Reduction of systematic errors by empirical model correction: impact on seasonal prediction skill

Abstract

Recent studies indicate that the atmospheric response to anomalies in the lower boundary conditions, e.g. sea surface temperatures, is strongly dependent on the atmospheric background flow. Since all general circulation models have long-term systematic errors it is therefore possible that the skill in seasonal prediction is improved by reducing the systematic errors of the model. In this study sensitivity experiments along this line are made with an empirically corrected dynamical model for which the systematic errors are reduced substantially and the dynamical variability has become more realistic than for the original model. As a measure of seasonal prediction skill, correlation of temporal anomalies between modelled and observed data has been determined. The corrected model shows improved skill in the Southern Hemisphere in general—on average a 20–30% improvement for the Southern Hemisphere compared with the original model. In the Northern Hemisphere skill is improved in some areas, but in other areas the skill of the original model is better. On average there is no improvement for the Northern Hemisphere. Also, pattern correlations have been determined for the following areas: the Northern Hemisphere, the Southern Hemisphere, the tropics and Europe. The general picture is that the two model versions are very similar in the Northern Hemisphere and in the tropics. For Europe the results of the two models are rather different, but no model can be said to be better than the other. In the Southern Hemisphere it is again seen that the correlations are higher for the corrected model than for the original model.