Maybe I'm overly distrustful of computer models (projecting my poor programming skills, maybe), but isn't energy dissipation of that type pretty chaotic? I'm wary that they over-smoothed some of the interactions to keep the models from "blowing up" (technical term) or diverging too much.
Remember that the weather people get second tier supercomputers, after the nuclear weapons guys, and eventually they're probably going to get bumped down to third by the AI people. The training clusters they run now aren't the right architecture for weather, so that's why they aren't second right now.