Interactive Effects of Prior Distributions and Likelihood Function on Bayesian Model Updating of Slope Reliability

The selection of the prior distribution of geotechnical parameters and likelihood function is crucial for the Bayesian model updating of slope reliability, but the rationale behind the selection requires further investigation. This paper explores several prior distributions and likelihood functions frequently used in the Bayesian model updating of geotechnical systems. Two illustrative slope examples are adopted to examine the interactive effects of prior distributions and likelihood functions on the posterior distributions of geotechnical parameters and updating of slope reliability. Scenarios involving both homogeneous and spatially variable soils are explored. The results indicate that in a homogeneous soil medium, the choice of the likelihood function exerts a more profound impact on the Bayesian model updating of slope reliability than the selection of the prior distribution. In contrast, in spatially variable soils, the choice of prior distribution has a noticeable impact on posterior distributions. However, the effects of the likelihood function on the Bayesian model updating of slope reliability depend on the choice of the prior distribution. If there is sufficient prior knowledge, the choice of the likelihood function significantly influences the Bayesian model updating. If there is insufficient prior knowledge, resulting in a prior uniform distribution with conservative bounds, the choice of the likelihood function has negligible effects.