Probabilistic analysis of rainfall-induced shallow landslide susceptibility using a physically based model and the bootstrap method

Physically based landslide susceptibility analysis is widely used for landslide prediction owing to its high predictive capability. However, due to limited information and the spatial variability of slope materials, this approach involves uncertainty. To quantify the uncertainty, probabilistic analysis has been adopted. However, for accurate implementation of probabilistic analysis, it is important to have sufficient data for an evaluation of statistical parameters of random variables. Probabilistic landslide analysis for a regional area is associated with difficulties because of limited data. The bootstrap method, which was adopted in this study, is known to be effective in dealing with uncertainty caused by insufficient data. The bootstrap method combined with the point estimate method (PEM) was proposed to overcome the limitations of previous bootstrap methods, the results of which did not provide a single value of failure probability. The proposed method was applied to a practical case, and the probabilistic approach using Monte Carlo (MC) simulation was also applied for comparison. The analysis showed that the performance of the bootstrap–PEM method was superior to that of the MC simulation. In addition, by comparing analysis results obtained with and without correlated variables, this study found that the cross-correlation between cohesion and the friction angle affects the analysis results. Therefore, the proposed approach based on the bootstrap sampling method that can readily evaluate and handle the cross-correlation presents an advantage over probabilistic analysis in that cross-correlation between input parameters can be involved in physically based susceptibility analysis.