Deterministic Generation of Broadband Ground Motions! with Simulations of Dynamic Ruptures on Rough Faults! for Physics-Based Seismic Hazard Analysis Zheqiang Shi and Steven M. Day! Department of Geological Sciences San Diego State University SCEC Rupture Dynamics Code Validation Workshop March 14, 214
Introduction Improving earthquake ground-motion estimates is a key step in seismic hazard assessment that aims to reduce earthquake impact, especially for regions with high seismic risk such as southern California. Three essential elements in making accurate ground motion predictions: (1) Accurate earthquake source description (2) Accurate structure model (3) Accurate wave propagation calculation (attenuation) Several hybrid approaches have been developed to simulate broadband ground motions, in which LF (<1 Hz) is generated deterministically (kinematic/dynamic rupture) which HF (>1 Hz) is generated stochastically [e.g., Liu et al., 26; Graves and Pitarka, 21; Mai et al., 21]. We developed a fully deterministic approach to generate broadband ground motions up to ~1 Hz by accounting for fault roughness in fully 3-D dynamic rupture simulations. UCERF3!2
Introduction Geometrical Complexities at all Scales SCEC CVM v4 Candela et al. (29) kilometer scale meter scale millimeter scale Effects of fault roughness (geometric irregularities) (1) Static stress distribution (EQ nucleation) (2) Dynamic rupture process (EQ propagation and termination) (3) High-frequency radiations (strong ground motion) (4) Heterogeneous distribution of fault slip and slip rates!3
Introduction Strike-Slip planar free surface ( x 2 = ) km km rough fault surface x 1 x 2 nucleation center at (x 1, x 2 ) = ( km, 12 km) Shi and Day (213) Dip-Slip x 2 x 1 Dip hanging-wall station footwall station Rake W L hypocenter λmin: minimum roughness wavelength α: amplitude-to-wavelength ratio!4
Rough-fault Simulations of Strike-Slip (SS) Events km rough fault surface planar free surface ( x 2 = ) km SORD Δx ~ 2 m ~9.4 billion cells 2-sec rupture simulation ~ hours on 16,384 cores self-similar roughness λ min =8m x 1 x 2 nucleation center at (x 1, x 2 ) = ( km, 12 km) Shi and Day (213) On the fault: Rate-and-state friction Off the fault: Drucker-Pager Plasticity RSD Parameter a, b RSD Parameter v w..1.1.2.2 2. 1. 1. 3 3 4 Depth (km) 6 8 1 12 a b v w 1 2 2 1 14 16 18 2!
(SS) Ground Motions x3 (km) 4 Fault-Parallel a (m/s ) 6 3 3 4 4 Fourier Spectral Amplitude (m/s) 1 1 1 1 1 1 2 (a) fault-parallel fault-normal vertical 1 1 1 1 1 Frequency (Hz) 2 x 1 = +9 km = +9 km = +7 km = + km = +3 km = +1 km = 1 km x (km) Fault-Parallel.21.31.31.7.73.4 Fault-Normal.16.21.21.48 1.3 1.48 Shi and Day (213) Vertical.22.28.27.32.49.4 Normalized Fourier Spectral Amplitude 1 1 1 fault-normal (b) elastic plastic x3 (km) 1 1 1 1 x 1 (km) 1 1 2 3 4 fmax = 3 km.36 = km.31 = 7 km.34 = 9 km.29 2 3 4 6 7 8 9 1 Time (sec) 1.8.63.79.42.46.26.31.24 2 3 4 6 7 8 9 1 2 3 4 6 7 8 9 1 Time (sec) Time (sec) 1 1 1 1 1 Frequency (Hz) 2!7
(SS) Effects of Medium Scattering 3 KM WITH WITHOUT 21 KM 21 KM 3 KM Hypocenter Hypocenter 1... m/s -.2..2 m/s 1... m/s -.2..2 m/s Cui et al. (213)!8
(SS) Ground Motions - PGAs Peak Ground Accelerations x3 (km) PGA [g] 1 (a).8 1.7.6..4.3 1 1 fault-parallel.2 1 1 1 1 2 2 3 3 4 4 PGA [g] 1 (b) 1.6 1 1.4 1.2 1.8.6 1.4.2 fault-normal 1 1 1 1 1 2 2 3 3 4 4 x3 (km) 1 1 (c) PGA [g] 1. 1.4.3.2.1 1 1 vertical geometric mean 1 1 1 1 1 1 2 2 3 3 4 4 1 1 1 1 2 2 3 3 4 4 x 1 (km) x 1 (km) (d) PGA [g].9.8.7.6..4.3.2.1 Shi and Day (213) *Without off-fault plasticity: PGAs have similar pattern but on average ~% larger in magnitude!9
(SS) Ground Motions - Attenuation Site-Corrected GMRotD Response Spectra Compared to the 28 Next Generation Attenuation (NGA) GMPE Curves Site Amplification: SH plane-wave response of the generic rock structures representative of western North America rock sites [Boore and Joyner, 1997] Site Attenuation: e πκf with site an elastic loss exponent (defined by Anderson and Hough [1984]) κ =.4 sec Half-Space Model Spectral Acceleration [g] Spectral Acceleration [g] 1 1 1 1 1 1 Simulation GMPE + τ GMPE medians Period =.1 sec CY AS BA CB Period = 1. sec Period =.3 sec Period = 3. sec 1 1 1 1 1 1 1 1 1 2 1 1 1 2 Distance (R jb ) (km) Distance (R jb ) (km) Shi and Day (213) Mw = 7.23 CY: Choi and Young (28) BA: Boore and Atkinson (28) AS: Abrahamson and Silva (28) CB: Campbell and Bozognia (28)!1
Rough-fault Simulations of Dip-Slip (DS) Events x 2 x 1 Dip hanging-wall station footwall station self-similar roughness λ min =8m Rake W L hypocenter Northridge EQ Geometry Parameters θ = 4º λ = 1º W = 24.9 km L = 2 km Ztop = km Znucl = 17.47 km W = 24.9 km W L = 2 km!11
Rough-fault Simulations of Dip-Slip (DS) Events x 2 x 1 Dip hanging-wall station footwall station self-similar roughness λ min =8m Rake W L hypocenter On the fault: Linear Slip-Weakening Off the fault: Drucker-Pager Plasticity f 3 3 s 2 d Δx ~ 2 m ~14.3 billion cells 2-sec rupture simulation ~4 hours on 32,768 cores Dc 1 2 1!12
Rough-fault Simulations of Dip-Slip (DS) Events 1 1. 11 1 9. 9 9. 8. 9 8 6. 1 7. 7 6 6. 2 7 6. 6. 4 4. 4. 2 4 3. 6 3 2. 1. 2 1. 3 6.. 4. 3 1 1 2 ) Mw = 6.74!14
Rough-fault Simulations of Dip-Slip (DS) Events!1
(DS) Ground Motions!16
(DS) Ground Motions - PGAs 3 2 Strike Parallel PGA [g].4.4 3 2 Strike Normal PGA [g] 1.2 2.3 2 1 1.3 1.8 (km) 1.2.2 (km) 1.6.1.4 1.1. 1.2 1 2 3 x 1 (km) 1 2 3 x 1 (km) (km) 3 2 2 1 1 1 Vertical 1 2 3 x 1 (km) PGA [g] 1.6 1.4 1.2 1.8.6.4.2 (km) 3 2 2 1 1 1 Geometric Mean 1 2 3 x 1 (km) PGA [g]..4.4.3.3.2.2.1.1.!17
(DS) Ground Motions x 1 = 3 km.12.6.1.23.18.22.23.16.33 Fourier Spectral Amplitude (m/s) 1 1 1 1 1 strikeparallel strikenormal vertical.24.24.21.14.1.24.4.33.24.16.2.33.18.1.11 1 2 1 1 1 1 1 Frequency (Hz).8.14.11 6 8 1 12 14 16 6 8 1 12 14 16 6 8 1 12 14 16 Time (s) Time (s) Time (s )!18
(DS) Ground Motions - Hanging-wall Effect Hanging-wall Effect: increase in ground motion observed for sites located in close proximity to the rupture plane on the HW side compared to sites situated on the FW side. 1 T =.1 sec T =.3 sec T = 1. sec T = 3. sec GMRotD 3 2 2 1 Spectral Acceleration [g] 1 1 (km) 1 1 2 3 2 2 1 1 1 1 Distance (Rx) (km) 1 1 1 1 2 2 3 x 1 (km) Rx: horizontal distance from the top edge of the rupture measured perpendicular to the fault strike. Depth (km)!19
(DS) Ground Motions - Hanging-wall Effect T =.1 sec T =.3 sec T = 1. sec T = 3. sec 1 1 Distance (Rx) (km) Hanging-wall Effect: increase in ground motion observed for sites located in close proximity to the rupture plane on the HW side compared to sites situated on the FW side. 1 2 3 2 2 1 1 1 1 Spectral Acceleration [g] Depth (km) 1 Donahue and Abrahamson (213) Rx: horizontal distance from the top edge of the rupture measured perpendicular to the fault strike.!2
Discussion We can successfully simulate broadband ground motions (frequencies up to ~1 Hz) deterministically over a large spatial domain (fault length ~1 km) by including fractal fault roughness (from 1 2 m to 1 m) in our 3-D dynamic rupture model. The rupture irregularities along the rough fault lead to ground motions of considerable complexities with ground accelerations showing extensive high-frequency content. In the case of strike-slip event, characteristics of site-averaged synthetic GMRotD response spectra, including the distance and period dependence of the median values, absolute level and intra-event standard deviation, are generally comparable to appropriate empirical estimates (when station effects have been removed), throughout the period range.1-3. sec. In the case of dip-slip event, the general shape of the distance dependence of the synthetic GMRotD agrees well to the current GMPEs at the period of ~.2 sec. In addition, we observed interesting period-dependence of the HW effect not constrained in the current GMPEs. Simulation data from these fully physics-based simulations can add to the development of GMPEs (especially in cases where empirical data are spare and inadequate) and contribute to better physical understanding of GMPEs. This class of rough-fault model may provide a viable representation of the ground-motion excitation process over a wide frequency range in a large spatial domain, for various faulting mechanisms (strike-slip and dip-slip), with potential application to the numerical prediction of source- and path-specific effects on earthquake ground motion. This can be a significant step towards fully physics-based seismic hazard analysis.!21
Performance of SORD 16 12 Runtime/step (s) TACC Ranger (8M elements per core) ALCF Intrepid (1M elements per core) ALCF Vesta (1M elements per core) 16 Ideal Obtained 13. 8 7TFlops Speedup 8 4 3.96 7.78 4 1TFlops 2 1.99 4TFlops 1 4 16 64 26 124 496 16384 636 Cores 2 4 8 16 Thread Number Near ideal weak scaling on BlueGene/P and BlueGene/Q in pure MPI mode (no multi-threading) Good OpenMP strong scaling on a single node on BlueGene/Q!22