It is widely recognized that the initial ensemble describes the uncertainty of the variables and, thus, affects the performance of ensemble-based assimilation techniques, which is investigated in this paper with experiments using the Community Earth System Model (CESM) and the Data Assimilation Research Testbed (DART) assimilation software. Five perturbation strategies involving adding noises of different patterns and with/without extra integration are compared in the observation system simulation experiments framework, in which the SST is assimilated with the ensemble adjustment Kalman filter method. The comparison results show that for the observed variables (sea surface temperature), the differences in the initial ensemble lead to different rate of convergence in the assimilation, but all experiments reach convergence after three months. However, other variables (sea surface height and sea surface salinity) are more sensitive to the initial ensemble. The analysis of variance results reveal that the white-noise perturbation scheme has the largest RMSE. After excluding the effect of the white noise perturbation scheme, it can be found that the difference in the effect of different initial ensembles on the SSH with only assimilated SST is concentrated in the region of the Antarctic Circumpolar Current, which is related to the spread of the covariance between the SSH and the SST.

With the rising global demand for renewable energy sources, a great number of offshore wind farms are being built worldwide, as well as in the northern South China Sea. There is, however, limited research on the impact of offshore wind farms on the atmospheric and marine environment, particularly tropical cyclones, which frequently occur in summertime in the South China Sea. In this paper, we employ the Weather Research and Forecasting (WRF) Model to investigate the impacts of large-scale offshore wind farms on tropical cyclones, using the case of Typhoon Hato, which caused severe damage in 2017. Model results reveal that maximum wind speeds in coastal areas decrease by 3–5 m s⁻¹ and can reach a maximum of 8 m s⁻¹. Furthermore, the wind farms change low-level moisture convergence, causing a shift in the precipitation center toward the wind farm area and causing a significant overall reduction (up to 16%) in precipitation. Model sensitivity experiments on the area and layout of the wind farm have been carried out. The results show that larger wind farm areas and denser turbine layouts cause a more substantial decrease in the wind speed over the coast and accumulated precipitation reduction, further corroborating our findings. Significance Statement: This study holds significant implications for developing offshore wind farms in tropical cyclone-prone regions like the South China Sea. By focusing on Typhoon Hato as a case study, the research sheds light on the previously understudied relationship between large-scale offshore wind farms and tropical cyclones. The observed decrease in coastal wind speeds and altered precipitation patterns due to wind farm presence highlights the potential for mitigating cyclone-related risks in these regions. Additionally, the study’s sensitivity experiments underscore the importance of careful planning and design in optimizing wind farm layouts for maximum impact reduction. This research contributes vital insights into sustainable energy infrastructure development while minimizing environmental and meteorological risks in cyclone-prone areas.

Coupling the Weather Research and Forecasting (WRF) model with wind farm parameterization can be effective in examining the performance of large-scale wind farms. However, the current scheme is not suitable for floating wind turbines. In this study, a new scheme is developed for floating wind farm parameterization (FWFP) in the WRF model. The impacts of the side columns of a semi-submersible floating wind turbine on waves are first parameterized in the spectral wave model (SWAN) where the key idea is to consider both inertial and drag forces on side columns. A machine learning model is trained using results from idealized high-resolution SWAN simulations and then implemented in the WRF to form the FWFP. The difference between our new scheme and the original scheme in a realistic case is investigated using a coupled atmosphere–wave model. The results show that the original scheme has a lower power output in most of the grids with an average of 12 % compared to the FWFP scheme. The upstream wind speed is increased slightly compared to the original scheme (<0.4 m s⁻¹), while the downstream wind speed is decreased but by a much larger magnitude (<1.8 m s⁻¹). The distribution of the difference in turbulent kinetic energy (TKE) corresponds well to that of the wind speed, and the TKE budget reveals that the difference in TKE in the rotor region between the two schemes is mainly due to vertical wind shear. This demonstrates that the FWFP is necessary for both predicting the wind power and evaluating the impact of floating wind farms on the surrounding environment.

A new scheme is developed for floating wind farm parameterization (FWFP) in the Weather Research and Forecasting (WRF) model. The impacts of the side columns of a semi-submersible floating wind turbine on waves are firstly parameterized in the spectral wave model (SWAN) where the key idea is to consider both inertial and drag forces on side columns. A machine learning model is trained using results of idealized high-resolution SWAN simulations and then implemented in the WRF to calculate the frictional velocity at the turbine site. This frictional velocity is passed to the Fitch wind farm parameterization to form the FWFP. The difference between our new scheme and the original Fitch scheme in a realistic case is investigated using a coupled atmosphere-wave model. Results indicate that the FWFP can increase total power output of wind farms by over 5% in the high wind speed stage due to significant wave height attenuation caused by large-scale floating turbines. The turbulent kinetic energy decreases within the wind farm, with the greatest drop of 0.4 m2s-2 at the top of the turbine. The impact of the new scheme can spread to the top of the atmospheric boundary layer. The proposed new scheme will help forecast wind energy and explore the potential impacts of large floating wind farms.

Coupling Weather Research and Forecasting (WRF) model with wind farm parameterization can be effective in examining the performance of large-scale wind farms. However, the current scheme is not suitable for floating wind turbines. In this study, a new scheme is developed for floating wind farm parameterization (FWFP) in the WRF model. The impacts of the side columns of a semi-submersible floating wind turbine on waves are firstly parameterized in the spectral wave model (SWAN) where the key idea is to consider both inertial and drag forces on side columns. A machine learning model is trained using results of idealized high-resolution SWAN simulations and then implemented in the WRF to form the FWFP. The difference between our new scheme and the original scheme in a realistic case is investigated using a coupled atmosphere-wave model. Results indicate that the original scheme underestimates the power output of the entire floating wind farm in the winter scenario. On average, the power output of a single turbine is underestimated by a maximum of 694 kW (12 %). The turbulent kinetic energy decreases within the wind farm, with the greatest drop of 0.4 m² s⁻² at the top of the turbine. This demonstrates that the FWFP is necessary for both predicting the power generated by floating wind farms and evaluating the impact of floating wind farms on the surrounding environment.