Reappraisal of reliability of a slope through hybridisation of regional and site-specific soil shear strength information

Uncertainty in soil parameters is usually characterised by probability density functions (PDFs), with the influence on system performance represented through the probability of failure. Difficulties in selecting representative PDFs for a project often arise from scarcity of site-specific information, even with ample previous knowledge and test data of similar soil types in the region. This paper proposes an approach to rationally assimilate regional and site-specific information. A newly-compiled regional database of shear strength information for saprolitic soils in Hong Kong is presented, based on results of multi-stage consolidated-undrained triaxial tests. A hierarchical Bayesian model is fitted to the regional database, followed by a Bayesian updating model that produces posterior predictive distributions of shear strength parameters. The posterior estimates incorporate site-specific features into regional information, leading to profound impacts on the evaluation of failure probability for a slope case. To further illustrate the significance of data hybridisation, four semi-hypothetical scenarios are created using the same slope geometry, by assuming that distributions of shear strength parameters are completely known at the site. With the proposed approach, the estimated failure probability approaches the true value with increasing amount of site-specific data, and is more robust than adopting regional data or site data alone.