Figure 1: CISM computational pathways for Probabilistic Seismic Hazard Analysis (PSHA). Pathways involving the earthquake rupture simulator RSQSim are marked in red, and those involving the ground-motion simulator CyberShake are marked in blue. The standard empirical PSHA pathway is shown in black. The CISM project has recently demonstrated the ability to execute all of the pathways shown here, including the full-physics pathway, RSQSim-CyberShake-GM (Callaghan et al., 2018).

Overview

In 2015, SCEC received a three-year grant from the W. M. Keck Foundation to construct a Collaboratory for Interseismic Simulation and Modeling (CISM) that will provide a unique environment for developing large-scale numerical models that can simulate sequences of fault ruptures and the seismic shaking they produce. The goal of CISM is to equip earthquake scientists with HPC-enabled cyberinfrastructure that creates a new generation of comprehensive, physics-based earthquake forecasts using the California fault system as their primary test bed.

CISM provides a new computational framework for earthquake forecasting (schematized in Figure 1). The standard empirical PHSA pathway is compared to the fully physics-based pathway, combining earthquake simulations, which account for the physics of earthquake nucleation and stress transfer, with ground-motion prediction models derived from simulations of seismic wave excitation and propagation. It is a workflow-oriented cyberinfrastructure with common tools for integrating various types of scientific software modules provided by different research teams into well-structured forecasting models that can be calibrated against existing data and tested against observations within the Collaboratory for the Study of Earthquake Predictability (CSEP).

Research Priorities

1. Assemble interdisciplinary teams to create system-specific models for time-dependent earthquake forecasting that are comprehensive, physics-based, data-calibrated, and prospectively testable.
2. Implement high-performance, workflow oriented cyberinfrastructure that supports research on comprehensive physics-based models by facilitating earthquake simulation and data assimilation.
3. Engineer CISM based on community data-exchange standards, equip CISM with common tools for integrating heterogeneous sets of rupture and ground-motion simulators provided by difference research teams, and couple CISM into high-performance computing environments.
4. Provide a computational environment in which scientists can develop numerical rupture and ground-motion simulators using California as the primatary test bed, and combine time-dependent rupture forecast models with ground motion prediction models to forecast earthquake exceedance probabilities at sites in California.
5. Retrospectively calibrate and prospectively evaluate CISM forecasting models in the Collaboratory for the Study of Earthquake Predictability (CSEP).

Research Accomplishments

2018 Highlights

• In a paper published in Science Advances, Shaw et al. (2018) compare time-independent hazard maps for California computed using the PSHA pathway, UCERF3→NGA-GMPE→IM, which is the state’s current operational forecast, with maps computed using the physics-based earthquake simulator, RSQSim→NGA-GMPE→IM. Both forecasts used the UCERF3 fault geometry and fault slip rates. The agreement between the physics-based model and the empirical model is surprisingly good (see Figures 7-9). The ability to replicate statistically-based seismic hazard estimates by a physics-based model cross-validates the methods, and provides a new approach for estimating seismic hazards. Figure 2 illustrates the correspondence of the simulator and the hazard models on a standard measure of shaking used in the national seismic hazard maps (the PGA 2% in 50-year exceedance, which is the peak ground acceleration expected to be exceeded at the 2% probability level in a 50-year time-period, expressed in fractions of gravitational acceleration).
  

  Figure 2: Maps of shaking hazard in earthquake simulator compared with UCERF3 hazard model for California. Maps and histogram show standard hazard measure of peak ground acceleration PGA 2% in 50yr exceedance. Because the simulator has only on-fault type events, only those types of events from the UCERF3 hazard model are included in the comparison. (a) UCERF3. (b) RSQSim model. (Shaw et al., 2018)

• The first hazard curves have been computed using from the full-physics pathway given by the upper row of Figure 1, RSQSim→CyberShake→GM (Callaghan et al., 2018; see Figure 12). This achievement is notable because it represents the first-ever PSHA calculation derived entirely from physics-based models, in this case the RSQSim rupture simulator and the CyberShake ground-motion simulator. These physics-based models have the potential to improve hazard estimates at high intensities and low probabilities, which is important in assessing the seismic risks for critical facilities. The calculation of a complete physics-based PSHA model for the Los Angeles region, which is being done on two of the nation’s largest supercomputers, NSCA Blue Waters and OLCF Titan, should be completed this fall.
• A new paper has been published describing a set of 13 PSHA models generated by the pathway UCERF2→CyberShake→GM, which now includes models for Central California as well as the Los Angeles region (Jordan et al., 2018). These results, which were presented at the 11th National Conference on Earthquake Engineering in June, 2018, are now being used by engineers to estimate the long-period hazards to tall buildings and other extended structures. The CISM group will be collaborating with the USGS and SCEC to develop CyberShake models for the San Francisco region.
• A framework for testing PSHA models that are probabilistically complete (i.e., specify epistemic uncertainties) has been developed by Marzocchi and Jordan (2017, 2018). Plans to test UCERF3 and RSQSim forecasts within the Collaboratory for the Study of Earthquake Predictability (CSEP) are being implemented.

2016 Highlights

• The earthquake simulator RSQSim was ported to NCSA’s Blue Waters supercomputer and test simulations were performed to verify the performance of RSQSim with the new California fault model which was set up using the UCERF3 fault model discretized into ~250,000, 1km2 triangular fault elements.
• A computational interface was developed between the OpenSHA software and RSQSim. This allows for direct comparisons between RSQSim simulated catalogs and the UCERF3 models, and for the coupling of RSQSim and ground-motion prediction equations to perform probabilistic seismic hazard analysis. OpenSHA is the computational platform for the UCERF3 models, and was used to calculate the California component of the 2014 National Seismic Hazard Maps produced by the USGS. These calculations include the best available Ground Motion Prediction Equations (GMPE’s) for hazard mapping. Additionally, SCEC-VDO, a 3D visualization platform, was modified to visualize the high-resolution UCERF3 fault model and animate the simulated catalogs.

2015 Highlights

• We hosted a large public project kick-off meeting at the SCEC Annual Meeting on September 13, 2015. The main outcomes of the workshop were: 1. The confirmation of RSQSim as the main earthquake simulator for the CISM software package, which was going to be used in conjunction with the Uniform California Rupture Forecast 3 (UCERF3) model for the development of long-term earthquake catalogs; and 2. The creation of topic-targeted working groups with key leaders identified by the SCEC CISM management team.

Outreach

CISM and SCEC UseIT

For three years, several CISM participants were mentors for the SCEC Undergraduate Studies in Earthquake Information Technology (UseIT) program. The Grand Challenge for the Summer 2017 interns was focused on a key CISM problem: developing earthquake forecasts and hazard maps using long-term simulated earthquake catalogs generated by RSQSim. We were awarded an NCSA Education Allocation of 25,000 node hours on the Blue Waters supercomputer for the interns to run large, RSQSim simulations. Using Blue Waters, the interns High Performance Computing (HPC) and Forecasting Teams generated several half-million-year simulated catalogs, and developed conditional probabilities of large-earthquake sequences for three initial-event scenarios for the San Andreas Fault (magnitude M6 on the Parkfield section, M7 on the Mojave section, M6 in Bombay Beach), which they compared to the probabilities given by the official UCERF3 model. The Virtual Display of Objects (VDO), and Hazard and Risk teams created animations of seismic activity, as well as maps of expected ground motions, economic losses, and human casualties for the three multi-event scenarios.

The UseIT program, which is separately funded by an NSF/CISE REU-site grant, UseIT sponsors large, diverse cohorts of undergraduates from around the country to work with CISM scientists on physics-based earthquake forecasting in California. The integration of the UseIT program into CISM has provided these undergraduate students with valuable experience in the process of real, large-scale scientific projects, as well as an understanding of earthquake hazard, earthquake physics, and the applications and capabilities of high performance computing. The 2016 and 2017 UseIT interns presented their work with RSQSim at the SCEC Annual Meetings in Palm Springs.

Select Publications

Field, E. H., & Milner, K. R. (2018). Candidate Products for Operational Earthquake Forecasting Illustrated Using the HayWired Planning Scenario, Including One Very Quick (and Not‐So‐Dirty) Hazard‐Map Option. Seismological Research Letters, 89(4), 1420-1434. doi: 10.1785/0220170241. SCEC Contribution 8033

Jordan, T. H., & Others (2017). CyberShake models of seismic hazards in Southern and Central California. Proceedings of the 11th National Conference in Earthquake Engineering, (in preparation). SCEC Contribution 7994

Marzocchi, W., & Jordan, T. H. (2017). A Unified Probabilistic Framework for Seismic Hazard Analysis. Bulletin of the Seismological Society of America, 107(6), 2738-2744. doi: 10.1785/0120170008. SCEC Contribution 7269

Marzocchi, W., & Jordan, T. H. (2018). Experimental concepts for testing probabilistic earthquake forecasting and seismic hazard models. Geophysical Journal International, 215(2), 780-798. doi: 10.1093/gji/ggy276. SCEC Contribution 8131

Roten, D., Olsen, K. B., Day, S. M., & Cui, Y. (2017). Quantification of Fault-Zone Plasticity Effects with Spontaneous Rupture Simulations. Pure and Applied Geophysics, 174(9), 3369-3391. doi: 10.1007/s00024-017-1466-5. SCEC Contribution 8863

Shaw, B. E., Milner, K. R., Field, E. H., Richards-Dinger, K., Gilchrist, J. J., Dieterich, J. H., & Jordan, T. H. (2018). A physics-based earthquake simulator replicates seismic hazard statistics across California. Science Advances, 4(8). doi: 10.1126/sciadv.aau0688. SCEC Contribution 8093

Sheng, Y., Denolle, M. A., & Beroza, G. C. (2017). Multicomponent C3 Green’s Functions for Improved Long‐Period Ground‐Motion Prediction. Bulletin of the Seismological Society of America, 107(6), 2836-2845. doi: 10.1785/0120170053. SCEC Contribution 7365

Schorlemmer, D., Werner, M. J., Marzocchi, W., Jordan, T. H., Ogata, Y., Jackson, D. D., Mak, S., Rhoades, D. A., Gerstenberger, M. C., Hirata, N., Liukis, M., Maechling, P. J., Strader, A., Taroni, M., Wiemer, S., Zechar, J. D., & Zhuang, J. (2018). The Collaboratory for the Study of Earthquake Predictability: Achievements and Priorities. Seismological Research Letters, 89(4), 1305-1313. doi: 10.1785/0220180053. SCEC Contribution 8036