Probabilistic back analysis for rainfall-induced slope failure using MLS-SVR and Bayesian analysis

Measurement and model uncertainties in soil parameters account for the difference between slope behaviour in the field and expected behaviour. The probabilistic back analysis is an effective approach to quantify these uncertainties in soil parameters. A new methodology for probabilistic back analysis is proposed to evaluate the uncertainties in soil parameters for observed data for slope. The proposed methodology implements multi-output least square support vector regression (MLS-SVR) to replicate the numerical model for slope under precipitation. This methodology also utilises a multi-objective genetic algorithm and Bayesian analysis to estimate updated statistics of soil parameters for observed data for slope. The rainfall-induced slope failure at Malin, Pune, India, in 2014 is used as a case study to validate the proposed methodology. The mean values of soil parameters are updated using multi-objective genetic algorithm for the expected values of safety factor. The uncertainties in soil parameters are estimated using Bayesian analysis. The updated statistics of input parameters suggest that matric suction governs the slope behaviour under rainfall precipitation. The results of the study suggest that continuous updating of the observations reduces the uncertainties involved in soil parameters. It is noted that the values of safety factor calculated using updated parameters are consistent with the slope failure observed in the field. Hence, results of the study can be used for the reliability-based design of slopes and the provision of remedial measures.