Performance assessment of HurryWave in the North Sea using climate datasets

Abstract

The North Sea is a highly dynamic environment where accurate wave modeling is essential for coastal safety, offshore operations, and long-term climate adaptation. Recently, policy in the Netherlands has shifted toward reliance on SEAS5-based datasets to estimate storm loads, creating a demand for computationally efficient wave models capable of producing long climatologies with sufficient accuracy. HurryWave, a new third-generation spectral wave model developed by Deltares, promises an order-of-magnitude faster performance than SWAN due to simplified physics and numerics. This thesis investigates whether such efficiency compromises wave modeling accuracy in the North Sea.

HurryWave was evaluated against buoy measurements and ERA5 reanalysis data from 1950–2023. Model performance was assessed for significant wave height (Hm0), peak period (Tp), and mean wave direction (θm), with additional analyses on storm events, and spectral representation. Results show that HurryWave reproduces significant wave height with skill comparable to ERA5, achieving root mean square error (RMSE) values of 0.35-0.55 m and scatter index values of 0.12-0.20 across stations in shallower water near the Netherlands. However, in deeper areas such as at North Cormorant, HurryWave without boundary conditions underestimates wave height by a large amount with a scatter index of 0.36. This is mitigated by including a larger domain or by adding boundary conditions. Additionally, using this model setup, large wave heights are underestimated, even at stations in shallower water. This is mainly due to the general underestimation of high wind speeds in the ERA5 input compared to KNMI observations. Peak period (Tp) was consistently underestimated using this setup, with RMSE values of 1.5-2.5 s and negative biases of -0.5 to -1.5 s. Wave direction was the weakest-performing parameter, with RMSE frequently exceeding 30° and large directional scatter during storms. This is due in part to both, the absence of triad modeling which would have an effect on energy distribution to lower frequencies and in transferring energy to other directions, and the inaccuracies in the ERA5 wind data. Additionally the DIA approximation used in modeling quadruplets can lead to errors in fast turning wave fields, contributing to the error in direction. During the 2013 “Sinterklaas storm”, HurryWave captured overall storm evolution but underestimated offshore peak Hm0 by up to 1.0 m and showed narrower spectral energy distributions compared to buoy observations. The analysis of other spectral wave parameters revealed that HurryWave performed better when modeling mean period than the peak period with it performing best in modeling the mean period of the negative first order spectral moment Tm-10 with an RMSE of up to 0.24s compared to the 1.5 to 2.5s RMSE observed for Tp during the Sinterklaas storm.

The findings highlight HurryWave’s strength in reproducing significant wave height, making it suitable for applications such as long-term SEAS5-based climatologies. However, limitations in extreme wave heights, peak period, wave direction, and spectral realism caution against its sole use in operational forecasting or safety assessments that depend on extreme storm accuracy. This study concludes that HurryWave is a promising complement, not replacement, to established models, aligning well with the SEAS5-driven shift toward risk-based, long-horizon coastal policy, while still requiring refinement for storm-scale hazard analysis.