Cloudy Crystal Balls

By David Bjerklie

Climatologists regularly issue confident warnings about impending atmospheric disasters. The secret of their wizardry: sophisticated computer models, which are no more than mathematical representations of the world's climate and the conditions that scientists think may contribute to a specific phenomenon like, say, ozone depletion. Unfortunately, when all the variables are fed into the computer, the predictions can fail miserably to match reality.

Take the Antarctic ozone hole, for example. Before it was discovered, climate modelers trying to simulate ozone loss in the atmosphere had not yet factored in the presence of ice clouds in the Antarctic stratosphere. Thus their models failed to predict the existence of the ozone hole. After the hole was finally stumbled upon two years ago, Susan Solomon, a chemist at the National Oceanic and Atmospheric Administration in Boulder, and Rolando Garcia, of the National Center for Atmospheric Research, plugged more numbers into NCAR's computer model to account for the Antarctic ice clouds. Bang! The hole appeared.

Does that mean, as one critic put it, that models projecting climatic change are "just the opinion of their authors about how the world works"? Not necessarily. That the model eventually proved accurate, if only in hindsight, was a tribute to the powers of computer climate models -- and a demonstration of their shortcomings. The models attempt to reduce the earth's climate to a set of grids and numbers, then manipulate the numbers based on the physical laws of motion and thermodynamics. The sheer number of calculations involved is mind-boggling. A three-dimensional model, for example, requires more than 500 billion computations to simulate the world's climate over one year.

Not surprisingly, the earliest models in the 1960s were hopelessly simplistic. The earth's surface was often reduced to one continent with one ocean, fixed cloud cover and no seasons. But as computing power grew, so did the complexity of climate modeling. Continents were added. So were mountain ranges, deeper oceans and surface reflectivity.

Even so, climate modelers admit, building a completely realistic mock earth is an impossibly tall order. "You divide the world into a bunch of little boxes," explains Michael MacCracken, an atmospheric scientist at Lawrence Livermore National Laboratory. The size of the geographic box -- the degree of detail called for -- limits the model. Smaller grids dramatically increase the number-crunching power required. "The state of the art would be to get down to small areas so we can say what's going to happen in Omaha," says Livermore's Stanley Grotch. "The models just aren't that good yet."

Why, then, do scientists trust them? How do they assess their accuracy? "You compare them with reality," explains Princeton Climatologist Syukuro Manabe. "How well do they reproduce the movement of the jet stream, the geographical and seasonal distribution of rainfall and temperature? You can also reproduce climate changes from the past. Eighteen thousand years ago, there was a massive continental ice sheet. Given the conditions that we know existed, can we reproduce accurately the distribution of sea-surface temperatures then? The answer is, We can do this very well. It gives you some confidence." Large-scale phenomena can be modeled more easily than those affecting small areas. So when it comes to the global warming produced by the greenhouse effect, for example, the outlines are predictable but the specifics are not. Says Manabe: "All we can say is that maybe the mid-continental U.S. becomes dryer."

A major drawback of computer models is that the various data do not necessarily behave as a system. Coaxing ocean currents to interact with the atmosphere is no small matter. For starters, oceans heat and cool far more slowly than the atmosphere. "We've had a hard time coupling the two systems," admits Manabe. "Even though the atmospheric model and ocean model work individually, when you put them together, you get crazy things happening. It's taken us 20 years to get them together, and we're still struggling."

+ Offsetting the obvious weaknesses of climate models, says Warren Washington, who developed the model now used at NCAR, is one significant advantage. "They are experimental tools that allow us to test our hypotheses," he says. "We can ask such questions as 'What happens when a big volcano like El Chichon goes off?' and 'How much will the earth warm up by 2030 if we continue to dump CO2 into the atmosphere?' "

Models can also describe the effects of climatic phenomena that have never been seen. In 1983 a group of scientists that included Cornell's Carl Sagan calculated what would happen if the U.S. and the Soviet Union fought a nuclear war. Their conclusion: the dust and smoke from burning cities would blot out enough sunlight to plunge the land into a "nuclear winter" that would devastate crops and lead to widespread starvation.

The problem with their model was that it ignored such key factors as winds, oceans and seasons. When NCAR's Stephen Schneider and Starley Thompson ran the numbers through their agency's three-dimensional computer model, they found that the winter would be more like a "nuclear autumn." Schneider says the less dramatic conclusion does not change the fact that "nuclear autumn is not going to be a nice picnic out there on the rocks watching the leaves change color." Despite the limitations and omissions of climate models, he argues, scientists cannot afford to ignore their predictions. They are, he concedes, a "dirty crystal ball. The question is, How long do you wait to clean the glass before you act on what you see inside?"

With reporting by J. Madeleine Nash/Chicago