New Advances in CyberShake PSHA Models

Scott Callaghan, Philip J. Maechling, Christine A. Goulet, Kevin R. Milner, Mei-Hui Su, Robert W. Graves, Kim B. Olsen, Yifeng Cui, Brad T. Aagaard, Kathryn E. Wooddell, Albert R. Kottke, Bruce E. Shaw, & Thomas H. Jordan

Submitted August 15, 2019, SCEC Contribution #9717, 2019 SCEC Annual Meeting Poster #297 (PDF)

Poster Image: 
SCEC has developed the CyberShake software platform to perform 3D physics-based probabilistic seismic hazard analysis (PSHA). To reduce computational cost, CyberShake uses reciprocity: Strain Green Tensors are convolved with slip time histories for hundreds of thousands of individual events from an earthquake rupture forecast (ERF). Synthetic seismograms are post-processed to obtain intensity measures, which are combined with event probabilities to produce hazard curves. CyberShake PSHA results from hundreds of locations are then interpolated to produce a regional hazard map. In the last year, we completed a study for a new region and developed a version based on earthquake simulators.

We expanded CyberShake to a large Northern California region including the San Francisco Bay Area. PSHA calculations up to 1 Hz were performed for 869 locations as part of CyberShake Study 18.8, running for 128 days on the NCSA Blue Waters and OLCF Titan systems. To support simulation volumes that included most of California, we used UCVM (Small et al. 2017) to tile three separate 3D community velocity models into a composite statewide model and applied smoothing along interfaces to minimize unrealistic reflections and refractions. To improve the representation of the near-surface velocity structure in tomographically-derived models, we inserted a geotechnical layer (GTL) in the top 500 meters by applying the Ely (2010) method on Vs30 values from the Wills et al. (2015) map.

We find that results from CyberShake Study 18.8 differ from those obtained with the NGA-West2 ground motion models. For example, CyberShake shows higher hazard in the southern San Joaquin Valley, especially at longer periods, likely due to basin amplification from the representation of the valley in the velocity model. Hazard estimates are also higher compared to those obtained in CyberShake Study 17.3, likely because Study 17.3 used a higher minimum Vs and no GTL.

We also performed CyberShake hazard calculations using an alternative ERF. The RSQSim earthquake simulator was used to generate synthetic earthquake catalogs for California. Events with M≥6.5 are selected from the catalog to create an extended ERF and used to perform CyberShake hazard calculations for multiple sites, creating the first fully physics-based hazard curves for Southern California. We find that RSQSim CyberShake curves generally show close agreement to those using UCERF2, suggesting this approach has promise for future work.

Key Words
seismic hazard analysis, workflows, high performance computing

Citation
Callaghan, S., Maechling, P. J., Goulet, C. A., Milner, K. R., Su, M., Graves, R. W., Olsen, K. B., Cui, Y., Aagaard, B. T., Wooddell, K. E., Kottke, A. R., Shaw, B. E., & Jordan, T. H. (2019, 08). New Advances in CyberShake PSHA Models. Poster Presentation at 2019 SCEC Annual Meeting.


Related Projects & Working Groups
Computational Science (CS)