CyberShake: bringing physics-based PSHA to central California

Scott Callaghan, Philip J. Maechling, Christine A. Goulet, Kevin R. Milner, Robert W. Graves, Kim B. Olsen, & Thomas H. Jordan

Published August 15, 2017, SCEC Contribution #7725, 2017 SCEC Annual Meeting Poster #303 (PDF)

Poster Image: 
The SCEC CyberShake computational platform is a collection of scientific software integrated with workflow tools which performs physics-based probabilistic seismic hazard analysis (PSHA) using 3D deterministic wave propagation simulations. CyberShake performs PSHA by first populating a velocity mesh using SCEC community velocity models, via UCVM queries. This mesh is input to the AWP-ODC-SGT code, which simulates a wavefield of Strain Green Tensors. An earthquake rupture forecast (ERF) is extended by varying hypocenters and slips on finite faults, generating about 500,000 events per site. Seismic reciprocity is used to calculate synthetic seismograms, which are processed to obtain intensity measures (IMs) such as RotD100 and duration metrics such as Arias duration. The IMs are combined with ERF probabilities to produce hazard curves. PSHA results from hundreds of locations across a region are interpolated to produce a hazard map.

In 2017, SCEC performed CyberShake Study 17.3, expanding into Central California for the first time. Seismic hazard calculations were performed at 1 Hz at 438 sites, using both a 3D tomographically-derived central California velocity model (CCA-06) and a regionally averaged 1D model (CCA-1D). For simulation volumes extending outside of Central California, we included other SCEC velocity models and developed a smoothing algorithm to minimize reflection/refraction effects along interfaces. CyberShake Study 17.3 ran for 31 days on NCSA's Blue Waters and ORNL's Titan supercomputers, burning 21.6 million core-hours, producing 285 million two-component seismograms and 43 billion IMs, and utilizing end-to-end CyberShake workflows on Titan for the first time.
Our results demonstrate that CyberShake can be successfully expanded into new regions, and lend insights into the effects of directivity-basin coupling associated with basins near major faults such as the San Andreas. They are especially informative about epistemic uncertainties in basin effects, which are not well parameterized by depths to iso-velocity surfaces, a common input to GMPEs. In particular, we observe in the 3D results that basin amplification for sites in the southern San Joaquin Valley is less than for sites in smaller basins such as around Ventura.
We will present CyberShake hazard estimates from the 1D and 3D models, compare results to those from previous CyberShake studies and GMPEs, discuss our new workflow capability on Titan, and describe our future plans.

Key Words
probabilistic seismic hazard analysis, high performance computing, 3D simulations, scientific workflows

Citation
Callaghan, S., Maechling, P. J., Goulet, C. A., Milner, K. R., Graves, R. W., Olsen, K. B., & Jordan, T. H. (2017, 08). CyberShake: bringing physics-based PSHA to central California. Poster Presentation at 2017 SCEC Annual Meeting.


Related Projects & Working Groups
Community Modeling Environment (CME)