Biases in sea surface temperature are associated with different methods of measurements (basically bucket and intake), which also respond in different ways to atmospheric conditions. On average, measurement with an uninsulated canvas bucket introduces an error of about 0.5° C due to evaporative cooling. Folland and Hsiung (1986) [see also Folland 1991] have developed a physical model for a typical canvas bucket that predicts the amount of evaporative cooling as a function of the atmospheric conditions and exposure time of the bucket. This model has been used effectively to correct the bias in bucket observations (Farmer et al. 1989). However, Farmer et al. (1989) have used the bucket correction method only for pre-World War II observations. Observations after that time were assumed to be almost exclusively measured by intake or by insulated buckets. This is consistent with the fact that the largest discontinuity in SST due to changes in measurement procedure (approximately one-half degree rise in SST around 1940) took place prior to our analysis period (Bottomley et al. 1990, Kushnir 1994).
Recently, the Voluntary Observing Ships Special Observing Program for the North Atlantic (VSOP-NA, Kent et al. 1993a) has found that observations of SST from insulated buckets and hull contact sensors are reliable. A constant bias of 0.35° C was found in observations from ships' intake thermometers compared to those measured by bucket.
The unreliability of the bucket/intake indicator in COADS and the lack of reliable information about the kind of bucket used make this correction impractical from individual observations. Therefore, we do not include a sea surface temperature correction in this version of our calculations.