T. Spangehl et al.: Intercomparing the quality of recent reanalyses for offshore wind farm planning 123 4 Discussion and conclusions Detailed meteorological data and information is required to assess the potential wind farm profitability. In this study, new model-based regional reanalyses and a regional downscaling simulation are investigated with regard to their applicabil- ity for offshore wind farm planning. Different reference data sets are used for quality assessment. In-situ measurements of the wind speed and wind direction at heights near 100 m of the research platform FINO1 are used to analyse data of high temporal resolution. The analysis focuses on the time period from 2004 to 2009 prior to the installation of sur- rounding wind farms to avoid the influence of wake effects on the FINO1 measurements (e.g. Podein et al., 2022). Con- sistent biases are found for this undisturbed period that do not vary much over time. Larger variations between the different products are evident for other time periods. Here the ques- tion arises how uncertainties on longer time-scales and fur- ther back in time can be addressed (e.g. Krieger et al., 2021). In summary, the choice of the analysis period provides a pro- found basis for further quality assessment. The observed annual mean wind speed varies between 9.65 m s?1 in 2004 and 10.29 m s?1 in 2007 during the anal- ysis period. Such a difference in wind speed translates to a 21 % difference in wind energy density since wind en- ergy content of the air flow increases with the third power of the wind speed. ERA5, COSMO-REA6, CERRA and HoKliSim-De capture the interannual variations of the wind speed during this period. The regional reanalyses COSMO- REA6, CERRA and the regional downscaling simulation HoKliSim-De show only small biases and resemble the dis- tribution of the hourly wind speed measurements (10 min averages used). By contrast, ERA5 slightly underestimates the wind speed measurements near 100 m at FINO1. The re- gional downscaling simulation HoKliSim-De shows smaller biases than ERA5 and reveals added value with respect to the representation of extreme wind speeds. Wind direction and its variability are known to affect wind-farm power output (Porté-Agel et al., 2020). Wind roses are used to analyse the representation of the wind di- rection and respective wind speed. All model-based products tend to slightly underestimate southwesterly wind directions and to overestimate wind directions from West to Northwest. The same systematic deviation is reported for wind atlas sim- ulations with WRF (Hahmann et al., 2020). COSMO-REA6 is comparatively close to the observed wind roses and shows only small directional biases. Wind power plants are shut down when cut-out wind speeds are exceeded to reduce maintenance costs and to avoid damage (Gliksman et al., 2023). Analysis of the wind- storm CHRISTIAN, which led to extreme wind speeds of more than 35 m s?1 on 28 October 2013 at FINO1, illustrates how far the model-based products are able to capture such an extreme event. The amplitude and timing of the peak is cap- tured by COSMO-REA6, CERRA-fc, ERA5 and HoKliSim- De. COSMO-REA6 slightly overestimates the maximum wind speed whereas ERA5 shows a slight underestimation. In CERRA-EDA the peak is shifted by 1 h but the amplitude is within the ensemble spread. While HoKliSim-De captures the timing and amplitude of the peak, NEWA fails to capture the windstorm at FINO1. Therefore, the choice of the down- scaling area might affect the quality of the results, besides data assimilation methods and model formulation. Finally, various satellite-based gridded data products are used for the evaluation of the near-surface wind speed. The analysis is performed based on monthly averages for an area around the German EEZ of the North Sea. The analysis fo- cuses on the ten-year time period from 2008 to 2017. In addition, the year 2018 is analysed. Consistent results are found using two different reference dataset from CMEMS and CMSAF. For the ten-year period the median of the bias distribution is close to zero for ERA5 and COSMO-REA6. A systematic shift towards positive bias values is analysed for CERRA. By contrast, HoKliSim-De shows a shift to- wards negative bias values. Analogue biases are found for 2018. Here evaluation results for COSMO-R6G2 are simi- lar to COSMO-REA6 which provides first indication for the applicability of the new product. Large biases are found for monthly averages at individual grid boxes indicating that fur- ther aspects such as the sampling and retrieval methods and associated uncertainties of the satellite data needs to be con- sidered for a more detailed evaluation. Also, wake effects need to be considered (e.g. Akhtar et al., 2022) which are not explicitly accounted for by the reanalyses and simulations we use. Moreover, additional analysis of the vertical wind profile is required. For example, as shown by Spangehl and Tinz (2021) for wind farm site investigations near FINO1, the vertical wind speed profiles obtained for COSMO-REA6 and ERA5 are in good agreement but the ERA5 median of the wind speed shows slightly lower values at higher levels. Future evaluation work should therefore make use of avail- able data from measurement campaigns with ground-based lidar instruments (https://pinta.bsh.de, last access: 6 Novem- ber 2023) to improve our knowledge on the vertical wind profile. Appendix A: FINO1 windroses for DJF, MAM, JJA, SON The wind roses of the model-based products (COSMO- REA6, ERA5, CERRA, and HoKliSim-De) are similar to the observed wind roses for all four seasons (Fig. A1), yet there are some systematic differences. In DJF the south-westerly wind directions occur most frequently in the model-based products, which is in accordance with the FINO1 observa- tions. However, all model-based products tend to slightly underestimate the occurrence of south-westerly wind direc- tions. By contrast they show a slightly too frequent occur- rence of the wind directions from sector W to sector NNW https://doi.org/10.5194/asr-20-109-2023 Adv. Sci. Res., 20, 109–128, 2023