Fully Coupled Simulations of Megathrust Earthquakes and Tsunamis in the Japan Trench, Nankai Trough, and Cascadia Subduction Zone

摘要

Subduction zone earthquakes can produce significant seafloor deformation and devastating tsunamis. Real subduction zones display remarkable diversity in fault geometry and structure, and accordingly exhibit a variety of styles of earthquake rupture and tsunamigenic behavior. We perform fully coupled earthquake and tsunami simulations for three subduction zones: the Japan Trench, the Nankai Trough, and the Cascadia Subduction Zone. We use data from seismic surveys, drilling expeditions, and laboratory experiments to construct detailed 2D models of the subduction zones with realistic geometry, structure, friction, and prestress. Greater prestress and rate-and-state friction parameters that are more velocity-weakening generally lead to enhanced slip, seafloor deformation, and tsunami amplitude. The Japan Trench's small sedimentary prism enhances shallow slip, but has only a small effect on tsunami height. In Nankai where there is a prominent splay fault, frictional parameters and off-fault material properties both influence the choice of rupture pathway in complex ways. The splay generates tsunami waves more efficiently than the decollement. Rupture in Cascadia is buried beneath the seafloor, but causes a tsunami that is highly complex due to the rough seafloor bathymetry. Neglecting compliant sediment layers leads to substantially different rupture behavior and tsunami height. We demonstrate that horizontal seafloor displacement is a major contributor to tsunami generation in all subduction zones studied. We document how the nonhydrostatic response of the ocean at short wavelengths smooths the initial tsunami source relative to commonly used approach for setting tsunami initial conditions. Finally, we determine self-consistent tsunami initial conditions by isolating tsunami waves from seismic and acoustic waves at a final simulation time and backpropagating them to their initial state using an adjoint method. We find no evidence to support claims that horizontal momen