Abstract: Using NCEP ensemble forecast products and continuous experiments, the GRAPES global ensemble-3DVAR, which is a new hybrid assimilation system is studied. Of all the experimental schemes, the GRAPES global ensemble-3DVAR is found to have the smallest analysis and forecast errors when the ensemble sample number is 60 and the ensemble weight is 0.5. However, above 150 hPa, the analysis error exceeded that of the 3DVAR. Therefore, the coupling of the static covariance and the ensemble covariance are also considered for upper layers, which is different from that of the original design. Our experiments revealed that this improvement could resolve the above-mentioned problem of large errors in the upper layer analysis.