Supercomputer to improve UK weather The Met Office, home of UK weather forecasting, is getting a shiny new supercomputer from IBM. The £33m contract with IBM, which covers maintenance, support and a mid-life upgrade, will run until 2013. The machine should have total peak performance of 1 Petaflop by 2011 - some 30 times more powerful than the Met's current …
With a small w.
If they had windows in their offices they could have regional observers look outside and see whether it's raining or not, and modify their computer-based regional short term forecasts accordingly.
The Met Office already have national live (ish) rainfall radar coverage across the UK, and by watching the progress of the rain in the previous few hours you can make a very good guess as to what the short term forecast should say about rain. But they don't seem to do this, judging by the number of times the forecasts I watch have said "dry for the next 4+ hours" whilst the radar has said "rain coming your way Real Soon Now" (usually followed by "it's raining, men").
So maybe radar's too high tech for the Met, maybe they need windows.
And then there's the Met Office web site itself, which since last year's redesign is absolutely totally completely utterly useless on the kind of web device people use when out and about ie when the weather forecast is *important* (mobile phone, PDA, etc). Try it, you won't like it.
Met Office rainfall radar: http://www.metoffice.gov.uk/weather/uk/radar/index.html
Meteox rainfall radar (includes a prediction of where it will rain in the next few hours): http://www.meteox.com/h.aspx?r=&jaar=-3&soort=loop1uur
"Once up and running the computer is expected to be the second most powerful in the UK, behind the University of Edinburgh's Hector machine. It will also be in the top 20 supercomputers worldwide"
What about those really powerful computers that certain doughnut shaped Government facilities have near J11 M5 - hold on, someone seems to be banging on my door................
Super Whizzy graphics and a fully interactive presentation are all well and good but er pardon me, your weather forecasts are crap.
Before 'the office' had any super computers at all, the weather forecasts were much better.
I think that 'the office' are trying to be too accurate and so will never get it right.
Bring back old Bob and his lucky pine cone!
Paris because she maight get a wash and blow dry if the weather forecast is right
What I don't get about this though is the "low-probability high-impact" bit. With my limited understanding of meteorology I'd expect that high-impact weather is pretty much always plain as daylight. "And the weather forecast: There's a five percent chance that a hurricane will pop up above the Isle of Man tonight, so you'd better brace yourselves."
"The Met Office will also use the machine for researching climate change."
Shouldn't that read...
The Met Office will also use the machine for _causing_ climate change.
How much power do you think this thing uses? And I bet it's on all day and all night, running its simulations.
....to improve the current forecasting. Now I know this is about the Met Office, but thats where Auntie gets their forecasts from.
The 24 hour forecast and the 5 day forecast on the beeb bear no relation to each other at all. The local beeb weather at 10:30 pm and the national beeb weather for your your region at 10:28 pm are mostly completely different. I appreciate that weather forecasting is nigh on impossible, but before investing 33m of our money in a nerw computer, can we at least invest a few pence in some polish people to check that the 85 different arms of the beeb are at least saying the same thing. Or display a note at the bottom of the screen saying, "This is our best guess - please disregard other forecasts." Grrr.....
Uh, who hasn't heard of the observing network? I suspect these "weather observers" use windows (or doors as egress to the scary "outside", scorning the dangers of paedophiles, axe murderers and Al Qaida in the selfless persuit of knowledge) to view the weather and report on it.
Unfortunately, the view outside doesn't yet contain any tachyon emissions so the view to the future (forecasting) is not amenable to the "look out the window" method.
"Before 'the office' had any super computers at all, the weather forecasts were much better."
So you've kept a record of how often the weather forecast has been right and how often it's been wrong, then? I'm pretty sure that the Met Office would like those figures... Unless you pulled them from your arse.
Actually, "low probability high impact" events can be hard to predict. If you recall the floods in the South of France several years ago, they were due to quite localised weather conditions. At the time the predictive model for the weather over the next day did not show this, and so they were totally unprepared. A few years later they had more compute power so could up the resolution of the grid used in the predictive model, and they revisited the data. The new higher resolution correctly predicted the localised conditions, and if it had been available at the time, some warning could have been given. So there you go.
So does this mean that
www.bbc.co.uk/weather
www.metcheck.com
www.weather.co.uk
Might all actually tally with each other?
(Darn, they are all pretty similar at the moment, first time in history. Although none of them mention the rain outside my window right now.)
I've lost count of the number of countries I've visited where the forecast has been something like: "If you live here there is a 40% chance of rain." The met office don't think we can cope with that, they seem to think we can only cope with certanties. So they tell us that it either will or won't rain. Of course this massively increases the chances of them being wrong.
If you look at there lovely high resolution map and see that it shows rain for 9am over my house and it doesn't rain at that time then the forecast is wrong. If, OTOH, they said there was a 60% chance of rain over my house at 9am and it didn't rain then the forecast would be sort of 40% right.
The more granular they try to make the forecast the more likely they are to be wrong.
The biggest problem I have is their arrogance. They will seldom admit to being wrong. They just rabbit on about how accurate their forecasts are no matter how much evidence you present to the contrary. And if they do admit to being wrong they always qualify it with the curious statement "it isn't a prediction it's a forecast". Well my dictionary contains the entry:
forecast n. - a prediction, esp. as to the weather.
So a forecast is a prediction and the Met Office are often wrong.
Occasionally I like to record what mondays five day forecast says is going to happen on friday and then check back on every forecast up to and including friday morning. This is, I suspect, how they always manage to claim to be right. They change it so often that one of those forecasts will have been right.
New Scientist published a study that compared the Met office's output with a simple "much the same as yesterday's" blanket forecast, and found the latter to be more accurate. Je reste mon valise.
Oh, and Aimee, you poor confused person, post first - think later. Any ful know that with a shiny new computer the Met office can start offering some better weather forecasts, and by christ we could do with it.
Didn't the Met Office use Mr Fish's infamous faux pas as an excuse to bin their old Cray in the 80's and install a stonking great liquid nitrogen cooled, go-faster striped jobbie at vast expense only to decide it wasn't good enough after all, binned that and went back to Cray?
Paris? 'cos I reckon she'd love a stonking great liquid nitrogen cooled, go-faster striped jobbie too :-)
Simulated, of course.
Met Office - official UK home of climate. What a limp excuse for a new box, but even Met men must have their jollies, one supposes.
The rest of us just have weather - or, after removal of the e, war. So this new box will presumably be called - George.
Here's a machine they could do better forecasts with:
http://cgi.ebay.co.uk/Zambretti-Weather-Forecaster-Very-Accurate-NEW_W0QQitemZ120288556277QQcmdZViewItem?hash=item120288556277&_trkparms=72%3A984%7C39%3A1%7C66%3A2%7C65%3A12&_trksid=p3286.c0.m14.l1318
Shuffle a pack of cards with icons of rain, sun, snow, hail, hellfire, sleet, wind on them.
Get Michael Fish to pick a few each day live on Telly. Terry Wogan can hold the deck of cards and make unfunny old fart quips.
That'll be about as accurate as a super computer, far cheaper and with the added possibility of some really exciting weather when Michael Fishes guesswork backfires and we get a Tsunami on the Tyne.
We'll also want to know when there'll be leaves on the railway tracks (thus bringing the rail network to a halt) and if we'll get a light sprinkling of snow (bringing the road networks to a halt) and the possibility of more than 3 hours sun in one go (bringing the entire country to a halt)
Mr. Fish can pick those potential showstopper options out of a large hat, live on Telly each evening at 8.
Paris, because when the weather is bad, she goes tits up.
They replaced the Cray system in 2004, and they're replacing the NEC now with one due to run until 2013? Get through them a bit, don't they? I'm sure they could make do with what they've got. There are plenty of internet sites that do weather predictions. Granted, I am starting to think perhaps I should expect the opposite of what the Beeb tells me. Last week it was predicting the rain would stop after Sunday!
x
I seem to remember one of the Universities in California a few years back had designed a computer model to predict rain. They ran it for a year and found it something like 85% accurate and were really pleased until someone pointed out that just saying "No rain" every day would have been over 80% accurate for that year.
Sorry, we're not going to be able to predict weather until that butterfly in Japan quits flapping its wings.
It's not the oomph in their computers, it's the modelling they use that needs looking at. A day's correct forecast is most unusual, and even current conditions occasionally elude the Met Office. I usually get my Met Office forecasts from the BBC, which is pretty convenient, but rarely right.
Try typing in the name of your town and check that forecast and the current conditions, then try entering your postcode instead. The range of temperatures, rain or sun can be breathtakingly wrong.
If the Met Office says sunshine, I'll get my coat.
If your model is predicting rain and you say no rain 100% of the time so it doesn't predict any day that it rains then it's 0% accurate. Measuring performance is complicated as it needs to take into account what you are trying to predict and give that weight. All the silly examples people have given (such as tomorrow's weather will be the same as today's) would perform much worse than the Met Office forecasts.
It's true.. the weather today is likely to be the same as yesterday, the question is how good is the "same as yesterday's" forecast when the weather isn't the same as yesterday? That's right, it's 0% accurate, which is impossible for the Met Office forecast to be less accurate than so your comment is tripe.
Unfortunately, "same as today" gets less and less accurate geometrically.
75% right for a one-day forecast
56% two days
42% three days
23% five days.
a model forecast is 85% and gets less accurate depending on how predictable the system is, so is generally linearly (or sub-linearly less accurate:
85% right for a one-day forecast
75% two days
65% three
45% five
(figures are a little pessimistic).
So a five-day forecast is better than saying "the same as today" is in three days.
Try living in Exeter! At the very least you'd expect the forcast for the city hosting the Met Office to be the best in the country, yet loads of times they've said it'll be sunny, and only a mile away from their office it's pissing it down, and not just briefly!
Apparently they don't actually have a weather station at the Exeter site itself! Perhaps that's where they should concentrate their money.
@Mark: "Unfortunately, "same as today" gets less and less accurate geometrically. 56% two days, 42% three days, 23% five days."
And for climate models?
"The scientific evidence is overwhelming — our climate is changing."
http://www.metoffice.gov.uk/research/hadleycentre/
Give these clowns another £35m!
In your imagination - you're a superhero!
So hello again, Mark from the Met Office. Er, Climate Center?
It makes me feel good to know my taxes are going towards non-productive labour who can sit at a PC and type insults to El Reg readers all day. Twattery at the public's expense.
Why is the AC so concerned about people working for the Met Office? Why do they consider working for the Met Office bad? Why do they jump to conclusions about people working for the Met Office (especially when they are wrong)?
Stephen kept saying that it didn't matter who he worked for, he was just a physicist who was concerned and interested. Well, I'm an astrophysicist and concerned and interested. My fourth year project was modeling the boundary layer of a white dwarf accreting binary, so I know a *little* about fluid dynamics in discrete models.
So why is it important to some people to tell everyone someone works for the Met Office? Even if I DID work for them, why would that make a difference?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
