Abstract
Numerical simulations of processes including bed deformation, vegetation dynamics, and movement of the water’s edge in natural flooded rivers generally necessitate high computational costs attributable to use of both lots of numerical meshes and small time increment. Therefore, drastic time reduction in such computations has recently been one of the challenging issues in computational hydraulic engineering. For this study, we applied a GPGPU acceleration technique to flood flow simulations in the Asahi River with a river diversion. Then, we examined both computational efficiency and numerical accuracy of GPGPU, in comparison with conventional CPU computations. Results showed that the discrepancy between these simulated results was considerably small, regarding both the water levels and river discharges. Therefore, the accuracy of GPGPU is sufficient for use in river engineering tasks. Additionally, it is found use of more meshes improves the efficiency of GPGPU, although GPGPU computations bring disadvantages if a computation domain comprises much less meshes.
