate the coupled influence of the non-stationary characteristic of shear strength parameters and saturated hydraulic conductivity, as well as water level drawdown velocity, maximum drop height and scale of fluctuation on the slope failure probability. Results show that the adopted framework can estimate the low-level probability of slope failure with high accuracy and efficiency. It is found that the velocity and maximum height of water level drawdown have a significant effect on the unsaturated slope stability. Furthermore, it is recommended that the depth-dependent non-stationary soil properties be considered in most cases to ensure a more accurate result.