Bayesian Model Updating and Model Selection of Ground Motion Models for Crustal Earthquakes in Colombia

Ground motion models (GMMs) are fundamental for probabilistic seismic hazard analysis and probabilistic seismic risk analysis. This study presents the first Bayesian analysis of the performance of the GMM by Arteta et al. (2023) relative to global GMMs widely used in Colombia. This GMM was not utilized within the framework of Colombia's National Seismic Risk Model. This study provides a justification for its inclusion in a future update of the MNRS, since it demonstrates capabilities similar to those of the global models frequently used in Colombia under the ergodic assumption. First, three global models were regionalized using Bayesian model updating (BMU) with a regional strong‐motion database. This allowed us to obtain probability distributions for each regression coefficient of the GMMs. Next, the out‐of‐sample predictive capability of the regionalized models was compared using the Widely Applicable Information Criterion, as it is a fully Bayesian method that utilizes the full posterior distributions of each coefficient obtained in BMU. This study demonstrates how the probability distributions of GMMs can be used to make predictions for specific scenarios while exploring epistemic uncertainty more effectively than traditional residual analysis. The whole set of GMMs used in this study achieved similar performance for spectral periods between 0.01 and 5 s. These results provide a basis for assigning weights in the logic tree for future updates of Colombian seismic hazard and risk models.