Weather forecast accuracy

[lethargic_man]

One of the subjects with which I disagree with livredor is the utility of weather forecasts. livredor holds they're useless and there's no point paying any attention to them; by contrast, I rely on them to know whether to take waterproofs with me when I'm cycling to work, and they rarely let me down.

After the last time we disagreed over this, I thought: this is something that can be objectively proved one way or the other: I can write a little program which fetches the weather forecast from the Met Office website, sees how many times the Met Office change it, and then gets the recently-observed weather data to tell whether yesterday's forecasts are accurate or not; and then puts the data together into reports with pretty graphs and things.

This sat on my to-do list for rather a long time, but eventually percolated to the top, and I started writing such a program. An hour and a half in, I thought: this is silly; I'm sure somebody else must have done this already. And indeed they had. (And so, in this age we live in, there's never any point doing anything because not only has someone done it already, but they've put the results up on the World Wide Web where it's only a google away.)

Anyhow, the results were surprising (for me at least). I was expecting a higher accuracy than 55% for the one-day forecast. (And this page seems to say the four-day forecasts are more accurate than the one-day ones; I think this shows they're not using enough data. OTOH, what I remember from looking at other pages last week seems to generally corroborate this one, so it's not complete rubbish.) On the other hand, it does show the forecast is rarely completely wrong.

The other surprising thing is that they said the BBC was more accurate than the Met Office. Aren't the BBC supposed to get their data from the Met Office? In any case, I find it's generally the other way around: The two don't always agree, and I tend to go straight to the Met Office website to get my data from the horse's mouth. (Bet you didn't know horses could forecast the weather...)

Comments

[khiemtran.livejournal.com]

Hmm. Two weeks wouldn't seem to be a big enough sample size. Also, to gauge usefulness, you probably need to weight unpredicted rain and falsely predicted rain differently.

[lethargic-man.livejournal.com]

Now, can I be bothered to google a more reliable example, or do I have to take the lazy option and finish my program myself...?

[iddewes]

I was told AccuWeather is the best.

[nancylebov.livejournal.com]

Forecast Advisor does comparisons for US weather predictions.

[rysmiel.livejournal.com]

i think we went through a phase in which weather forecasts were more accurate as the tech increased, and then became less so again as we pushed global weather into a more chaotic state, and it would seem possible to me that your and livredor's different takes on the utility of the things are to some extent correlated with that timing and when you becamde aware of the things.

One and two day forecasts here tend to be a lot more accurate than in Ireland; I read this as a feature of the difference between a continent and a maritime island.

[bluepork.livejournal.com]

I find that metcheck.com is the best.

I have known for a long time that, in this country at any rate, if you assume that the weather tomorrow will be the same as the weather today, you will be more than 50% accurate, which is higher than a lot of proper forecasts. The reason is that the weather in the UK tends to be dominated by systems and fronts which take a few days to pass. You then only need to rely on the forecasters to identify when the systems are passing.

Also, I find that local forecasts are often silly. Much better to look at the national picture and spot the fronts, rather than getting bogged down in what is, statistically, noise.

That, I think, is why the one day forecasts are often wrong too, btw.

I also find that the short term forecast is often correct, but the timing is inaccurate. Eg rain in 8 hours means rain in anything from 6 to 15 hours.

Anyway, at least you only need to forecast the weather. Try adding surf prediction in as well...

rachat de credit

[(Anonymous)]

Vous avez de bons points il, c'est pourquoi j'aime toujours verifier votre blog, Il semble que vous etes un expert dans ce domaine. maintenir le bon travail, Mon ami recommander votre blog.

Mon francais n'est pas tres bon, je suis de l'Allemagne.

Mon blog:
Credit immobilier (http://fr.wikipedia.org/wiki/Crédit_immobilier) ou Rachat de credit (http://www.rachatdecredit.net) pour locataire