Bayesian hierarchical and measurement uncertainty model building for liquefaction triggering assessment

Abstract This study examines the details of creating and validating an empirical liquefaction model, using a worldwide cone penetration test (CPT) liquefaction database with the intent of incorporating the rigor found in predictive modeling in other fields and addressing shortcomings of existing models. Our study implements a logistic regression within a Bayesian measurement error framework to incorporate uncertainty in predictor variables and allow for a probabilistic interpretation of model parameters when making future predictions. The model is built using a hierarchal approach to account for intra-event correlation in loading variables and differences in event sample sizes. The model is tested using an independent set of recent case histories. We found that the Bayesian measurement error model considering two predictor variables, normalized CPT tip resistance and cyclic stress ratio decreased model uncertainty while maintaining predictive utility for new data. Hierarchical models revealed high model uncertainty potentially due to the database lacking in high loading non-liquefaction sites. Models considering friction ratio as a predictor variable performed worse than the two variable case and will require more data or informative priors to be adequately estimated. The framework developed is flexible and can be extended using different methods of predictor variable selection, model function forms, and validation processes.