Conveners: Annemarie Baltay (USGS ESC) and Rachel Abercrombie (Boston University)
Dates: November 4, 2021 (online)
SCEC Award and Report: 21114

The motivation for the SCEC Technical Activity Group (TAG) for Community Stress Drop Validation Study is focused on understanding the nature and causes of discrepancies in earthquake stress drop, as well as where random and physical variability arises. We observe differences in stress drop estimates from different researchers, even when applying the same underlying method. We also observe differences between methods and data selections, even when applied by the same researcher. These differences are often larger than the calculated uncertainties, making it hard and confusing to use these measurements for ground motion prediction and to study earthquake source physics.

In this context, the main goals for the TAG are to use a common data set of records from the 2019 Ridgecrest earthquake sequence (consisting of over 12,000 events of M1 to M7.1) to address the questions: 

• How do differing methods and model assumptions affect stress drop estimates? How do different researchers approach similar methods? 
• How do data quantity, quality, selection and processing affect stress drop estimates? 
• How do physical source (mechanism, depth, radiation pattern, directivity), path (geometrical spreading, attenuation), and site (soil conditions, site attenuation) features affect the estimates?
• What measurements, and uncertainties, would be most useful for the broader community?

WORKSHOP OVERVIEW: Workshop #1 “Stress Drop Validation – Planning and Preliminary Results” was held as a Zoom meeting on November 4, 2021. Anyone currently working on, or interested in, the science questions noted above and/or learning about Ridgecrest earthquake studies was encouraged to apply to participate, especially researchers and young investigators. We received 132 registrations, and 101 participants from 14 countries on 5 continents (see Workshop Participants below). We asked each registrant to fill out a survey indicating their interest in stress drop, what they hoped to get out of the workshop, if they planned to use the common data set to estimate stress drop, if they intended to submit a SCEC proposal under the TAG, and other demographic information. We solicited preliminary results estimating stress drop using the common dataset, and used those results in the workshop to start discussion (more details on those actual results below). As one overarching goal for the workshop was simply to facilitate communication between different research groups and disciplines, and to build (and maintain) project momentum, we experimented with formats in different sessions, built in lots of time for discussion and breakout rooms, and had fewer long talks. We also worked to ensure a diverse group of speakers and moderators, prioritizing early-career researchers and gender equity. The decision to include both observational seismologists making measurements, and users of those measurements was clearly successful with both groups contributing and learning from one another. The workshop also proved beneficial to the many students who participated, looking to understand the problems and meet the community as they begin work in this field.

Figure 1. Initial submitted stress drop results from 11 research groups (legend), vs catalog magnitude. Generalized Inversion Technique (GIT)/ Spectral Decomposition (SD) methods are shown with diamond symbols; empirical Green’s function (eGf) spectral ratio methods shown in dots; eGf finite fault in stars; and other methods in triangles

PRELIMINARY RESULTS: The common dataset was made available in late spring 2021. Prior to the workshop, all researchers were invited to analyze the data and provide estimates of stress drop, corner frequency, or other similar source parameters. Approximately one month prior to the workshop, participants were asked in with registration form if they plan supply stress drop estimates and those who indicated “yes” or “maybe” and provided them with a template Excel worksheet to submit their results by October 25, 2021.

Initial results were received from 11 research groups (and included 14 methods) – many more than expected. TAG leaders Abercrombie and Baltay made some initial comparisons and analysis of the submitted results. Not surprising, direct comparison of the reported stress drops reveals considerable scatter, with perhaps some stronger correlation between results using similar methods (Figure 1).

Figure 2. Initial submitted results for 22 events which were analyzed by 8 or more (out of 14) methods, plotted as estimated corner frequency vs catalog magnitude. Solid black dots are the log10 averaged across all methods; colors show the different estimates. Some events (smallest ML) show relatively less variability as compared to some of the larger events.

This initial analysis was intended to guide the TAG activities over the next year. One idea was to identify a subset of events to focus on, to narrow the scope from the 12,000+ events in the common data set. To that end, a set of 22 events were identified for which stress drop was estimated by 8 or more of the 14 submitted initial results (Figure 2). Some events have less variability between estimates (e.g., the smallest magnitude event) while others have larger variation between methods.

WORKSHOP PROCEEDINGS: The workshop was divided into three sessions, each with a specific focus and somewhat different format (see Agenda below).

Session 1 started with a welcome from Annemarie Baltay and a data overview from Taka’aki Taira. Rachel Abercrombie gave an overview of the TAG’s goals in light of recent published work and framing the problem. Research groups that submitted initial results presented their methods in a series of lightning talks, divided into four method types: (1) Spectral Decomposition/Generalized Inversion - Inverting a set of spectra simultaneously for source, site, and path effects, then model fitting source spectra; (2) Spectral Ratios/eGf frequency domain – Co-located eGf event used to isolate source from path and site effects, then model fitting source spectra: frequency domain; (3) Source time functions/eGf time domain – Co-located eGf event used to isolate source from path and site effects, then model fitting source time functions: time domain; and, (4) Other – Arias intensity; filtered amplitudes, single station. Rachel Abercrombie then showed some comparisons of the submitted preliminary results for discussion.

Session 2 kicked off with a keynote talk by Colin Pennington entitled Comparisons of Stress Drop Results: a Focus on the Prague, OK Sequence. This work, where Pennington used different methods to estimate stress drop in comparison with some published studies, got the group thinking about how to proceed with the Ridgecrest stress drop validation study. Workshop participants self-selected into two groups to focus discussions on the main methods: spectral decomposition and spectral ratios. Afterwards, participants separated into 12 smaller groups to discuss “What can we measure reliably, and to what extent?”. They were reshuffled again into 12 different groups to answer, “As a community, what should we focus on for the validation study?” Each breakout discussion was led by discussion leaders who captured comments and themes in a shared document, later used by Baltay and Abercrombie for future planning. Session 2 closed with a discussion with all participants exploring what we can and should measure and the priorities and targeted activities for the next year. Key themes from this session, which were later honed in session 3, included: (a) a need to identify other parameters and information to provide during results submission; (b) identification of a subset of events for focused study; (c) further collaboration between different groups using similar approaches to guide more critical analysis choices; (d) potential for a related simulation study using synthetically generated records (so we know the “real” stress drop); and (e) need for more frequent meetings to coordinate progress.

Session 3 focused jointly on the needs and priorities of “users” of stress drop estimates, and looking forward to the next year of the TAG. The session began with four short presentations on the topics of (1) Stress Drop in Ground Motion presented by Behzad Hassani; (2) Static Stress Drop for Scaling Relations in Seismic Hazard by Bruce Shaw; (3) Stress Drop in Dynamic Rupture Simulations by Ruth Harris; and (4) Stress Drop in Source Physics by Valere Lambert. These talks generated excitement and stimulated questions about the different applications of stress drop, leading to four breakout discussions hosted by the session presenters and assigned moderators. The breakout discussions explored the applications of stress drop and also future directions for the TAG. We then regrouped in the main session room and continued developing ideas for future directions.

DISCUSSION TOPICS: The various methods or approaches and themes discussed can be expanded on in future meetings. These include:

• Focus on magnitude range, etc that we are confident in. Determine a subset of events for all to focus on.
• What frequencies are being fit? More information needs to be reported.
• What can be said about events that “failed” a method?
• Consistency of terminology (e.g., eGf vs. constraint function or empirical correction spectrum)
• How to best report uncertainties? Sources of uncertainty: model selection, inversion, data, different records, different eGf, different assumptions, etc.  and real variability in earthquake recordings
• Most consistent results are seen for events in the 3-7km range. How can we address any depth dependencies in the methods?
• Considerable discrepancies exist in the moment and magnitude estimates. How are people defining sheer modulus, radiation pattern, etc?
• Tradeoffs between corner frequency and moment, source and path effects
• Should we also compare our site and path corrections?
• What are the assumptions between methods that estimate corner frequency versus those that go directly to stress drop? What about the difference between Brune and Madariaga stress drops?
• Effect of different eGfs or reference/constraint events
• Can we use synthetically (dynamic rupture) generated earthquake-station records to use in the analysis, with known stress drops?
• Improve link between estimated stress drops here and those used in dynamic rupture simulations, etc.
• Can consistent quality control and quality assurance help?
• Source complexity not fit by the Brune model – How is that treated in different methods? Should we expect agreement between methods for events that are “simple” (aka more Brune-like)?

FUTURE DIRECTIONS: The group collectively decided on several thrusts going forward, generally divided into scientific and organizational directions.

Scientific Directions

• Continue analysis of stress drop. Researchers who submitted preliminary results will refine their results based on feedback from the workshop. Those not able to submit results in November will work towards submitting results for Workshop #2. Many researchers became aware of others using similar and different methods, and have already reached out to initiate collaborations.
• Determine a subset of ~50 events for all groups to focus on. These events will include the 22 already identified (of approximately M3-4 already analyzed by 8 or more methods) and augment with some larger and smaller events to ensure an even distribution of hypocentral depths, recording stations, etc. Participating groups that only analyzed a smaller number of events will be encouraged to prioritize these  to ease comparisons.
• Consider generating synthetic earthquake records, with known stress drop, to use in the inversions.

Organizational Directions

• Convene two workshops in 2022 (one virtual and  one in person, preferably associated with the SCEC Annual Meeting).
• Host monthly or bi-monthly topical meetings (via Zoom) on methods, user needs, and data processing starting January 2022.
• Notify PIs Baltay and Abercrombie if any group  plans to submit SCEC Proposal under the TAG umbrella.

PARTICIPANT FEEDBACK: Following the workshop, the TAG leaders Baltay and Abercrombie gathered feedback from presenters and moderators on the workshop content and format to see what worked best. People were unanimously happy and excited about the workshop, felt they learned a lot, and were able to connect with many colleagues, new and old. Everyone enjoyed the lengthy discussion times and breakout rooms. It was noted that two of the 12 breakout rooms had too few people, making discussion somewhat awkward. In the future, we will ensure breakout rooms have more evenly distributed participants to facilitate interactions.

WORKSHOP AGENDA
All times below are Pacific Daylight Time (PDT or UTC-7). Presentation videos and slides 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.

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.