Statistical Correction for Dynamical Prediction of 500 hPa Height Field in Mid-high Latitudes

In terms of the Météo France model data of DEMETER project, the performance of ensemble forecast system at 500 hPa height field of winter in mid-high latitudes (20°—90°N) is studied, then both optimum subset regression (OSR) and dynamical analogue prediction (DAP) method are used to improve the model prediction. First, empirical orthogonal function (EOF) analysis is applied to investigate the observed 500 hPa height field of 1958—1991. The time coefficients of different modes for the numerical model are calculated by projecting the model data onto the observed EOF basement. Then the performance of the model modes based on empirical orthogonal function (EOF) of observations is examined by calculating the anomaly correlation coefficient (ACC) between the time coefficients of the leading 10 model and observed EOF modes. Next, the optimum subset regression (OSR) experiential model is established to advance the model prediction on the modes, which are predicted by the numerical modes with very low skill (i.e., low skill modes). Finally, the mean time coefficients of 5 observed similarity years on each low skill mode are substituted for those of the model prediction, where the similarity year is defined as its time coefficient estimated by OSR has minor difference from that of the prediction year. In this way, the analogue method is employed to correct the model prediction on OSR basis, namely, OSR-based analogue method. The results suggest that the prediction ability of the mode accounting for less variance may be higher than the mode with more variance, such as the 2nd and 3rd EOF modes have low skill but with large variance contribution to total variance of the model field. OSR fails in advancing the model prediction. The DAP method based on OSR (DAP-OSR) shows a possibility of improving the prediction techniques with ACC increasing 0.1 by correcting the bad modes of model while OSR fails.Correcting the dynamic prediction by combing the advantages of the numerical models and statistic methods, the nonlinear analogue method based on linear OSR shows a possibility of improving the prediction techniques by correcting the EOF modes, which are predicted by the numerical modes with very low skills. However, since the numerical model has a poor capability in representing the 2nd and 3rd EOF modes of the observation which account for large percent of total variance, and the forecast ability can not be improved effectively because the model prediction information is not enough or incorrect. Therefore, it is necessary to make further analysis on the samples of the modes, predicted with low skill by the numerical model, and the corresponding external forcing. The external forcing might be more effective to improve the correction for such modes with low skill.