← Back to Sessions

Ergodic and Non-Ergodic Ground-Motion Models for Small Magnitude Earthquakes in the San Francisco Region

Description: 

An important issue in the propagation of seismic waves at high frequencies into numerical simulations is the lack of knowledge of the 3-D velocity structure at short wavelengths. Empirical ground-motion data include the effects of the 3-D crustal structure at high frequencies, but to have enough observations, record- ings from small-magnitude events are used. To use the small-magnitude data to evaluate path effects requires a reference ergodic ground-motion model (GMM) to remove the average scaling and a non-ergodic GMM to identify regions with strong path effects.

In this study, we focus on the small magnitude earthquakes in the San Francisco region within 50km of the Hayward fault. From all the recordings available in the NCEDC in that region, only the recordings with a signal-to-noise ratio S/N > 3 are selected. The resulting dataset includes about 5,000 recordings from 346 events recorded at 151 stations, with magnitude 1.5 to 4 at rupture dis- tance Rrup < 100km. From this dataset, an ergodic GMM is developed, which includes the traditional magnitude, distance and depth scaling, and the source and site random effects. We also include a novel distance-dependent random effect on the source which removes the path effects that are sometimes mapped in the source random effect in traditional analysis. From the ergodic residu- als, a non-ergodic GMM is developed which captures the spatial variability of the source, site and path effects. The coordinate-specific source, site and paths effects show spatial correlation patterns which can be accurately modeled via Gaussian Processes (GP) that allow us to extrapolate them in space at unobserved source and site locations conditionally to observations and under physical constraints. As a result, the standard deviation of the residuals at high frequencies is reduced by a factor of 2 from the ergodic GMM compared to the non-ergodic GMM (1.1 to 0.52 at 15Hz). Going forward, the residuals of the ergodic GMM can be used as inputs to machine learning to develop non-ergodic path effects at high frequencies that can be compared to the results of the non-ergodic GMM based on GP.

Session: How Well Can We Predict Broadband Site-Specific Ground Motion and Its Spatial Variability So Far? - II

Type: Oral

Date: 5/1/2024

Presentation Time: 11:15 AM (local time)

Presenting Author: Maxime

Student Presenter: No

Invited Presentation: 

Authors