WAVEWATCH III Accuracy and Efficiency within Coupling Framework

It is well demonstrated that the accuracy of the simulated nearshore hydrodynamics is highly dependent on the grid resolution, the inputs from the large-scale and transoceanic activities and the maturity of the numerical models in resolving physics in such complicated geometry, where various interdependent phenomena are taking place. To meet all the needs in a unified solution, large scale unstructured grids with high resolutions near the coast can be deployed in the individual components of a coupled system of various numerical models, where dynamic data exchange is happening. However, these models have their own constraints in terms of computational feasibility and assumptions and considerations in the model equations. In addition, a key element in the coupling between models with different spatiotemporal resolutions and computational loads is run time equivalency of the system components. It is known that the spectral wave models (i.e. WAVEWATCH III) are relatively expensive compared to the surge models (i.e. ADCIRC) leading to a significant slowdown of the coupled system. Moreover, very high resolution grid size for the wave model is required in breaking zone and inundated overland areas to match the observations, which makes the grid larger and model even slower due to CFL constraints in the explicit solver. In these regards, substantial improvements in the WW3 model are required.

Recent developments in the WAVEWATCH III on unstructured grids have pushed the limits of the model in terms of grid size and computational efficiency on HPC environments. A new parallelization based on domain decomposition algorithm with an implicit solver is implemented in the model [1]. The model is validated against the pre-existing parallelization algorithm and explicit solver for a laboratory experiment and severe events in the east coast of the US (Hurricanes Ike 2008 and super storm Sandy 2012 [1,2,3]) from stand alone and wave-surge coupled simulations. In this work, the model scalability and performance together with the limits of each parallelization and solver are evaluated. In addition, large scale computations for the Hurricane Irma 2017 is carried out. The model is validated at point source observations and along satellite tracks. The footprint of the error, introduced from the upstream atmospheric model (HWRF) is analyzed.

References:

[1] Abdolali A., Roland, A., Van Der Westhuysen, A., Meixner, J., Chawla, A., Hesser, T., Smith, J.M. and M. Dutour Sikiric (2019), Large-scale Hurricane Modeling Using Domain Decomposition Parallelization and Implicit Scheme Implemented in WAVEWATCH III Wave Model, Ocean Modeling.

[2] A. Abdolali, Roland, A., Van Der Westhuysen, A., Meixner, J.., Chawla, A., Hesser, T. and J. M. Smith (2019), Towards Alignment of Computational Efficiency and Accuracy of WAVEWATCH III, 26th meeting on Waves In the Shallow water Environment, Jyozankei, Hokkaido, Japan, 12-16 May 2019

[3] Moghimi, S., Vinogradov, S., Myers, E. P., Funakoshi, Y., Van der Westhuysen, A. J., Abdolali, A., Ma, Z., & Liu, F. (2019). Development of a flexible coupling interface for adcirc model for coastal inundation studies. NOAA technical memorandum NOS CS ; 41, . doi:https://doi.org/10.25923/akzc-kc14.