Reassessing biases and other uncertainties in sea surface temperature observations measured in situ since 1850: 2. Biases and homogenization

Abstract

New estimates of measurement and sampling uncertainties of gridded in situ sea surface temperature anomalies are calculated for 1850 to 2006. The measurement uncertainties account for correlations between errors in observations made by the same ship or buoy due, for example, to miscalibration of the thermometer. Correlations between the errors increase the estimated uncertainties on grid box averages. In grid boxes where there are many observations from only a few ships or drifting buoys, this increase can be large. The correlations also increase uncertainties of regional, hemispheric, and global averages above and beyond the increase arising solely from the inflation of the grid box uncertainties. This is due to correlations in the errors between grid boxes visited by the same ship or drifting buoy. At times when reliable estimates can be made, the uncertainties in global average, Southern Hemisphere, and tropical sea surface temperature anomalies are between 2 and 3 times as large as when calculated assuming the errors are uncorrelated. Uncertainties of Northern Hemisphere averages are approximately double. A new estimate is also made of sampling uncertainties. They are largest in regions of high sea surface temperature variability such as the western boundary currents and along the northern boundary of the Southern Ocean. The sampling uncertainties are generally smaller in the tropics and in the ocean gyres.