Home  /  SCEC Workshops  /  SCEC Workshop: Multi-scale seismic velocity models—Imaging and validation studies

SCEC Workshop: Multi-scale seismic velocity models—Imaging and validation studies

Conveners: Carl Tape, Cliff Thurber, and Yehuda Ben-Zion

SCEC workshops were held in October 2018 and September 2019 on the future development of SCEC Community Velocity Models (CVMs). The goals of those workshops were to: (1) assess the current status of the CVMs for California; (2) define key information needed to improve and/or develop CVMs; (3) propose viable workflow strategies for CVM development. The third in this series of SCEC CVM workshops was held virtually on September 1, 2020, with 63 participants. The main goals of the workshop were to inform the community interested in the improvement and/or use of the SCEC CVMs of recent developments and state-of-the-art methodologies relevant to the CVMs, and to identify key research areas that should be incorporated into the 2021 SCEC Science Plan and/or the next SCEC earthquake science center proposal to NSF.

The workshop conveners, Yehuda Ben-Zion, Carl Tape, and Cliff Thurber, organized the workshop into four sessions:

  • Session 1: Multiscale models - estimation and merging
  • Session 2: Regional models
  • Session 3: Local and specialized models and shallow crust
  • Session 4: Future research directions

The sessions consisted of two to five pre-recorded presentations, during which a chat window was available for asking questions and raising issues.  Each presentation was followed by a question and answer period, and each session concluded with an open discussion. The sessions had a moderator to control the flow of the session and a reporter to capture the key points of the presentations and subsequent discussion. This report is based largely on the notes provided by the reporters.

Session 1: Multiscale models - estimation and merging

This session was focused primarily on methodology. The main topics discussed during this session were: (1) model resolution and uncertainty quantification, (2) embedding of localized higher-resolution models into CVMs, (3) model validation, (4) and choice of model parameterization. For all, there appear to be multiple choices, and no generally agreed upon answers, particularly given finite resources and different application interests.  A. Fichtner (ETH- Zurich) presented practical ideas for estimating resolution and iteratively improving a large-scale model efficiently.  D.-P. van Herwaarden (ETH-Zurich) described the parameterization philosophy of the Collaborative Seismic Earth Model using basis functions, which allows model refinement via the addition of new basis functions combined with the stochastic gradient descent method to update the model. Y. Capdeville (CNRS France) discussed the use of homogenization for transforming very high-resolution models into a lower resolution version that still captures the necessary effect on the seismic wavefield up to a specified frequency. R. Ajala (LSU) presented an approach for smoothly blending a new higher-resolution localized model into the regional-scale CVMs.  Muir (Caltech) described a 'level set' inversion approach for imaging basins by including the boundary position as part of the inverse problem.

Some of the points raised during the discussion included distinguishing issues related to the forward problem (simulation) versus the inverse problem (tomography), the important role of basins and their boundaries and the geotechnical layer, the gap between arrival-time tomography and waveform tomography, how to reconcile overlapping models, and the limitations of homogenization. There was perhaps general agreement that future CVMs should work towards (1) making uncertainties clearer, (2) allowing flexibility in the use of embedded smaller-scale models, (3) doing 'smarter' smoothing or 'blending' when merging models, (4) validating models with independent datasets, and (5) respecting a priori/known geologic constraints better. Questions arose in discussing all of these points regarding the practicality of these challenging goals, and what is feasible during SCEC5 versus longer-term goals. Some specific questions included: What are the goals of the CVMs - prediction of only bandlimited seismic data, or something more general? What data are needed to validate models, particularly to be independent of the data used in model construction?  Do we have the computational power to do the desired calculations, especially resolution and uncertainty quantification? How do we best incorporate multiple sources of information into CVMs?

Session 2: Regional models 

This session focused on the well-established CVM-H models, regional models obtained with new methods and/or new combinations of data, and model validation. The main topics discussed during this session were: (1) incorporation of constraints from boreholes and the SCEC Community Fault Model on the internal structure of basins and defining the basin boundaries, (2) methods for forward simulations and tomographic inversions applied to southern and northern California, and (3) comparisons of models and model validation. A. Plesch (Harvard) described the approach in creating CVM-H models and the plan to make the basins modular so that new basin models could be tested readily. H. Fang (MIT) presented a flexible, stochastic approach to model parameterization and tomographic inversion to overcome the issue of bias from subjectively choosing a particular model parameterization and applied it to southern California. A. Rodgers (LLNL) described improved forward simulation capabilities and the assessment of models for the San Francisco Bay Area, as well as preliminary efforts on model updating using long-period data. Y. Lu (USC) presented work on validation of multiple southern California models, emphasizing the strong effect of the near-surface structure on simulated waveforms even at longer periods.

Some of the recurring themes in the discussion were model uncertainties, model merging, the importance of the near-surface and geotechnical layer, basin geometries and their representation or preservation in new/updated models, the effect of sharp basin boundaries on waveforms, the push to higher frequencies, and the potential value of homogenization to facilitate extending modeling to higher frequencies.

Session 3: Local and specialized models and the shallow crust

This session focused on high-resolution models, the structure of the shallow crust from various methods, and the effect of shallow structure on seismic waveforms. The main topics discussed during this session were: (1) new velocity models for different regions obtained with different tomographic methods, (2) the effect of shallow structure on wave incidence angles and on the seismic wavefield at different frequencies, and (3) ways to validate model results. M. White (USC) described the rapid estimation of a tomographic model for the Ridgecrest region using an automated arrival-time picker and H. Fang's stochastic tomography method. E. Berg (Utah) presented a tomographic model for southern California that used a variety of data types covering a broad range of frequencies to illuminate both the very shallow and deeper structure. S. Park (Caltech/U Chicago) presented a method to use wave arrival polarization (i.e., incidence angle) to infer the near-surface velocity. A. Juarez (USC) described observations and simulations of wave propagation that demonstrated the strong effect of the near-surface structure on the seismic waves, even at relatively long periods (e.g., 10 s), including significant temporal changes.

The main recurring themes in the discussion were validation and testing, from validating features in a 3D velocity model via independent information or simulation to testing unusual distributions of relocated earthquakes using double-difference to validating arrival polarization with dense array data to testing the relative waveform sensitivity to 3D structure versus the near-surface, and testing the trade-off effects between shallow and deep structure on the wavefield.  Some questions and issues raised included how can fine-scale resolution of the near-surface be obtained over broad regions, how topography is treated in various approaches and how significant is it to the simulations or tomographic modeling, the spatial scale of a model needed for reliable wavefield results, the potential of homogenization to aid wavefield modeling and inversion, and the importance of scattering.

Session 4: Future research directions 

This session featured one presentation on efficient storage and access of large models and one on a complete workflow for tomographic updating of a velocity model. B. Aagaard (USGS) described his "GeoModelGrids" scheme for efficient model storage and an accompanying application programming interface (API) for efficient model access. Some aspects of the scheme are a regular "logical grid" that is mapped to a physical grid (with topography) which is stored in a self-describing HDF5 file format. B. Chow (Victoria U. of Wellington) presented an automated adjoint tomography workflow and demonstrated its application to New Zealand's North Island. The workflow is open source, with mostly Python components, and it can run fully automated after an initial set-up.

After a brief question and answer period on the two talks, a wide-ranging discussion ensued. Topics included the shallow structure, heterogeneity length scales, homogenization and effective medium approaches, model merging, model evaluation and validation, and uncertainty quantification.  A scheme for model delivery, documentation, and open archiving is also important. The collection of more data, especially densely sampled data, is essential for making significant progress on improving the CVMs, especially in terms of their representation of the shallow structure. All of these aspects of CVM development need to be incorporated in the 2021 SCEC Science Plan and/or the next Earthquake Science Center proposal.

Workshop Agenda

Presentation videos may be viewed by clicking the links below. PLEASE NOTE: Files are the author’s property. They may contain unpublished or preliminary information and should only be used while viewing the talk. Only the presentations for which SCEC has received permission to post publicly are included below.

08:00 - 08:10 Workshop, Introduction, Goals (PDF, 800KB) Cliff Thurber
08:10 - 10:15 Session 1: Multiscale models—estimation and merging
Moderator: Carl Tape    Reporter: Victor Tsai
08:10 - 08:30 Practical and emerging methods for resolution analysis and uncertainty quantification (VIDEO) Andreas Fichtner
08:30 - 08:50 The Collaborative Seismic Earth Model: Bridging the scale gap from local to global scale seismology (VIDEO) Dirk-Philip van Herwaarden
08:50 - 09:10 Combining elastic models of different scales: a homogenization perspective (VIDEO) Yann Capdeville
09:10 - 09:30 Discussion All
09:30 - 09:45 UCVM development and low-frequency validation of hybrid Earth models near the southern San Andreas fault (VIDEO) Rasheed Ajala
09:45 -10:00 A level-set approach to parsimoniously updating the SCEC CVMs (VIDEO) Jack Muir
10:00 - 10:15 Discussion All
10:15 - 10:45 Break  
10:45 - 12:00 Session 2: Regional models
Moderator: Christine Goulet    Reporter: Pieter-Ewald Share
10:45 - 11:00 Modular sedimentary basin models in California to support next generation CVM developments (VIDEO) Andreas Plesch
11:00 - 11:15 Regional seismic velocity models for Southern California based on travel time tomography with Poisson Voronoi cells parameterization (VIDEO) Hongjian Fang
11:15 - 11:30 Earthquake ground motion simulations and 3D Earth models in northern California and beyond Artie Rodgers
11:30 - 11:45 Validation of seismic velocity models in southern California with full-waveform simulations (VIDEO) Yang Lu
11:45 - 12:00 Discussion All
12:00 - 13:30 Lunch break (and viewing of pre-recorded talks)  
  Transfer-Learning Aided Meter-Scale Double-Difference Tomography Chengping Chai
  Updates on SCEC Unified Community Velocity Model (UCVM) Software and Future Directions (VIDEO) Phil Maechling
  Autocorrelations and Surface-Wave Reflections (VIDEO) Robert Clayton
13:30 - 15:30 Session 3: Local and specialized models and shallow crust
Moderator: Cliff Thurber      Reporter: Ben Heath
13:30 - 13:45 Seismic velocity structure in the Ridgecrest, CA region from local earthquake traveltime tomography (VIDEO) Malcolm White
13:45 - 14:00 Joint inversion of Rayleigh wave phase velocity, ellipticity, and receiver functions: Expanding to the shallower and deeper Crust (VIDEO) Elizabeth Berg
14:00 - 14:15 Body-wave polarization for constraining near-surface structure (VIDEO) Sunny Park
14:15 - 14:30 Effects of shallow velocity reductions on 3D propagation of seismic waves Alan Juarez
14:30 - 14:45 Discussion All
14:45 - 15:30 Group Discussion (on Sessions 1-3 and Lightning Talks)  
15:30 - 16:00 Break  
16:00 - 17:00 Session 4: Future research directions
Moderator: Yehuda Ben-Zion     Reporter: Cliff Thurber
16:00 - 16:10 Geomodelgrids: A storage scheme and API for multiscale, rasterized Earth models Brad Aagaard
16:10 - 16:20 An automated workflow for adjoint tomography applied to New Zealand's North Island (VIDEO) Bryant Chow
16:20 - 17:00 Discussion  
   - Review action items from previous workshops Cliff Thurber
   - Model archival and dissemination (e.g., IRIS EMC) All
   - Validation: benchmark data sets and CME machinery All
   - FWI platform for improvements at long length scales (>1s period) All
   - Strategy for developing multi-scale models needed for high-frequency simulations All
   - Future research plans for CVM TAG Carl Tape / Cliff Thurber
17:00 Adjourn  

Workshop Participants


The Southern California Earthquake Center is committed to providing a safe, productive, and welcoming environment for all participants. We take pride in fostering a diverse and inclusive SCEC community, and therefore expect all participants to abide by the SCEC Activities Code of Conduct.