[S-SS07] Seismic wave propagation: Theory and Application

[SSS07-P14] Theoretical seismic waveform calculation with an accuracy of 1.6 seconds by the spectral element method and its supplication to the attenuation structure at the base of the outer core

*Seiji Tsuboi¹, Rhett Butler² (1.JAMSTEC, Center for Earth Information Science and Technology, 2.University of Hawaii at Manoa)

Keywords:Theoretical Seismograms, Spectral-Element Method, Inner Outer Core Boundary structure

We have performed global seismic wave propagation calculations using the spectral element method, which is a type of finite element method, for a realistic earth model. Last year, we reported that we have performed the simulation with an accuracy of 1.6 seconds by using the Earth Simulator (ES4) system. In the calculation of the spectral element method, the division of the earth model divides the entire earth into six quadrangular pyramids, and each quadrangular pyramid is divided into finer quadrangular pyramids and assigned to individual CPUs of the supercomputer to perform the calculation. In this calculation, the theoretical seismic waveform propagating globally with an accuracy of 1.6 seconds was calculated by dividing it into 244.7 billion grid points. The parameters NEX and NPROC indicating the division of the spectral element method in this case are 2656 and 83, respectively, and the total number of cores used in the calculation is 41,334 and the ES4 vector engine (VE) is 5168. The grid point spacing in this mesh is 0.94 km on average. For this scale of calculation, it took about 30 minutes CPU time to calculate the mesh and 4 hours 40 minutes CPU time to calculate the theoretical seismic waveform for 23 minutes. The size of the mesh is about 41 Tbyte. The calculation used NEC's MPI as a flat MPI, and the effective performance according to the MPI Program Information was 1.13 PFLOPS, and the vectorization rate was 99%. This effective performance is about 8.8% of the theoretical peak performance of 5168VE. The Earth's internal structure model used in the calculation is transversly isotropic PREM (Dziewonski and Anderson, 1981) for the radial symmetric structure model, and s362ani (Kustowski et al., 2006) for the mantle three-dimensional structure. In addition, the ellipsoidal shape is adopted and atteuation is taken into consideration, but the effects of gravity and rotation are not taken into consideration. As in our JpGU2022 presentation, we approach this analysis of the basal outer core boundary (BOC) with antipodal waveform data in the distance range 179.0°–180° to test the hypothesis whether propagation at the BOC is commensurate with diffraction and/or refraction. We use two thin LVZ models —a 20 km thick layer, and a negative gradient from 50 km thick to the IOCB—with a velocity of 10.0 km/s. Synthetic modeling of the thin, low velocity structures requires higher resolution (1.6 sec) parametrization to achieve necessary detail. We synthesized the antipodal data at the Qiongzhong (QIZ) station in China due to the April 17, 2009 earthquake in northern Chile (Mw6.1). Comparing the two Q models, they are nearly identical. This suggests that the simple approximation of the effect of increased attenuation at the BOC is more complex than can be ascribed to simple layers.