SCEC Award Number 21039 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title A workflow for homogenization of seismic velocity models in Southern California
Investigator(s)
Name Organization
Yehuda Ben-Zion University of Southern California
Other Participants Yang Lu (postdoc), Mei-Hui Su (SCEC developer) and Malcolm White (student)
SCEC Priorities 3a, 4a, 4b SCEC Groups Seismology, CXM, CS
Report Due Date 03/15/2022 Date Report Submitted 04/29/2022
Project Abstract
One significant limitation of 3D wavefield simulations is modeling very low seismic velocities in the shallow subsurface. The computational requirements increase exponentially by increasing mesh resolution and modeling low material speeds. We aim to implement the 3D seismic velocity fast-Fourier homogenization (FFH) technique coupled with the SCEC Unified Community Velocity Model (UCVM; Small et al., 2017) software to facilitate efficient computations. The goal is to make easily accessible effective-medium velocity models. Toward this goal we: (1) Installed the 2D homogenization program of Capdeville et al. (2010) in a high-performance computer and ran 2D homogenizations in parallel in 8, 16, and 24 cores. To test the simulations, we used simple models, including a 1D layered model and a model of circular inclusion on a homogeneous half-space. The input models were isotropic, with three parameters characterize every point: density and P- and S-wave velocities. The number of samples necessary to compute the effective media makes the required memory too large to fit a single computer shared memory node in the RAM. Thus, multistage runs divide the domain into smaller sections that can run in parallel. (2) The outputs of FFH are the anisotropic elastic constant tensor and density properties at any domain location. These values should be adapted to the input needed to be read by the wave equation solvers. As such, we are working on installing and testing the spectral element wave-propagation software code package SEM46 to be able to run parallel wavefield simulations using effective media velocity models from FFH.
Intellectual Merit The research contributes to development of techniques for homogenizing multi-scale 3D velocity models with strong gradients near the surface and large faults. The homogenization process is an important step for developing integrated multi-scale velocity models in southern California. The developed methodology will allow rigorous down sampling of velocity models that retain the in-formation on larger scales. The homogenized version of the multi-scale models will allow efficient large-scale computations.
Broader Impacts The project contributes to the development of integrated multi-scale community velocity models. The developed workflow will be made available with easy access to SCEC researchers. The resulting velocity models can be used in many different studies by multiple SCEC groups and will contribute directly and indirectly to most SCEC5 science questions. The project contributes to the education and research experience of a PhD student.
Exemplary Figure Figure 1

Illustration of the homogenization procedure for a complex 2D velocity model [After Capdeville and Cance (2015)].