A data-driven, multi-scale sediment velocity model for Southern California

Yi Liu, Grigorios Lavrentiadis, & Domniki Asimaki

In Preparation 2026, SCEC Contribution #15499

We develop a Bayesian Gaussian process (GP) model to characterize the near-surface shear-wave velocity in the Los Angeles Basin. The proposed approach represents the residuals relative to CVM-S4.26 as a conditional random field and integrates both stationary and spatially varying kernel components to simultaneously capture localized geological variability effects and large-scale basin structure. Shear-wave velocity measurements from geotechnical site investigations and sonic log testing are jointly incorporated to constrain both shallow and deep velocity structures. A two-step training strategy was adopted to improve optimization stability and parameter interpretability. The trained GP model was subsequently applied to representative geological cross sections across the Los Angeles Basin.

Key Words
Community Velocity Model, Near-surface Data-driven Velocity Model, Los Angeles Basin, Gaussian Proces

Citation
Liu, Y., Lavrentiadis, G., & Asimaki, D. (2026). A data-driven, multi-scale sediment velocity model for Southern California. Journal of Geophysical Research: Solid Earth, (in preparation).

Related Projects & Working Groups
Ground Motions, Seismology, Community Earth Models (CEM)