THE graph flashed up on the screen for only a few seconds, but it set alarm bells ringing. Had I read it right? The occasion was a workshop on climate change at the UK's Met Office in Exeter. If carbon dioxide in the atmosphere doubled from its pre-industrial level, the graph suggested, global warming would rise far above the widely accepted prediction of between 1.5 and 4.5 °C. The real warming could be as high as 10 °C. Surely some mistake? Too much wine at lunch? But no. This was for real.

Till now, climate modellers' forecasts of future warming have resembled the famous bell curve, with the most likely result of doubling CO2 being a temperature increase of about 3 °C, and with declining probabilities on either side for a narrow range of higher and lower temperature rises (see Graph). But not in this case. The graph, shown by James Murphy of the Met Office's Hadley Centre for Climate Prediction, had a long "tail" at the higher end, reaching up to 6, 8 and even 10 °C.

Temperature rises of this much would have serious implications. CO2 is expected to reach double its pre-industrial levels within a century if we carry on burning coal and oil in what economists call a "business-as-usual" scenario. Nobody has seriously tried to work out what this extra warming would mean for the planet or human society. But it would certainly not mean business as usual.

First, a health warning. Murphy was not making a firm prediction of climatic Armageddon. But nor was this a Hollywood movie full of impossible science. The high temperatures on the display, he said, "may not be the most likely, but cannot be discounted".

Nor was Murphy alone with his tail. He showed a projection by David Stainforth from the University of Oxford that suggested a possible warming of 12 °C or more. This new generation of scarily skewed distributions will start turning up in the journals soon. They arise because modellers have for the first time systematically checked their models for uncertainties and discovered that they have an Achilles' heel: clouds.

While clouds have always been regarded as one of the biggest uncertainties in calculations of global warming, they are turning out to be far more of a wild card than anyone imagined. The fear is that global warming will either reduce how cloudy the planet is, or significantly change the type of clouds in the sky, and their influence of the planet's radiation budget. This could amplify global warming more than so far anticipated. Being mostly of an age to remember 1970s Joni Mitchell songs, the climate scientists say they have "looked at clouds from both sides now", and they don't like what they see.

Later this month, many of the researchers at the Exeter workshop will sit down again in Paris to begin work on the UN's fourth global assessment of climate change, which will be published by the Intergovernmental Panel on Climate Change (IPCC) in 2007. The sessions will discuss how sensitive the climate system is to infusions of CO2 and other greenhouse gases. If the evidence presented at the Exeter meeting holds true, the UN will have to ratchet up its predictions of global warming, and in particular warn that their worst-case scenarios have just not been worst-case enough.

The climate's sensitivity to warming depends critically on feedbacks that may amplify or damp down the initial warming. If you double the amount of CO2 in the atmosphere, the direct greenhouse effect is only about 1 °C. Not much to worry about. But climate scientists expect the warming to trigger a series of feedbacks, of which the three biggest, at least in the next few decades, will be from ice, water vapour and clouds.

Take ice. As the world warms, snow and ice from polar caps and mountain glaciers melts and is replaced by open water, bare rock, tundra and forests. As this happens, the surface of the Earth becomes darker and absorbs more radiation from the sun. This positive feedback is already evident in much of the Arctic, where warming in recent decades has happened faster than elsewhere. But it will also warm up the global atmosphere.

Water vapour, like CO2, is a potent greenhouse gas. Without it our planet would freeze. But what will happen to water vapour as the world warms is not as clear-cut as with ice. A warmer surface will certainly cause more water to evaporate. And, though some sceptics disagree, this will probably increase the amount of water vapour in the atmosphere. That again will amplify warming.

In the standard climate models extra water vapour in the air will at least double the direct warming effect of CO2. Add the impacts of water vapour and ice together and we are close to climate scientists' central prediction - a warming of about 3 °C for a doubling of CO2.

But it's when we come to the third feedback mechanism that things get really sticky. Clouds are clearly linked to water vapour. A lot of water vapour in the air eventually forms clouds. During their short lives, clouds produce both positive and negative feedbacks. We all know that during the day, they can keep us cool by reflecting the sun's harsh rays. And at night they keep us warm, acting like a blanket that traps heat rising from the ground. But which of these effects wins out depends a lot on the height at which the clouds form, their depth, colour and density.

Researchers still know surprisingly little about how many and what sort of clouds are up there. Last year, for instance, it emerged that there may be vastly more heat-trapping cirrus clouds in the upper atmosphere than anyone had thought. Some studies suggest that, taken globally, the cooling and warming effects of clouds may largely cancel each other out. But nobody is sure. And small changes in either the area of cloud cover or the types of clouds that form could change things radically. So for modellers of our future climate there are two issues. Will global warming change clouds? And will the changes produce positive or negative feedback on the climate?

A first guess would suggest that extra evaporation and water vapour in the atmosphere will make more clouds. But it may not be so simple. Higher evaporation rates in the heat of a greenhouse day may "burn off" clouds without them ever producing rain. Equally, clouds may "rain out" more quickly, leaving clearer skies rather than cloudier ones. The fear is that clearer skies will amplify, rather than damp down global warming.

And there is growing evidence that this clear-skies effect could already be under way. One of the foremost experts on clouds and climate, Bruce Wielicki of NASA's Langley Research Center, has found that there are fewer clouds these days in the tropics. Since the mid-1980s, he says, the rising and descending motions of air that cover the entire tropics, for example in the Hadley circulation cells (see Diagram), appeared to increase in strength. The result was faster formation of storm clouds in areas where the air rises - what meteorologists call the inter-tropical convergence zone - but with the clouds raining out more quickly, which left the rest of the tropics drier and less cloudy.

New research showing a decrease in "earthshine", the sunlight reflected from the Earth onto the moon, is still controversial (New Scientist, 5 June, p 10), but seems to confirm both that the Earth's cloud cover is falling and that the reflectivity of clouds plays a vital role in controlling the planet's radiation budget, says Peter Cox, head of climate chemistry at the Met Office.

Wielicki is still cautious about what is behind the clearer tropical skies, but many others see them as strong evidence of global warming. And this matters a great deal because an estimated two-thirds of global water-vapour feedback, and probably an equal proportion of the cloud feedback, take place in the tropics. So, clearer tropical skies could bring a major positive feedback to global warming.

Even if global warming is not the cause of clearer tropical skies, Wielicki says his findings show that we are being complacent if we think clouds are an unchanging feature of the world. Not only are they highly variable but their potential effects on climate are poorly understood. The extent to which clouds control the planetary thermostat may be between two and four times greater than previously thought, he says. Most disturbing, he says, is that "climate models used to forecast the effect of global warming have so far failed to pick up on this".

For example, if you run the standard models backwards to "hindcast" the 1980s and 1990s, they fail to predict what happened to clouds or the radiation budget. "Since clouds are thought to be the weakest link in predicting future climate change, these new results are unsettling - the models may be more uncertain than we had thought," says Wielicki.

It is this uncertainty that Murphy, Stainforth and others set out to explore. Murphy took a standard climate model and tweaked it to reflect the range of uncertainties we have for characteristics such as cloud cover, cloud lifetime and cloud thickness. The model responded by delivering much higher probabilities of greater than expected warming. Murphy says that "variations in cloud feedback played a major role" in the predictions of higher temperatures - the tail in his graph. His results are due to be published soon in Nature. And fellow modeller Rob Colman of the Bureau of Meteorology Research Centre in Melbourne, Australia, agrees. "I suspect it is strong positive cloud feedbacks that make up the bulk of the tail."

Of course, part of the problem could just be that there are too many bad models producing scary results. Till now, IPCC scientists have included the predictions of all available general circulation models in their estimates of future climate. But for the next assessment, they want to weed out the poor models that cannot reproduce short-term climate changes.

According to Susan Solomon, chair of the IPCC's science working group, the biggest difference between models that give high estimates of global warming and those that offer lower predictions is how they handle cloud feedbacks. But which are right? By and large, says Myles Allen of the climate dynamics group at the University of Oxford, the models being rejected in the weeding process are those that predict low warming. Murphy agrees: "Models at the low end have a lot of unrealistic representations of clouds. The weeding process suggests higher temperatures."

On the face of it, we should be able to narrow down our estimates of the sensitivity of the climate to the greenhouse effect and cloud feedbacks by checking with the real world. The past century has seen a rise in global temperatures of around 0.6 °C, most of it during the past 50 years when we have good records of CO2 concentrations in the atmosphere. But such analysis is not so simple.

For one thing, the impact of greenhouse gases on air temperature is not instant. There are delays of several decades in the system while, for instance, heat is redistributed between the atmosphere and the oceans. Then there is natural temperature variability, caused by changes in solar radiation and volcanic eruptions. The final complication is that greenhouse gases are not the only source of human influence over atmospheric temperatures.

As well as pumping gases into the atmosphere, we are also filling it with huge volumes of microscopic particles, mostly from burning forests, crop waste and fossil fuels. Depending on their characteristics, these aerosols can scatter or absorb solar radiation and may influence the formation, colour and reflectivity of clouds. The precise nature of their involvement in global temperature has been hotly disputed for a decade. But most researchers now believe that the dominant effect of these aerosols is to suppress warming by shading the planet.

"We are dealing with a coiled spring, with temperatures being held back by aerosols," says Solomon. "If you shut off aerosols, temperatures would increase rapidly, but we don't yet know exactly how coiled the spring is." The best guess until recently was that this "parasol effect" was holding back a quarter of the warming so far, or about 0.2 °C. But critics say this calculation is little more than a guess. The first efforts at directly measuring the parasol effect suggest the spring may be much more tightly coiled. In an assessment last year, Nobel prize-winning atmospheric chemist Paul Crutzen argued that aerosols could be disguising between half and three-quarters of present warming (New Scientist, 7 June 2003, p 7). That suggests the coiled spring is already holding back warming of anything up to 2 °C. "The two major pollutants have been almost cancelling each other out," says Cox.

This is doubly bad news. First because it shows that cleaning up aerosols would release a burst of warming. But secondly, it suggests that the climate system is much more sensitive to greenhouse gases than we thought. Crutzen's estimate would put the true warming effect of doubling CO2 at between 7 and 10 °C, which Murphy's graph predicts, albeit at a low probability.

Some climate scientists find these new figures disturbing not just for what they suggest about the atmosphere's sensitivity to greenhouse gases, but also because they undermine existing predictions. Uncertainty about those predictions is stopping politicians from acting to halt global warming. So, they argue, even suggesting that the model results are less certain could be politically dangerous.

But other climate scientists fear creating a spurious certainty about climate change. Since we don't know what the future holds, they say, we shouldn't claim to know. These people see the predictions of climate models as less like a weather forecast and more like a bookmaker setting odds for a high-stakes horse race. There are no "dead certainties". They say that humanity has to act prudently and hedge its bets about future climate change in the absence of certainty. We will, they argue, never be able to see through the clouds, and politicians will just have to accept that.

If you would like to reuse any content from New Scientist, either in print or online, please contact the syndication department first for permission. New Scientist does not own rights to photos, but there are a variety of licensing options available for use of articles and graphics we own the copyright to.