In 2004 the Indian Ocean tsunami once again showed the world the destructive capabilities of these natural disasters. In this work this practical problem was abstracted to a simplified problem of waves propagating over complex geometry, specifically the processes involved in a solitary and leading depression wave interacting with a bottom seated semi-cylinder over uniform depth and shoaling topography. A purpose written Arbitrary-Lagrangian-Eulerian finite element code to solve the two-dimensional Navier-Stokes equation was developed. It utilised high performance libraries PETSc, Hypre and Triangle to perform simulations with approximately 8 million degrees of freedom across 256 cores on UCL’s supercomputer, Legion. The code was validated against five benchmark test cases and mesh convergence was shown. The code exhibited linear weak scaling and superlinear strong scaling and with this capability, these problems could be analysed using two tools; firstly, global conservation measures such momentum and energy. The rate of change of these integral measures were related to forces on obstacles and boundaries and agreement with model estimates was found even when the wave/boundary interactions were complex. Secondly, flow diagnostics such as the rate of strain and vorticity were used to characterise the velocity field. The large scale capability also allowed the free surface boundary layer to be captured which revealed an (weaker) oscillatory nature of that found at a rigid boundary. The code could not handle breaking and inundation, therefore experiments were carried out to study a depression wave as it interacts and breaks on a beach. Using a novel mechanism for generating isolated depression waves, general features such as extent and speed of recession and surge where characterised and it is noteworthy that the shoreline recession, a defining feature of shoaling depression waves, which was seen during the 2004 tsunami, was captured.
|University College London (University of London)
|Electronic Thesis or Dissertation
Page generated in 0.0054 seconds