Abstract: Reliability sensitivity analysis provides a rational vehicle to shed light on effects of different uncertain parameters on slope failure probability when only a limited number of data is available. This paper develops an efficient reliability sensitivity analysis method for slope stability. The proposed approach combines direct Monte Carlo simulation (DMCS) and probability density reweighting method (PDRM) to, efficiently and accurately, calculate slope failure probabilities for different cases considered in sensitivity analysis. In the proposed approach, random field theory is used to model geotechnical spatial variability. Local spatial averaging technique is applied to constructing the joint probability density function of geotechnical parameters with relatively low dimensions, which is directly used in PDRM. Finally, the proposed approach is illustrated using a design scenario of James Bay Dyke. Results show that: only one run of DMCS is needed in the proposed approach, avoiding repeated generation of random samples and performing slope stability analysis. This saves a large amount of computational efforts, and significantly improves computational efficiency for DMCS-based reliability sensitivity analysis. Using joint probability density function of geotechnical parameters derived from local spatial averaging in PDRM gives proper estimates of slope failure probability and avoids biased estimates, making PDRM feasible in reliability sensitivity analysis of slope stability in spatially variable soils.

Key words: slope stability, reliability sensitivity analysis, spatial variability, random field, local spatial averaging