Abstract
In this study, we propose a data-driven method using the Deviance Information Criterion (DIC) to select the most suitable earthquake ground motion model (GMM) for application in probabilistic seismic hazard analysis (PSHA). The standard deviation (sigma) of the GMM is an important parameter for PSHA and plays an important role in data-driven methods. The main advantage of the proposed procedure is to introduce the posterior sigma as the key quantity for objectively ranking different candidate models against a given earthquake ground motion dataset. In the context of the Bayesian statistical framework, sigma is then determined for a given GMM based on the observed ground motions and at the same time takes into account the misfit of the GMM predictions to the observed ground motions. This feature addresses issues associated with other ranking methods where in some cases a considerable bias between the GMM predictions and the observed ground motions is effectively ignored. On the contrary, the DIC considers the influence of these two factors together by ranking models more favorably when they are associated with smaller bias and the determined sigma is close to the actual variability of the ground motions in the region under study. We submit the DIC method of this study as a useful and objective method for evaluating the performance of a GMM to a given dataset. This has potentially important application for PSHA when using multiple GMMs and either logic tree or backbone approaches are required to handle epistemic uncertainty in an appropriate manner.
•A new data-driven method using the deviance information criterion (DIC) is proposed.•The main advantage of the proposed procedure is to introduce posterior sigma.•The method optimizes selection of GMMs using the Bayesian statistical framework.•DIC is adjusted for posterior inference based on Markov chain Monte Carlo algorithm.