| Abstrakt: |
The current state of the practice in probabilistic seismic hazard analysis (PSHA) employs ergodic ground motion models (GMMs), which assume that the ground motion variability observed in a global database is the same as the variability in ground motion at a single site-source combination. However, the fast-growing empirical ground motion databases indicate significant regional differences in ground motions due to repeatable and systematic source, path, and site effects. These systematic effects, which are spatially correlated, are not consistent with the ergodic assumption, promoting the transition to non-ergodic GMMs for PSHA. In this study, we use Gaussian processes with different covariance functions to model the spatial correlation structures of systematic source, path, and site effects for the Ridgecrest area. We compare the proposed correlation models for Ridgecrest with those previously developed for the ANZA array in terms of predictive performance. We also evaluate the effects of the cell-specific attenuation approach on the spatial correlation structures of path effects. We find that the spatial correlation of systematic source and path effects is best characterized by anisotropic non-stationary covariance functions in Gaussian processes. We also find that the cell-specific attenuation approach with squared grids has limitations in predicting path effects and does not affect the correlation structures significantly for the Ridgecrest database. Finally, comparisons with the ANZA array suggest that the spatial correlation structures of path effects derived from the ANZA array may be transferable to the Ridgecrest area, potentially due to their similarity in crustal heterogeneity. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink