Abstract: The GRAPES ensemble Kalman filter (EnKF) is able to assimilate the conventional observations in batches. The design and tentative experiment of the EnKF are introduced in part I. The experiment results of assimilating the regional radiosonds show that the reasonable analysis of the GRAPES EnKF is obtained and the system can be practically applied to the operational forecast. In this paper, the regional analysis and ensemble prediction experiments of the GRAPES EnKF are carried out, and the results are compared with the fields of the 3DVar assimilation system in sigma level. The results indicate that the analysis of the EnKF has the competitive advantage over the 3DVar analysis and the predicted precipitation of the EnKF is more similar to that of the observational station. The forecasting errors have been examined, which shows that the distribution of error correlation and error variance are flowdependent.