Introduction

A substantial effort has gone into the development of the two "standard" SCEC CVMs for Southern California, currently CVM-H version 15.1.0 (Shaw et al., 2015) and CVM-S version 4.26 (Lee et al., 2014). Incremental improvements are being made to CVM-H over time, but CVM-S model is essentially frozen at the current version. SCEC has also developed a Central California Area CVM, current version CCA-06. SCEC's Unified CVM framework (Small et al., 2017) provides access to these models as well as other models covering different regions of California, including the San Francisco Bay Area and the entire state. To make further progress on these CVMs, open-source workflows are needed to evaluate and/or update these models, methods need to be developed to merge new high-resolution localized models into the existing models, strategies to integrate constraints from other geophysical, geotechnical, and geological sources are required, and methods for assessing model uncertainty must be developed. This TAG will facilitate research initiatives at the intersection of seismology, ground motion, and computational science. 

Research Priorities

The proposed goals for the CVM TAG are:

• Develop open methods for improving SCEC CVMs;
• Develop methods for CVM assessment and validation, including providing uncertainty estimates;
• Expand the participation of SCEC researchers in work related to improving and utilizing the CVMs.

CVM TAG Workshop

The tasks for the workshop are (1) to reach final consensus on the mission and goals statement of the TAG (a draft of which will be circulated in advance of the workshop), (2) to refine and prioritize the identified action items, and (3) to coordinate plans for submission of related SCEC 2020 proposals as well as future proposals to other funding agencies. The primary outcomes expected from the workshop are a completed mission and goals statement for the CVM TAG, including prioritized action items, and plans for submission of coordinated and collaborative proposals in response to the 2020 SCEC Science Collaboration Plan and other funding opportunities.

CVM TAG Mission and Goals - DRAFT

A Technical Activity Group for the coordination of SCEC5 research on Community Velocity Models

Mission Statement
The mission of this Technical Activity Group (TAG) is to promote and coordinate efforts to improve SCEC Community Velocity Models (CVMs).  This TAG will help galvanize efforts to develop and apply an open framework for the assessment, validation, and improvement of CVMs.

Problem Definition and Relationship to SCEC Research Priorities
A substantial effort has gone into the development of the two "standard" SCEC CVMs for Southern California, currently CVM-H version 15.1.0 (Shaw et al., 2015) and CVM-S version 4.26 (Lee et al., 2014).  Incremental improvements are being made to CVM-H over time, but CVM-S model is essentially frozen at the current version.  SCEC has also developed a Central California Area CVM, current version CCA-06. SCEC's Unified CVM framework (Small et al., 2017) provides access to these models as well as other models covering different regions of California, including the San Francisco Bay Area and the entire state.  To make further progress on these CVMs, open-source full waveform modeling and tomography (FWT) workflows are needed to evaluate and/or update these models, methods need to be developed to merge new high-resolution localized models into the existing models, strategies to integrate constraints from other geophysical, geotechnical, and geological sources are required, and methods for assessing model uncertainty must be developed.  This TAG will facilitate research initiatives at the intersection of seismology, ground motion, and computational science.
 
Within SCEC5, this TAG is complementary to - but not overlapping with - research activities carried out in the Technical Activity Group on nonlinear effects in the shallow crust.  This TAG also directly supports efforts in the following areas highlighted in the 2019 SCEC Science Plan:

Computational Science

Community Models (CXMs)

• Develop tools that can accelerate community building of new (or existing) community models.  These may include but are not limited to the design of versatile data-structures and application libraries (software elements) for data manipulation and integration, meshing and gridding, and visualization of community models.
• Develop tools that can help integrate different community models between themselves and/or with simulation software.  These may include but are not limited to the design software interfaces capable of connecting in efficient manners HPC simulation codes with community models.

Seismic wave propagation

• Develop tools and implement procedures to validate SCEC community fault and seismic velocity models as applied in inverse and forward problems.
• Develop and improve existing software tools and algorithms with application to HPC that accelerates and advances high-frequency simulation methods, while contributing to solve standing research problems as stated in the Ground Motions interdisciplinary group priorities. 

Tomography

• Develop new or adapt existing forward modeling software to solve full 3D tomography (F3DT) problems using HPC resources.  Computational Science research in this area should help diversify current dependency on existing software and provide alternatives for SGT and F3DT verification, as done in dynamic rupture and forward ground motion simulation.
• Develop efficient and sustainable computational procedures to facilitate the assimilation of regional waveform data in the SCEC community velocity models (CVMs); and integrate F3DT and inversion results into existing CVMs.

Community Models

• Community Velocity Model (CVM)
  • Integrate new data (especially from the Salton Sea Imaging Project) into the existing CVMs with validation of improvements in the CVMs for ground-motion prediction.
  • Quantify uncertainty in the CVM structure and its impact on simulated ground motions.
  • Develop efficient and sustainable computational procedures to facilitate the assimilation of regional waveform data into the CVMs. 

Community Modeling Environment

• Improve the accuracy of community velocity models (CVMs), through the development of techniques that may involve, for example, the development of 3D tomography codes as well as the integration of geology constraints into CVMs.  Proposals are also sought to improve the methodologies used for integration of models from different sources and scales within UCVM.

These are challenging tasks, but a coordinated effort by interested SCEC researchers will facilitate progress.

Technical Activity Group Research and Coordination Plan

Goals

• Develop open methods for improving SCEC CVMs;
• Develop methods for CVM assessment and validation, including providing uncertainty estimates;
• Expand the participation of SCEC researchers in work related to improving and utilizing the CVMs.

Research Thrust Areas

The October 2018 workshop, "The Next Leap Forward for SCEC CVMs," consisted of three CVM review presentations (Plesch on CVM-H, Goulet on CVM-S, and Maechling on the UCVM framework), 13 "lightning" talks on new data sources, new tomography methods, model validation and uncertainty, and computational aspects, a plenary discussion on the lightning talk topics, and a closing discussion on what tasks need to be done and how to accomplish them.  The group identified a number of action items, which are listed here ordered by my judgment on priority.  The first set of action items are proposed to be completed by the end of SCEC5:

• Develop end-to-end full 3D tomography (F3DT) code and workflow
• Develop an approach and tools for integrating new models into current models and examining quality of the modified model for validation (through data analyses and 3D visualization of model properties)
• Develop strategies for dealing with topography in creating, modifying, or comparing models
• Make detailed comparisons between the current CVM-S and CVM-H models (velocity values, depths to key isosurfaces, power spectra, etc.)
• Develop and share SCEC ambient noise Green's function results (from SCEC’s Stanford and UW-Madison teams and other sources) 
• Relocate earthquakes in the current CVM-S and CVM-H models for Community Fault Model assessment
• Submit decimated versions of the current CVM-S and CVM-H models to the IRIS EMC

The second set of action items are somewhat lower in priority and may be initiated but likely not completed during SCEC5:

• Develop approaches for assessing model uncertainty
• Explore approaches for determining near-surface structure and fine-scale heterogeneity
• Explore strategies for imbedding high-resolution near-surface structure and fault zone models into CVMs
• Pursue the potential of joint geophysical inversions to improve CVMs
• Request suite of real and synthetic data used and developed by Taborda et al. (2016)
• Establish libraries of (1) data used to develop SCEC CVMs and (2) real and synthetic Green's functions
• Incorporate Salton Sea experiment data into next round of model updates
• Work to increase availability of continuous strong motion data 

These action items will be refined and priorities will be revised at the TAG workshop.  The expectation is that a number of inter-related and coordinated proposals will subsequently be submitted to SCEC and other funding agencies (USGS, NSF, and others) to carry out research directly targeting these action items.

Improving SCEC CVMs

A key distinction between CVM-H and CVM-S is that the former incorporates topography whereas the latter treats the Earth as flat.  Given that topography affects wave propagation, and other considerations, updating CVM-S to include topography is an essential goal.  Topography should be an integral part of the forward and inverse modeling codes that form an open-source workflow for CVM updating.

In principle, the same workflow for updating CVM-S could be applied to CVM-H, but in practice, a key concern is assuring that an updated model adheres as closely as possible to "hard" constraints on the structure, such as from borehole and reflection data.  Some potential approaches for this include explicitly keeping parts of the model fixed or adopting a Bayesian strategy with tight a priori probability distributions for model areas with "hard" constraints and broader a priori probability distributions elsewhere.  Similarly, it is not clear how one best imbeds a local higher-resolution model, for example obtained from a dense array study, into the larger scale CVMs.  Another issue with CVM-H is that it is fundamentally a Vp model, with Vs obtained via the empirical relations of Brocher (2005).  In contrast, the CVM-S Vs model is probably better constrained than the Vp model because of the preponderance of relatively long period waveform data used to develop the current CVM-S model.
 
CVM Assessment, Validation, and Uncertainty

Determining the value, accuracy, and uncertainty of a CVM is an underexplored avenue of research.  These issues are all inter-related.  Over what frequency range can a CVM be used to predict desired observables, and to what accuracy?  What are reasonable estimates for the uncertainty of CVM parameter values?  How well are the models resolved?  For tomographic models derived from body-wave arrival time data, the forward and inverse problems are sufficiently simple (assuming path lengths and data frequency range are such that finite-frequency effects can be ignored) that reasonable estimates of model parameter uncertainty and resolution can be obtained.
 
Expand Participation in CVM Research

Development of an open-source work flow is the primary key to opening up opportunities for expanded participation in CVM development.  Strategies for merging high-resolution local models into the larger-scale CVMs are also needed in order for imaging results from new dense array studies to be integrated into the CVMs.  Once these capabilities are developed and tested, any SCEC researcher would be able to pursue work related to the CVMs.  One could also then contemplate setting up a SCEC summer intern program centered on the CVMs.

Milestones (2019-2021)

Year 4

• Make detailed comparisons between the current CVM-S and CVM-H models.
• Develop and share SCEC ambient noise Green's function results.
• Relocate earthquakes in the current CVM-S and CVM-H models for Community Fault Model assessment.
• Submit decimated versions of the current CVM-S and CVM-H models to the IRIS EMC.

Year 5

• Develop end-to-end full 3D tomography (F3DT) code and workflow.
• Develop an approach and tools for integrating new models into current models and examining the quality of the modified model for validation.
• Develop strategies for dealing with topography in creating, modifying, or comparing models.

References

Brocher, T. M., (2005), Empirical relations between elastic wavespeeds and density in the Earth's crust, Bull. Seism. Soc. Am., 95, 2081-2092.
Lee, E.-J., P. Chen, T. H. Jordan, P. B. Maechling, M. A. M. Denolle, and G. C. Beroza (2014), Full-3-D tomography for crustal structure in Southern California based on the scattering-integral and the adjoint-wavefield methods, J. Geophys. Res. Solid Earth, 119, 6421-6451, doi:10.1002/2014JB011346.
Shaw, J. H., A. Plesch, C. Tape, M. P. Suess, T. H. Jordan, , G. Ely, E .Hauksson, J. Tromp, T. Tanimoto, R. Graves, K. Olsen, C. Nicholson, P. J. Maechling, C. Rivero, P. Lovely, C. M. Brankman, and J. Munster (2015), Unified Structural Representation of the southern California crust and upper mantle, Earth Planet. Sci. Lett., 415, 1-15, doi: 10.1016/j.epsl.2015.01.016.
Small, P., Gill, D., Maechling, P. J., Taborda, R., Callaghan, S., Jordan, T. H., Ely, G. P., Olsen, K. B., & Goulet, C. A. (2017). The SCEC Unified Community Velocity Model Software Framework. Seism. Research. Lett., 88, 1539-1552, doi:10.1785/0220170082.
Taborda, R., S. Azizzadeh-Roodpish, N. Khoshnevis, and K. Cheng (2016), Evaluation of the southern California seismic velocity models through simulation of recorded events, Geophys. J. Int., 205, 1342-1364, doi:10.1093/gji/ggw085.