The document slaps down theories which hold that volcanoes and variations in the sun's output are a far bigger influence on climate in the recent past and near future than any man-made pollution. But there remain huge, intractable uncertainties about when and where the warming will occur, how it will change rainfall and weather, how much harm it can do us. The best estimate is for an average temperature rise of between 1C and 3.5C midway through the next century.
One complication: pollution from the burning of fossil fuels is cooling large areas of the planet's surface as well as warming it. Emissions of carbon dioxide (CO2) trap heat in the atmosphere that would otherwise escape into space, but sulphate aerosol - from burning coal and gas - acts as an atmospheric heat shield.
The hope is that one day supercomputers will be able to model most of the complexity and credibly predict the fate of an increasingly polluted earth. These models are among the most awesome, grandiose and beautiful applications of computer science. Their ability to simulate the behaviour of the oceans and atmosphere over a period of centuries has greatly improved over the past 15 years, but there is far to go.
Scientists at the Hadley Centre for Climate Prediction have just taken delivery of their new Cray C90 supercomputer, also used by the Met Office for routine weather forecasting. Last month they began a series of modelling experiments which will continue through the summer. The centre, at Bracknell, Berkshire, funded by the Department of the Environment and the Met Office, is among a dozen institutes around the world up to the task. Its findings are fed to the IPCC.
Starting from around 1860, the Cray will make its own calculations of how the atmosphere and oceans have responded to gradually rising emissions of man-made greenhouse gases. It will be interesting to compare its version of worldwide weather and climate in 1994 with the reality. Then the supercomputer will march on into a future in which concentrations of 'greenhouse gases' continue to rise as the world burns more and more coal, oil and gas.
Experiments using the centre's previous supercomputer showed that after 70 years of steadily rising greenhouse gas concentrations (1 per cent extra a year) average temperatures went up by 1.5C. But this simulated global warming was highly uneven. In some places there had actually been a slight cooling (see graphic), while parts of the Arctic had warmed by 6C.
The new supercomputer's most heroic feat is to simulate the behaviour of the oceans and the atmosphere and the way they influence one another. There are complex flows of energy and water going on between the two, with the oceans slowing any warming or cooling trend in the air.
It takes the Cray one day to work through two years of simulated time. The model it runs sees the continuum of air and sea as 200,000 evenly spaced points while time is broken down into instants half an hour apart. At each of these points, for each instant, it calculates pressure, temperature, wind or current direction.
If the machine can model the natural climate reasonably well, it should be able to simulate one perturbed by pollution. The scientists factor in the heat-trapping effect of rising concentrations of CO2 and other greenhouse gases caused by decades of rampant economic and population growth.
Early supercomputer models used to let their imaginary climates establish an equilibrium, then doubled the CO2 concentration instantly and waited to see what new equilibrium was established. This was highly unrealistic, for rather than abruptly vandalising the climate system humanity is gradually corrupting it. CO2 concentrations have been rising since the 19th century, slowly at first but accelerating gradually to today's rate of half a per cent a year. On present trends, carbon dioxide levels will reach twice their pre-industrial level of 280 parts per million by around 2070.
The Hadley scientists hope to get more credible predictions by inputting this gradual, accelerating build-up of greenhouse gases which reflects recent history and the likeliest future.
They are driven to cheating by their models' imperfections. It cannot simulate some of the subtleties of energy flow between the atmosphere and oceans. For instance, its resolution is too coarse to paint a true picture of what the North Atlantic's Gulf Stream is up to. So, across the oceans' entire surface there have to be an annual cycle of 'flux corrections' which compensate for the model's errors and prevent it drifing away from what is observed in the real world.
The Hadley scientists are just beginning to model the cooling effect of aerosols. These are microscopic particles of sulphate, dust and water produced by the burning of fossil fuels as well as nature - volcanoes, sea spray, forest fires and even plankton. They appear as a haze in the lower atmosphere.
Sulphate areosol cools directly, by reflecting incoming sunlight into space, and indirectly by 'seeding' clouds and boosting the number of water droplets they contain. The clouds are then whiter and brighter and also reflect more incoming sunshine back to space.
Computer modelling of these two cooling influences is in its infancy. But they may well mask some of the global warming which has already occurred and explain why supercomputer simulations of global warming predict a greater warming to date than appears to have been the case.
Clouds, which could have a critical influence on the progress of any man-made global warming, are just as confusing. 'Things seem to be even more complicated than they were perceived to be five or 10 years ago,' Dr Tony Slingo, who heads research into cloud and aerosol problems, says.
A warmer atmosphere would probably contain more water vapour - itself a greenhouse gas - causing further warming. But more vapour could mean more clouds, which have both heating and cooling effects. They trap some of the Earth's outgoing heat radiation, but also reflect incoming sunlight, preventing solar radiation from heating the lower atmosphere and ground.
Ice clouds above 35,000ft have an overall warming effect. Dense low cloud is a coolant during daytime, as anyone who has felt the chill that follows its blotting out of the Sun will know. We need to know the fate of clouds in a warming world, but supercomputers can, as yet, give little help.
None the less, the models are becoming more accurate. Just as well, because it costs taxpayers pounds 14m a year to run the Hadley Centre. But if politicians and the public demand rapid, precise information about how the climate will change they will be disappointed.
'It's a longer term problem than we may have imagined originally, but then I would say that, wouldn't I?' Dr Slingo says.