Catch up on stories from the past week (and beyond) at the Slashdot story archive


Forgot your password?

Japan Plans 30-Year Supercomputer Forecasts 200

BaltikaTroika writes "According to a ministry representative, 'Japan is planning ultra long-range 30-year weather forecasts that will predict typhoons, storms, blizzards, droughts and other inclement weather.' Maybe they should tell their secret to my local weatherman, who usually can't even get tomorrow's weather right. Whatever happened to chaos?"
This discussion has been archived. No new comments can be posted.

Japan Plans 30-Year Supercomputer Forecasts

Comments Filter:
  • Chaos? (Score:5, Funny)

    by TheAngryMob ( 49125 ) on Wednesday July 19, 2006 @10:05AM (#15742974) Homepage
    Whatever happened to chaos?

    Pfft. Chaos is so predictable.
    • Re:Chaos? (Score:5, Insightful)

      by neonprimetime ( 528653 ) on Wednesday July 19, 2006 @10:15AM (#15743062)
      Early warning could enable the government to allocate money and resources to potential disaster areas before disaster strikes.

      Now this would be total chaos. WTF are they thinking? Oh ... the supercomputer says we're going to get hit with a Tsunami in 2024, oh please oh please government start giving us money now so we can squander it early!
    • Pfft. Chaos is so predictable.

      It is funny how predictable it is that every time there's a story about long-range forecasting, someone will bring up Chaos Theory...

      • Re:Chaos? (Score:3, Insightful)

        by mrxak ( 727974 )
        Because that's what forcasting is all about. We don't know enough of the variables so we have to have to make best estimates on a chaotic system.

        Any predictive computer will ultimately fail, because you can't compress the universe into a computer smaller than that universe, and we are unable to figure out every equation that's being calculated anyway. You might get data that's "good enough" for 30 years, but the deviation will only increase with time. That's why weather predictions are generally only good 3
        • Yes, but *climate* predictions can be accurate for much longer. And there are intermediate levels that fall between climate and weather that are more easily predicted than local weather (and especially than micro-clime weather).

          This is a popularized version of an article that was probably originally written in Japanese. And I doubt that the translator was a climate modeling specialist.

          (OTOH, this *could* just be pork-barrel politics.)
          • Here is a story with a little more background on Earth Simulator's modeling prowess. Near the end, it states that it can predict and track typhoons with very high accuracy:

  ,,1 517946,00.html []

            • But he's predicting the paths of those typhoons in real time, not typhoons predicted to occur 10 years from now.

              The predictions won't be at a fine grained level...even if they manage to get the modeled area down in size to an acre (and they'd like even finer) they won't be able to escape chaotic effects. But many chaotic effects can be "summarized", as temperature "summarizes" the speeds of atoms within a particular piece of material. So you may not be able to predict details (a typhoon will strike land n
        • Any predictive computer will ultimately fail, because you can't compress the universe into a computer smaller than that universe,

          Not if the universe is infinitely complex!
    • Pfft. Chaos is so predictable.

      It's funny 'cause it's true.

      Chaotic processes can be predicted with great accuracy for short time into the future, but can't be predicted a long time into the future.

      Random processes cannot be predicted in the immediate future, but generally can be predicted a long time into the future.

      Not only that, but chaotic processes can be controlled with minimal force and be predictable forever.
      • Re:Chaos? (Score:2, Insightful)

        Chaotic processes can be predicted with great accuracy for short time into the future, but can't be predicted a long time into the future.

        That entirely depends upon the process in question and the selection of initial variables.

        Sensitivity to initial variables and deviation from the expected path are what makes chaotic functions fun.

        For some equations and parameters the expected path can be estimated with great precision, however move a fraction to the left and they will spin wildly out of control.
        Just look
    • OTOH, this sounds to me, like the very predictable scenario of "the uncuttable budget". Having lived in Japan for 3 years I learned that government budgets, once granted, are inviolable for *eternity*. Why do you think they still do MagLev research in spite of every one else in the world having long since abandoned it? The budget for MagLev research is uncuttable. Until that money can be absorbed by another department in some face-saving way, MagLev research will continue.

      Probably there's a budget item
    • 40 years of darkness, earthquakes, volcanos... the dead rising from the grave... human sacrifice, dogs and cats, living together...

      mass hysteria!
  • by oudzeeman ( 684485 ) on Wednesday July 19, 2006 @10:07AM (#15742993)
    your weatherperson is trying to be fairly specific. I admit to not reading the article, but I do know a little about computer simulation, and I would guess they are looking at larger trends in temperature and storm patterns. Not trying to accurately predict daily temperatures and precipitation like your weatherman (who interprets/puts a local spin on data (s)he gets from noaa).
  • by Foobar of Borg ( 690622 ) on Wednesday July 19, 2006 @10:09AM (#15743006)
    Maybe they should tell their secret to my local weatherman, who usually can't even get tomorrow's weather right.

    Actually, those days are pretty much gone now. With all the latest computational models for weather, as opposed to what was essentially pattern matching before, I find that the weather forecasts on the whole are pretty accurate out to a few days. As for 30 years, I would be more than a little skeptical since you even have to account for things like solar flares and sunspots, or you get small inaccuracies that will grow more massive the further out you get. But, with the new Hello Kitty Supercomputer Center, perhaps they are able to account for this in their computations.

    • by Jerf ( 17166 ) on Wednesday July 19, 2006 @10:20AM (#15743094) Journal
      It might depend where you are.

      In Michigan, sure, sometimes they get a week right.

      On the other hand, sometimes they're so far off you can barely recognize the week. What seems to happen is a lot of storms stall that they don't expect, or they expect something to stall and it doesn't.

      Probably the funniest was the recent "hurricane" over Michigan (about a month ago), which even made Fark. This storm complex stalled for a week and change, and basically every day of the week, the prediction was that it would move away by tomorrow.

      Michigan seems to be at a meeting point for storm systems coming from the West, cold air coming from Canada, and wet, moist air coming from the Gulf. Predicting which will "win" for any given day seems to give the models fits. For example, the worst winter storms for us are when the cold Canadian air meets the warm, moist Gulf air, but predicting exactly where they will meet and drop all the snow seems to have an error bar of several hundred miles (i.e., for a prediction of hitting Lansing, smack dab in the middle of the lower peninsula, you're looking at it actually hitting anywhere from mid-Ohio to the top of the UP.) I've noticed that for predicting precipitation, you're almost better off just watching a couple of hours of the radar loop and making your own prediction.
      • by NichG ( 62224 ) on Wednesday July 19, 2006 @12:18PM (#15744012)
        This gives me an idea for an interesting analysis. I wonder if you take all the weather predictions for the last 20 years or so and compare with the actual weather, if you'd see any patterns when you plot a map of the error as a function of location (and perhaps isolate it to the weather during a particular time of year). If there are particular locations which end up being tipping points, then that tells you something about the dynamics and where you need the highest resolution when you're building your models.

        Probably someone has already done this though.
  • by djupedal ( 584558 ) on Wednesday July 19, 2006 @10:09AM (#15743008)
    If there is one thing the Japanese know how to do, it is gather information. And with a few thousand years of weather logs to work from, they became quite handy at accurate short range weather prediction years ago, with nothing more than an abacus.

    What we have here is the 'bullet-train syndrome' at work, where they don't just move from weeks to months, or months to years...they jump to decades. Hubris aside, this is very typical of the Japanese culture and a natural 'next step', actually.
  • RTFA, submitter (Score:5, Informative)

    by Anonymous Coward on Wednesday July 19, 2006 @10:10AM (#15743012)
    The results will help establish predictable routes for typhoons and identify areas that are recurring targets for heavy rains, abundant snow, high waves, heavy winds, scorching heat or crop-threatening droughts.
    They're not trying to forecast weather 30 years in the future, they're just looking for statistical trends in locations where hurricanes and such are more likely to occur, based on predictions of the overall global climate. Things like "there are probably going to be 20-40% more typhoons off the east coast of Japan in 10 years", not "watch out, Tokyo is going to be hit by a tsunami on August 12, 2032".
    • Tokyo is going to be hit by a tsunami on August 12, 2032

      I better cancel my holiday right away. Thanks for the warning.
    • Even the article got it wrong, it starts with " Japan is planning ultra long-range 30-year weather forecasts", and later states "Japan's science ministry hopes to calculate long-term patterns in the interaction of atmospheric pressure, air temperatures, ocean currents and sea temperatures", i.e. climate modelling.

      This is nothing new either. Earth Simulator has been used for these things for many years, and was the worlds fastest supercomputer for several years.
    • "Tokyo is going to be hit by a tsunami on August 12, 2032".

      Whaaat? I was gonna vacation in Tokyo that August. Oh well, Fiji it is.
  • Seems like the submitter didin't even bother to read the article.
    ... Japan's science ministry hopes to calculate long-term patterns in the interaction of atmospheric pressure, air temperatures, ocean currents and sea temperatures, ...
    The weather forcast for tomorrow has nothing to do with long-term patterns. Anyway, why oh why do I complain, this is /. after all ...
  • Actually Useful (Score:5, Interesting)

    by Ignignot ( 782335 ) on Wednesday July 19, 2006 @10:12AM (#15743033) Journal
    Everyone is going to talk about how the buttefly effect [] makes this useless, and that is true for any sort of instantaneous weather. However, there are many things that affect weather cycles that are much more predictable. First is El Nino/La Nina [] which oscillates every few years. Then there are other oceanic oscillators that operate on a decade or longer cycle. Also there is solar output [] and human output. Add all of these up and you may be able to predict the frequency and severity of storms, the probablility of different weather patterns, etc. You will be able to plan for these events which will be 30 years down the road, and be able to do something about them - like build buildings capable of withstanding stronger typhoons, or rising sea levels, or what have you.

    But never, in no way, will someone be able to tell you if it will rain in 3 weeks, let alone 30 years. I've studied the accuracy of forecasts quite a bit (as an energy analyst), and you can't get much better than climatology once you go 2 weeks out.
    • Re:Actually Useful (Score:2, Interesting)

      by Gospodin ( 547743 )

      So here's something I'm honestly curious about that maybe you could answer: Why did weather forecasting recently go from 5-day or 7-day forecasts to 10-day? Did we get better at prediction, or did we just get more tolerant of error? This change just happened in the past couple of years

      • Re:Actually Useful (Score:4, Informative)

        by Ignignot ( 782335 ) on Wednesday July 19, 2006 @10:27AM (#15743142) Journal
        If you are referring to forecasts in the United States, there are several different forecasts provided by the government which provide the baseline to basically everything you see on TV or hear on the radio or whatever. There is a short term forecast called the AVN/NAV which is intended to aircraft, so they will know how to schedule flights. This has a forecast which provides information on 3 hour intervals and is updated many times a day. Next there is the MRF, which goes out 12 days or so (it has been a few years since I looked at it directly) and is a daily forecast, updated several times a day. It is intended for general use and is basically what you see every day. There are some commercial vendors that put their own spin on things, and plenty of specialized forecasts for things like hurricanes, etc. However, these two are the most important forecasts for anyone in the United States, and have been around since the 90's at least. What you may be seeing is a "keeping up with the Joneses" approach to TV weather forecasting. If one station has 9 days, and the other has 8, which one are you going to watch? While that last day may have no accuracy whatsoever, people would still tend to watch one over the other I think.
    • Re:Actually Useful (Score:4, Insightful)

      by lawpoop ( 604919 ) on Wednesday July 19, 2006 @10:25AM (#15743133) Homepage Journal
      " can't get much better than climatology once you go 2 weeks out."

      I heard a great quote somewhere along the line: "It isn't decided that far in advance".
      • Occasionally there will be a more stable weather pattern which increases the accuracy of the forecast window to make longer range forecasts useful, but typically that's true. However, there are some longer predictions that you can make easily - I can tell you that there will probably be much fewer hurricanes this year than last year, both because last year was extreme, and because the water in the gulf is cool. Things like that don't change rapidly.
      • I believe I read somewhere that the effects of quantum mechanics make forcasts beyond 17 days physically impossible, even if you know the exact position and velocity of every molecule in the atmosphere and had a supercomputer the size of the universe and 10^N years of computation time.
    • RTFA and you get this quote "Just like the daily forecast, we can't give a percentage for how accurate they are"

      It's just climate research and bad reporting. Nothing new there.

    • Re:Actually Useful (Score:2, Interesting)

      by rm999 ( 775449 )
      Can you please explain why the butterfly effect wouldn't mitigate the accuracy of longterm forecasts compared to using statistical analysis of past data? I'm not saying I disagree with you, I just don't understand the reasoning.
      • The buttefly effect is a problem for a forecast for a particular day or short period, but over a longer period of time it cannot defeat climatology. There are forces in the atmosphere that work to keep the weather in an equilibrium. These are not effected by the butterfly effect in a large way. This is why you can say that it will probably snow this winter, while you can't say if it will snow on Christmas. Since these forces themselves actually change, and it is often possible to estimate how they will
      • Disclaimer: IANAM.

        I believe the reason this would "work" is because they're looking at general models. Current methods of weather forecasting let us predict small pockets of weather with acceptable accuracy. What these researchers seem to be trying to do is try and generalize over a larger period of time. Although weather forecasts aren't able to determine "City X will have scattered showers and a temperature of X degrees Fahrenheit" past a two-week threshold (or so), extending the time to decades may sti

      • If they're doing this correctly, they aren't looking for individual points or even individual curves. They're (hopefully) trying to see the whole butterfly [].

        Any particular chaotic equation with a stable set of forcing constants [] will end up with a semi-predictable structure. The problem is that the weather's input forces are changing. Even so, you should be able to solve how those changes distort the overall shape, with sufficient computing power.

    • Mostly right, but this bit caught my eye:

      But never, in no way, will someone be able to tell you if it will rain in 3 weeks, let alone 30 years. I've studied the accuracy of forecasts quite a bit (as an energy analyst), and you can't get much better than climatology once you go 2 weeks out.

      It's possible that someday it'll be possible to tell that it'll rain in three weeks. What won't be possible is to tell you *how much* and on *what days*. It's generally much easier to predict the rough sequence of events t

  • by TechDogg ( 802999 )
    will that thing be able to predict when Godzilla will strike again? I think that Japenese people need that kind of information, IMHO.
  • How are they going to choose between the multitude of different climate-altering theories?

    Take just for example the world's temperature: are we going to have another Ice Age or a Hot Age? Just choosing one of them changes drastically the results of such experiment.

    The data they are using for such experiment is, I believe, reliable (since it is mostly historical data), but the question here is not which dataset to use as input but rather to which function should this input be applied.
  • by Anonymous Coward
    The difference between predicting tomorrow's weather and the wether over the next 30 years is precision. For tomorrow's weather, you want specifics about whether it is going to rain within a pretty narrow window. If these folks are only aiming for general trends over the next 30 years (e.g. "we expect a 3-5 dry spell starting in about 2 years", "we predict 30% fewer large hurricanes in the late 2010's compared with the early 2010's" etc) it's a different issue. Nobody is claiming they will be able to s
    • Even if I can't predict what will happen on the next pull of the slot machine, I can still predict that if I play for 12 hours straight I'm pretty much going to end up broke.

      Yet if you play the right machines (the ones programmed to pay out more often) and you only play progressive machines, you're right that over a 12-hour period you're likely to end up broke. But over several lifetimes, you're likely to end up positive. The hard part is the discipline -- only playing the good machines, and sitting tight

  • Yes another person (Score:4, Insightful)

    by sholden ( 12227 ) on Wednesday July 19, 2006 @10:17AM (#15743075) Homepage
    Who can't see that climate and weather are two different things.
  • by LordKazan ( 558383 ) on Wednesday July 19, 2006 @10:18AM (#15743084) Homepage Journal
    check the accuracy of the national weather service forecasts - they tend to be highly accurate (temperature +-5 degrees F, other conditions very high accuracy)

    accuracy tends to extend very well out to the 3-day period and acceptably well to the 7-day
  • because there is so much more to the weather than just analyzing atmospheric trends. For instance, are they taking into consideration the fact that volcano eruptions can play a large part in changing conditions? Maybe their computer will predict eruptions too. Are they taking in considerations of anomolous behaviors from the sun, such as solar flares, etc. that may influence patterns? Or maybe the effect a metorite has while passing through the atmosphere. Now we're predicting things that are independe
  • What the hell? (Score:3, Insightful)

    by Capt'n Hector ( 650760 ) on Wednesday July 19, 2006 @10:19AM (#15743086)
    "Whatever happened to chaos?" That's the whole point. That's why you need a supercomputer. From the article:

    The results will help establish predictable routes for typhoons and identify areas that are recurring targets for heavy rains, abundant snow, high waves, heavy winds, scorching heat or crop-threatening droughts.

    This seems very reasonable. They're not trying to predict the weather on the third Tuesday in March, 2025, they're trying to establish long-term trends.
    • But isn't the whole point of chaos that you can't predict the outcome no matter how may supoercomputers you throw at it? Those decimal places have to stop at some point in a computer, and the digits beyond that are the seeds of eventual chaos.
      • I don't think that's "the whole point" of chaos. There's probably a number of points, such as demonstrating that the limits of the knowable, while real, are not fixed, or providing a source of creative new combinations of matter or ideas, or making that fuzzy sound in a guitar chord, or giving me a cool job title.
      • Chaos != Entropy. There IS a difference. The "whole point" of chaos theory is that you CAN extract predictibility from a seemingly entropic system.
  • by MarkusQ ( 450076 ) on Wednesday July 19, 2006 @10:24AM (#15743125) Journal

    It all depends on your assumptions. Look at Venus. The weather there is dead simple to predict. Heavily overcast, highs in the mid 900's, with poisonous smog in low lying areas through the weekend.

    The only reason the Earth's weather seems hard to predict now is that we haven't (yet) experienced a run-away feedback loop. If you posit that we're starting into one, making accurate daily forecasts thirty years out will be much easier than sticking around to see how well you did.


  • A Few Things (Score:5, Informative)

    by Hoplite3 ( 671379 ) on Wednesday July 19, 2006 @10:29AM (#15743154)
    1) The computer will be doing CLIMATE modeling, not weather prediction. That's a different bird. It's like the difference between the average score on a test and your score on the test. Or like describing the flow of heat, but not knowing the underlying collisions that result in the transfer of energy.

    2) Higher precision does help you model chaotic systems longer, but... If you run your model until the difference between your prediction and the actual system is larger than a tolerance, the time when this happens is called the horizon time. If you improve your accuracy (let's say your computer system is perfect and errors only occur in getting the initial state right), you only improve the horizon time as the LOG of your improvement. In an age where quadratic methods are just adequate in scientific computing, this is unbearable.

    3) Another weather (not climate) prediction option is to use a statistical cohort model. Such a model just takes in data and tries to predict what will happen next based on past trends. It doesn't know any physics, and can take a while to train. This means that the cohort you train in London is useless in Paris. Such "models" often beat physical models in predictive ability, but don't give any insight into why. If you want to fly a plane, they're fine. If you want to do science, see (1) or (2).

    Also, this computer is way, way cooler than the one predicting nuclear bomb blasts. But that's, just like, my opinion, man.
    • Such "models" often beat physical models in predictive ability, but don't give any insight into why. If you want to fly a plane, they're fine. If you want to do science, see (1) or (2).

      I'm a pilot, and I'm confused...

      Care to elaborate?
      • Sure. A condition of confusion is inherent in being a pilot. You can become less confused by switching to another hobby, such as being a logician, or a gynecologist. Did that help?
      • I'm a pilot, and I'm confused... Care to elaborate?
        In aviation you only care about accuracy. A cohort model of weather prediction provides a more accurate forcast, but gives absolutely no insight into the why and how of it. For a pilot, that's fine. A pilot only wants to know if the weather is going to be conducive to flying. Scientists don't really care what the weather is, their job is to figure out why it's that way.
        • Gotchya. It's worth noting, though, that even general aviation pilots are trained in a good deal of basic meteorology, and not at all in cohort analysis. Yes, the goal is 'merely' to obtain a functional knowledge of what it will be like, but you will frequently see pilots checking temperature/dewpoint spreads, looking at doppler images, keeping eyes on fronts and whatnot; you won't see them with a hundred printed pages of cohort analysis.

          It may be what the METARs are made from, but the average private pilo
          • I think you misunderstand me. The point of cohort analysis versus the more physical differential equation models is that they take known conditions and predict future ones. For short times -- a few days -- both can predict the weather on a grid of locations very well, but neither can be done practically by hand. The data your referring to (doppler images, etc) are the input to these algorithms. If you're getting information from these programs, its coming to you as forecasts -- generalized to cover large
  • by mikael ( 484 ) on Wednesday July 19, 2006 @11:02AM (#15743410)
    According to this website on paleoclimatology, there are some long period weather oscillations such as: []

    the El Niño -Southern Oscillation (ENSO) [] - 6 to 18 months,

    the Pacific Decadal Oscillation [] (PDO) - 20 to 30 years

    the Pacific-North American Oscillation [] (PNA) - 3 to 10 years

    the The North Atlantic Oscillation [] NAO - 5 to 10 years

    the Artic Oscillation [] (AO)- 5 to 10 years

    the Antartic Oscillation [] (AAO) - 5 to 10 years

    Paleoclimatologists have the records of weather condifions going back thousands of years using information such as tree rings [], snow, lava, and seed deposits.

    If the researchers could develop a long timescale atmospheric simulator that could replicate this data, then maybe they could predict general trends 30 years into the
    future. Although unpredictable events such as earthquakes and volcanos) make things
    bit harder, although they will probably run a large number of possible scenarios
    before making any conclusions.
  • Whether forecasts try to accurately predict what will happen.
    Climate forecasts try to accurately predict the probability that something will happen.

    For those of you that don't understand the butterfly effect, it is an ustable element in an overall stable system. Sahara isn't going to become a rain forest just because a butterfly start flapping. So what are climate forecasts for? Obviously not planning your 50th birthday party a few decades in advance.

    Hydro power: More rain, less rain, more unstable
  • My comptuer can do 30-year weather predictions just fine.
    You have to understand, though, that weather prediction is different than climate prediction.
    You should also realize that my computer is a little slower than the Earth Simulator.

    So, contingent on funding, I'll need a little time.
    I think I can have one ready in about 35 years.
  • This sounds incredibly stupid.
  • It's all pretty easy to predict after the nuclear war:

    year 16: bleak
    year 17: bleak
    year 18...

    Seriously though, more and more scientists and even politicians are waking up to the fact that humankind is constantly changing its environment. Some are saying that the small rises in global temperatures these past couple years may have triggered the increase in hurricane activity and strength we are seeing. Makes sense to me - from a laymans point of view higher temperature = more energy = stronger storm.

    As we co

  • I don't understand you guys sometimes with all your griping about weather predicitions and TV news and all.

    The factors that affect weather predicitions on a small time scale are different from those that affect weather on long time scales. In fact, 0-48 hours is relatively easy to forecast based on extrapolating local conditions and observations. Long-term trends (on the order of magnitude of years) are also relatively easy to forecast (see The Farmer's Almanac, El Nino and Typhoon cycles etc. for evide
  • Not bad science, just bad journalism.

    Chaos has not gone away, but the objective of the project is not to perdict specific weather events.

    Climate is the aggregate statistics of weather. The fact that we have a word "climate" indicates that such statistics are predictable to some extent. The Japanese are planning to try to get as far as is possible in predicting climate. This is not a thirty year weather prediction, and they know it.

    The fact that there is a language barrier and probably an incompetent journal
  • > The results will help establish predictable routes for typhoons and identify areas
    > that are recurring targets for heavy rains, abundant snow, high waves, heavy winds,
    > scorching heat or crop-threatening droughts.

    In other words: What are probable areas where these phenomena occur and what are the most probable paths for those phenomena that are moving.

    The reason they take a 30 year period is not that they want to predict the weather 30 years in advance (that's ridiculous), but that they want real
  • Everybody, point your vacuum cleaners west!
  • I was about to say "WTF" to this story, then I stopped and thought about it. Clearly they are NOT modeling weather as such, but atmospheric patterns. This isn't even climate modeling; they just want to know what will happen if the sea starts to act in such-n-such a manner at some point. It's the same sort of data mining + modeling that gave us our understanding of the ENSO phenomenon in the southern Pacific. This stuff works.

    I took a look at a map of sea currents and noticed that the Japanese Current is a w
  • "Whatever happened to chaos?"

    Chaos was defeated in FF1, eesh.

Today is a good day for information-gathering. Read someone else's mail file.