Testing how well the model works and using it to answer questions are intertwined activities, says Kotamarthi. Because the complexity of global processes involves so many variables and uncertainties, a "validated model" is rarely an option. "The preferred word around here now is 'evaluated' model," he says.

Wuebbles, Kotamarthi, and their collaborators recently finished "probably one of the harshest tests MOZART was ever subjected to," says Kotamarthi. It took him aboard a NASA jet on missions across the south Pacific as an "onsite modeler." A modified DC-8 passenger plane with air inlets and instrument stations for 30 to 40 scientists, the plane flew up to 40,000 feet and made spiral-staircase descents to 1000 feet above the ocean at sites along its route. "The idea is you are trying to sample the vertical profile of trace gases at various altitudes," Kotamarthi says.

The air of the remote south Pacific is supposed to be squeaky clean, and it generally was. "But the scientists were surprised by fingers of polluted air—"dirtier than maybe in the LA city area on a bad day," says Kotamarthi. These fingers seemed to extend thousands of miles in strips only a few thousand feet thick.

On the ground, Kotamarthi and fellow modelers compared the flight measurements to MOZART-2 predictions for the same region and time of year. The concentrations of ozone differed by about 30 percent, Kotamarthi says. But the model tracked seasonal changes. It also predicted dirty fingers like those the scientists encountered, identifying them as the listing plumes of fires in subequatorial Africa or South America that farmers light seasonally to clear rainforests and grassland for cultivation.

To see the simulated plumes, the modelers had to double the burning emissions they had first assumed. But scientists' best estimates of the dirtiness and extent of such fires are just rough averages, says Kotamarthi, and they relate to 1991. The fires that produced the plumes NASA detected on its 1996 and 1999 south Pacific flights may have generated more pollution, he says. The model was further handicapped by having to work from generic seasonal winds created by a computer.

While MOZART-2 can't compute global chemistry from weather measurements alone, the modelers have begun using artificial weather steered by real data for specific years. A precise assessment of how MOZART-2 and the Pacific measurements disagree awaits the results of these runs, Kotamarthi says. But he says the team has already acquired confidence in using the model to answer basic questions.