Bayesian approach for probabilistic site characterization assimilating borehole experiments and Cone Penetration Tests

• Data classification of compression modulus: direct data and indirect data.
• Maximum entropy theory depicts the local uncertainty of indirect data.
• Bayesian algorithm combining geostatistical interpolation to provide posterior prediction at unsampled location.

This study investigates a Bayesian approach for assimilating data from borehole experiments and Cone Penetration Tests (CPTs) for site characterization (e.g., soil compression modulus). The spatial variability of compression modulus is depicted by random field theory. A Bayesian inverse modeling method is established by combining prior information, in the form of empirical knowledge, with observations, in the form of borehole experiments and in-situ CPT profiles to calculate posterior estimates of compression modulus at unsampled locations. The approach classifies all relevant data as either direct or indirect, and uses geostatistical tools in a Bayesian algorithm to estimate the posteriors. The local uncertainty of indirect data is analyzed using maximum entropy theory and the Bayesian algorithm simulates the posterior prediction of unsampled locations given direct data, plus integrating out the local uncertainties on joint Probability Density Function (PDF) of indirect data. To validate its effectiveness, Bayesian inverse modeling method is applied to a shallow load bearing soil layer located at the National Convention and Exhibition Center in Shanghai. Spatial estimates are refined in comparison to those from Kriging method. It is concluded that the above approach appears promising as a framework for site characterization given multiple types of observation.