Impact of Initialization on Ensemble Prediction of Intraseasonal Oscillation during Boreal Summer

Abstract

The Impact of initialization and perturbation methods on the ensemble prediction of the boreal summer intraseasonal oscillation was investigated with the 20-year hindcast predictions from a coupled general circulation model (CGCM). The three perturbation methods used in the present study included the lagged averaged forecast (LAF), the Breeding, and the empirical singular vector (ESV) methods. The hindcast experiments were performed with a prediction interval of every 10 days for an extended boreal summer (May- October) season over a 20-year period. The empirical orthogonal function (EOF) eigenvectors of the initial perturbations depended on the individual perturbation methods. The leading EOF eigenvectors of the LAF perturbations exhibited large variances in the extratropics. The Bred vectors with a breeding interval of three days represented a local unstable mode moving northward and eastward over the Indian and western Pacific region, and the leading EOF modes of the ESV perturbations represented the planetary-scale eastward moving perturbations over the tropics. By combing the three perturbation methods, a multi-perturbation (MP) ensemble prediction system for the intraseasonal time scale was constructed, and the effectiveness of the MP prediction system for the Madden and Julian Oscillation (MJO) prediction was examined in the present study. The MJO prediction skills of individual perturbation methods were all similar, however, the MP‐based predictions had a higher correlation skill in predicting the real-time multivariate Madden-Julian Oscillation (RMM) indices compared to those of the individual perturbation methods. The predictability of the intraseasonal oscillation is sensitive to the MJO amplitude and to the location of the dominant convective anomaly at the initial state. The improvement in skill of the MP prediction system was more effective during periods of weak MJO activity.