And today's forecast is . . . chaotic: Bill Burroughs explains why the weather gets stuck in a mood, and so defies prediction

In case you hadn't noticed, it has been raining a lot recently. Total rainfall from September to November was 34 per cent above the long-term average for England and Wales. December was wetter still - 57 per cent above average.

Every winter the weather 'gets stuck' in one of a limited number of well-defined patterns. Here in Britain the most recognisable of these is the mild and wet westerly conditions associated with a deep depression close to Iceland. In late November, however, the pattern changed to the cold easterly weather that occurs when high pressure lurks over Scandinavia.

For around three-quarters of each winter the atmospheric circulation falls in one of four or five of these patterns. Once established, a given regime may persist for several days or longer. During such a quasi-stationary situation the weather behaves in a more predictable manner. But when the regime breaks down, the change is often rapid and unexpected.

This relative stability punctuated by sudden and less predictable changes has profound implications for both day-to- day weather forecasting and predicting the nature of future change. It means the accuracy of weather forecasts will vary substantially with the changing global circulation patterns. It also influences how the climate responds to perturbations, such as the build-up of carbon dioxide due to the burning of fossil fuels - responses that may be more subtle than computer models suggest.

The response of the atmosphere to changes in its physical properties is non- linear. So as one parameter changes, others alter in a way that is not in direct proportion to this change. This non-linear behaviour is central to Chaos Theory, and the problems of weather forecasting and climate prediction are classic examples of this discipline.

In the case of weather forecasting, this means that the quality of the forecasts is sometimes highly sensitive to the uncertainties in measuring the initial state of the atmosphere. This variation in performance tends to reflect whether the atmosphere is in transition between quasi-stationary states or is stuck in one mode. But because of uncertainty about the switch between the states it is impossible to tell by inspection whether a change will occur during the forecast period and hence whether the forecast will be good or bad.

One way to tackle this problem is to see how the predictions behave when slightly different starting conditions are used to reflect the uncertainty about the current state of the atmosphere. If, with a subtle range of starting conditions, the ensemble of forecasts look remarkably similar up to 10 days ahead, then there is a good chance that they are on the right track. If, however, each forecast diverges significantly after a few days, then clearly the atmosphere is in a less predictable mood.

At the European Centre for Medium- Range Weather Forecasts near Reading, Tim Palmer has been researching ensemble forecasting for many years. The centre now produces blocks of 33 ten-day forecasts each weekend. These show that it is possible to get a much better fix on when the weather is in a predictable regime, and this will place increasing pressure on forecasters to provide a statement on the quality of their output.

When modelling climatic change, the effects of non-linearity are best examined in a different way. If the climate is subjected to a small perturbation - say the result of natural fluctuations in ocean currents or the build-up of atmospheric carbon dioxide - its impact will vary depending on the state of the atmosphere. When it is stuck in a well-defined regime it may be of little consequence, but when in transition it could have quite an effect on which state the atmosphere next settles into. So even if the different regimes basically remain unaltered by the perturbation, the proportion of time each quasi-stationary state lasts could shift substantially and with it the global climate.

The consequence of this interpretation, as Dr Palmer expounded recently in the magazine Weather, is to suggest that the impact of a given increase in carbon dioxide may not necessarily be a proportionate global warming. Intuition suggests this perturbation would make warmer regimes more probable. In northern winters this perturbation effectively tilts the climate in favour of spending more time in the mild westerly regime. But this is not a foregone conclusion, and the reverse, in which the colder blocked pattern prevails, cannot be ruled out.

The test for the global computer models (which predict that increased carbon dioxide leads to global warming) is whether they can stimulate the statistics of different quasi-stationary patterns. In practice, they do not rise to this challenge well. So there is a suspicion that they are producing an over-simplified incremental response that may not reflect the real consequence of the non-linearity of the climate.

This conclusion is not merely a theoretical hypothesis. Recent results from new ice cores drilled in Greenland have suggested increasing evidence of chaotic behaviour in the climate. In particular they have produced dramatic insights into climatic variations during the warm period before the last Ice Age.

Because the temperature during this interglacial period, known as the Eemian, was higher than at present, it had been assumed that it might be a good model for predicting the consequences of global warming. What seems to have emerged from the new observations was wholly unexpected. Instead of a relatively stable warm climate, there appear to have been three different climatic states. Shifts between these states took place suddenly, within a decade or so, sometimes involving average temperature changes of 10C or more, and lasted anything from 70 to 5,000 years.

Measurements of parallel variations in dust level in the ice cores suggest that the fluctuations in climate were associated with significant switches in atmospheric circulation patterns. It is postulated that these shifts may have been driven by sudden alterations in the circulation of the oceans. But, whatever the explanation, they provide a chilling reminder of how non-linear responses to climatic perturbations can be amplified by the changing prevalence of atmospheric circulation regimes. They also indicate that, while computer models remain the most sophisticated analysis we have, it is not safe to assume that the response of the global climate to the build-up of carbon dioxide will be gradual, nor that it is absolutely certain to lead to warming.

(Photograph omitted)

Comments