The Computational Infrastructure for Geodynamics (CIG), in collaboration with Lawrence Livermore National Laboratory (LLNL), held a workshop last year that focused on computational seismology. The workshop provided training in seismic waveform processing, visualization, and high-performance computing (HPC) waveform simulation. Fifty-five predominantly early-career participants—graduate students and postdocs—from the United States and 16 other countries attended. The workshop was the first of its kind to feature full access to HPC resources for research-grade example problems.
A few trends served as the backdrop for the meeting. Seismic data processing software and numerical codes for HPC simulation of the seismic wavefield have evolved substantially in recent years. The volume of seismic data has greatly increased as instrumentation and technology have advanced and barriers to deploying sensors and transmitting data have fallen. Capabilities for HPC simulation of seismic waves in realistic 3-D Earth models have also greatly increased for source and Earth structure studies, driven by advances in numerical methods and computer programs as well as the inexorable growth in computing power. These trends indicate that HPC simulations of seismic waves will become more common in research and seismic network operations.
The 5-day workshop had several elements. The main objective was practical hands-on tutorials on the workflow involved in simulating seismic waveforms. Keynote lectures by leading researchers described how seismic simulations are advancing understanding of earthquake hazards and Earth structure. These lectures also outlined challenges that must be overcome to maximize the potential of HPC simulations and to ensure that the methods gain broader use. Participants shared their current work in poster and lightning talk sessions.
Workshop tutorials gave participants hands-on experience using four open-source codes for waveform processing and simulation. ObsPy is a Python-based software package for accessing, processing, and visualizing seismic waveforms, event data, and metadata. Three methods for computing synthetic seismograms were covered. Instaseis computes seismograms for radially symmetric models and runs on a laptop. Two codes compute seismograms in 3-D Earth models on parallel computers: SW4, a Cartesian finite difference code developed at LLNL, and SPECFEM3D, a spectral element code developed by a large team led by Princeton University. Participants learned how to run these codes and then processed and visualized the results.
LLNL provided access to 7,200 cores of the Quartz HPC cluster for 3 days during the workshop, enabling participants to simultaneously run simulations on hundreds of processors. During the workshop, the 19 September 2017 M 7.1 Mexico earthquake occurred. Participants ran a simulation of this (see leadoff image) and other earthquakes with SPECFEM3D and visualized the results to exercise their newly acquired skills.
The workshop was sponsored by CIG, a National Science Foundation–funded geoinformatics project to develop, disseminate, and maintain numerical codes for geodynamics. The Seismological Society of America also provided support. LLNL provided the venue, local organization, and HPC cycles.
Details, including links to presentations and tutorials, can be found on the workshop website.
—Arthur J. Rodgers (email: [email protected]), Geophysical Monitoring Program, Lawrence Livermore National Laboratory, Livermore, Calif; Lorraine J. Hwang, Computational Infrastructure for Geodynamics, University of California, Davis; and Louise H. Kellogg, Earth and Planetary Sciences, University of California, Davis