When world experts in various aspects of weather forecasting get together, it only serves to emphasise the complexity of the problem they are tackling.

The physics of climate sounds a daunting subject. As a conference held by the Royal Meteorological Society and the Institute of Physics at the Royal Society last week showed, "daunting" is an understatement.

The key to understanding climatic change is to produce computer models that accurately simulate the global climate. The challenge is to establish which physical processes matter and how they interact. In the opening address, Sir John Houghton, co-chairman of Working Group 1 of the Intergovernmental Panel on Climate Change, identified the most difficult areas. First, there are the feedback processes associated with the build-up of greenhouse gases in the atmosphere, including the vertical distribution of water vapour and associated changes in cloudiness, which we do not fully understand. Then there is a need to know more about the dynamics of the oceans and their links with the atmosphere. Finally there is the fundamental question of estimating the natural variability of climate in the absence of human activities, and whether dramatic changes, which have not occurred for some 10,000 years, could be triggered by current events.

These issues were brought into sharp focus by Carl Wunsch of MIT, who examined how our view of the oceans has changed. The historical view was of a slowly moving system, which acted as a giant flywheel in steadying the climate. Large-scale studies now show a more complicated situation, with turbulent motion on all timescales, and at all depths. This suggests that the oceans are as chaotic as the atmosphere, and their behaviour years ahead is unpredictable and capable of sudden shifts.

The problem of clouds was spelt out by Bob Charlson of the University of Washington. Apart from the great difficulty in defining their natural properties, there are major uncertainties about the impact of human activities, whose net effect is likely to make clouds more extensive, longer lasting and able to reflect more sunlight, leading to some cooling of the climate.

John Mitchell of the Hadley Centre presented results of simulated atmospheric- ocean analyses which contained impressive representations of the climate. While human activity may lead to a temporary cooling, in the long run the warming due to the build-up of greenhouse gases will be dominant. To validate this work, better observations of the oceans are essential. To improve the models, three principal areas matter: clouds, clouds and clouds.

Behind all these issues looms the fundamental uncertainty of natural variability. In summing up, both John Mitchell and Brian Hoskins noted the difficulties posed by quasi-cyclic fluctuations: El Nino in the equatorial Pacific every few years, and the North Atlantic Oscillation, which exerts a significant influence on winter temperatures across the northern world, are challenges the models must confront.

Overall, the picture is of steady progress, with bigger and better models exploiting huge increases in computer power to boost our understanding of how the global climate can change in the future. But lurking behind this comfortable facade is the uncomfortable recognition that we may be nowhere near to getting to the bottom of what really matters in the physics of the climate. It could easily spring some nasty surprises, when we least expect it.

William Burroughs is the author of: `Does the Weather Really Matter? The social implications of climate change' (CUP, pounds 16.95).