122 T. Spangehl et al.: Intercomparing the quality of recent reanalyses for offshore wind farm planning Figure 6. As Fig. 5 (top row panels), but for year 2018. Results for COSMO-R6G2 are shown in addition. Table 4. Maximum and minimum bias of monthly mean near-surface wind speed [m s?1] of grid boxes embedded in geographical area covering the German Exclusive Economic Zone of the North Sea. Biases are calculated for the various model-based products using two satellite based data products as reference. The time period is 2008–2017. Product Scatterometer and Model (e5), HOAPS, CMSAF CMEMS ERA5 1min =?3.1, 1max = 0.9 1min =?1.5, 1max = 1.9 COSMO-REA6 1min =?3.7, 1max = 2.2 1min =?1.5, 1max = 2.5 CERRA-an 1min =?3.1, 1max = 2.0 1min =?1.2, 1max = 3.0 HoKliSim-De 1min =?3.7, 1max = 1.9 1min =?1.4, 1max = 2.3 slightly lower mean near-surface wind speeds when com- pared to the other reference data set thereby favouring more positive bias values (Fig. 5, Table 4). The HOAPS dataset does not cover the full study area due to the coarser spatial resolution and missing values near the coastline. The CMEMS dataset combines scatterometer measure- ments with ERA5 and is therefore not purely based on mea- surements. HOAPS is purely satellite-based and therefore provides independent data, e.g. for evaluation of COSMO- REA6 where no satellite data is assimilated. However, HOAPS is subject to sampling uncertainties due to sparser data coverage which is clearly reflected by the larger range of biases (Fig. 5, top row panels and Table 4). Maps of the spa- tial bias distribution facilitate the interpretation of the Box- Whisker plots. The multi-year average bias shows only mod- erate spatial variations (Fig. 5, row 2 to 5 panels). Larger biases are seen along the coast for the Scatterometer and Model (e5), CMEMS dataset. As noted earlier HOAPS does include missing data near the shore. From the spatiotemporal analysis we conclude that consistent results are found for the median bias indicating comparably small deviations of the near-surface wind speed for ERA5 and COSMO-REA6 from the satellite-based reference datasets, a slight shift of the bias distribution towards positive values for CERRA and a minor shift towards negative values for HoKliSim-De. 3.2.2 Evaluation of bias for 2018 Results for 2018 partly confirm findings for 2008–2017. For example, CERRA shows a tendency towards positive bias values. Moreover, a shift towards negative bias val- ues is clearly seen for HoKliSim-De (Fig. 6). Compared to the analysis for 2008–2017 the median bias of ERA5 and COSMO-REA6 is slightly more negative when using Scat- terometer and Model (e5) (CMEMS) as reference (Fig. 6, left panel). The median bias of both ERA5 and COSMO- REA6 is close to zero when choosing the HOAPS dataset as reference (Fig. 6, right panel). Note that both ERA5 and COSMO-REA6 show comparatively low annual mean wind speed values at 100 m when compared to other model-based products at the position of FINO1 (cf. Sect. 3.1). Results for COSMO-R6G2 are similar to COSMO-REA6 for both reference datasets (Fig. 6). The median bias of COSMO-R6G2 is slightly negative for the Scatterometer and Model (e5) (CMEMS) reference and close to zero when choosing HOAPS as reference. Therefore, results for COSMO-R6G2 provide first indication of the suitability of the new product. Adv. Sci. Res., 20, 109–128, 2023 https://doi.org/10.5194/asr-20-109-2023