Weather forecasting

[Brian Ellis]

Very tricky thing, but I've had neighbours beta testing my weather forecasting for a few weeks. The reported accuracy is excellent, better than the official ones. The modelling is based on my own weather station for the local microclimate, along with METAR, GSF and other sites. The forecasts are calculated and displayed with the aid of three different softwares. I also have it in French. I hope to put it also in Greek, but I'm having difficulty with UTF-8 characters in php scripts.

[The PIT]

I make Monthly predictions for Sheffield and seem to be doing all right at the moment. As ever though it will go tits up sooner or later.

Since I use Net Weather quite a lot I can tell you then place is littered with people who've got a few months or weeks right and then suddenly it goes pair shaped.

One of the more interesting ones is mooncasting as ever people using this get a few hits and then go strangley quiet when they get huge misses. However what excuse can they give when the predicted weather doesn't turn up. The Moon didn't spin right.

[Brian Ellis]

As the weather is dynamic, it is constantly changing. It therefore needs enormous resources of data to follow the evolution. A lot of these, up to high altitudes, are available on the Internet. This is why two forecasts at 1-hour intervals can vary, even for the short-term ones. The bloke who wrote one of my softwares has been modelling weather in different climates from all over the world for over 20 years. He is a mathematical physicist. Just to give you an idea, I download >6 Mb of purely numerical data every hour. Of these, there are 462 buoys, 274 synoptic weather stations, about the same METAR stations etc. There is nothing from the moon! At the present state of the art, five days is about the absolute max that can be forecast with any pretence at accuracy. Anything else is just guesswork based on annual cyclic trends. For example, an anticyclone usually hits the UK sometime mid Nov to early Dec, lasting ~5-10 days, producing very cold, clear weather, often with a temp inversion. No one can tell more than a week beforehand when it is likely to arrive. 3 days before, one can guess reasonably at the barometric pressure, but no one can predict how long it will stay before a warm front pushes it away until about 3-5 days before it happens. That is why long range forecasts (>5 days) are very iffy. Historical data is not used with the software I've described.