Parameterization

Parameterization is the substitution of a simplified model for unresolved processes in a larger-scale simulation. In climate modeling, parameterizations represent subgrid-scale phenomena — clouds, turbulence, convection, boundary layer dynamics — that cannot be explicitly resolved at the grid spacing of global models. The parameterization is not a theory of the unresolved process; it is a functional relationship between the resolved variables and the effects of the unresolved processes on those variables.

The methodological risk of parameterization is the parametric fallacy: the assumption that a process can be adequately represented by a small number of tunable coefficients when the actual physics is high-dimensional and state-dependent. Parameterizations are typically calibrated against observations or high-resolution simulations, but they extrapolate poorly to conditions outside their calibration range. A convection scheme tuned for present-day climates may fail catastrophically in a nuclear winter scenario or a paleoclimate state. The stochastic parameterization movement attempts to address this by replacing deterministic closures with random processes, but the fundamental problem remains: parameterization is an act of epistemic compression, and compression always discards information.