Bayesian inference for coping with geotechnical model uncertainty: a unified HP2O framework

Geotechnical model uncertainty (GMU) is unavoidable due to model idealizations and assumptions. Incorporating GMU into engineering-decision making can be formally achieved under a Bayesian framework. There are a variety of pathways for Bayesian inference considering GMU as well as uncertainties associated with input parameters of geotechnical calculation models, providing, potentially, different inference results given the same priori and observation. The differences arise from different probabilistic modelling choices, formulations of likelihood function, and data utilisation. However, few studies offer explicit explanations of these differences, impeding thorough understanding of results obtained from different Bayesian methods considering GMU and their applications. This is due, at least partially, to the absence of a unified interpretation framework for existing Bayesian inference pathways considering GMU. This study aims to develop a unified framework, namely HP2O, for bookkeeping these pathways and revisiting them from a theoretically rigorous and coherent perspective. Three pathways of Bayesian inference are considered under the proposed framework. It is shown that HP2O enables rational explanations and improved understanding of Bayesian inference results across different pathways from a unified viewpoint, and, hence, allows for their coherent comparison. The findings were demonstrated using a deep excavation example together with a well-studied semi-empirical model.