The ensemble transform Kalman filter (ETKF) is an effective ensemble prediction initial perturbation scheme and is widely used.However,the finite ensemble sample size,the same ensemble member setting in ETKF and the forecast model error may make the two remote state variables have higher spurious correlation,thus affecting the quality of ETKF ensemble perturbation.The reason why ETKF generates spurious correlation is that each ensemble member is an estimate of the atmospheric state,while the degree of atmospheric freedom is too high,and the limited ensemble members are difficult to fully express.On the other hand,due to the effect of the forecast model error,the same size of members may lead to convergence of different ensemble members in the prediction process,resulting in spurious correlation.In order to solve this problem,the localization of ETKF,called LETKF,is proposed.By means of localization,the spurious correlation of error variance can be truncated in the localized radius,thus improving the quality of error variance.That is to say,only the observation data in the local radius are absorbed and aimed at a grid point,and the observation outside the radius is not taken into account so as to avoid the spurious correlation at a distance.Based on the GRAPES regional ensemble prediction system (GRAPES REPS),the localization scheme of ETKF initial perturbation is developed on the basis of the ETKF initial perturbation scheme,in order to solve the problem of the range spurious perturbation and the divergence of the filter in the regional ensemble prediction.Through the continuous experiments for 7 days,this paper analyzes the ensemble prediction quality of LETKF initial perturbation scheme from the case of rainstorm and multiple scoring methods of ensemble prediction.Results show that the localization scheme of ensemble transform Kalman filter initial perturbation in regional ensemble prediction can more reasonably capture the physical structure of the rapidly growing analysis error,and more accurately reproduce the linear and nonlinear propagation and evolution characteristics of the forecast error in the numerical model.The localization scheme can improve the quality of forecast and increase the accuracy of precipitation forecast,especially for the forecast of magnitude of light rain,moderate rain and rainstorm.Compared with the existing regional ensemble prediction business system GRAPES REPS,the regional ensemble prediction produced by LETKF initial perturbation scheme has obvious advantages.In general,the LETKF initial perturbation scheme can improve the quality of regional ensemble prediction.