02.18.08

The Third Branch of Science?

I recently judged a grade school science fair. Students were being taught a very simple and straightforward model of "the scientific method". I'm sure they didn't fully understand it, which is all for the best because it is pretty much wrong, but I'm concerned that the teachers believed it.

In this view, you start the scientific process with an idea, which you formalize into a testable hypothesis. Then you test the hypothesis, and you report your result. Of course the real process of discovery is far more convoluted than that. Nevertheless, it does capture the idea that there is a theoretical branch of science and an experimental branch of science.

Some people these days are saying that computing has become so important to science that it constitutes a third branch. Even though computationally intensive science is what occupies my time, I am not sure that this is the right way to think about it philosophically. To some extent computing brings the theoretical and experimental branches closer together.

There are some sciences where the experimental method is extremely limited. Earth sciences are among them, but the quintessential example is astrophysics. We simply can't afford to be monkeying around with stars to test our theories about how they work!

These sciences have traditionally replaced experimentation with observation. We can't play around with stars but we can observe a great huge number of them. Accordingly we can make certain predictions about how stars behave and test them against observations.

In meteorology, we have only one earth, but we have many cycles of the weather. Meteorology thus progressed by making observations and then making predictions based on the observations. Over time, the predictions got measurably much better (though people still tend to remember the mistakes!) This is a measure of how the understanding of the short term behavior of the atmosphere improved over time.

Climatology, as I explained in an earlier article, is closely related to meteorology but has a different purpose. It is not to predict the individual events of our weather but the trends. This of course yields some difficulties. The longer the trend, the fewer the observations. Things that happen over more than a couple of centuries have few or no direct measurements. (The thermometer was not invented so long ago in the grand scheme of things.)

As in astrophysics, climatology suffers from few direct observations and an incapacity to do experiments. Does this mean we shouldn't think about such things? In the case of climate, we don't have that luxury. Both our civilization and our environment depend closely on climate, and there's plenty of evidence that we are starting to disrupt that climate. We have no choice but to do the best we can.

In fields where experimentation and direct observation are limited, computational science is especially important. Computational science is about simulation. Once we know the equations which describe a physical system, we can program a computer with that description and watch the simulated system just as we would watch a real system.

This is a tradeoff; in fact our understanding is imperfect, so the simulated system won't behave exactly like the real system. (In fact, more often than not it doesn't behave even remotely like the real system, but you don't hewar about these back-to-the-drawing-board efforts. Perhaps it might be better if failures were better documented to prevent others from going down certain blind alleys repeatedly, but in general failed efforts are just abandoned.)

If the imperfect simulation looks somewhat like the observable parts of the real system, though, it has tremendous benefits. The model is perfectly observable. We can investigate phenomena in ways that could never be affordable to measure in the real world.

There's more we can do. We can experiment with these systems. We can alter the code to represent alternative physics and thereby test hypothesis of why certain things are happening.

Here's an example. In a recent talk at the University of Texas, Z. Liu of Wisconsin described some experiments he did with two different climate models in tracking down something called a Pacific Multidecadal Oscillation, a sort of long-term wobble in the climate of the Pacific. Because there is so much interest in the equatorial zone in the Pacific (due to El Nino) most attention to this phenomenon had been in the tropics, but Liu observed that there was no known process in the tropics of the right time scale. There is such a process in the far north Pacific, called "baroclinic Rossby waves" the mathematics of which I won't trouble you with.

Liu wanted to test his idea that these very slow waves are crucial to the PMO. He found some climate models that display a PMO in their statistics, and put a giant simulated sponge in the North Pacific that would leave the rest of the system as unimpeded as possible but would suppress the baroclinic Rossby wave. Sure enough, the PMO went away in the model. A couple of further experiments confirmed that putting the sponge elsewhere had little effect. He repeated this with two distinct model codes.

This constitutes strong support for Liu's hypothesis. Is it "experimental" support? That's a matter of nomenclature. I think it is. Others, who don't are forced to call what Liu (and I) do a "third branch". It's just nomenclature though. It's hard to argue that Liu's point isn't supported by these efforts.

To the people building and using climate models, this is the sort of thing they are for. Using the output to represent the future is important qualitatively. It is important for the public and the governments to understand the great size and impact of the changes that we are expecting in the future.

The models are not as yet good enough to yield reliable enough predictions for regional planning. Can they ever be? Perhaps not in time. Should we make the effort to try? I think so.

Even if you argue that this effort is futile though, you should still have respect for the role models already play in improving our understanding. Similar things are happening in other sciences as well.

Tags: computing, scientific method

+ Add Comment