By Richard Black
Environment correspondent, BBC News website
Global temperatures did not dip sharply in the 1940s as the conventional graph shows, scientists believe.
They say an abrupt dip of 0.3C in 1945 actually reflects a change in how temperatures were measured at sea.
Until 1945, most readings were taken by US ships; but after the war, UK vessels resumed measurements, and they took the sea's temperature differently.
Writing in the journal Nature, the researchers say this does not affect estimates of long-term global warming.
In the 1940s, there was no universally accepted way of measuring sea temperature.
British vessels typically lowered a bucket over the side of the ship, pulled it up full of seawater, and put a thermometer in.
US vessels, on the other hand, usually had thermometers in place near the pipe where water was pumped in to cool the engines.
The bucket method is known to produce readings slightly lower than the actual temperature, because a fraction of the water will evaporate as it is hoisted on board.
The conventional temperature history shows a sharp fall around 1940
But engine intake measurements can produce higher temperatures than are actually there.
In the war-ravaged seas of the early 1940s, about 80% of the readings put into supposedly global datasets came from US ships. But after the war, the British research effort ramped up sharply, and soon half of the readings were coming from British vessels.
Researchers have now taken the sea temperature record and extracted all the natural variations from the El Nino/La Nina cycle and from weather patterns. When they did, the abrupt fall of 0.3C showed up clearly - but only in data from oceans, not from the land.
"I think the reason this hadn't been found before was that the abruptness of this change only became clear once you took out of the data the natural variability associated with El Nino and 'noise' from weather," lead researcher David Thompson from Colorado State University in the US told BBC News.
Other sharp dips in the 20th Century are associated with volcanic eruptions; but for this, there is no natural explanation that the researchers can find.
The answer, they suggest, has to lie in the ship-borne measurements.
The favoured explanation for the mid-century dip had been a rise in the industrial output of aerosols, tiny particles which act to cool the planet, much like dust from a volcano.
That might still prove to be a factor.
Measurements nowadays are made using sophisticated floats
Corrections for this measurement switch have not yet been applied to produce a new graph of 20th Century temperatures - that work is ongoing at the UK Met Office - but as the land temperature record shows a flattening of the upwards trend from the 1940s to the 1970s, clearly something did change around the 1940s to ameliorate the warming.
"It perhaps suggests that the role of sulphate aerosols, that cooling effect, was less powerful than we thought," said Mike Hulme from the University of East Anglia (UEA), who was not involved in the study.
"And perhaps the solar effects that had been dominant in the 1920s and 1930s weakened; so the downturn in solar forcing is part of the story."
Professor Hulme said the new findings might help to improve computer models of climate, leading to better forecasts for the decades ahead.
According to the Intergovernmental Panel on Climate Change (IPCC), changes in aerosol concentrations in the atmosphere are the second biggest human influence on the climate (after greenhouse gases).
But modelling their influence is a difficult business, and uncertainties persist. A better evaluation of their influence in the recent past ought to lead to better models.
This Nature paper is the latest piece in the giant-sized jigsaw puzzle of compiling a global temperature record from the vast range of measurement programmes that the world has seen over the centuries.
"The study highlights how climate records need to be pieced together from measurements that were not designed to measure long-term trends, and that the corrections required are an ongoing effort," said UEA's Phil Jones, who was on the study team.
"It is just as vital to know how the measurements were taken as the values themselves."
And Mike Hulme emphasised that the new work did not cast doubt on the longer-term, century-scale pattern of rising temperatures, attributable to a changing combination of solar variability, natural climate cycles and, increasingly, human-driven emissions of greenhouse gases.
"I suspect there will be people who want to say it discredits the whole dataset, and that's not the appropriate response," he said.
"The appropriate response is that it's a sign that climate science is in a healthy state. Science is never closed, it's always open for re-examination and scrutiny and testing, and this is a very good example at what science is good at doing and what it must be allowed to do."