Partially Non-Ergodic Ground-Motion Model for Subduction Regions using the NGA Subduction Database

This report presents a summary of the development, evaluation, and comparison of a new subduc-tion ground-motion model (GMM), now known as Kuehn-Bozorgnia-Campbell-Gregor (KBCG20) model. This GMM was developed as part of the Next Generation Attenuation for Subduction Regions (NGA-Sub) program using a comprehensive compilation of subduction interface and in-traslab ground-motion recordings and metadata compiled in the NGA-Sub database. The KBCG20 model includes ground-motion scaling terms for magnitude, distance, site amplification, and basin amplification. Some of these terms are adjustable to accommodate differences between interface and intraslab earthquakes, and differences among seven subduction-zone regions for which data were compiled as part of the NGA-Sub program. These regions include Alaska (AK), Central America and Mexico (CAM), Cascadia (CASC), Japan (JP), New Zealand (NZ), South America (SA), and Taiwan (TW). Some of these regions are further divided into sub-regions to account for differences in anelastic attenuation between the subduction forearc and backarc, and differ-ences in breakpoint magnitude (the magnitude at which magnitude scaling rate decreases) between segments of a larger subduction zone. This study uses an innovative Bayesian regression approach to incorporate informative prior distributions of model coefficients and formally estimate the uncertainty in their posterior esti-mates. The posterior distributions of coefficients together with their co-variance matrix can be used to estimate epistemic uncertainty in the median ground-motion predictions for a given earth-quake scenario. Partial non-ergodicity was achieved by accounting for the regional differences in overall amplitude (constants) of prediction, anelastic attenuation, linear site amplification, and basin amplification. Because of the expanded database and innovative regression approach that includes median, aleatory variability, and epistemic uncertainty models, this new GMM represents a significant improvement in the understanding and prediction of subduction ground motion. Fur-thermore, the Bayesian approach used to develop the model will facilitate update of this innovative GMM as new data become available.