Download presentation
Presentation is loading. Please wait.
Published byArthur Hart Modified over 9 years ago
1
Simulations of Large Earthquakes on the Southern San Andreas Fault Amit Chourasia Visualization Scientist San Diego Supercomputer Center Presented to: Latin American Journalists July 11, 2007
2
Global Seismic Hazard Source: Global Seismic Hazard Assessment Program
3
Expansion of urban centers in tectonically active areas is driving an exponential increase in earthquake risk. Growth of Earthquake Risk Growth of cities 2000-2015 Source: National Geographic Increasing Loss Slide: Courtesy Kim Olsen
4
Structural vulnerability Risk Equation Risk = Probable Loss (lives & dollars) = Hazard Exposure Fragility Faulting, shaking, landsliding, liquifaction Extent & density of built environment Slide: Courtesy Kim Olsen
5
Seismic Hazard Analysis Definition:Specification of the maximum intensity of shaking expected at a site during a fixed time interval Example:National seismic hazard maps Intensity measure: peak ground acceleration (PGA) Interval: 50 years Probability of exceedance: 2%(http://geohazards.cr.usgs.gov/eq/) Slide: Courtesy Kim Olsen
6
“HAZUS’99 Estimates of Annual Earthquake Losses for the United States”, September, 2000 The FEMA 366 Report U.S. annualized earthquake loss (AEL) is about $4.4 billion/yr. For 25 states, AEL > $10 million/yr 74% of the total is concentrated in California 25% is in Los Angeles County alone Slide: Courtesy Kim Olsen
7
Southern California: a Natural Laboratory for Understanding Seismic Hazard and Managing Risk Tectonic diversity Complex fault network High seismic activity Excellent geologic exposure Rich data sources Large urban population with densely built environment high risk Extensive research program coordinated by Southern California Earthquake Center (SCEC) under NSF and USGS sponsorship Slide: Courtesy Kim Olsen
8
1994 Northridge When: 17 Jan 1994 Where: San Fernando Valley Damage: $20 billion Deaths: 57 Injured: >9000 Slide: Courtesy Kim Olsen
9
Slip deficit on the southern SAF since last event (1690): 315 years x 16 mm/year = 5.04 m -> M w 7.7 1857 M 7.9 ~1690 M 7.7 Major Earthquakes on the San Andreas Fault, 1690-present1906 M 7.8 146+91-60 yrs 220±13 yrs Slide: Courtesy Kim Olsen
10
TeraShake Simulation Region 600km x 300km x 80km Spatial resolution = 200m Mesh Dimensions 3000 x 1500 x 400 = 1.8 billion mesh points Simulated time = 4 minutes Number of time steps = 22,728 (0.011 sec time step) 60 sec source duration from Denali 3D Crustal structure: subset of SCEC CVM3.0 Near-surface S-wave velocity truncated at 500m/s, up to 0.5 Hz
11
Computational Challenge!
12
TeraShake-2 Data Flow TS2.dyn.200m 30x 256 procs, 12 hrs, TG IA-64 GPFS Okaya 200m Media Okaya 100m Media 100m Reformatting 100m Transform 100m Filtering 200m moment rate SDSC IA-64 TS2.dyn.100m 10x 1024 procs, 35 hrs Initial 200m Stress modify Initial 100m Stress modify TS2.wav.200m 3x 1024 procs, 35 hrs NCSA IA-64 Datastar p690 Datastar p655 Visualization Analysis Network TG IA-64 GPFS-wan NCSA-SAN SDSC-SAN Velocity mag. & cum peak Displace. mag & cum peak Seismograms Registered to Digital Library SRB SAM-QFS HPSS Datastar GPFS Slide: Courtesy Yifeng Cui
13
Challenges for Porting and Optimization Before Optimization Code deals up to 24 million mesh nodes Code scales up to 512 processors Ran on local clusters only No checkpoints/restart capability Wave propagation simulation only Researcher’s own code Mesh partition and solver in one Initialization not scalable, large memory need I/O not scalable, not portable After Optimization Codes enhanced to deal with 32 billion mesh nodes Excellent speed-up to 40,960 processors, 6.1 Tflop/s Ported to p655, BG/L, IA-64, XT3, Dell Linux etc Added Checkpoints/restart/checksum capability Integrated dynamic rupture + wave propagation as one Serve as SCEC Community Velocity Model Mesh partition separated from solver 10x speed-up of initialization, scalable, memory reduced MPI-I/O improved 10x, scaled up to 40k processors Slide: Courtesy Yifeng Cui
14
Data from TeraShake 1.1 Scalar Surface (floats) 3000 x 1500 ie 600 km x 300 km =17.2 MB per timestep 20,000 timesteps 3 variables Vx, Vy & Vz Velocity components Total Scalar data = 1.1 TB Scalar Volume (floats) 3000 x 1500 x 400 ie 600 x 300 x 80 km^3 =7.2 GB per timestep 2,000 timesteps 3 variables Vx, Vy & Vz Velocity components Total Vol data = 43.2 TB Other Data – check points,etc Grand Total = 47.4 TB Aggregate Data : 160 TB (seven simulations)
15
Visualization MovieMovie (1.5 mb)
16
Comparative Visualization MovieMovie (11 mb)
17
PGV (NW-SE Rupture) PGV (SE-NW1 Rupture) Scenario Comparison
18
Topography Deformation MovieMovie (11 mb)
19
Glimpse of Visualization MovieMovie (65 mb)
20
Visualization Over 130,0000 images Consumed 40,000 hrs of compute time More than 50 unique animations
21
Does Viz work?
23
TeraShake Results NW-directed rupture on southern San Andreas Fault is highly efficient in exciting L.A. Basin Maximum amplification from focusing associated with waveguide contraction Peak ground velocities exceeding 100 cm/s over much of the LA basin Uncertainties related to simplistic source description. Extremely nonlinear dynamic rupture propagation Effect of 3D velocity structure: SE- NW and NW-SE dynamic models NOT interchangeable Stress/strength/tapering - weak layer required in upper ~2km to avoid super-shear rupture velocity Dynamic ground motions: kinematic pattern persists in dynamic results, but peak motions 50-70% smaller than the kinematic values due to less coherent rupture front TeraShake-1TeraShake-2 Slide: Courtesy Yifeng Cui
24
Summary TeraShake demonstrated that optimization and enhancement of major applications codes are essential for using large resources (number of CPUs, number of CPU-hours, TBs of data produced) TeraShake showed that multiple types of resources are needed for large problems: initialization, run-time execution, analysis resources, and long-term collection management TeraShake code as a community code now used by the wider SCEC community Significant TeraGrid allocations are required to advance the seismic hazard analysis to a more accurate level Next: PetaShake! Slide: Courtesy Yifeng Cui
25
References Chourasia, A., Cutchin, S. M., Olsen, K.B., Minster, B., Day, S., Cui, Y., Maechling, P., Moore, R., Jordan, T. (2007) “Visual insights into high-resolution earthquake simulations”, IEEE Computer Graphics & Applications (Discovering the Unexpected) Sept-Oct 2007, In press. Cui, Y., Moore, R., Olsen, K., Chourasia, A., Maechling, P., Minster. B., Day, S., Hu, Y., Zhu, J., Majumdar, A., Jordan, T. (2007), Enabling very-large scale earthquake simulations on parallel machines "Advancing Science and Society through Computation", International Conference on Computational Science 2007, Part I, Lecture Notes in Computer Science series 4487, pp. 46-53, Springer Olsen, K.B., S.M. Day, J.B. Minster, Y. Cui, A. Chourasia, M. Faerman, R. Moore, P. Maechling, and T. Jordan (2006). Strong shaking in Los Angeles expected from southern San Andreas earthquake, Geophys. Res. Lett. 33, L07305,doi:10.1029/2005GRL025472
26
TeraShake Collaboration Large Scale Earthquake Simulation on Southern San Andreas 33 researchers, 8 Institutions Southern California Earthquake Center San Diego Supercomputer Center Information Sciences Institute Institute of Geophysics and Planetary Physics (UC) University of Southern California San Diego State University University of California, Santa Barbara Carnegie-Mellon University ExxonMobil Slide: Courtesy Marcio Faerman
27
Acknowledgements Southern California Earthquake Center (SCEC) San Diego Supercomputer Center (SDSC) Funding: National Science Foundation
28
Thanks for your patience Q&A Websites: http://www.sdsc.edu/us/sachttp://www.sdsc.edu/us/sac (Computation) http://epicenter.usc.edu/cmeportal/TeraShake.htmlhttp://epicenter.usc.edu/cmeportal/TeraShake.html (Seismology) http://visservices.sdsc.edu/projects/scec/terashakehttp://visservices.sdsc.edu/projects/scec/terashake (Visualization)
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.