A framework for specifying prior distributions for the parameters of CPT-based empirical equations

Bayesian data analysis has been praised as a suitable approach for analysing geotechnical data, partly due to the possibility of including information other than (usually limited) project-specific data in the form of prior distributions. However, the geotechnical literature on specifying informative prior distributions is relatively thin. With the increasing application of the cone penetration test (CPT) across large offshore projects, an application that could benefit from such prior distributions is establishing CPT-based empirical equations for predicting soil strength/stiffness parameters for site characterisation. That is, at early stages of site investigation with no or limited data, prior predictions can be used; as the site investigation advances, the predictions can be updated considering newly obtained site-specific data. This paper discusses hierarchical Bayesian modelling (HBM) as a meta-analysis approach for specifying prior distributions for parameters of such empirical equations from existing databases. Using a CPT and laboratory test database compiled for statistical analysis of such data, it is demonstrated why the Hierarchical Bayesian modelling approach is necessary to obtain realistic statistical representations of the database, and subsequently representative prior distributions that can be used as starting points for future projects.