Bayesian Approach for Probabilistic Site Characterization Using Cone Penetration Tests

This paper develops a Bayesian approach for probabilistic site characterization (i.e., on both stratigraphy and soil properties) using cone penetration tests (CPTs). The available site information prior to the project (e.g., existing geological maps, geotechnical reports, and local experience) is used in the Bayesian approach as prior knowledge, and it is integrated systematically with results of CPTs that are performed deliberately for the project. The inherent spatial variability of soil is modeled explicitly by random field theory. The proposed approach contains two major components: a Bayesian model class selection method to identify the most probable number of statistically homogenous layers of soil and a Bayesian system identification method to estimate the most probable layer thicknesses and soil properties probabilistically. Equations are derived for the Bayesian approach, and the proposed approach is illustrated using a set of real CPT data obtained from a site in Netherlands. It has been shown that the proposed approach correctly identifies the number and thicknesses/boundaries of the statistically homogenous layers of soil and provides proper probabilistic characterization of soil properties. The Bayesian approach provides a means to identify the statistically homogenous layers progressively by gradually zooming into local differences with improved resolution, and it also contains a mechanism to determine when to stop such zooming. In addition, a sensitivity study is performed to explore the effect of prior knowledge. © 2013 American Society of Civil Engineers.