THE COMPARISON EXPERIMENTS OF SV AND LAF INITIAL PERTURBATION TECHNIQUES USED AT THE NMC ENSEMBLE PREDICTION SYSTEM

The Ensemble Prediction System (EPS) at NMC is briefly described. It mainly includes such seven parts as the data pre-processing, optimal interpolation analysis, the forecasting model (T106L19), the computation of initial perturbations based on singular vectors, post-processing, the generation of ensemble products and the system monitoring. The configuration of EPS compromises one control forecast started from the unperturbed analysis and 31 perturbed forecasts started from the perturbed analysis. Then the lagged average forecasting (LAF) scheme with 12 members is proposed and implemented at the SW-1 computer. The comparison experiments are conducted by using different initial perturbation techniques. One is SV (singular vectors), the other is LAF. The results show that the performance of SV is better than that of LAF in most of the verified regions and lead times, both for the anomaly correlation coefficients and the root mean square errors.