JOSEPH Hobson Jagger was fond of a little flutter, and was a frequent visitor to the Salon Prive of the casino at Monte Carlo. During one such visit in 1886, Jagger suspected that one of the roulette wheels had a faulty spindle, and having watched the proceedings for a week or so he began to bet on the numbers which he knew to be turning up with more than average frequency. He won two million francs in eight days, and immortality as the "Man who Broke the Bank at Monte Carlo".
In recent times, meteorologists have been using a similar method to assess the likely accuracy of their predictions. They do not of course, suspect that the world has a faulty spindle, but they know from experience that when they produce a forecast by computer, the forecasts are more reliable in some circumstances than in others sometimes the atmosphere is "predictable", while at other times "chaos" sets in very quickly. Methods are being developed, however, which allow the computer not only to provide a forecast, but also to give an estimate as to how accurate it is likely to be.
The technique is called "ensemble forecasting". The computer is first provided with the real weather observations, and a prediction of the future pressure and temperature pattern is obtained in the normal way. The process is then repeated 30 or 40 times, and in the case of each member of the "ensemble", some of the weather observations on which it is based are very slightly altered.
If the end result the forecast for, say, three days time is very similar for each "run" of the ensemble, it is seen as an indication that the atmosphere is in a "predictable" state, and the forecast is likely to be correct. If on the other hand the results differ widely, the atmosphere is seen as unstable and very sensitive to the absolute accuracy of the initial information. In this case, the input of the computer must be treated with caution.
The technique is often called the Monte Carlo method presumably because it is reminiscent of roulette players like Mr Jagger who watch the wheel spin a great many times to see if it has a bias. But it can be carried even further. If, for example, the ensemble consists of 10 "runs", and nine of the resulting forecasts indicate rain and only one does not, it is a reasonable inference to take the probability of rain in these circumstances as 90 per cent. In this way the computer can be used to indicate the probability of various weather events over a wide area.