Seismic Simulations of

1

• Seismic Simulations of
• Explosions and Earthquakes Computing Grand
  Challenge Symposium

Arthur Rodgers Chemistry, Materials, Earth and
Life Sciences Directorate
Lawrence Livermore National Laboratory, P. O. Box
808, Livermore, CA 94551
This work performed under the auspices of the
U.S. Department of Energy by Lawrence Livermore
National Laboratory under Contract
DE-AC52-07NA27344
IM 358050
2
Our Effort

• This team effort involved several people
• Kathleen McCandless
• Computer Application and Research, Computations
• Anders Petersson, Bjorn Sjogreen, Stefan Nilsson
• Center for Applied Scientific Computing,
  Computations
• Jeff Wagoner, Phil Harben
• Atmospheric, Earth and Energy Division, CMELS
• Rich Cook, Liam Krauss, Becky Springmeyer
• IM and Graphics Group,Computations
• Bill Walter and Steve Myers
• Geophysical Monitoring Program, AEED NP-Div
• Dave McCallen
• NP-Division, Global Security, PD
• We had an allocation of 50,000 CPU-hours/week on
  Thunder
• Also, several DAT weekends

3
Summary

• This allocation enabled progress in seismic
  simulations in several areas
• Nuclear explosion monitoring (NNSA/NA-22)
• Oct. 9, 2006 North Korean nuclear test
• Hydroacoustic wave reflection/conversion
• Earthquake ground motion (USGS)
• San Francisco Bay Area 3D model
• Hayward Fault scenario earthquakes
• It also enabled development of advanced features
  of the WPP elastic wave propagation code
• LDRD project 05-ERD-079
• We attracted support from NNSA/NA-22 for a
  demonstration calculation on BlueGene/L

4
Seismic simulations are computationally intensive

• Numerical (finite difference or element)
  algorithms
• discretize a 3D volume of the earth into grid
  points
• require a certain number of grid
  points/wavelength
• fmax vmin/?min and ?min nh
• step through time explicitly, time step ?t
• high frequencies require small h and high
  velocities require small ?t.
• We typically want to model
• large volumes (many wavelengths)
• high-resolution (frequency)
• Weve used two codes
• FD (WPP, LLNL)
• SEM (SPECFEM3D, Caltech)

h
5
We are striving for ever larger domains and
higher resolution (frequency)
6
Nuclear Explosion Monitoring (NNSA/NA-22)

• NEM requires analysis of signals resulting from
  wave propagation phenomena
• Seismic
• waves in the solid earth
• Hydroacoustic
• waves in the ocean (SOFAR channel)
• The physics of these phenomena are generally well
  understood and can be modeled.
• However, we do not know the material properties
  of the earth to the scale-length required to
  model the full bandwidth of observations

7
We modeled seismograms from the 9 October, 2006
North Korean Nuclear Test

• Seismograms at Beijing (BJT) showed
• signals at BJT (1100 km) are weak
• large amplitude surface waves
• energy on transverse component
• - possibly due to sympathetic earthquake

surface waves
We wanted to know if model(s) of 3D structure,
including sedimentary basins, can predict the
observed wavefield.
8
3D model predicts the observed energy
partitioning - consistent with explosion source
data simulation
Explosion source in 3D model predicts refracted
energy on transverse component
Explosion source in 1D model predicts no energy
on transverse component
9
The sparse hydroacoustic network requires maximum
information be extracted reflections
However, reflections have lower amplitudes
Hydrophones in the Indian Ocean
reflection
direct
Reflections can help locate events, or may be
provide the only detection when direct wave is
blocked.
10
We modeled the hydroacoustic reflection from the
Seychelles Plateau
Incoming wave is reflected by bathymetry
land
450 million points h50 m 125 x 100 x 4.25
km Ran in 5,000 CPU-hours
ocean
ocean
11
Earthquake modeling in the SF Bay Area

• We have been modeling earthquakes in the Bay Area
  with the USGS (Menlo Park) since 2005
• Evaluation of a 3D geologic/seismic model
• In press at BSSA (Rodgers et al., 2008)
• October 31, 2007 Alum Rock earthquake
• Did you feel it?
• Simulations of the 1906 SF earthquake
• In press at BSSA (Aagaard et al., 2008)
• Currently, working on simulations of a M 7.0
  Hayward Fault earthquake
• Presented at AGU Fall 2007

12
We evaluated the USGS 3D seismic model of the San
Francisco Bay Area
USGS 3D Model
We compared simulated and observed seismograms
for moderate (M 4-5) earthquakes, 2000-4000
CPU-hours/run
Simulated seismograms are late (?t lt 0) relative
to observed. Model is too fast! However, its
being fixed
13
October 31, 2007 Alum Rock Earthquake
MD
WC
Oak
Tri-Valley
LLNL
SF Bay
East
West
Diablo Range
SC Valley
South
14
October 30, 2007 (M 5.6) Alum Earthquake was the
largest since 1989 (M 6.9) Loma Prieta
Recordings at Wente
LLNL code and USGS model can accurately predict
ground motions, including future large earthquakes
Shaking in Livermore was 1 g Large
earthquake (Mgt6.5) expect gt 50 g
15
1906 SF simulation, f 0.5 Hz68,000 CPU-hours
200 km
550 km
40 km deep
16
Were working with the USGS on Hayward Fault
scenario earthquakes - the mostly likely next EQ
One ShakeMap for M 7
Simulation of the 1995 Kobe, Japan earthquake on
the Hayward Fault Mesh refinement reduces
effort to 8000 CPU-hours
Slip distribution along fault
17
Conclusions and Future Directions

• This allocation allowed us to advance and
  demonstrate seismic modeling capabilities for a
  number of applications
• The results from the last twelve months have
• enabled important new science that was not
  possible with routine LC access
• attracted interest from sponsors
• Future directions
• BlueGene/L port demo
• Improve seismic velocity models
• Including waveform methods
• Perform higher resolution simulations
• Use suites of simulations to bound uncertainty