There's a story on wattsupwiththat about the UK MET office's hot new computer. The headline of the story is:
Met Office supercomputer: A megawatt, here, a megawatt there, and pretty soon we’re talking real carbon pollution
Of course Anthony Watts is saying that tongue in cheek, because carbon dioxide is plant food, not pollution. So the truth is the MET office has one of the GREENEST buildings in the UK. CO2 is nutrient. We should be glad of that small bounty as they go about proving the uselessness of their forecasts.
Here's the facts about numerical modelling: chaos principles guarantee that within a relatively short time the results are nonsense. No amount of speed or computer memory can change that. It's a mathematical fact. Three weeks out is probably the best possible, no matter what computing resources you throw at it.
So if you want to do better, you have to use entirely different principles. For example, a theory of how the sun influences climate might tell you that a year on, the climate will be constrained to a colder (or warmer) regime. You still won't know whether the weather on day X will be hot, cold, wet, dry, but you will be able to forecast the sort of weather happening around about that time. This kind of forecast simply doesn't need the kind of mind-boggling computing power in that building. You might even be able to do it with pencil and paper. They are two entirely different kinds of calculation. And the fact that the MET office are gearing up to push numerical techniques beyond anywhere they have gone before (and they have already hit the limits imposed by chaos) tells me they are a bunch of incompetents. Either that or shysters.
One of my goals on this site is to make any claims I might make intelligible to intelligent lay people—meaning non-scientists in this case. So what exactly is the difference between these two kinds of calculations? Let's take some examples.
Say I have a bow and arrow, and I fire an arrow into a forest— an extensive thicket of trees— aiming at a specific known spot. Wherever the arrow lands or hits a tree, I will go to that spot and fire a second arrow just to the right of the second-nearest tree. I will repeat this five times. Let's say you have a lot of information about my archery skills, wind speed, temperature, anything else you think is necessary to predict where I will finally end up. Do you think you could predict my final position?
I would guess, most likely not. There will be some inaccuracy in calculating my first shot, and from then on, each subsequent shot amplifies the original error and gets its own error added on. After even five shots, I think my final location would be impossible to predict in even the vaguest of terms, and no amount of extra computing power will make things any better because the multiplication of error is inherent in the calculation.
So that is an example of chaos. And note that better faster computers allowing you to forecast a million steps of my arrow shooting won't get you any more sensible an answer, it will just make you look even sillier (not that anyone outside the UK MET office actually is that silly, of course!).
What is an example of the other kind of calculation? Scientists will tell you that the Moon was once much closer to the Earth than it is now, and the Earth was spinning faster. This is kind of like a standard exercise in astronomical physics, and I recall sitting down with a real pencil and real paper and working it out for myself once. How is it I can tell you how the Moon's orbit altered over millions of years, and yet I can't predict tomorrow's weather? This second problem depends on what I call "outer envelope" calculations. The Earth and Moon must conserve angular momentum, so if the Moon's orbit changes in a particular way, the spin of the Earth needed to conserve angular momentum must do exactly the appropriate thing to keep angular momentum constant. Nextly, we know the tides are dissipating energy, so the energy of the Earth-Moon system must have once been higher. By putting in a higher energy and solving both the energy and angular momentum equations, we can tell what must have happened. No way I could tell you exactly where the Moon was in the sky on Jan 1, 400,000,000B.C., of course, but that is a chaos-affected question, not a question the depends on simple physical laws.
So what do we learn from the news of the MET office's computer purchases? One simple thing: they are trying to do the impossible. It is always easier to crunch more numbers than to actually think creatively about a problem.