On the feasibility of data assimilation for uncertainty modelling in geotechnics.

Paper in proceeding, 2023

Advanced constitutive models are usually employed in geotechnical applications to achieve higher fidelity solutions consequently demanding more model parameters. However, due to limited data availability, characterisation becomes difficult leading to increased uncertainties in the prediction which is quite common in geotechnical problems. In view of this, simpler models are preferred but sometimes are not robust enough to capture a complex geotechnical system. In this paper, we examine the issue of system uncertainty and model selection to gain a qualitative insight regarding the question of whether a simple model can capture a complex system when augmented with a data assimilation procedure, namely Ensemble Kalman Filter, while maintaining model fidelity and robustness. Results indicate that data assimilation can help capture the behaviour of the system even if the model complexity does not match that of the in-situ geotechnical system considered. The calibrated parameters can still capture the behaviour (be it simple or complex) beyond the assimilation window, however, for system with time dependent behaviour, longer monitoring time is required to enable simple models to reasonably capture the creep settlements demonstrating that a simple model would not always be sufficiently robust for modelling alternate scenarios that substantially change the complex systems behaviour.