Abstract:
Ground motion models (GMMs) are a critical component of seismic hazard analysis. Ergodic assumption is employed by typical GMMs for ground motion prediction using a number of physically meaningful functions and parameters for source, path and site effects. If the parameters are identical, the prediction by these models would neglect the spatial variation of source characteristics, propagation media quality, and site conditions. This simplification can lead to significant biases for a given "source-path-site" system and become one of the main reasons limiting prediction accuracy. To address this limitation, this study further introduces spatial modeling techniques based on the typical ground motion prediction equations. The spatial variation of the attenuation in path effects is evaluated by cutting the study area into geographic cells. The spatial variation of source effects, path effects, and site effects is modeled as Gaussian stochastic processes and an exponential kernel function is selected to characterize the spatial variation discipline. This framework is capable of predicting ground motions for specific source-path-site system. Using 5957 ground motion recordings from the shallow crustal earthquakes in Japanese subduction zone, a nonergodic ground motion model is developed. Bayesian inference is employed to estimate the non-ergodic ground motion terms. The results show that incorporating nonergodic effects allows the model to more accurately capture the spatially non-uniform variation character of source effect, ground motion attenuation and site effect. Compared to ergodic models, the nonergodic model achieves significant reductions in between-event and between-site standard deviations of up to 40%, and a maximum reduction in total standard deviation of 26%. In regions with sparse recordings, the epistemic uncertainty remains relatively large. These findings provide valuable insights for the development of nonergodic ground motion models in China.