Data assimilation for Bayesian updating of predicted embankment response using monitoring data

Abstract

Data Assimilation (DA) algorithms has been successfully employed in geotechnical problems to jointly estimate the state of the system and model parameters, however, the impact of the field monitoring setup on the performance of DA is often overlooked. In this paper, the impact of the field monitoring setup on the performance of DA is studied. The Ensemble Kalman Filter is used as the DA algorithm as part of a synthetic experiment which includes a fully coupled hydromechanical numerical model of an embankment constructed on soft ground. The results of the assimilated parameters show different rate of convergence toward their synthetic true value which corroborates well with the results of the global sensitivity analysis performed in this study. In order to investigate the difference in influence between the quantity and type of measurement, different monitoring strategies were chosen in this study. The results indicate that the effective friction angle and Poisson’s ratio are better estimated when the horizontal displacement is included along with the vertical displacement in the observation space of the DA procedure. Finally, the strong correlation between observation type and parameter convergence is independent of the type of initial prior knowledge, but strongly depends on the measurement location.