Abstract:The proper initial perturbation structure is the core of constructing ensemble prediction, and the quality of initial perturbation directly affects the quality and overall performance of ensemble forecasting.This study focuses on the uncertainty of the initial value, then analyzes and reveals its spatial physical structure and the spatiotemporal evolution characteristics of the initial disturbance in ensemble prediction.Therefore, this paper provides an objective basis for the rational construction of the initial disturbance in ensemble prediction.In this study, based on the prediction field of the ECMWF, the T639 global ensemble forecast system in China and the GRAPES regional ensemble forecast system, the physical structure and evolution characteristics of wind perturbation in the three ensemble forecasts are revealed by analyzing the initial disturbance component, structure of ensemble spread, and evolution of the perturbation energy.The analysis results show that most of the initial perturbation are located near the main weather systems, and the perturbation has the characteristics of flow-dependence.In addition, the ensemble spread and Total Perturbation Energy present a developing state over forecast hours.Meanwhile, the lower atmosphere is dominated by the Internal Perturbation Energy, while the higher atmosphere is dominated by the Kinetic Perturbation Energy, and the Kinetic Perturbation Energy is dominated in the evolution process.The evolution of the ensemble spread is also closely related to the evolution of the weather situation.This reflects the dependence of the perturbation structure on the flow pattern from another angle.The results confirm that the regional ensemble prediction can reflect more mesoscale and small-scale disturbance information than the global ensemble prediction.The perturbation structure of the ECMWF is more reasonable in the global ensemble prediction system, but the prediction products of T639 is more applicable for China.Compared with the ECMWF, the domestic ensemble prediction system has the drawback of insufficient high-level spread.