We describe a physically based empirical technique for correcting historical sea surface temperature measurements for time-varying biases. The corrections are based on models of heat and moisture transfers from uninsulated (canvas) and partially insulated (wooden) sea temperature buckets exposed on deck. One of the canvas bucket models is tested using measurements on board ship and published wind-tunnel measurements. The method gives geographically and seasonally varying bias corrections through the period 1856 to 1941. The corrections are fairly insensitive to uncertainties such as the size of the bucket or the details of its exposure on deck. A discussion of the history of sea surface temperature observations provides a background to the procedure.

The resulting globally and seasonally averaged sea surface temperature corrections increase from 0.11 degC in 1856 to 0.42 degC by 1940. The corrections are compatible with recent measurements made at sea of the errors of canvas buckets. Global and hemispheric time series of corrected sea surface temperature and night marine air temperature data show good agreement: more detailed verifications of the corrections will be reported elsewhere.