T. Spangehl et al.: Intercomparing the quality of recent reanalyses for offshore wind farm planning 119 wind directions in COSMO-REA6, which is consistent with the measurements (Fig. 3e). Similarly, the wind roses of CERRA, ERA5 and HoKliSim-De are in good agreement with observations. The underestimation of the wind directions at south-westerly sec- tors (WSW, SW, SSW) and overestimation at the sectors W to NNW is somewhat more pronounced in CERRA (Fig. 3b) than in COSMO-REA6. However, CERRA shows the ab- solute maximum at sector SW which is consistent with ob- servations. Moreover, CERRA resembles but underestimates the observed secondary maximum at sector E. ERA5 shows a clear underestimation of the relative frequency at sector SSW. On the other hand, populations at sectors W to NNW are slightly overestimated when compared to measurements. The relative maximum at sector E is only weakly pronounced in ERA5 (Fig. 3c). HoKliSim-De shows a slight underesti- mation at sectors SSW to WSW, which is most pronounced at sector SSW (Fig. 3d). By contrast, populations at sectors W to NNW are noticeably overestimated by HoKliSim-De. The secondary maximum at sector E is captured by HoKliSim- De but less pronounced than in the FINO1 observations (Fig. 3e). Further quality assessment is provided by analysis of the different seasons (Appendix A). One outstanding feature is the systematic underestimation of south-westerly wind di- rections by the model based products. A systematic under- estimation of south-westerly wind directions is also reported by Hahmann et al. (2020) for NEWA simulations with WRF. A more detailed interpretation of results is beyond the scope of the present study. 3.1.3 Windstorm CHRISTIAN The correct representation of extreme wind speeds in the re- analysis systems is shown for storm CHRISTIAN as an ex- ample: CHRISTIAN developed over north-western and cen- tral Europe in late October 2013 and, with a travelled dis- tance of 1200 km in 12 h, was classified as a fast-moving storm (Deutscher Wetterdienst, 2013). Across Europe, there were at least 15 deaths, severe destruction and traffic chaos. Affected areas were around the English Channel, in the Netherlands, north-western Germany, parts of Denmark and southern Sweden (CEDIM, 2013). On 26 October, CHRIS- TIAN developed as a secondary cyclone, of the low-pressure system BURKHARD over the western Atlantic (Deutscher Wetterdienst, 2013). On 27 October, the first gusts with wind speeds of up to 133 km h?1 were recorded along the Bre- ton coast. On 28 October, CHRISTIAN continued to track north-eastward and crossed the south of Great Britain with a core pressure of 977 hPa. With a further decrease in pres- sure to 968 hPa, CHRISTIAN then moved towards the north- west coast of Denmark, triggering the highest wind speeds in the North Sea between 13:00 and 14:00 UTC (Deutscher Wetterdienst, 2013). Peak wind speeds in northern Germany and Denmark exceeded 190 km/h. In the course of 29 Oc- tober, CHRISTIAN continued as winter storm with hurri- cane strength and moved towards southern Sweden and Fin- land. The extreme wind speeds at noon on 28 October in the North Sea are also reflected in the various reanalyses in Fig. 4. It shows a comparison of hourly wind speed at 100 m at FINO1 between COSMO-REA6, CERRA and CERRA- EDA, ERA5, HoKliSim-De, NEWA and the observations. The time series of 7 d in Fig. 4a emphasizes that except NEWA all products are very well able to represent ampli- tude and phase of the storm passage. NEWA significantly un- derestimates the maximum wind speeds and also reaches the peak too early. Figure 4b shows 28 October in more detail. Here, the 10 min observations show that there is a short-term wind decrease to 30 m s?1 between the peaks at 11:00 and 13:00 UTC. This pattern cannot be reflected by the reanal- yses because of the only hourly temporal resolution of the available data. The maximum wind intensity is best repro- duced by the regional products COSMO-REA6 and CERRA, where COSMO-REA6 even slightly overestimates the maxi- mum wind speeds. ERA5 and HoKliSim-De, however, show a slight underestimation. Figure 4b also points out a 1 h tem- poral shift between the deterministic and ensemble-based CERRA product, that has to be investigated further. 3.2 Comparison against satellite-based products Biases of monthly mean near-surface wind speed are anal- ysed using a selection of grid points embedded in a geograph- ical area covering the German EEZ of the North Sea (3.0 to 8.5? E, 53.5 to 56 degrees North). 10 m wind speed is used from models. Two different satellite-based data products are used as observational reference. The approach enables an as- sessment of the near-surface wind field. The first product (WIND_GLO_PHY_L4_MY_012_006, hereafter referred to as Scatterometer and Model (e5)) incor- porates scatterometer observations used to correct for persis- tent biases in hourly ERA5 model fields. The second prod- uct (HOAPS version 4.0+ extension, hereafter referred to as HOAPS, CMSAF) is based on passive microwave sensor measurements. Due to coarser spatial resolution and missing data near the coast the HOAPS dataset does not cover the complete study area chosen for this analysis. For the covered area HOAPS shows a slightly lower 10 m wind speed average when compared to the Scatterometer and Model (e5) refer- ence dataset from CMEMS (?0.18 m s?1 for 2008 to 2017). Biases are calculated and discussed separately for the ten year period 2008 to 2017 and for the year 2018. The analysis for 2018 incorporates in addition results for COSMO-R6G2. The year 2018 is characterized by a comparably low value of the annual mean wind speed at FINO1 as indicated by reanalyses (cf. Sect. 3.1). https://doi.org/10.5194/asr-20-109-2023 Adv. Sci. Res., 20, 109–128, 2023