Modular Filter-based Approach to Ground Motion Attenuation Modeling

Abstract

Previous-generation ground-motion attenuation relations ( e.g. , Abrahamson and Silva 1997; Boore et al. 1997; Campbell 1997; Sadigh et al. 1997) were designed to predict ground motion at distances less than 100 km from an earthquake fault. As probabilistic and deterministic seismic hazard analysis concepts have become more prevalent in performance-based seismic design, these relations have been used in practice much beyond their distance limitation. Realizing the engineering needs for predicting ground motions at distances farther than 100 km, the Next Generation Attenuation (NGA) relations project targeted a distance range up to 200 km (Power et al. 2006). For seismic hazard analysis in the western United States (WUS), 200 km is a sufficiently large distance to quantify design-basis hazard level from known active faults due to relatively fast attenuation of ground motion. For the central and eastern United States (CEUS), ground motion attenuates more slowly, yet design of critical infrastructures in the CEUS ( e.g. , nuclear power plants) requires computation of seismic hazard from distant large events. Thus, the recently initiated NGA-East project for CEUS set a goal to design ground-motion attenuation models applicable to 1,000 km. The objective of this article is to revisit the commonly used empirical approach to ground-motion attenuation modeling and compare it with an alternative modular filter-based approach that can be effectively used for predicting ground motion at near- ( 100 km). In this latter approach, each filter is calibrated separately to represent a certain physical phenomenon affecting seismic radiation from the source. We demonstrate that the modular filter-based approach provides accurate (that is, expected median prediction without significant bias) and efficient (that is, relatively small standard error of prediction) predictions. We also present our peak ground acceleration (PGA) based predictive model for …