Bayesian Regression Models for Predicting Undrained Shear Strength from Piezocone Penetration Tests

This paper presents Bayesian regression models for predicting undrained shear strength from piezocone penetration test (CPTU) measurements. A database of laboratory tests from high-quality block samples in clays and the corresponding CPTU measurements are re-analyzed. Attention is paid to reporting and communicating the results of regression analyses, particularly statistical parameter uncertainty, a task that is often performed poorly in geotechnical engineering literature. Four regression models are fitted to the data, and some preliminary goodness-of-fit checks are discussed. Furthermore, the predictive accuracy of the models is compared. The models are then used in an example of predicting undrained shear strength from CPTU data at a new location. It is discussed how such probabilistic predictions relate to the notion of characteristic values prescribed in geotechnical design standards. Finally, updating regression models based on site-specific data is briefly discussed.

Keywords: Undrained shear strength, CPTU, Bayesian data analysis, uncertainty quantification.