Author(s): Greg Collecutt, Bill Syme
Linked Author(s): Greg Collecutt
Keywords: Shallow water equations, finite volume schemes, high performance computing, model convergence, TUFLOW
Abstract: High Performance Computing (HPC) hardware is moving towards computation of accelerator devices such as GPU and Xeon Phi. These devices now enable multi-million cell hydraulic models to be solved in reasonable times, and have brought supercomputer performance to the office desktop. TUFLOW's new HPC 2D Shallow Water Equation (SWE) solver was developed with the GPU accelerator target in mind, and utilises an explicit finite volume scheme that parallelises effectively across thousands of computational cores. Along with conservation of volume, momentum, and energy, mesh size convergence is an important feature of any numerical solution scheme, and it is important for the modeller to understand how the model results change with mesh resolution. The modeller can then choose the mesh size for production runs, knowing the likely magnitude of the residual discrepancy between the production results and fully converged results. The new TUFLOW scheme is detailed, and two examples for which experimental results are available to benchmark the numerical results are presented. The first example is steady state sub-critical shallow water flow around a 90� bend, Malone et. al. (2008). The second example is the United Kingdom Environment Agency (UK EA) Test 6a, UK Environment Agency (2013), in which a reservoir breaches into a confined channel with an obstruction. This model is highly transient with supercritical flow and moving hydraulic shocks. The convergence of the model results against mesh resolution and time-step is examined for these test cases, using 1st and 2nd order spatial interpolation options. Comparisons of model results against experimental data are made, and conclusions regarding �best practice� mesh resolution are drawn
Year: 2017