“We can’t provide just the potential vorticity at 10 millibars or the geopotential height at 500 millibars,” says NCAR/ESSL head Guy Brasseur in describing the ideal Earth system model. “We have to provide what’s of interest to people.” To Brasseur and others, this means models that can resolve clouds, ocean eddies, wildland fires, and cities—in other words, models that can peer decades into the future and generate results on the scale of a few kilometers, just as today’s best weather models do for the next three or four days.
“We have applications requiring 1- to 2-kilometer grids for our core,” says NCAR’s Bill Skamarock. He’s part of a loose-knit set of about two dozen weather and climate modelers, drawn equally from inside and outside NCAR, that’s scouting out possible routes toward ultra–high-resolution global models. The ideal pathway isn’t yet clear, but the group anticipates that computer power may not be the roadblock it’s been in the past. “We have the machines, or we will soon, on which we can run these models,” says Skamarock.
One intriguing piece of technology is the super-fast graphical processing units (GPUs) now common in high-end video gaming systems such as the Microsoft Xbox, Sony PlayStation, and Nintendo Wii. Until now, GPUs haven’t been applied in atmospheric modeling, but according to NCAR’s John Michalakes, they have the potential to run as much as 40 times faster than standard chips. “There may be real potential for applying these to the more computationally intensive parts of our code,” he says.
At NCAR’s Computational and Information Systems Laboratory, Rory Kelly and José Garcia are recoding atmospheric models for GPUs. In addition, they’re developing general-purpose mathematical libraries that modelers can use on accelerators in the future. When they recoded one computationally intensive part of a model for inverting data on solar magnetic fields, it ran 10–20 times faster than the baseline code. Kelly and Garcia also worked on the radiation calculations in an older version of the Community Atmosphere Model (CAM), speeding them up by about 15–20 times.