Quasi-Region-Specific Liquefaction Probability Prediction Based on a Liquefaction Database

This study aims to propose a novel model for quasi-region-specific liquefaction probability prediction. For this purpose, this study compiles a liquefaction database that contains liquefaction and nonliquefaction records from different regions. This study further develops a novel hierarchical Bayesian model (HBM) to learn the interregion and intraregion characteristics in the database, including the region-uniqueness behavior of the model factor of the liquefaction safety factor (FS) (defined as the ratio of the actual FS to the calculated FS). The trained HBM is then used to develop a prior model for the target region, and this prior model is further updated by the target-region data into a posterior (or quasi-region-specific) model. The quasi-region-specific model can predict the probability of liquefaction occurrence for future cases of the target region, which is illustrated using a real case study. The performance of the quasi-region-specific model is compared with other models using a comprehensive cross validation, which shows that the quasi-region-specific model outperforms these models in general.