The answer depends on what you want to predict. Extending day-to-day forecasts weeks and months in advance is impossible. Analyses of numerical weather- forecasting methods by Edward Lorenz, a founding figure of chaos theory, show that even with far more accurate measurements of the current state of the atmosphere and oceans, and unlimited computer power, the limit of reliable predictions would be 10 to 14 days. This is because tiny uncertainties in the initial measurements are magnified in the computation process.
The chaos barrier seems, however, to be at odds with the tantalising glimpses of order in the broad features of seasonal weather that have led some meteorologists to persevere with longer-range forecasts. The underlying proposition is that the climate has a memory, and that weather can leave patterns on the slowly varying components of the global climate system.
For instance, if the summer and autumn produce above- or below- average sea surface temperatures in the Atlantic, early and persistent snow cover over Canada and northern Russia, or extensive pack ice off the coast of Greenland, this can influence the sort of winter we have in Britain.
Such far-flung harbingers of winter can be combined with local information. Looking for previous examples of years when the summer and autumn weather followed the same broad pattern in the same areas can be illuminating. The sequences of weather in the British Isles often reflect wider global patterns. We can use nearly three centuries of regional temperature and rainfall records to seek out the best fits.
This use of instrumental observations is a more sophisticated form of the experience enshrined in folklore. A heavy crop of berries on trees and shrubs is a product of the weather during preceding months. In the same way, the early arrival of fieldfares, redwings and waxwings tells us something about the advance of winter over Scandinavia and Siberia.
At a more scientific level, the US Weather Service has been analysing broad weather patterns for more than 25 years, to prepare 90- day forecasts. Its winter forecasts have proved quite successful, while those for autumn have been worthless. But even the winter predictions have been mixed. The results for the eastern half of the continent have been good, while those for the other side of the Rocky Mountains have been virtually useless.
These patchy results may explain why the UK Meteorological Office does not publish seasonal forecasts. But Robert Ratcliffe, formerly head of the long-range forecasting group at the Met Office, together with his colleague, the late Ned Davis, produced summer and winter forecasts for Weather magazine for a number of years with a fair degree of success. This year Mr Ratcliffe and I collaborated to prepare a forecast for the current winter.
Combining all available meteorological evidence for 1992 produced a reasonably consistent picture. We had a relatively warm but rather wet summer, which started well but tailed off badly. This downward pattern continued in September, and October was particularly cold. Past years with similar patterns of summer and autumn weather were often followed by winters that featured lengthy and intense cold spells.
Other features of the autumn weather circulation patterns in the northern hemisphere pointed in the same direction. Furthermore, the summer of 1992 was much cooler than in other recent years because of the cumulative effects of the eruption of the volcano Pinatubo in the Philippines in June 1991. This cancelled out the effects of global warming and made conditions more like the years used to guide our thinking.
On the basis of the available evidence, we concluded that it would be a variable winter, but with a prolonged cold spell most probably around January, conditions that frequently result in freezing fog and dangerous road conditions in Britain. The very cold spell over Christmas fitted our forecast. It was cold enough in many areas for the Department of Social Security to pay out extra benefits - much colder in central and northern England than in the South-east - but we will have to see what the second half of winter brings before claiming success.
One year we identified as having a similar pattern to 1992 was 1837. That year was followed by a cold winter, with January 1838 featuring some of the lowest temperatures ever recorded in southern England. It became known as Murphy's winter, after the publisher of an almanac who correctly forecast a hard winter - and made a small fortune from sales. We hope that our efforts to establish a firmer scientific footing for seasonal forecasting will produce more frequent successes.
'Weather' is published monthly by the Royal Meteorological Society, 104 Oxford Road, Reading, Berkshire RG1 7LJ, at pounds 24 a year.