How good are the weather forecasts?

The BBC weather service offers countless weather forecasts on their web site. One of the most useful features being their localised weather reports. This gives a five day prediction of the weather for any location in Britain. For example, here's what the weather in Cambridge is prophesised to be.

Problem is, just how good are these predictions? After once using the BBC's weather page to book myself a nice sun-drenched Friday off work, only to find it pissing down with rain on the day, I was inspired (well... ok I got bored sometime later on) to write a small computer program to check up on the BBC. Are their five day weather predictions useful, or are they no good at all? Is the super computer in the MET office accurately modelling the world's climate, or is it resting its brain and picking out sun and rain symbols at random? We will find the answer!

So how do we tell if they're right?

Well, rather than scurry around the whole of the UK in order to check their weather divinations, I've chosen to take the easy route and use the BBC's own web pages. The five day weather predictions also include today's weather. I'm making the assumption that predicting today's weather is dead simple, so the BBC couldn't possibly get this wrong. So every day, a computer script reads the latest five day forecast, and compares today's weather with what the BBC said it would be in their previous forecasts.

Over time, as this program collects more and more of the weather predictions, we'll be able to see how often the weather forecasts were right.

What are the results?

The tables below show the accuracy of the BBC's weather forecasts, listed by time. E.g. as I write this, the table below shows that the weather forecast for Cambridge one day ahead was 53% accurate. In other words, the BBC's guess about tomorrow's weather in Cambridge was right roughly half of the time.

It's not so useful to just show the numbers for the BBC's weather accuracy on their own. You need something to compare them to. So, the tables also show the accuracy of another really simple weather predictor - guessing that tomorrow's weather will be the same as today's. Hopefully the BBC weather will be better than this?

Bournemouth

Days monitored: 4093
Days ahead12345678
BBC accuracy56 %46 %42 %38 %30 %28 %26 %25 %
BBC (yearly)56 %46 %41 %32 %29 %28 %25 %24 %
My accuracy33 %28 %27 %26 %26 %25 %24 %23 %

Cambridge

Days monitored: 4150
Days ahead12345678
BBC accuracy55 %45 %41 %36 %34 %31 %30 %31 %
BBC (yearly)56 %45 %41 %33 %32 %33 %33 %33 %
My accuracy33 %28 %26 %25 %25 %23 %25 %24 %

Cardiff

Days monitored: 4139
Days ahead12345678
BBC accuracy57 %47 %41 %36 %31 %27 %27 %26 %
BBC (yearly)59 %42 %37 %34 %31 %29 %28 %29 %
My accuracy33 %28 %26 %25 %25 %24 %23 %23 %

Durham

Days monitored: 4063
Days ahead12345678
BBC accuracy53 %43 %38 %34 %31 %28 %28 %30 %
BBC (yearly)64 %51 %45 %37 %32 %30 %30 %32 %
My accuracy32 %29 %27 %25 %25 %24 %25 %25 %

Ely

Days monitored: 4150
Days ahead12345678
BBC accuracy56 %46 %41 %37 %34 %33 %29 %30 %
BBC (yearly)59 %54 %43 %35 %29 %33 %32 %32 %
My accuracy34 %28 %26 %26 %25 %24 %24 %23 %

Glasgow

Days monitored: 4137
Days ahead12345678
BBC accuracy54 %44 %41 %35 %31 %29 %27 %26 %
BBC (yearly)58 %45 %41 %36 %32 %32 %30 %29 %
My accuracy32 %27 %24 %25 %24 %23 %23 %22 %

London City

Days monitored: 4150
Days ahead12345678
BBC accuracy56 %47 %43 %38 %35 %32 %30 %31 %
BBC (yearly)59 %48 %43 %37 %36 %35 %32 %32 %
My accuracy33 %27 %27 %26 %25 %25 %25 %25 %

Nottingham

Days monitored: 4149
Days ahead12345678
BBC accuracy55 %45 %40 %36 %32 %30 %30 %30 %
BBC (yearly)59 %48 %38 %33 %31 %29 %28 %28 %
My accuracy32 %28 %27 %24 %24 %25 %24 %24 %

Southampton

Days monitored: 4150
Days ahead12345678
BBC accuracy57 %48 %43 %38 %30 %27 %28 %26 %
BBC (yearly)59 %48 %43 %32 %26 %27 %30 %26 %
My accuracy33 %28 %27 %26 %25 %26 %24 %24 %

York

Days monitored: 4149
Days ahead12345678
BBC accuracy54 %45 %40 %36 %34 %30 %29 %32 %
BBC (yearly)55 %52 %44 %38 %35 %31 %28 %33 %
My accuracy31 %28 %27 %26 %24 %27 %24 %24 %

How do I measure 'accuracy'?

Very simply! I take the BBC's weather icons and compare them, using a bit of leeway. So if the prediction is 'Partly Cloudly', then 'Sunny Intervals' is also considered equivalent. Likewise, 'Light Showers', 'Light Rain' and 'Drizzle' are all considered close enough to be an accurate forecast.

The 'yearly' accuracy is based on the last 365 days of results, to show if the BBC is getting better or worse over time.

I'm working on adding a better scoring system, since this accuracy measurement is a bit rubbish really...

So? Are weather reports prophetic or pathetic?

So what have we learned? Can we fault the prescience of the BBC's five day forecasts? Yes. The weather predicted by the BBC for four days time was just 30-40% accurate - i.e. they get it wrong around two times out of three.

My alternate weather prediction scheme isn't stunning either - it gets a fairly consistent 25-35% accuracy, for all lengths of time. This is surprising - it means that the weather in four days time is just as likely as tomorrow's to be identical to today's. (If you can follow that and figure out a better way to phrase it, please let me know!) It is reassuring that the BBC do better than this simple prediction technique!

The moral of this is don't use the forecast to book time off work! :)

Links to more weather sites


Comments? Email me 7th April 2003