A coupled probabilistic site characterization method for estimating soil stratification and spatial variability using multiple-source site investigation data

Probabilistic site characterizations, which primarily involve delineating soil stratification and assessing the spatial variability of soil properties, is crucial for geotechnical reliability analysis and risk assessment. Geotechnical site investigation typically produces sparse, multiple-source and multiple-type data. However, most current site characterization methods can only handle a single type of data such as cone penetration test (CPT) data and address the stratification uncertainty and the spatial variability of soil properties separately. To overcome these limitations, this study proposes a coupled probabilistic site characterization method that integrates the Markov random field and Gibbs sampler methods to simultaneously quantify both types of uncertainties. The Markov random field (MRF) is used to estimate the distribution of the soil stratification while the Gibbs sampler is used to construct the probability density function (PDF) of soil properties for each soil stratum. The Bayesian method is employed to account for the interaction between soil stratification and the spatial variability of soil properties. The proposed method is demonstrated through application to a virtual site and a real project site in Hong Kong. Results show that the proposed method effectively leverages limited multi-source site investigation data, considering the interactions between stratigraphic uncertainty and the spatial variability of soil properties. It delivers higher prediction accuracy of soil stratification compared to the Markov random field alone and models the spatial variability of soil properties more effectively.