Earthquakes: The Quake Project
Jacobo Bielak, Omar Ghattas,
Volkan Akcelik, Hesheng Bao, Ioannis Epanomeritakis, Loukas Kallivokas,
Eui Joong Kim, David O'Hallaron, Jonathan Shewchuk, Tiankai Tu, Jifeng Xu,
Carnegie Mellon University
Volkan Akcelik, Hesheng Bao, Ioannis Epanomeritakis, Loukas Kallivokas,
Eui Joong Kim, David O'Hallaron, Jonathan Shewchuk, Tiankai Tu, Jifeng Xu,
Carnegie Mellon University
CMU scientists used the PSC's T3E and now up to 3000 processors on our
terascale system, Lemieux, for problem sizes up to 100 million elements, to
investigate how soil composition affects ground motion during earthquakes.
Simulation of 1994 Northridge Earthquake Aftershock
These first images and animations show their simulation of a 1994 Northridge
earthquake aftershock. This area is an alluvial basin- soil and soft rock
contained by a bowl shaped space within denser rock.
The following animation shows a surface view of displacement amplitude and
is composited over landsat data (provided by Bill Harbert, U of Pittsburgh)
to show the local terrain.
This volume rendering shows how the shock wave radiates upwards until it
reaches less dense rock, then becomes amplified as it enters the basin.
These next two animations show
visualizations of simulated ground motion of an earthquake generated along a
vertical (strike-slip) fault. Oblique view from above and below ground, showing
horizontal velocity amplitude pattern in the vicinity of the fault, at
increasing times after onset of earthquake. Earthquake source originates
near the far edge of the computational domain and travels along the fault
toward the near edge. Seismic waves travel both perpendicularly to the
fault and downstream from the epicenter. At first, ground motion is
strongest in the direction parallel to the fault (initial red crescent
moving away from fault), and later in the perependicular direction
(initial green cresecent becoming red as it moves downstream). Peak
fault-normal velocity is twice as large as peak fault-parallel velocity.
The two views below show the results of inversion of synthetic surface seismograms for
a portion of the San Fernando Valley, California. The inverse problem is
solved on a 1293 grid using an acoustic wave propagation forward model.
Each animation compares the Inverted (left) with Target (right)
compressional wave velocity. (wave velocity km/s, domain 32x32x16 [depth]
km)
This image shows
inversion of synthetic surface seismograms for soil stiffness of San
Fernando basin model. (read left->right, top->bottom) First three images
show progression of inverted models. Final image shows target basin.
Next is animation of surface response from a simulation of the 1994 Northridge
earthquake in an 80x80x30 km3 Los Angeles Basin model. Color indicates
magnitude of horizontal component of velocity.
Lastly a simulation of the 1995 Kobe, Japan main shock
More information:
- Getting Ready for the Big One, Projects in Scientific Computing, 1997
- Big City Shakedown, Projects in Scientific Computing, 2003
- The Quake Project at Carnegie Mellon University
- High Resolution Forward and Inverse Earthquake Modeling on Terascale Computers, SC2003 (PDF)