Abstract: Using the GRAPES-REPS (Global and Regional Assimilation and Prediction Enhanced System-Regional Ensemble Prediction System) model and ETKF (Ensemble Transform Kalman Filter) initial perturbation method, ensemble forecast experiments at 10 km and 15 km horizontal resolutions were conducted from 1 to 15 June 2015. By analyzing ETKF characteristics of initial perturbation components, variance accuracy, kinetic energy spectrum, perturbation energy evolution, daily variation and ensemble spread, root mean square error, characteristics of initial perturbation structures and its growth rate in the regional ensemble prediction system of GRAPES-REPS are revealed in the present study. The preliminary experimental results indicate:(1) The ETKF initial ensemble perturbations can maintain even distribution to the orthogonal and uncorrelated ensemble directions, while the value of α parameter and inflation factor of the ETKF component are stable. (2) The perturbation field derived from the ETKF initial perturbation schemes is mainly large-scale with a flow dependent structure. The internal perturbation energy and the kinetic perturbation energy are the prevailing perturbation energy in the lower and higher levels, respectively. In addition, the ensemble perturbation can effectively capture the structure of forecast error. (3) The total energy of initial perturbation and ensemble spread can keep appropriate growth rate at all forecast lead times. However, the growth rate of ensemble spread is lower than that of root mean square error, indicating that there still exist problems of insufficient ensemble spread. (4) The increase in horizontal resolution can increase the large-scale perturbation spectrum energy at the middle and high levels, and contribute to the improvement of ensemble forecast results at isobaric levels and near the surface, such as forecast of temperature and wind. It is worth pointing out that the internal perturbation energy in the lower levels shows obvious diurnal variation characteristics in GRAPES-REPS, especially over the Qinghai-Tibet Plateau. It is necessary to further explore the rationality of initial perturbation structure over the Qinghai-Tibet Plateau.