Ask a dozen meteorologists what the weather will do next week and you might get a dozen different answers. Why?

Accuracy has improved immensely over the last couple of decades, with forecasts for three days in advance reckoned to be as accurate as day-ahead forecasts were 20 years ago.

But at a time when we are being bombarded from all sorts of directions with confident predictions for the entire winter ahead, it is important to remember that there are limits.

Seasonal forecasts are notoriously difficult to nail down unless the atmosphere is giving us a very strong signal, and the skill is rather different from the one required to produce a forecast for a few days or a week and a half ahead. These are not entirely reliant on but do owe a great debt to numerical weather predictions – the “models” which are produced by super computers crunching hordes of numbers that mathematically describe the state of the atmosphere.

The initial conditions as best as they can be modelled are introduced, and then thermodynamic equations are applied to produce a picture of the estimated state of the atmosphere at certain times ahead – one or three hours in the short term, six or 12 hours further ahead.

You will note some uncertainty there – “initial conditions as best as they can be modelled”, and “estimated state”. This is because the atmosphere is a vast and complicated entity, whose exact state at any one time is impossible to gauge however many observations there are on land, at sea and upwards via weather balloons or satellites. To get it spot on, the position and movement of every single particle in the air would have to be known.

This is why models, and by inference forecasts, can go wrong. Any errors or even slight inaccuracies at the start of a forecast period can become hugely amplified as the model develops down the line. This is particularly true if the atmosphere is in a notably chaotic state.

It is a chaotic and delicate system anyway but sometimes more so than others. Hence the illustrative “butterfly effect” proposed by Edward Lorenz, although his first publication on the subject stated: “One meteorologist remarked that if the theory were correct, one flap of a seagull’s wings could change the course of weather forever.”

At the moment we would seem to be in one of those harum-scarum phases, with little agreement between models and a lack of confidence among forecasters. This is where their experience and knowledge comes into play but they can also make use of what they call “ensembles”. When a model is run it produces its “deterministic” forecast but at the same time it runs 50 more times, each with a slightly different starting position. This is called an “ensemble”.

If those different model runs give quite a similar result to one another then an assumption can be made that the atmosphere is relatively stable, and a degree of confidence can be levied in the forecast.

However, if they start to diverge wildly then clearly there is too much instability for there to be much sureness, and forecasters need to weigh the probabilities and impart such uncertainty.