back to article Met Office wants better supercomputer to predict extreme weather

The UK's Met Office needs bigger and better supercomputers if it is to confidently and accurately predict the weather and give emergency services a longer lead time for extreme weather conditions, a government group said today. The Science and Technology committee of MPs advised in a report on the Met Office that, despite the …

COMMENTS

This topic is closed for new posts.

Page:

  1. Graham Dawson Silver badge

    Actually I'd say its bad rep is due to the fact that it has a proven track record of getting long-term and even short-term predictions blisteringly wrong. This is nothing to do with the amount of computing power available, but everything to do with the models they use to make their predictions, models based on assumptions that are (as their track record shows) demonstrably wrong.

    Which presumably means they're really just trying to get a prestige project to justify their budget in the face of continuing government cuts.

    1. Tim Parker

      @Graham Dawson

      "Actually I'd say its bad rep is due to the fact that it has a proven track record of getting long-term and even short-term predictions blisteringly wrong. This is nothing to do with the amount of computing power available, but everything to do with the models they use to make their predictions, models based on assumptions that are (as their track record shows) demonstrably wrong."

      Blimey - I hadn't realised it was that simple, how marvellous ! If you could just pop over there and let them know what they're doing wrong, I think we'd all be really grateful. Lovely.

      1. Anonymous Coward
        Coat

        Re: @Graham Dawson

        I'd like to see the range increase from what the weather has been doing today (which strangely enough I already know) to include maybe, you know, a forecast? Not a description of what the last couple of days have been like and a report on the drought (by a reporter with an umbrella and no sense of irony). Don't get me started on what "unsettled" is supposed to mean. It *ought* to mean unpredictable and changeable but the met office seem to think it means rain.

        Mine is the one that has seen temperatures soaring across the uk today but with rain in places before becoming unsettled.

      2. David Beck
        Thumb Down

        Re: @Graham Dawson

        Huh?

        You got data, you got a model and you got a verifiable result, weather. If the data are good and the result is wrong, this would suggest the model is wrong. you don't need to know how to improve the model to validate it. That's the trick with modelling, you can tell when you are wrong without actually knowing why.

        Seems the Met's problem is admitting the model is bad, so either they want to "adjust" the data or complain about the interpretation of the result, anything other than revise the model.

    2. Code Monkey

      I'd say they have a bad rep because a lot of people don't understand probability. So if they say there's an 80% chance of rain, then if it doesn't rain people say they're wrong.

      1. smudge
        Boffin

        What does 80% chance" mean?

        Someone above said: "I'd say they have a bad rep because a lot of people don't understand probability. So if they say there's an 80% chance of rain, then if it doesn't rain people say they're wrong."

        I DO understand percentages and probability, but I've never understood what "80% chance of rain" means. Does it mean:

        a) that if I step outside during the day, there is an 80% chance that I will get wet - ie that it will be raining for 80% of the time

        or

        b) there is an 80% chance that it will rain on at least one occasion during the day?

        I think they probably mean b), but then lots of other questions arise - mostly around how much rain there will be and how long it will last. There's a big difference between 80% chance of a 5 minute shower and 80% chance of a few hours solid rainfall.

    3. Gary F
      Unhappy

      Upgrade the software, not the hardware

      I agree that their models are unreliable and proves that long term weather predication is not an exact science. Just how much computing power do they need to get the weather right? Surely they're not suggesting if they had, say, 10x more processing power their predictions would be more accurate?

      If they improved their science and enhanced their models and formulas it would make a significantly greater improvement on predictions than simply making their current software produce poor predictions more quickly! I don't trust the Met Office predictions for periods greater than 48 hours.

      1. Douglas Lowe

        Re: Upgrade the software, not the hardware

        I trust Met Office predictions past 48 hours - but then I use them to get a general idea of what the weather will be like over a period of several days, rather than trying to determine when this evening I'm least likely to get rained on walking the dogs.

        Also, increasing the processing power will allow the use of higher resolution models - which are needed if you're attempting to predict (very) localised extreme weather events. Increasing the efficiency of (already well optimised) code will not make the same increase in resolution possible.

      2. NomNomNom

        Re: Upgrade the software, not the hardware

        I am no expert but as far as I am aware they run the weather model multiple times changing the initial conditions of each run very slightly and the forecast is based on the statistical spread of all the model run results rather than a particular one. Initial conditions being slightly different can make a big difference to the results. If they just do one run they can't know how common it's path is. Another anecdote: I heard that (but I might remember the details wrong) that big storm that Michael Fish was famously involved in was not predicted because at the time they were just using a single run, and the initial conditions happened to not produce the storm. If they had done multiple runs with different starting points they'd have spotted a fair few of them generated a massive storm.

      3. NomNomNom

        Re: Upgrade the software, not the hardware

        forgot to mention the point. Another reason they want more computing power might be because they want to do more model runs for each forecast.

      4. Steve Taylor 3
        Flame

        Re: Upgrade the software, not the hardware

        > Surely they're not suggesting if they had, say, 10x more processing power their predictions would be more accurate?

        Of course they bloody well are! Do you have any idea at all about how their models work?

        They always involve dividing the world into a series of coarse gridsand performing calculations on the values which (approximately) model the weather of one grid square/cube. Finer grids mean more accuarate calculations but also a much greater need for computing power.

        1. David Beck
          Thumb Down

          Re: Re: Upgrade the software, not the hardware

          Actually it only means that they will have a more refined output from the model. If the model or data are wrong the result will be a finer grained error, not an improvement in the quality of the forecast.

          Doesn't anyone on this forum know anything about modelling?

    4. Lars Silver badge
      Pint

      @Graham Dawson

      Computer power does make a difference as the atmosphere can be divided in smaller and smaller pieces thus increasing the possibility for a more accurate estimate.

      Weather forecasting is not an easy task and the models are not fool proof but I do believe they are the same globally and improving.

      I do remember a discussion between two professors in that field where one guy said that things will get better when the computers are ten times more powerful and the other guy replied - yes, but then you will still ask for ten times the power.

  2. JimmyPage Silver badge
    Flame

    The real problem ...

    is that the Met Office's job seems to back up the Climate Change brigade, not predict the weather. Remember those successive long range "barbecue summer" and "mild winter" forecasts ? And how the Met Office got arsey and stopped issuing long-range forecasts because "the media misrepresented them" ?

    There is now so much *big* money tied up in climate change that whatever the truth is, it has fallen victim to entrenched positions.

    I guarentee the Met Office will get it's supercomputers, if it puts on the request "to investigate climate change".

    1. Douglas Lowe
      FAIL

      Re: The real problem ...

      The "barbecue summer" forecast of several years ago (which went so spectacularly wrong) *was* one which was blown out of all proportion by the news media. The actual forecast the met office was a 60% chance of a hot, dry, summer - it was pure wishful thinking (and the blatant disregard for honesty which our tabloid media displays so widely) which changed this into the "barbecue summer" prediction.

      1. JimmyPage Silver badge
        Flame

        Re: Re: The real problem ...

        It wasn't that single forecast. It was the fact that a succession of long-range forecasts were issued which supported the claims of the Climate Change brigade, but turned out to be hopelessly wrong - last winter being the nadir.

        Climate Change has ceased to be a scientific debate anymore. Science - right, wrong, sensible, stupid, good bad is not relevant anymore. It has become a political issue. With big companies, with lots of revenue on the line involved. For proof, look at the UK, where piddling around with windmills and solar panels to the exclusion of nuclear means we are likely going to see phased power cuts within the decade. Look at the howls from the well-heeled Tories as the FIT were slashed last year.

        1. NomNomNom

          Re: Re: Re: The real problem ...

          "It was the fact that a succession of long-range forecasts were issued which supported the claims of the Climate Change brigade"

          Apart from the BBQ summer one what did they get wrong? They dropped their long-range forecasts before the first winter. I think you are basing conclusions on a woefully small sample size.

        2. Code Monkey

          Re: Re: Re: The real problem ...

          You seem to be confusing "global climate" with "the weather where you are".

          I'm not making a comment on whether or not climate change is genuine (I'm yet to be convinced by either camp) - just on your straw man argument.

      2. Graham Marsden
        Stop

        Re: Re: The real problem ...

        And yet every news report on BBC News 24 today has started with something like "remember the barbecue summer that wasn't?" as if it was the Met Office that got it wrong, rather than the media misunderstanding the fact that a 60% chance means that there's a 40% chance of it *not* happening.

      3. fremsley

        Re: Re: The real problem ...

        Revisionist nonsense. I fully support the Met Office and think they do a great job in most things, but the "barbecue summer'' phrase was uttered twice by their Chief Forecaster in a UKMO-made video.

        The fault lied in the Chef Executive [one assumes] sanctioning their PR department to put pressure on scientists to use from experimental stochastic forecasting and make concrete predictions of relatively rare events- I can remember one good (i.e. barbecue) and 3-4 fair summers in my life, and I'm 44.

  3. ElNumbre
    Trollface

    New Computamabobs...

    They seem to want new super computermabobs every 18 months or so. Why don't they just build one that can be upgraded? Clusters and the like?

    <<Sent from my 5 year old Dell Laptop that won't be upgraded as it still does the job>>

    1. Anonymous Coward
      Anonymous Coward

      Re: New Computamabobs...

      Cost.

      At this price range, power consumed is significant.

      After about 2 years, its typically cheaper to buy a new HPC machine than run the old one, in terms of electricity.

      Example: when we upgraded our cluster back in 2008, we went from a 3 TF machine that consumed 120 kW of electricity to a 30 TF machine consuming 110 kW.

      Now you _can_ upgrade the machine (we did in 2010) by adding better processors, but by 3-4 years, everything else in the system is getting old, too: the internal infiniband networks in particular.

      And then there's the miracle of warranty engineering, and the way everything starts to break magically at 2years + 1 day old.

    2. Silverburn
      Happy

      Re: New Computamabobs...

      Your Dell lasted 5 years??? You sure it's a Dell?

      1. Anonymous Coward
        Anonymous Coward

        @Silverburn

        well, my Dell desktop lasted for more than 7 years and still does the job. Except that, for a month or two now, it sometimes pulls a greek, i.e. it doesn't do anything at all.

        1. Anonymous Coward
          Anonymous Coward

          Re:new Computamabobs

          Well if extra power is so critical just put up a windy mill on the roof of the building. All the power you'll ever need all the time and you won't be adding to climate change.

          Mwahaahhaaa!

          1. Anonymous Coward
            Anonymous Coward

            Re: Re:new Computamabobs

            How does having some flour made on your roof help with supercomputing?

            1. ToddRundgren

              Re: Re: Re:new Computamabobs

              "How does having some flour made on your roof help with supercomputing?"

              Camberwik Green versus Trumpton, discuss!

        2. Anonymous Coward
          Anonymous Coward

          Re: @Silverburn

          My Dell home desktop works fine .....8 years and counting....

          It has an easy life though, since I don't need the latest gaming and sound whossnames in a machine, just run MS office, do some surfing, email, handle some dig cam jpegs etc....bit of extra Ram....sorted.

          Laptop though ...18 months, then phut.

      2. ToddRundgren

        Re: Re: New Computamabobs...

        Don't even think about buying a "serious" facility from Dell, it will be disastrous. Ask UCL, Imperial, Manchester etc etc etc. Actually asked the users, rather than the people who procured it!

    3. I. Aproveofitspendingonspecificprojects
      FAIL

      Re: New Computamabobs...

      @ElNumbre

      Correct me if I am wrong but it took more than 18 months to get the last one up and running.

      When they got it running properly, they started asking once more. It's a finance war against the mandains, similar to the military ones.

      The biggest problem with Superdoopers is they consume inordinate amounts of energy queuing up algorithms. What is really needed is a device to save that 98% energy flux heating the environment with what is something like 10 megawats of waste.

      Either that or find out what actually causes weather.

      Nobody seems to be working on that one.

    4. ToddRundgren
      Mushroom

      Re: New Computamabobs...

      They don't buy clusters, (apart from small ones), because they like being an IBM, or Cray or CDC, or NEC customer and to therefore feel really important.

      They are civil servants for heavens sake, what do you expect?

  4. Amonymous Ocward
    Go

    Distribute it?

    Offer a program for people to use idle cpu power like the folding@home and other systems do. There's your new supercomputer for a fraction of the price, and keeps upgrading itself for you.

    They have 60,000 followers on twitter, there's probably a few thousand willing to do this, if not more.

    1. Anonymous Coward
      Anonymous Coward

      Re: Distribute it?

      That's a very unreliable and incredibly energy in-efficient way of supercomputing.

      Besides, I doubt that the MOD would be very happy about distributing their code to all and sundry.

    2. Martin Gregorie

      'Idle' computer power isn't so idle these days

      Now that modern CPUs can adjust their clock speed and number of active cores to match the load the situation with regard to 'free CPU cycles' has changed. Back in the day of P3 and P4 chips the CPU ran at the same speed and used the same amount of power regardless of whether it was working hard or twiddling its thumbs, so it made sense to donate the idle cycles to something like SETI@home or MalariaControl.

      However, now there are no 'free CPU cycles' because the clock slows down and cores are stopped instead of being left to spin in idle.

      I've just upgraded my old P3 box, which ran at a fixed speed, by replacing it with a dual core Athlon system, which does all that good frequency stepping and shuts down the second core when its not needed. I haven't measured its power consumption yet, but the effect is highly audible: once the OS had been installed and tweaked to match my workload it was pleasantly quiet - until I enabled BOINC, which masterminds the use of 'free cycles' to, in my case, run SETI@home and MalariaControl. The box promptly ramped up to full power, complete with howling fans that could be heard throughout the house. After three days of this I realised that I wasn't donating 'free cycles' at all, but instead I was paying for each one of the little buggers through the electricity meter and wear and tear on the PC. I've now deleted BOINC and the apps it managed. Result: peace, quiet and reduced running costs.

      1. mark 63 Silver badge

        Re: 'Idle' computer power isn't so idle these days

        good points Martin.

        Maybe SETI and such should adjust the interface so you can select how many of your hard earned noisy cycles you want to donate

        1. Daniel Evans

          Re: Re: 'Idle' computer power isn't so idle these days

          There is (was?) a climate modelling project out there - but as with all these things, it was designed to give a client a chunk of modelling data ("model 1830-2030 using these initial parameters") - and it didn't care if you took a month to do it. The MET want to run single iterations of their model, but in time for the evening's forecast, using what I'd imagine is "a lot" of data, which can't be split up into areas (easily, anyway) - you can't just model North America on one client and Europe on the other.

      2. ToddRundgren

        Re: 'Idle' computer power isn't so idle these days

        @Martin Gregorie

        Why and how has SETI got into this discussion?

        Climate models use huge amounts of I/O period

    3. Anonymous Coward
      Anonymous Coward

      Re: Distribute it?

      I have knowledge of the typical queues on Met Office's current supercomputers. I can tell you they are never idle, there is always a queue of work waiting.

      Unfortunately, the type of work that is done by weather forcasting systems needs huge network bandwidth (it's not the nice compartmentalised units of work like folding@home). The Power6 systems that were installed early in 2009 have 8 planes of DDR Infiniband in a fat tree configuration, giving an aggregate point-to-point bandwidth of 20GB (that's bytes, not bits) to/from all compute nodes simultaneously. I don't think you would be able to match this using the type of networks home and desktop systems use.

      1. Andrew Jones 2
        WTF?

        Re: Re: Distribute it?

        Surely - that post right proves then - that throwing more hardware at the situation is not going to make things any better? Maybe they won't be happy until they have the largest computer known to man but we still complain that it isn't enough to forecast the weather.

  5. Adrian Challinor
    WTF?

    Helps if you actually forecast

    Last time I watched a weather report it was all about how cold it was in some god forsaken part of Blighty I wasn't in, and hadn't been in. If I had been there, I would have known because I was there. As it was, I didn't care.

    Neither did I care for all the photo's showing me icicles on a fence, or roof. Because, honestly, I have never seen them before.

    How about going back to forecasts with real symbols, not these cartoon characters. And isobars - just because the soap-opera watching, day time TV brigade find anything more complex to understand doesn't mean the rest of us can't.

    And don't even get me started on how this affects global warming. You are a weather forecasting team. Forecast the weather and leave the rest well alone.

    So - Met Office - if you want to get some more toys, get back to your job, do some forecasts about the weather.

    1. Peter Gathercole Silver badge

      Re: Helps if you actually forecast

      Try the localised detailed forecast from the Met Office's web site, rather than relying on some summary forecast where the poor forecaster has to cram some approximation of the weather for a region the size of several counties into a few 10's of seconds.

      You might as well try to describe the colour of London as seen from Google Earth in 10 words or less.

      1. Arfur Smiff
        Thumb Up

        Loosed for words

        "You might as well try to describe the colour of London as seen from Google Earth in 10 words or less."

        Grey and brown?

      2. Homard
        Happy

        Re: Re: Helps if you actually forecast

        Easy : smoke grey

  6. Anonymous Coward
    Anonymous Coward

    Wouldn’t we all!

    UK boffins, both Uni’s and RCUK are crying out for new HPC and a replacement for HECToR, but things are really stalling and one of the main BIG problems is funding, as capital funding for both Uni’s and RCUK have been frozen.

    And as has been mentioned climate modelling is not just done by the Met Office! NERC has a national capability in climate modelling and Uni’s such as UEA also play a big part.

  7. Anonymous Coward
    Anonymous Coward

    Non Story...

    I spoke to my partner about this story this morning who works for the Met Office - Basically what's happened here is that the current main supercomputer is EOL for its primary use and need to be replaced, this is standard and an ongoing situation. The current machines will all be re-used for other tasks, they're not about to be chucked out, they will be replaced by more up-to-date hardware.

    The main story here is that the MOD no longer fund (as much of) the Met office and that money for the computers needs to be found from elsewhere.

  8. 0laf
    Black Helicopters

    Extreme weather prediction?

    Normally extreme weather occurs shortly after the plebs start to stamp their feet about excessive green taxation.

    I don't think they need a super computer to predict that.

  9. Anonymous Coward
    Flame

    Evidence please.

    @Dawson:

    Evidence to back that up?

    The UK Met Office is independently audited on its record and reliability (as are most Met offices).

    Have a look e.g. in its annual reports for references. Typically it returns a 10-15 fold benefit to the country : crops, cost saved by knowing of incoming events, etc.

    On seasonal and long-term forecasting, the problem is lack of input data, mostly, particularly ocean data (yes, models will be wrong, too, but unless you know what the ocean currents were, you can't compare observations and model to diagnose the model ...). This has been changing dramatically in the last 5-10 years with Argo floats, new satellites for salinity, sea surface height, ...

    We can now do seasonal forecasting for certain areas of the globe 3-7 years out. Not Europe, yet, but we have good reason to believe it can be done. Seasonal and decadal forecasts have been visibly getting better year-on-year. I've more experience with the ECMWF and other European climate models that the recent HadGEM ones from UKMO, but they've all been improving, with increases in computing power enabling us to get certain features right.

    When sources like the Daily Mail and Telegraph diss the UKMO because they don't like climate science, and then go on to predict -20 Novembers and blizzard headlines (which, didn't happen, not that they noticed), I tend to call BS.

  10. Anonymous Coward
    Anonymous Coward

    make greater use of probabilistic information in their weather forecasts

    They mean treat the viewers as adults? Not likely!

    We're all 3 years old really and demand that we have sensation and will throw a tantrum if you either give us too much information, or lie to us because you've not given enough information.

    WRRAAH!

    1. Peter Gathercole Silver badge

      Re: make greater use of probabilistic information in their weather forecasts

      The way the weather is presented in the media is controlled by the media, not the Met Office.

      It's true that some, like the BBC, actually use Met Office forecasters, but the presentation style is normally controlled by the broadcaster, not the forecasters.

      If you want to see isobars, look at the Met Office web site (try the link http://www.metoffice.gov.uk/weather/uk/surface_pressure.html directly), not the forecast on the telly.

    2. Anonymous Coward
      Anonymous Coward

      Re: make greater use of probabilistic information in their weather forecasts

      I should be all in favour of probabilistic forecasting, except this is the country where people have trouble with concepts like 'greater than' and 'less than'

      http://www.guardian.co.uk/education/mortarboard/2007/nov/08/minusagraspforfigures

Page:

This topic is closed for new posts.

Other stories you might like