According to the myth — and a classic movie — Jason and his shipmates on the Argo suffered more than their fair share of reverses before bringing home the Golden Fleece. The international team of oceanographers working on the ambitious ARGO programme can surely sympathize. Instrumentation flaws have been discovered that undercut the programme's most striking finding to date. The problems have reinforced calls for caution when interpreting 'real-time' environmental data.

ARGO is an array of floats drifting freely around the oceans, 2,800-strong so far and still growing. Each float shuttles back and forth between the surface and the ocean depths measuring temperature and salinity, then sending data home via satellite when it is at the surface.

In 2006, data from the array led a team of scientists to the surprising conclusion that the world's oceans had cooled during 2003–05 — exceptionally warm years in terms of global surface temperature. The team published its findings in Geophysical Research Letters1. Such apparent cooling was seized on by people keen to highlight the uncertainties in forecasts of global warming2.

The ARGO array of almost 3,000 floats measures ocean temperature and salinity worldwide.

That cooling has now been shown to be an artefact. In some of the buoys — they are manufactured in separate batches — a software glitch caused the temperature and salinity data to be associated with the wrong depths. When the problem data are excluded from the analysis, the cooling trend drops below the level of statistical significance.

Novel observation technologies can often be error-prone. In 1997, for example, the discovery of a software problem involving NASA's TOPEX/Poseidon satellite forced a number of groups to lower previously published estimates of global sea-level rise3. And until three years ago, calibration problems with data retrieved from different satellites led to an apparently striking discrepancy between temperature trends at Earth's surface and in the atmosphere4.

The ARGO problem is a reminder that rigid quality control is vital in such cases, says Keith Alverson, director of ARGO's parent agency, the Global Ocean Observing System at the United Nations Educational, Scientific and Cultural Organization in Paris. The ideal is to be able to compare a new analysis with an independent data stream, he adds.

The flaw occurred in a batch of floats fabricated at the Woods Hole Oceanographic Institution in Massachusetts. It was detected when the temperature profiles generated by ARGO were compared with historical data from the regions where the cooling seemed to be most pronounced.

The authors of the original report, led by John Lyman of the National Oceanic and Atmospheric Administration in Seattle, Washington, submitted a correction to Geophysical Research Letters on 9 April; this had not been published when Nature went to press. Apart from the spurious ARGO data, they report a newly discovered bias in temperature profiles from expandable bathythermographs (XBTs) that were also used for the analysis. Inexpensive XBTs have been a major data source for many oceanographic studies. The two problems concealed each other, says Josh Willis, an oceanographer at the Jet Propulsion Laboratory in Pasadena, California, and a co-author of the report: “When I first became aware of the problem I was really horrified.”

The ARGO Steering Team says that additional information sources should allow one-third of the biased profiles to be corrected exactly, and the remainder approximately. Until then, it is “not prudent” to use data from the ARGO array in combination with XBTs, the authors recommend in their correction. Other studies done using the data will need to be re-examined in this light.

Yet oceanographers remain excited about ARGO's chances of delivering true gold, saying it is the best thing to happen to the field in decades. “Factor it in and move on,” says Gavin Schmidt, a climate modeller at the NASA Goddard Institute for Space Studies in New York. But he adds that scientists do need to be more aware of things that can go wrong with new observation systems — particularly if anomalous results turn up.

So has the ocean warmed after all? Yes, but not by as much as had been thought before the flaw in the XBTs was found, says Viktor Gouretski, an oceanographer at the Alfred Wegener Institute of Polar and Marine Research in Bremerhaven, Germany. Gouretski has reanalysed historical data, correcting for the bias in the XBTs5. According to this work, the top 3,000 metres of the oceans warmed by some 0.03 °C between 1957 and 1996 — around one-third less than when the bias was not taken into account.

The effect of the more recent bias in the ARGO data has yet to be worked out, however. “It's always a work in progress,” says Schmidt.