General Circulation Models Global Weather Forecasting Models Limited Area Models

General circulation models (GCMs) are run at low resolution (compared to forecasting codes) for long times (many model years, and months of wall-clock time) from arbitrary initial conditions to reach a statistical equilibrium, i.e., the climate. Global weather forecasting models (GWFMs) are run at high resolution for five to ten days of model time from observed initial conditions to make deterministic forecasts. Limited area models (LAMs) are “macroadaptive”: A high-resolution grid over a hurricane or a continent is embedded, explicitly or implicitly, in a glob-al model of lower resolution. The spiraling-out-of-control complexity of these codes is forcing a convergence of species; the British Met Office Unified Model, for example, is run as both a global and a limited area model. Each model has a “dynamical core, ” which is the fluid mechanics, and the “physics, ” which is a very misleading name for everything else: a slew of subroutines for radiative transfer, photochemistry, air–sea interaction, boundary layer turbulence, the hydrologic cycle, cloud physics, and biogeochemical cycles too complicated to enumerate here. Consequently, GCMs now exceed a million(!) lines of code. These many process-es are violently nonlinear and strongly coupled. A GCM or GWFM is the computational embodiment of a bar fight in a John Wayne western. The code cowboys who wrangle these megacodes are acutely aware that they are herding not sheep, but rather a flock of subroutines, ornery and independent as a herd of professors. Mathematics department missionaries have often sallied forth to mathematize the meteorological primitives and spread the gospel of new algorithms. They are invariably surprised when the ungrateful savages tie them to a stake. As the bonfire is lit, they wail, “But it works for Bur-gers ’ equation!