The weekly 1-degree spatial resolution sea surface temperature (SST) analysis of Reynolds and Smith (1994, Journal of Climate) has been produced using both in situ and satellite data from November 1981 to present and has been widely used for climate monitoring, prediction and research. Problems in the satellite bias correction and the sea-ice to SST algorithm are shown and an improved version of the analysis is completed. The results show a modest reduction in the satellite bias, which leaves small global residual bias of roughly -0.05C. The major improvement in the analysis occurs at high latitudes due to the new ice algorithm where local differences between the old and new analysis can exceed 1C.

To better quantify these changes, additional comparisons with other SST products are included to show typical differences in the monthly products produced at different centers. These results show that these differences occur on large time and space scales with RMS differences, which may exceed 1C. In addition, globally averaged SSTs show persistent differences on decadal scales of roughly 0.05C. Because it is difficult to objectively determine which of the analyses are more correct, special versions of the analyses are computed in which approximately 20% of the buoy data are withheld. The withheld data are then used as independent data to quantitatively determine the analysis accuracy.