Estimation of Uncertainties in Soil Using MCMC Simulation and Effect of Model Uncertainty

Abstract

The simulation of field conditions for seismically induced slope failures incorporates model uncertainties, which account for the difference between simulated and observed slope behaviour. The quantification of this uncertainty is mandatory to understand the field response of the geotechnical system and make decisions for geotechnical systems. Previous studies have partially studied uncertainty for slope systems under seismic loading. To this aim, this study proposes a methodology based on probabilistic back analysis to estimate uncertainties in soil parameters considering the observed slope response under seismic loading. The proposed method involves support vector regression (SVR) model to map the relationship between soil parameters and seismically induced slope displacement. The SVR model is generated using the data from the numerical simulation of slope system under seismic loading using FLAC 2D. Further, the developed SVR model is used for probabilistic back analysis using Markov Chain Monte Carlo (MCMC) simulation. The Noto Hanto earthquake in 2007 and the subsequent slope failure along Noto Yuryo Road, Japan, are considered as a case study to validate the proposed methodology. The results of the case study show that the updated or inferred soil parameters have less variability than the prior distribution. Further, the uncertainties in the slope system influence the inferred soil parameters. Hence, a parametric study is conducted to investigate the effect of model uncertainty on the posterior statistics of soil parameters. The study results facilitate a better understanding of the slope deformation mechanism and the effect of model uncertainty on the updated statistics of soil parameters.