Item

Abstract

Current non-ergodic ground-motion models for path effects only capture the effects due to anelastic attenuation (δP2P_Q); however, at longer periods (T >1s), the majority of path effects is due to the 3D velocity structure. Using a Gaussian processes (GP) regression, Sung et al. (2023) found systematic path effects for T=3 sec spectral acceleration due to the 3D velocity of up to 0.5 natural log units.For modeling non-ergodic path terms using GP, a key issue is how to measure the difference between two ray paths for input into Gaussian process models. Sung et al. (2023) parameterized the path differences by the vector sum of the distance between the two source locations and the distance between two the site locations. However, this metric does not capture the main differences between ray paths pairs. We develop a new metric for measuring ray paths differences based on difference in azimuth, rupture distance, and site locations. This metric has a better physical basis, giving us more confidence in application to scenarios outside the range of empirical data. Compared to non-ergodic ground-motion models that do not include the path effect due to the 3D velocity structure, the aleatory variability is reduced from about 0.5 to 0.4 ln units. Additionally, we combine our model with efficient numerical methods (SKIP) that can accommodate large datasets with 100,000 of ground motions. Using SKIP, we can condition forward predictions of path effects at thousands of new source-site locations within a minute using traditional laptop computers with standard memory.