Super Weather Machine: Beating Out Chaos Theory

[3113]

Actually a supercomputer (from here):

This month, on a barren Wyoming landscape dotted with gopher holes and hay bales, the federal government is assembling a supercomputer 10 years in the making, one of the fastest computers ever built and the largest ever devoted to the study of atmospheric science....The sheer speed of Yellowstone [the computer] is designed to burst through the limits of chaos theory — the difference, allegorically, between predicting the odds of blackjack after playing five hands versus playing a million. The machine is expected to give scientists a clearer image of the state of the planet, and its future, revolutionizing the study of climate change, extreme weather events, wildfires, air pollution and more.

 

[Bramblethorn]

Nifty! Though "burst through the limits of chaos theory" seems like a bit of a misrepresentation - by my reading it's more like "works within the limits of chaos theory", but I guess that doesn't sound as sexy.

 

[fuckwaffle]

To bad a computer is only as good as its programming.

A miss-placed comma in a single line of code of a million line program can cause untold problems. From an image not displaying correctly or at all or to causing a rocket to exploded on the launchpad as the count down reaches zero.

Yet, if they have done it right, then it could be cool or...skynet!

 

[voluptuary_manque]

I think they're underestimating Chaos but what to I know? If they're right there are going to be a lot of miffed butterflies in the Amazon.

 

[CarlusMagnus]

To bad a computer is only as good as its programming.

A miss-placed comma in a single line of code of a million line program can cause untold problems. From an image not displaying correctly or at all or to causing a rocket to exploded on the launchpad as the count down reaches zero.

Yet, if they have done it right, then it could be cool or...skynet!

One of the issues here is the programming—but not in the sense of misplaced commas. The real killer is the model they use. The predictions will only be as good as the model that underlies the calculations. This is a problem that's often swept under the rug in public discussions of computer models—you can't make reliable predictions about how a system will behave until you have a full understanding of that system.

And it's pretty clear that the reporter didn't understand what the problem with mathematical chaos is. It isn't computing power, per se—though that's important. The show stopper here is sensitivity to initial conditions. All "good" weather models display this; it means that a difference in initial data of a few thousandths of a percent can cause major differences in predictions not very far down the line. (And that's what the "butterfly effect" metaphor is meant to describe.) The problem for the new supercomputer will be having enough, and accurate enough, measurements to overcome this little instability.

 

[Stella_Omega]

In other words the media, and therefore the public, will deride any conclusions this project comes up with because "What about if they misplaced a comma?"

That's the real chaos.

 

[mynameisben]

Actually a supercomputer (from here):

This month, on a barren Wyoming landscape dotted with gopher holes and hay bales, the federal government is assembling a supercomputer 10 years in the making, one of the fastest computers ever built and the largest ever devoted to the study of atmospheric science....The sheer speed of Yellowstone [the computer] is designed to burst through the limits of chaos theory — the difference, allegorically, between predicting the odds of blackjack after playing five hands versus playing a million. The machine is expected to give scientists a clearer image of the state of the planet, and its future, revolutionizing the study of climate change, extreme weather events, wildfires, air pollution and more.

Precise blackjack odds were determined from a computer simulation of billions of hands dealt as early as 1966 by programmer Julian Braun from original theory established by Professor Edward O. Thorp of M.I.T. The "supercomputer" Julian Braun used to make his calculations was an IBM 7044 mainframe , which had less than a millionth of the computing power of today's iPhone.

 

[Bramblethorn]

One of the issues here is the programming—but not in the sense of misplaced commas. The real killer is the model they use. The predictions will only be as good as the model that underlies the calculations. This is a problem that's often swept under the rug in public discussions of computer models—you can't make reliable predictions about how a system will behave until you have a full understanding of that system.

And it's pretty clear that the reporter didn't understand what the problem with mathematical chaos is. It isn't computing power, per se—though that's important. The show stopper here is sensitivity to initial conditions. All "good" weather models display this; it means that a difference in initial data of a few thousandths of a percent can cause major differences in predictions not very far down the line. (And that's what the "butterfly effect" metaphor is meant to describe.) The problem for the new supercomputer will be having enough, and accurate enough, measurements to overcome this little instability.

And the catch is, you can never provide accurate enough measurements to achieve that. Even if you sent out miniature meteorologists to measure the position and momentum of every single molecule in the air, Heisenberg's principle guarantees a minimum level of uncertainty in those measurements, and that tiny level of uncertainty is enough to put a crimp into long-range deterministic forecasts.

What I suspect the article was trying to get at (badly garbled by the journalist) is probabilistic forecasting. If you have a good model, and plenty of computing grunt, you can take thousands or millions of slightly-different approximations of the starting position and see how they turn out. If 95% of your runs predict rain in seven days, you can forecast a 95% probability of rain at that time. Obviously the more processing power you have, the better your accuracy.

(Like you say, it still relies on having a good model to work from; I've seen interesting work that involved taking several different meteorological models, comparing their accuracy, and using the resulting info to build hybrid models.)