SCEC Award Number 24116 View PDF
Proposal Category Individual Research Project (Single Investigator / Institution)
Proposal Title A Statewide Sediment Velocity Model: Development and Implementation of a Conditional Random Field
Investigator(s)
Name Organization
Domniki Asimaki California Institute of Technology
SCEC Milestones C1,2,3-1, A2-1, D3-2 SCEC Groups GM, Seismology, CEM
Report Due Date 03/15/2025 Date Report Submitted 07/27/2025
Project Abstract
 We present a non-parametric, data-driven near-surface velocity model for CA that can be used to populate the basin structures of the velocity models. The CA\_SVM trained here for Southern California, is developed as a conditional random field of uncertain trend function and uncertain fluctuations about the trend of the residuals relative to the SCEC CVM-S4.26, using Gaussian process regression with the superposition of multiple Gaussian random fields. In the first part of the project, we aggregated geotechnical and geophysical data from Southern California in a 1D model with hyperparameters that we calibrated using a Bayesian estimation; and in the second part, we extended the aforementioned 1D model to 3D by means of the \emph{separability} assumption, namely separating the fluctuations in the horizontal and vertical directions; and successively separating the latter into stationary and spatially varying kernels. In the future, we plan to integrate datasets with higher spatial resolution such as DAS inversion data to better capture the spatially varying components of the model, and work in collaboration with the SCEC IT personnel to implement the Statewide SVM on UCVM for future testing of basin effects.
SCEC Community Models Used Community Velocity Model (CVM)
Usage Description The model is developed as a GP of the residuals relative to the CVMS 4.26 in the near surface.
Intellectual Merit The model contributes by providing a data driven approach to refining the shallow crustal models of UCVM, in the case CVMS 4.26. Better shallow crustal models in the basins of Los Angeles, where this model has been trained, mean improved simulated ground motions and thus improved simulated hazard assessments especially in higher frequencies. As more data become available, especially local data like geotechnical boreholes, the model does not need to be retrained, it only needs to be conditioned on the new datasets, which makes incorporating the new data very efficient.
Broader Impacts Improved predictions of ground motions in the high frequencies means improved hazard assessment and more effective risk mitigation. The engineering community will increasingly use the SCEC products as we improve the accuracy of simulated ground motions through improved resolution of shallow crustal velocity models.
Project Participants Yi Liu (Caltech graduate student)
Grigorios (Greg) Lavrentiadis (Caltech postdoc)
DOMNIKI ASIMAKI (PI)
Andreas Plesch (Harvard, data sharing of sonic logs).
Exemplary Figure Figure 13: {top) Gaussian Process data-driven sediment velocity model (this work); (middle) tapered GTL model (Olsen and collaborators, personal communication); (bottom) UCVM 4.26 model for a cross section of the Los Angeles basin.

Yi Liu, Grigorios Lavrentiadis, DOMNIKI ASIMAKI
Linked Publications

Add missing publication or edit citation shown. Enter the SCEC project ID to link publication.