Scientists' use of computer models to predict climate change is under attack

The Web site Footballoutsiders.com, which uses computers to show fans hidden dimensions of pro football, uses a model with about 80 variables. It looks at a team's third-down conversions, the experience of its coaches, even the age of its defensive backs.

No crystal balls

How much cleaner would the Chesapeake Bay be if it had twice as many oysters?

The Environmental Protection Agency uses a model that divides the bay into 55,000 slices, and maps how pollution progresses through them, from upstream tributaries into the deeper waters of the Chesapeake. It could imagine thousands more oysters -- which filter water as they feed -- and watch cleaner water spread out via currents and tides.

But, some of the time, these electronic futures haven't come true.

The Footballoutsiders site predicted the Redskins would win 7.8 games in 2009. The real-world team won four. The EPA's Chesapeake Bay model has been criticized repeatedly for over-optimism, for creating a virtual bay that looked cleaner than the real one. Last month, another model's prediction was busted: a Georgia Tech professor's computer said Kansas would win the NCAA men's basketball tournament. The Jayhawks lost in the second round.

These and other models are only as smart as the scientists who build them -- they rely on data that scientists have gathered about the real world, and the accuracy of estimates about how all the factors fit together (Is an experienced coach more or less important than young defensive backs?).

They also depend on the computers running them. To accurately depict how individual clouds form and disappear, for instance, the computers that model climate change would need to be a million times faster. For now, the effects of clouds have to be estimated.

But scientists say complexity doesn't guarantee accuracy. The best test of a model is to check it against reality.

"We're never going to perfectly model reality. We would need a system as complicated as the world around us," said Ken Fleischmann, a professor of information studies at the University of Maryland. He said scientists needed to make the uncertainties inherent in models clear: "You let people know: It's a model. It's not reality. We haven't invented a crystal ball."

Scientists say they don't need models to know that the world is warming: There is plenty of real-world evidence, gathered since the mid-1800s, to suggest that. "There's no climate model in that conclusion," said Christopher Field, of the Carnegie Institution for Science in California.

There are more than a dozen such models running around the world: mega-computers whose job is creating a virtual Earth.

These usually combine a weather simulation with other programs that mimic effects of rain and sun on the land, currents in the ocean, and emissions of greenhouse gases. First, these models imagine all the factors interacting within a "grid box" -- an imaginary cube of land, water and sky that might be 60 miles long and 60 miles wide.