Reliability-based prediction of landslide using efficient Monte Carlo Simulation

Abstract

Several studies have attempted to develop thresholds for prediction of landslide initiation due to antecedent rainfall by using the physical-based approach. The approach offers a better understanding of the underlying process of the landslide mechanism. In the previous studies, however, determinate values of hydro-mechanical soil parameters are commonly considered, neglecting the uncertainties of the variables. Thus, this study aims to propose a relatively computationally, inexpensive reliability approach to address the uncertainties of the multivariate. The application of the proposed model is demonstrated by taking a case study from Sabah, Malaysia. The Latin Hypercube Sampling was adopted for efficient sampling, while the dependencies of the multivariate were modelled using a higher-order copula. The infiltration process and slope stability were assessed based on a fully-coupled seepage-deformation and shear strength reduction technique, respectively in a finite element programme. The probabilities of failure for various antecedent rainfall condition were computed by using Monte Carlo Simulation. In order to reduce the computational cost of the simulation, the Multilayer Perceptron regressor was applied to form the non-linear surrogate model for an implicit limit state function.