Model Errors and Nonlinearities in Numerical Weather Prediction

Carolyn A. Reynolds

Naval Research Laboratory, Monterey, CA

Many techniques for data assimilation, ensemble design, and adaptive observing in numerical weather prediction are based on the assumptions that 1) the model is perfect and 2) perturbation growth is linear. Here we explore the impact of making these assumptions when we know them to be false. The limitations of these assumptions are examined in various applications using output from an operational numerical weather prediction model along with simple-model results. Nonlinearity is quantified by examining the symmetry between positive and negative perturbations as they evolve in time. We consider whether linear tools can provide qualitatively useful results even when perturbation growth is significantly nonlinear. Model error is explored by examining the nature of short-term error growth, and comparing the growth of forecast errors with the growth of forecast differences (forecasts started 1 day apart).
The difficulties of characterizing and accounting for model error will be discussed.

Last Update: March 20, 2005