Earth Simulator Sees Green Light

burbs writes "Big Blue's dominator is getting closer to being turned on. The Earth Simulator in Japan is, supposedly, the world's fastest parallel-processing supercomputer. Designed for the Earth's weather, the computer should be able to predict climate for the entire planet for thousands of years in a short amount of time."

Simulate wheather for thousands of years?

[Saib0t]

I don't know much about meteorology, but isn't predicting the wheather a tad difficult?
Besides, to be able to precisely model the earth's climate, they would need to have measures for about every (mathematical) point of the earth at a given time, which is not possible... Unless they go for an approximation, and then chaos theory kicks in and their 'thousands of years in advance' prediction is worth nothing. (the butterfly - hurricane thing anyone?)

Am I missing something there?

Re:Simulate wheather for thousands of years?

[khym]

Even if the computer did know the exact state of every single atmospheric molecule on the planet, things like metiorites, earthquakes and volcanic eruptions would changes things around, at least by a tiny bit, and then the Butterfly Effect comes back. So a 1,000 prediction is pretty ify.

Re:Climate, not weather

[Kynde]

A number of posters are confusing climate modelling with weather prediction. Weather prediction -- working out if it will rain tomorrow -- is very difficult because weather systems are chaotic. Climate prediction, however -- working out how large an effect increased CO2 emmisions will have on global warming -- is easy by comparison... at least in theory.

A number of meteorologists are also confusing modelling with scientific modelling. Those larger scale climate models have little chance giving accurate predictions since there's absolutely no reason to assume that such models would not depend chaotically on the underlying small scale weather. Those forementioned fudge factors that climate models are plagued with are manifestation of just that.

Even in more strict science circles people tend to resort to finding trends when the system vanishes out of scope. It's essential that the causality and predictability are present. Otherwise people wind up doing research based crap correlations. For example, for several years there's been really strong correlation between the number of Babtist preachers and number of people arrested for drinking in public. There's jack causality present as the dominant effect is the fluctuations in the population of US.

Just because climate is a lot slower than weather allowing it to predicted for longer periods of time than weather and short term trends give reasonable short term predictions, just as it's possible for weather for a couple of days at a time, it's still chaotic.

It's easy to blow the model out of practicality and show how shitload of CO2 emissions will create greenhouse effect. A lot shittier task is to work with real world figures and again work with a chaotic system. Don't get me wrong though, I'm absolutely all for the Co2 emission regulations and all.

Confusion

[C_James_B]

As my Geography teacher never tired of telling us, climate != weather.
Climate is big and long-term. Weather is here and now. Not even the people who built that machine think it can predict world weather for one thousand years. There's just been a bit of a misunderstanding.

nonlinear equations

[xeeno]

I dunno about this one. Wouldn't you be solving some pretty harsh nonlinear d.e's in order to model weather correctly? AFIAK, the only real way to solve these things is to make assumptions about the boundary conditions and the parameters in the equations that simplify the equation in such a way as to make it solvable by means we do know. Not only that, you have mixed fluid equations - one equation for each type of particle in the atmosphere. You'll have boundary conditions on each layer of the atmosphere, and where things behave like plasmas you'll have an entirely new set of equations to consider. It sounds like there are way, way, way to many considerations that go into an exact solution of an atmospheric model. So, solving a 1000 years in advance seems ridiculous. I'd be happy if we could solve a day in advance!