For the 50 km-mesh, a few small islands could not be properly resolved due to the higher resolution

of the GSHHS data and these islands were therefore removed from the meshes at all resolutions. The number of tetrahedral elements changed by a factor of approximately four for a doubling in resolution, such that the 6.25 km resolution simulation contained nearly 64 times the number of elements as the 50 km resolution simulation (Table 3). Due to the increase in element count, the modern multiscale simulation was carried out on 540 cores on the Imperial HPC system. ON-01910 cost Run time was approximately 56 h for 15 h of simulated time. Note that no parallel scaling tests were performed to ensure maximum parallel efficiency. In general runtimes are proportional to the

number of elements, which in turn in proportional to the number of degrees of freedom. In addition to discretisation errors, the change in resolution has two consequences. One is an improvement in the representation of bathymetric data by the computational mesh and the second is a change in the position of the virtual wave gauges (see Section 4). The bathymetry used here is the GEBCO 1 arcminute data. This is equivalent to ∼∼1.8 km resolution at this latitude, so even the highest resolution mesh used in this mesh resolution experiment cannot resolve all bathymetric features. We interpolate the bathymetry to each vertex in our computational mesh using bi-linear interpolation. As the mesh is refined, more features are resolved. The second effect is the refinement selleck products of detector locations. In order for a detector to be contained with the mesh (i.e. not on land as represented by this coastline), the latitude and longitude position was converted to spherical Cartesian coordinates and the detector was then moved to the closest mesh vertex. Similar issues occur in other studies

(Bondevik et al., 2005). The multiscale mesh (Fig. 5) was constructed in a similar manner to those above. Resolution varied from 500 m to 50 km and resolution was dependent on bathymetry, Hessian (second-order gradient matrix) of the bathymetry, distance to coastline (see Lambrechts et al., 2008 for details) and distance from slide location. Distance from slide was determined by tracing the approximate slide locations through time and then using GDAL (GDAL STK38 Development Team, 2013) to generate a mesh with resolution of 2 km in a region in the slide area, and which smoothly increased to a mesh spacing of 50 km at 100 km distance from the slide region. Coastlines were generated from GSHHS. The UK, Ireland and neighbouring islands were generated using the full resolution dataset (which has an approximate 200 m resolution). All other coastlines were generated using the intermediate resolution data (which has an approximate 1 km resolution). Small, unresolved islands were removed from all coastlines.