A Bayesian clustered-multilevel updating for local undrained shear strength prediction of fine-grained soils

Undrained shear strength is a key geotechnical parameter for site-specific preliminary foundation design and slope stability analysis of fine-grained soils. However, at a specific project site, it is common to deal with sparse, limited, and incomplete direct observations of undrained shear strength. It is then common practice to rely on indirect methods provided by well-established generic correlations or by computing qualitative multivariate similarities with worldwide sites. Drawbacks of such approaches (e.g., introduction of high systematic uncertainties and problem of pertinence for the local project site) are well known but seldom assessed due to several sources and type of uncertainties involved. In this context, a multivariate cluster-based hierarchical Bayesian model, based on site-clustering, is proposed and applied for local undrained shear strength predictions to improve accuracy and precision of local geotechnical parameter prediction. The approach allows to integrate the abundant generic sites information available in literature under the concept of multimodal hierarchicity. Novelty of the work consists of eliciting and quantified site-clustering criteria according to analyst`s experience and knowledge which varies depending on site-specific design purpose.