Cross-validation of a GPU-based wave propagation solver and application to seismic waveform modeling of an M4.6 earthquake in Linthal, Switzerland

Abstract

<p>We cross validate a numerical solver for wave propagation in 3-D elastic media written on Graphical Processing Units (GPUs) against the semi-analytical solver for earthquakes in axisymmetric media, using seismic full moment tensors as earthquakes sources, variable earthquake source durations, and comparing observed with synthetic seismograms. The GPU-based solver is based on a numerical formulation of elastodynamic wave equation and can capture isotropic and anisotropic media. The algorithm simulates wave propagation in elastic media in three dimensions and at very high spatial and temporal resolution, and can compute entire wavefields within seconds (Alkhimenkov et al., 2021). For example, the multi-GPU code for elastic wave propagation can compute the entire wavefield of a 1000^3 model (1 billion grid cells) in 40 seconds. We achieve a close-to-ideal parallel efficiency (98% and 96%) on weak scaling tests up to 128 GPUs by overlapping MPI communication and computations. Seismic full moment tensors are routinely used to model a range of seismic processes, natural and anthropogenic, including earthquakes (shear slip), volcanic events, explosions, cavity collapses, landslides, etc. The analytical solver is based on a Thompson-Haskell propagator matrix for layered axisymmetric media (Zhu and Rivera, 2002), with moment tensors as seismic sources, with seismic sources at depths 10s of km below the surface and seismic stations at distances over 2000 km, and has been successfully used in various earthquake source studies (e.g. Alvizuri et al 2018). We validate the GPU-based wave propagation solver through numerical experiments in homogeneous and in layered media, and with observed and synthetic seismograms for an M4.6 earthquake in Linthal, Switzerland on 2017-03-06 with seismic stations at distances up to 30 km. The seismograms from the numerical solver match the analytic and observed seismograms (within frequencies 0.02-0.10 Hz). In future work we will apply the solver to study earthquake source generation, wave propagation in anisotropic media, and seismic source determination.</p><p>References<br>Alkhimenkov, Y., Räss, L., Khakimova, L., Quintal, B., & Podladchikov, Y., 2021. Resolving wave propagation in anisotropic poroelastic media using graphical processing units (CPUs), J. Geophys. Res., 126, doi:10.1029/2020JB021175.<br>Alvizuri, C., Silwal, V., Krischer, L., & Tape, C., 2018. Estimation of full moment tensors, including uncertainties, for nuclear explosions, volcanic events, and earthquakes, J. Geophys. Res. Solid Earth, 123, 5099–5119, doi:10.1029/2017JB015325.<br>Zhu, L. & Rivera, L. A., 2002. A note on the dynamic and static displacements from a point source in multilayered media, Geophys. J. Int., 148, 619–627, doi:10.1046/j.1365-246X.2002.01610.x.</p>