Earlier this year, a pair of earthquakes cost California about $200 million in property damage, and some U.S. earthquakes have stretched past $500 million, or even into the billions. With that much money on the line, predicting earthquakes is a serious business – and so is predicting the kind of damage they’ll do. At Berkeley Lab, researchers are using HPC to more accurately predict how Bay Area structures will respond to future earthquakes along the Hayward Fault, one of the most dangerous faults in the region. 

Previously, site-specific earthquake assessments were fairly rudimentary, relying on extrapolation of past earthquake records. By modeling a 7.0-magnitude earthquake along the fault at a high enough resolution to assess specific structures, the researchers are breaking new ground (so to speak).

David McCallen. Image courtesy of Berkeley Lab.

“This is the first time that we can use what we refer to as physics-based models to predict regional-scale ground motions and the variability of those ground motions,” said David McCallen, a senior scientist at Berkeley Lab and the team’s principal investigator. “It’s really a three-step process where we model all the steps: rupture, propagation through the Earth, and interaction between the waves and the structure at the site.”

The simulations remain a work in progress. Access to greater computing power enabled the researchers to resolve their earthquake simulations at five to ten hertz (vibrations per second) – up from one or two hertz on lower-power systems – but those higher-res simulations are taking 20-30 hours, McCallen said, and require the highest-power machines.

The team hopes that these current exercises will prepare them for the advent of the Department of Energy’s exascale systems, which will begin arriving around 2022. These systems are expected to produce the desired higher-fidelity earthquake simulations with a much shorter turnaround time. “We’re looking forward to getting a lot of speed up with these new advanced machines so that we can resolve very high frequencies but do it in maybe three to five hours,” McCallen said. “We need this because we need a lot of simulations to account for the uncertainty and variability in earthquake parameters.”

A sense that California is in for the next ‘big one’ looms over the research. The last time the Hayward Fault experienced an earthquake – almost exactly 151 years ago – several small towns were devastated. Now, the area is much more densely populated, and UC Berkeley itself sits on the fault – and the clock is ticking. 

“An earthquake on the Hayward Fault,” McCallen said, “is overdue.”

Read Berkeley Lab’s interview with McCallen here.

The post Predicting Earthquake Damage Using HPC appeared first on HPCwire.