Weather: Those magnificent men from the Meteorological Office

We all complain when the forecasters get it wrong, but the figures show that when it really matters, the imprecise science of weather forecasting yields impressively accurate results.

Although people had been trying to forecast the weather since ancient times, anything other than purely local forecasts were quite useless until the middle of the 19th century. It was Samuel Morse's invention of wireless telegraphy that made all the difference. Before that, no weather forecast, however accurate, could be transmitted quickly enough to have any relevance by the time it arrived. The weather it was predicting would always get there first.

A similar problem arose in the Fifties when meteorologists first tried to enlist the help of digital computers. Collecting and collating weather data from around the world, then feeding it into a computer's mathematical model of the aero- and hydrodynamic systems that swirl around the globe to produce our weather, the letting the machine chug through billions of calculations just to tell us whether to take an umbrella to work - it was all too much until recently for even the fastest machines. One could, in theory, provide an accurate forecast for tomorrow, but it could well take a week's computing time to do it.

Now we have the technology, and the catch phrase of modern forecasting is NWP - Numerical Weather Prediction. Feed in tens of thousands of observations worldwide, and the machine does the rest. But is it worth the effort and expense when, as we pointed out yesterday, today's weather is usually much the same as yesterday's?

In a memorable episode of The Magic Roundabout on television many years ago, Dougal, the shaggy dog, wandered into vision cluttered with a mountain of equipment. "What's that?" asked Florence, and Dougal replied with an avalanche of verbiage explaining all the technical merits of his state- of-the-art new camera. "Ooh," said Florence, sounding impressed: "Does it work?"

Much the same question might be asked of the technology now applied to the job of forecasting the weather. The crucial test is whether it can predict the days when the "much the same as yesterday" forecast breaks down. And the evidence shows that they do remarkably well.

I have been looking at the forecasts for London on 79 days in 1996 when the temperature varied by three Celsius degrees or more from that of the previous day. These were certainly the most difficult days of the year for forecasters to get right, yet for more than half of them their previous day's prediction of the midday temperature was within one degree of the observed figure.

Total scores for the 79 days showed an exact prediction on 18 occasions, a forecast that was just one degree adrift on 23 days, 16 that were two degrees out, 11 that were three degrees out, and only 11 forecasts missing their target by more than three degrees. While a "same as yesterday" forecast would have been out by an average of 4.3 degrees, the weather men's predictions differed from the true readings by an average of only 1.8 degrees. In a particularly changeable week from 22-27 July, the temperature started at 27C then went through a roller-coaster of 30, 25, 18 and 22 and 27. For those five days the predictions were 32, 24, 19, 23 and 24 respectively - a remarkable resemblance to the true figures.

For all the general variability of the weather, and the strong feeling that the dynamic system governing it could be susceptible to the total unpredictability predicted by Chaos Theory, the figures show that meteorology is on its way to becoming a precise science. Three-day-ahead forecasts are now as accurate as 24-hour forecasts were in the Seventies, so perhaps it is time to stop looking out of the window and start listening to the weather forecasts instead.