A Comparison between EnKF and MCMC-Based Bayesian Updating for Consolidation Settlement Prediction

Consolidation settlement is usually predicted based on the geotechnical parameters obtained by the field and laboratory tests. However, such predictions usually deviate from field monitoring data due to uncertainties in parameter selection. Bayesian methods provide an effective way to update the geotechnical parameters and improve the prediction accuracy by incorporating the monitoring data. Markov chain Monte Carlo (MCMC) is a popular method to derive the posterior distribution of soil parameters. It can achieve a rigorous sampling but generally requires tens of thousands of forward model calculations. Ensemble Kalman filter (EnKF), as an alternative, can efficiently deal with recursive updating based on the sequential monitoring data, but accompanied by some limitations due to its Gaussian assumptions and linear update. This paper evaluates the performance of EnKF and MCMC methods for identifying the parameters and updating the predictions of consolidation settlement through a laboratory test. The results show that the EnKF and MCMC can result in a consistent estimation of the mean values of the updated soil parameters and settlements, while the uncertainties obtained from EnKF are overestimated.

Keywords: Inverse analysis, ensemble Kalman filter, Bayesian updating, consolidation settlement