SCEC Award Number 23191 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Combining Multi-resolution Velocity Models with a Focus on the Ridgecrest Region using Machine Learning
Investigator(s)
Name Organization
Peter Gerstoft University of California, San Diego Kim Olsen San Diego State University
Other Participants Graduate student Zheng Zhou (funded in previous proposal)
SCEC Priorities 4b, 4d, 3d SCEC Groups Seismology, CS, CXM
Report Due Date 03/15/2024 Date Report Submitted 03/18/2024
Project Abstract
 Inspired by the progress in image editing and medical tomography fusion (James et al., 2014), we introduce a seismic tomography model fusion technique, which enhances the local detail structures and simultaneously preserves global smoothness in the combined model. Combining the physics-informed mechanism and the Markov random field model, we propose a probability graphical model (PGM) which captures the relation between subdomains with multiple resolutions, in terms of high-resolution (HR) and low-resolution (LR). By transferring the information from the HR regions, the details in the LR areas are enhanced by solving a maximum likelihood problem with prior knowledge from the HR areas. Evaluation tests on both checkerboard and a real fault zone model derived from the 2019 Ridgecrest, CA, earthquake demonstrate its efficacy.

Intellectual Merit Inspired by the progress in image editing and medical tomography fusion (James et al., 2014), we introduce a seismic tomography model fusion technique, which enhances the local detail structures and simultaneously preserves global smoothness in the combined model. Combining the physics-informed mechanism and the Markov random field model, we propose a probability graphical model (PGM) which captures the relation between subdomains with multiple resolutions, in terms of high-resolution (HR) and low-resolution (LR). By transferring the information from the HR regions, the details in the LR areas are enhanced by solving a maximum likelihood problem with prior knowledge from the HR areas. Evaluation tests on both checkerboard and a real fault zone model derived from the 2019 Ridgecrest, CA, earthquake demonstrate its efficacy.

Broader Impacts funded a graduate student
Exemplary Figure
Figure 1: (a) Station locations (triangles) and main faults (lines) surrounding the Ridgecrest area.
There are six dense sensor arrays across the main faults (A1-2 and B1-4). (b) Vertical cross-
sections of the shear wave velocity along the B1-4 station arrays from (top) surface wave dispersion inversion, (center) the 3D fusion model from dictionary learning, and (bottom) the PGM.