A paper last summer in Nature Climate Change asks a simple but important question: why do global surface warming trends from different climate labs agree so much, when the Sea Surface Temperature (SST) trends they depend on differ so widely? Four different SST data sets yield post-1982 warming trends that range from 0.108 to 0.184°C per decade. And oceans cover 74% of the world’s surface. Yet the combined land and ocean data sets only range from 0.185 to 0.192°C per decade, a difference of only 0.007°C per decade. How is it possible? You will be shocked to discover that climate scientists have ways to bury the uncertainty and make their calculations appear far more precise than the underlying data warrant. It’s that expert judgement again.

The first trick, according to the authors, is that producing a global temperature data set involves combining land and ocean data, and the data sets that use SST data with relatively high warming trends pick land records with relatively low warming trends, and vice versa. Hence they draw attention to:

“the slight apparent compensation between land and ocean temperature trends in the global temperature datasets, such that global temperature reconstructions with relatively higher SST trends tend to have relatively lower land temperature trends and vice versa.”

But they also note that scientists tend to avoid the data sets with the highest and lowest warming trends. This expert judgement as to how to discard data that might conflict with our preconceptions artificially suppresses the range of trends across data sets and makes the global results appear more certain than is justified. Thus the authors say:

“Our analysis demonstrates that large differences across datasets in satellite-era global SST trends exist, imply greater uncertainty in global mean temperature trends than previously apparent and impact interpretations of past and present climate.”

Ignoring this uncertainty, in turn, causes scientists to be, in their words, too certain of their own certainty:

“Interpretations that assume differences in trends between global temperature datasets to [sic] provide a reasonable estimate of true uncertainty are overconfident, as global temperature datasets do not account for the full range of trends in published SST reconstructions.”

The authors insist that these uncertainties matter. They discuss a few topics, like claims about years of extreme warming, and measurement of climate sensitivity, where use of the full range of SST data would show the results are less precise than is commonly claimed. And they insist that the usual climate data sets out there just aren’t accurate when it comes to measuring uncertainty:

“The agreement between global temperature datasets over the satellite era does not reflect the full range of plausible SST reconstructions, and consequently should not be interpreted as a definitive measure of confidence in the rate of global warming.”

About the only thing we can ever be truly confident of is that climate scientists will overstate the confidence in their data.