Preparing for the Big One: GPU-Powered Simulations Highlight Earthquake Hazard Zones

by Tonie Hansen

Editor’s note: This is the fourth of five posts profiling finalists for NVIDIA’s 2015 Global Impact Award, which provides $150,000 to researchers using NVIDIA technology for groundbreaking work that addresses social, humanitarian and environmental problems. Check out other posts here.

Earthquakes devastate.

This month marks the fourth anniversary of the 9.0-magnitude quake off Japan’s east coast which, with the ensuing tsunami, killed 16,000 and caused some $235 billion in damage. Haiti, Chile and Indonesia have each been with catastrophic earthquakes in the past five years.

Californians know another big one is a matter of time. But how big will it be? How bad will it be?

SCEC CyberShake image at 10Hz
Snapshots of 10-Hz rupture propagation and surface wavefield for a crustal model. Simulations on NCSA Blue Waters and ORNL Titan.

High performance computing is helping to develop seismic simulations in hazard zones that let us better understand the conditions that cause earthquakes, and how the earth behaves when they occur.

Researchers from the San Diego Supercomputer Center, at the University of California, San Diego, and other research institutions are involved in creating these simulations. At the center, computational scientist Yifeng Cui and his team developed a GPU-accelerated code to create highly detailed simulations of high-frequency seismic waves as they propagate through the earth.

This work has placed Cui and his team among five finalists for NVIDIA’s 2015 Global Impact Award. We award our annual grant of $150,000 to researchers using NVIDIA technology for groundbreaking work that addresses social, humanitarian and environmental problems.

“Substantially faster and more energy-efficient earthquake codes are urgently needed for improved seismic hazard analysis,” said Cui.

This code, known as AWP-ODC, is used by the Southern California Earthquake Center (SCEC) to simulate how earthquakes make the ground move. It’s a breakthrough for predicting ground motions that affect small homes and structures, which are more vulnerable to high-frequency shaking. Large structures, such as skyscrapers and highway overpasses, are at risk during long-period shaking.

The computational effort is a part of the large collaboration coordinated by the SCEC. With more people moving to cities in seismically active regions, the economic risks from a devastating earthquake are high, and getting higher.

The center used AWP-ODC to simulate multiple scenarios showing how the Los Angeles basin would react as tremors from a magnitude 7.8 quake shook the region like a giant bowl of jelly.

For the Pacific Northwest, a “megathrust scenario” simulation of a magnitude 9 quake made Seattle undulate like a roller coaster.

“The 3D treatment of seismic-wave propagation has the potential to improve seismic hazard analysis models considerably,” said Thomas Jordan, director of the SCEC. “That’s where the accelerating technology is particularly helpful.”

The GPU-accelerated AWP-ODC code is a workhorse engine used by the SCEC to create a seismic hazard model known as CyberShake.  It computes seismic hazards in California using 3D simulations and fault-rupture descriptions. These computational demands are intense, requiring parallel algorithms and high-throughput workflows.

CyberShake hazard model
Two CyberShake hazard models for the Los Angeles region calculated on NCSA Blue Waters. The model to the left assumes a simplified Earth structure; the model to the right is based on a realistic representation of the crust.

This past year, Cui and his team used the NVIDIA Tesla GPU accelerator-powered Titan supercomputer at Oak Ridge National Laboratory’s Leadership Computing Facility—and broke scientific and technical barriers.

Scientists, for the first time, could simulate ground motions from large fault ruptures to frequencies as high as 10 Hertz in a physically realistic way.

Running on NVIDIA GPUs alongside a complex geometry representative of the San Andreas Fault, the AWP-ODC code ran an estimated 5X faster than it would have on a traditional architecture.

Faster processing and lower power costs for seismic simulations help the project’s goal of calculating risks of more than 280 sites in southern California on the CyberShake map.

Creating a state-wide seismic hazard map using 3D waveform modeling is key to helping structural engineers design safer structures, to insurance companies plan coverage, and to service agencies  allocate emergency response resources.