Probabilistic prediction of soil parameters based on multi-source geotechnical test data

To address the challenge of determining reliable soil parameters in cases with limited underwater geotechnical investigation data, a database involving ten soil parameters is established. Based on a hierarchical Bayesian model (HBM), quantitative information is learned and extracted from the abundant data in the database. By incorporating the constructed database with the limited test data from the target site, a quasi-site-specific multivariate probability distribution of soil parameters is constructed to improve the prediction accuracy of unknown soil properties for the target site. The effects of the quantity and types of data on soil parameter estimation are also investigated. The results indicate that HBM can provide the statistical characteristics and probability distribution of the target soil parameters, and accurately estimate them even with limited and incomplete geotechnical data from the target site. Only a small amount of data is required to effectively reduce estimation uncertainty. Moreover, using the compressibility coefficient, compression modulus, liquid limit, and plastic limit to predict cohesion and friction angle can effectively reduce prediction uncertainty.