The overall conclusion that we could extract from this validation exercise was that the Tsunami-HySEA model performed well in all benchmark problems proposed, being the most difficult one the first benchmark. In BP1 turbulent effects and viscosity play an important role and crude shallow water models that only consider friction terms similar to Manning or Darcy law will find it difficult to get numerical results close to measured data. For the rest of the benchmarks no special problems were found. For BP2 -Hilo Harbor- good agreement is obtained for the control point and the tidal station for sea surface elevation. Depth average velocities show a good fit for the initial pattern of the signal (in u and v), becoming noisier for the u component at location HA1125 (see comments of other modelers on data sampling for this case). We analyzed sensitivity to friction and convergence when mesh is refined. Differences in maximum velocity between 20m, 10m and 5m mesh resolutions suggest model convergence. The simulation of the complete scenario from the source in the whole Pacific domain was also performed, obtaining also in this case a good fit for sea surface elevation signals at control point and at Hilo tide station. For the velocity time series, a good agreement of the initial phase of the signal and less oscillatory time series for the u component is obtained in this case. Sensitivity to mesh refinement showed minor effects for inner finer meshes and larger effects for the outer coarser ambient meshes. BP3 –Tauranga Harbor- was performed in the three requested configurations (only tsunami, tsunami+tide and only tide). The time series for these three configurations for the sea surface elevation at the four locations considered is very good and also a very good fit is obtained for the speed signal at the ACDP location. For BP4 two numerical resolutions were considered, three numerical schemes used and several values for the friction coefficient imposed. Due to the fact that a good agreement for the three variables considered (flow depth, velocity and momentum flux) were obtained for the three inner locations (B4, B6 and B9) while the flow depth, and consequently momentum flux, were clearly underestimated for the closer to coast location B1, this made us wondering about the spatial variability in the y-direction of the simulated variables. We found out that by slightly moving downwards the location of the sampling point for B1 location we could recover a signal much closer to observations for the sea surface and momentum flux (the velocity signal was good at both locations). At location B9, when low values for the friction parameter are considered (0.012), reflected waves can be observed both in flow depth and velocity signal. Finally, for BP5, shallow water equations without and with dispersion (Madsen & Sorensen model) were used, varying friction and resolution. The main conclusion is that dispersion is mandatory in this BP. The exact arrival time and shape is not well captured by the non-dispersive NLSW model, getting a much better agreement for the model with dispersion. Dispersive model results fit to observations is very good for points located in front of the obstacle, including where the process of breaking starts. Besides, at these locations in front of the obstacle, the effect of varying the friction is minor. For points behind the obstacle, results are sensitive to friction and to the breaking criteria used and a better or worse fit to observations depends on the value of friction and on the parameters used in the breaking criteria. For the velocity, at x=13m y=0m location, a good fit for the u component is obtained while the v component is not well captured.
As Tsunami-HySEA is composed of a family of numerical schemes, well adapted to different flow configurations, in general it has produced good results in all the benchmark problems without the need of any new implementation. The sole exception was for BP1, which we found the more difficult one, mainly due to the simplicity of the shallow water model that we used for this test. We think that a better parameterization of the viscous effects should be included in the model in order to obtain a good fit with experimental data. For the rest of the BPs, the use of Tsunami-HySEA was simple and straightforward. Besides, for the very first time we tested a dispersive version of the model and applied to BP5. For large scale, highly computationally demanding problems (as BP2, complete scenario), Tsunami-HySEA performs extremely fast, producing accurate simulations in very short computational times. For such problems nested meshes were used.
