It is well known that the atmosphere is a nonlinear dynamical system with chaotic characteristics,and small differences in the initial value of the numerical model may lead to completely different results.Ensemble prediction is a new generation of stochastic dynamic forecasting technique.It is based on the analysis of the initial value of the assimilation analysis to generate a set of normal distribution of the initial disturbance,thus it can be used to reflect the uncertainty in the assimilation analysis.The method by which to generate the initial set of disturbances is the core of ensemble prediction.The ETKF method is an initial perturbation technique that has been developed over the past 10 years,and has been widely used.Because the number of actual ensemble members is far less than the prediction of the model,the variance of the ensemble prediction model prediction may be underestimated,thus an amplification factor is introduced to adjust the magnitude of the ETKF.Observation mean square error has a major impact on the structure and initial perturbation to the regional Ensemble Prediction System of the China Meteorological Administration Numerical Prediction Center.In this paper we design three different sets of numerical simulations of the sensitivity tests of observed error from August 1 to August 29 2012.We then analyze the impact of the structure and initial perturbation on the initial perturbation field,and assess the difference of the total energy of vertical perturbation and ensemble forecast skill score by means of the GRAPES-MEPS(Global/Regional Assimilation and Prediction System,Mesoscale Ensemble Prediction System) of the China Meteorological Administration Numerical Prediction Center.In addition,we analyze a typical ensemble prediction rainfall in the Yangtze-Huaihe River Basin.The results indicate that with the observation mean square error reduced,the model variable temperature and initial perturbation wind increases,and the ensemble forecasting dispersion grows slightly better.The precipitation area tests show that the ensemble forecasting precipitation is more effective when the observation mean square error is smaller,in which case the ensemble mean total energy has a better growth and its vertical structure is more obvious.The smaller the mean square error of the observation error is,the larger the total energy of the set predicted perturbation generated by the ETKF scheme will be,which in turn affects the increase of the later disturbance energy.It is also found that the total energy of the low-level initial disturbance is slower,due to the non-uniform distribution of the total energy perturbation of the GRAPES regional set.Therefore,we can use the disturbance observation mean square error appropriately to reflect the impact of observation mean square error on the ensemble prediction,thereby improving the techniques of GRAPES-MEPS ensemble prediction.