The uncertainty ranges fit well with the expected quality characteristics reported in the literature. According to Gelsthorpe et al. (2000), determination of speeds in the range 4–24 m s−1 with an accuracy of 2 m s−1 (or 10%)

and directions with an accuracy of ±20 deg is required. These criteria are met in both comparisons up to the 18-hour forecast lengths. In the case of the 30-hour forecasts these criteria are exceeded only slightly for the wind speed, but not for the wind direction. According to Figa-Saldaña et al. (2002), the accuracy target for ASCAT winds generated by the OSI SAF is 2 m s−1 root mean square wind component error and 0.5 ms−1 bias for all speeds below 25 m s−1. The wind components of the HIRLAM and ASCAT presented in Table 2 show that the wind component statistics fit the required accuracy thresholds see more well. The RMS of wind components higher than 2 m s−1 is present only in the 30-hour forecasts. The bias of the components this website is lower than that required in all HIRLAM forecasts. In a comparison of the ASCAT and ECMWF analysis in northern cceanic

areas (30°N–60°N), Bentamy et al. (2008) determined standard deviations of 1.77 m s−1 for wind speed and 20 degrees for wind direction. According to Verhoef & Stoffelen (2010), the global

ASCAT-ECMWF standard deviation of difference for wind speed is 1.26 m s−1 and for wind direction is 15 degrees for the 25-km gridded product. The u wind component standard deviation of the ASCAT-ECMWF winds is 1.45 m s−1; the corresponding υ component is 1.63 m s−1. More recently and in line with these results, Hersbach & Janssen reported at the 2010 International Ocean Vector Winds Meeting (18–20 May 2010) a vector RMS difference of ~ 2.2 m s−1 in the Adenosine triphosphate Baltic (http://coaps.fsu.edu/scatterometry/meeting/docs/2010/_may/gridded/hersbach.pdf). HIRLAM wind speeds and directions show similar or slightly worse results over these ranges. Again, the HIRLAM model may contain smaller scales than ECMWF that are not well resolved by the physical parameterizations and the observing systems. Generally, 100-km scales evolve fast and need to be sampled densely in both time and space. To reduce the uncertainty in HIRLAM wind predictions, more observations over the Baltic may be necessary. The fact that the comparison of ASCAT and HIRLAM winds is generally in line with results from other similar studies confirms that the ASCAT 10-m winds are a reliable data source over the Baltic Sea, which is of great importance for marine and NWP communities operating in the region.