Explosions are the bread and butter of Lawrence Livermore National Laboratory (LLNL), one of the labs under the auspices of the National Nuclear Security Administration (NNSA). But not all explosions, of course, are nuclear – and now, researchers at LLNL and waveform simulation startup Mondaic have created a new, supercomputer-powered model to model the ground shaking that ensues after both explosions and earthquakes.
The model – an adjoint waveform tomography model – falls under the umbrella of seismic tomography, which works much like the computed tomography (CT) scans used to study the human body. In seismic tomography, shockwaves are substituted for x-rays, and the structures and functions revealed are tectonic rather than biological. The researchers’ new model, unlike most of its predecessors, includes 3D wave propagation simulations to understand the relationship between real-world seismograms and those internal structures of the planet.
“While other models of the western U.S. exist, this model is unique in that it is based on many more inversion iterations than previous models and provides much better fits to recorded seismograms,” said Artie Rodgers, a scientist at LLNL and lead author of the paper, in an interview with LLNL’s Anne Stark. “It also can provide more accurate estimates of seismic source characteristics by removing distortion due to unknown 3D Earth structure in previous models.”
“We produced a more detailed 3D structure of the crust and upper mantle for the purposes of improving predictive capabilities of 3D waveform simulations for applications such as source characterization and/or long-period earthquake ground motion simulations,” Rodgers added. “Waveform fits are remarkably better with our final model compared to previous models of the same region.”
To perfect the model, the researchers used Lassen, one of LLNL’s in-house supercomputers. Lassen, an IBM-Nvidia system, delivers 18.20 Linpack petaflops and placed 30th on the most recent Top500 list. On Lassen, the researchers ran more than 60,000 simulations covering 256 iterations of 72 earthquakes, matching the results to almost 100,000 seismograms.
The results of the study elucidated important subsurface relationships and resulted in improved model fits for ground shaking from explosions and earthquakes, particularly in the 20-second to two-minute range after the blast or quake. This time range, of course, is particularly relevant when monitoring for nuclear explosions. Next, the researchers are working on resolving to even finer timescales in the five to ten second range, or even lower.
To learn more about this research, read the coverage from LLNL’s Anne Stark here.