Wind forcing, source term and grid optimization for hurricane wave modelling in the Gulf of Mexico

Abstract

This study evaluates the performance of WAVEWATCH III model driven by different wind forcing products (ECMWF-forecast, ERA5, HRRR, and Copernicus), as well as the behavior of different parameterizations of the model's source terms controlling energy input and dissipation (ST4, ST6) and quadruplet wave-wave interactions (Discrete Interaction Approximation - DIA, Generalized Multiple DIA - GMD) during Hurricane Ida (2021). We also compare the performance of the model configured on uniform unstructured grids (30, 20, 10, 3, and 1 km) and conventional non-uniform unstructured grids. Key findings show that ECMWF-forecast and HRRR outperformed other wind forcing products in capturing wind speeds relative to buoys, satellite and the revised Atlantic hurricane database (HURDAT2) observations. However, all wind products tended to underestimate wind speeds above 20 m/s, with ECMWF and HRRR occasionally performing better for most wind speed values above 35 m/s relative to observations. The corresponding wave simulation results indicated that Ida's wave fields were better captured by model simulations with ECMWF and HRRR wind products, with biases of 2% against buoys in the Gulf of Mexico and 6% and 3% respectively against satellite data. The wave model source terms comparison results showed that simulations with ST4-DIAs exhibited superior performance in bulk wave analyses and were computationally efficient. Simulation using ST4-GMD with three quadruplets performed better than with five, while those with ST6-DIA and ST6-GMD showed the highest error in bulk metrics. We also highlighted limitations in bulk wave analysis by computing partial H_s and 1D spectra density differences between model and buoy for some selected source terms. This reveals consistent overestimation at the lowest frequency bin (0.05–0.1 Hz) and underestimation of the three higher frequency bins (0.1–0.15; 0.15–2; 0.2–0.49 Hz) with a mix of negative and positive energy density difference across different frequencies. The parameterization ST4-DIA with optimal wind adjustment parameter outperformed other configurations in the three high frequency bins having lowest HH, RMSE, and bias. Lastly, the grid comparison result showed a reduction in H_s bias from 30 km to 1 km grid configuration by up to 13% (over 4 m). The 3 km and 1 km grids generally showed similar results. Conventional unstructured grid with 95k nodes demonstrated comparable or slightly better performance than 1 km grid with 1.86 million nodes. Increasing model's resolution did not reduce wave biases when the wind field was misrepresented, and even the source term with the highest agreement with observations could not compensate for these wind uncertainties.