Because wind is a key parameter for the determination of the air-sea fluxes at the ocean surface, a great deal of effort has been devoted to sort out the problems with ship winds. Wind speeds reported by ships are either directly measured with anemometers or are estimated from the sea state. Instrumentation problems with anemometers are believed (or better, assumed) to be non-systematic and are expected to cancel out when spatial/temporal averages are taken. Estimated winds are somewhat subjective and depend on the skill of the observer. But even when a correct identification of the sea state is made, it still needs to be converted to wind speed through a Beaufort equivalent scale.
Since 1946, a Beaufort equivalent scale developed by Simpson (1906), combined with a well defined sea state devised by Petersen (1927), has been used by meteorological weather services (Roll 1965; Isemer and Hasse 1991). This scale is known as Code 1100 and is sometimes referred to as the old WMO scale. It is now widely accepted that the old WMO Beaufort equivalent scale results in systematic biases when compared to anemometer measured winds. There are a few alternative Beaufort equivalent scales (WMO 1970; Cardone 1969; Kaufeld 1981), and they all suggest that the old WMO scale underestimates the winds for Beaufort numbers less than about 6 and overestimates winds for Beaufort numbers greater than about 6. As pointed out by Isemer and Hasse (1991), this systematic error causes climatological wind speeds to be underestimated, and climatological wind speed standard deviation to be overestimated in the North Atlantic Ocean. Concerning long term variability, Cardone et al. (1990) have shown that a Beaufort scale correction, ship anemometer height adjustment, and stability correction strongly reduce artificial trends in wind speed, and drastically improve the agreement between measured and estimated winds.
Finally, most definitions of transfer coefficients are calibrated to work with winds from a reference level of 10 m above the ocean surface. Cardone et al. (1990) suggest that the average ship anemometer height is actually 20 m. Kent et al. (1993a) shows that during VSOP-NA a typical anemometer height is in the range 15-20 m, but on modern container ships the anemometer heights are about 30 m. A common approximation is to take the 20 m wind in place of the required 10 m. Since virtually all ship anemometers are higher than 10 m, this approximation introduces a systematic error. Isemer and Hasse (1991) estimated that the 10 m wind speed is on average 93%of the wind speed at 20 m. Calculations by the authors show that for light winds under stable conditions, the surface layer has strong shear and the wind at 10 m can be as low as 40%of the wind speed at 20 m. This result suggests that a careful stability dependent correction is warranted.
The ship type indicator, however, is often incorrect or missing in COADS/CMR-5. This is particularly significant in the 1980's when large numbers of drifting buoys were placed in the southern oceans. We have found that nearly all buoy observations in the COADS/CMR-5 1980's interim product are not identified correctly. The result is that a wind observation measured at 5 m is treated as if it were taken at 20 m. Winds in regions sampled mainly by drifting buoys will be biased toward lower values. This problem has apparently been fixed in COADS Release 1A and the correct identification of buoys will be included in future analyses.