Local Skill Prediction for the ECMWF Model Using Adjoint Techniques

Abstract

Two experiments are performed to predict the regional forecast skill over western Europe using a three-level quasigeostrophic model with truncation T21 (T21QG). The predictor of skill, the maximal forecast error over the area of consideration, is obtained assuming linear error growth, no model error, and use of the tangent and adjoint of the T21QG model. A small maximal error implies small error growth and therefore good actual forecasts. A large maximal error, on the other hand, may or may not be associated with a large actual forecast error depending on whether the analysis error projects strongly on the fast-growing modes. In the first experiment, the forecast trajectory is obtained directly using the T2IQG model. In the second experiment, actual 96-h ECMWF forecasts are interpolated every 12 h using the T21QG model. Both experiments indicate that the maximal growth can provide useful discrimination about very good and very poor forecasts. The improvement upon climatology is of the order 10%?15%. The maximal error growth may vary considerably. This is illustrated by showing the error growth for a westerly circulation and a blocking situation.