2:45 PM - 3:00 PM
[STT39-05] Highly detailed viscoelastic crustal deformation analysis of the Nankai Trough region using a 3D finite-element method accelerated with a data-driven approach
Keywords:Nankai megathrust earthquake, Low-viscosity zone, High-performance computing, seafloor observation
After a large earthquake occurs in a plate subduction zone, successive crustal deformation follows in several to tens of years. This postseismic crustal deformation is typically attributed to viscoelastic stress relaxation in the asthenosphere and afterslip after the mainshock. In recent years, with the improvement of the seafloor crustal deformation observation directly above the seismogenic zone, the necessary data for the advancement of estimation of interplate states and rheological structures have been accumulated. To realize such a sophisticated estimation, it is necessary to calculate the Green's function using a model that realistically reflects the heterogeneous underground structure. Although finite-element analysis has been considered suitable for the calculation of Green's functions, the computational cost is enormous, and thus it is required to reduce the computation cost.
To reduce the computation cost of viscoelastic finite-element analysis using highly detailed 3D underground structures, we have previously developed a data-driven method to accelerate the analysis (e.g., Murakami et al. 2023). The initial solution is obtained with high accuracy using a data-driven predictor based on previous time step results, which reduces the number of multi-grid solver iterations and thus reduces the computation cost. The effectiveness of the proposed method has been demonstrated on a CPU-based high-performance computer system, resulting in a 3.2-fold speedup compared to the state-of-the-art solver. Also, we developed a memory-saving data-driven prediction method for GPU-based systems and achieved an 8.6-fold speedup from the GPU-based state-of-the-art solver.
In this presentation, we will show the results of viscoelastic crustal deformation analysis for a highly detailed 3D model of the Nankai Trough region computed using the above method. Based on crustal structure data, the 3D finite element model is generated with the smallest element size ds = 500 m and with a target area of 2496 km × 2496 km × 1100 km. The generated finite-element model has 4.2×109 degrees of freedom. We calculated the 4-year viscoelastic response to the possible coseismic slip for several cases, each with a different Maxwell viscosity distribution, including the case of a low-viscosity lithosphere-asthenosphere boundary (LAB). The results show that for very low LAB viscosities (2.5×1017 Pas), very large landward displacement occurs on the offshore side. On the other hand, in the vertical direction, complex behavior is predicted with varying distributions of uplift and subsidence in different cases. The displacement rate decays rapidly, and thus seafloor crustal deformation observation during the first year is effective for detecting low-viscosity LAB.
To reduce the computation cost of viscoelastic finite-element analysis using highly detailed 3D underground structures, we have previously developed a data-driven method to accelerate the analysis (e.g., Murakami et al. 2023). The initial solution is obtained with high accuracy using a data-driven predictor based on previous time step results, which reduces the number of multi-grid solver iterations and thus reduces the computation cost. The effectiveness of the proposed method has been demonstrated on a CPU-based high-performance computer system, resulting in a 3.2-fold speedup compared to the state-of-the-art solver. Also, we developed a memory-saving data-driven prediction method for GPU-based systems and achieved an 8.6-fold speedup from the GPU-based state-of-the-art solver.
In this presentation, we will show the results of viscoelastic crustal deformation analysis for a highly detailed 3D model of the Nankai Trough region computed using the above method. Based on crustal structure data, the 3D finite element model is generated with the smallest element size ds = 500 m and with a target area of 2496 km × 2496 km × 1100 km. The generated finite-element model has 4.2×109 degrees of freedom. We calculated the 4-year viscoelastic response to the possible coseismic slip for several cases, each with a different Maxwell viscosity distribution, including the case of a low-viscosity lithosphere-asthenosphere boundary (LAB). The results show that for very low LAB viscosities (2.5×1017 Pas), very large landward displacement occurs on the offshore side. On the other hand, in the vertical direction, complex behavior is predicted with varying distributions of uplift and subsidence in different cases. The displacement rate decays rapidly, and thus seafloor crustal deformation observation during the first year is effective for detecting low-viscosity LAB.