Simulation-based Seismic Hazard Analysis Using CyberShake SCEC CyberShake Collaboration: Robert Graves, Scott Callaghan, Feng Wang, Thomas H. Jordan, Philip Maechling, Kim Olsen, Kevin Milner, En-Jui Lee, Po Chen, Edward Field What is CyberShake? CyberShake Results Next Steps
Probabilistic Seismic Hazard Analysis Few data epistemic uncertainty Much scatter aleatory variability Ground Motion Prediction Equations (GMPEs) Boore et al. (1997) Earthquake Rupture Forecast Ground-Motion Prediction Eqn Intensity Measures UCERF NGA GMPEs Response Spectra
Probabilistic Seismic Hazard Analysis aleatory variability Strasser et al. (2009) Earthquake Rupture Forecast Ground-Motion Prediction Eqn Intensity Measures UCERF NGA GMPEs Response Spectra
CyberShake Goals Improve long-term seismic hazard analysis by replacing empirical ground motion prediction equations (GMPEs) with physics-based simulations - Account properly for phenomena such as rupture directivity and basin effects - Predict full time-series of ground motion rather than simple intensity measures - Requires accurate 3D models and EQ rupture characterization Extend seismic hazard analysis to account for spacetime variations in earthquake probability - Provide a computational platform for operational earthquake forecasting
CyberShake Computational Platform Scenario based seismic hazard calculation, incorporating many thousands of scenarios - 3D waveform simulations using full kinematic rupture description Simulates ground motions for potential fault ruptures within 200 km of each site - 40,000 sources (M w > 6) in from UCERF2.0 (2008) Extends UCERF2.0 to multiple hypocenters and slip models for each source - 440,000 ground motion simulations for each site
Traditional Hazard Model Earthquake Rupture Forecast Ground Motion Prediction Eqn Intensity Measures Empirical PSHA model
CyberShake Hazard Model hazard maps hazard curves seismograms Extended EFR KFR AWP NSR Ground Motion Physics-based simulations Earthquake Rupture Forecast Ground Motion Prediction Eqn Intensity Measures Empirical PSHA model KFR = kinematic fault rupture model AWP = anelastic wave propagation model NSR = nonlinear site response
Computational Efficiency of Seismic Reciprocity To account for source variability requires very large sets of simulations 440,000 rupture variations Ground motions can be calculated at much smaller number of surface sites to produce hazard map About 350 in LA region Source 3 Source 1 Receiver Source 2 Use reciprocity to compute 3D Green s functions for all potential sources at each site Use representation theorem to convolve 3D GF s with rupture variations (relatively fast computation) M sources to N receivers requires M simulations M sources to N receivers requires 2N or 3N simulations M >> N Use of reciprocity reduces CPU time by a factor of ~1,000
Comparison of 1D and 3D CyberShake Models for the Los Angeles Region BBP-1D CVM-S4.26 1 2 2 3 4 1. 2. 3. 4. lower near-fault intensities due to 3D scattering much higher intensities in near-fault basins higher intensities in the Los Angeles basins lower intensities in hard-rock areas
NGA (2008) Attenuation Relations used in National Seismic Hazard Maps Epistemic Differences CyberShake (2010) Model NGA Campbell & Bozorgnia near-fault effects NGA Chiou & Youngs basin effects NGA Boore & Atkinson SA-3s PE = 2%/50 yr UCERF2, no background seismicity NGA Abrahamson & Silva
Wang & Jordan (2014, BSSA) Averaging-Based Factorization Dependence of Directivity Effects on Rupture Complexity 0.0 0.2 0.4 0.6 0.8 35 d r,k σ d maps (SA-3s) CS11 SC08 34 8 10 15 64 85 119 118 117 GP07 used in CS11 86 87 88 89 93 Model σ d GP07 raw 0.41 GP07-SC08 0.31 112 218 219 231 232 254 255 267 271 273
Wang & Jordan (2014, BSSA) Averaging-Based Factorization Dependence of Directivity Effects on Rupture Complexity 0.0 0.2 0.4 0.6 0.8 35 d r,k σ d maps (SA-3s) CS13a SC08 34 8 10 15 64 85 119 118 117 GP10 used in CS13a 86 87 88 89 93 Model σ d GP07 raw 0.41 GP07-SC08 0.31 GP10 raw 0.26 GP10-SC08 0.17 112 218 219 231 232 254 255 267 271 273
ABF Directivity-Basin Coupling Maps (M8 source; variable hypo; SA-3s corrected for SC08 directivity) Wang & Jordan (2014) coupling point coupling point ln d(r)
CyberShake Model Evolution CS11 : CVM-S4 CS14 : CVM-S4.26 Refined Velocity Model Refined Rupture Generator CS13 : CVM-S4 Improved knowledge leads to reduction of epistemic uncertainty
CyberShake 15.4 Study Los Angeles Urban Seismic Hazard Map (fmax= 1 Hz) 2s-SA 15.4 hazard map 3s-SA 15.4/14.2 ratio map Makespan of 38 days: 642,000 node-hours on Blue Waters and 426,000 node-hours on Titan 1 Hz CyberShake requires 33 times as much computational work as at 0.5 Hz, but only 7 times as many node-hours owing to improved efficiency of the CyberShake code base
ABF Variance Analysis Reduction in siteeffect & directivityeffect variance CS11- NGA08 CS13.b- NGA08 Residual Variance site path directivity magnitude source complexity
CyberShake Products Develop Urban Seismic Hazard Maps for LA Region - Input to building code recommendations (next talk) - Compliment USGS National Maps Extend CyberShake models to 1400 sites across California Develop statewide Unified Community Velocity Model (UCVM) Couple time-dependent UCERF3 to CyberShake Provide frequently updated time-dependent seismic hazard maps