Abstract: Due to limited observational data, the vortex fields in a global numeric al model are often too weak. However, in a regional model, the TC vortex fields are well developed because it has a higher resolution and more sophisticated physical process package which benefit typhoon simulation. Generally, the background fields for the regional model are from the global model forecasts. The well developed TC vortex in a regional model is unable to be inherited so as to occur in the initial field for the next forecast of the regional model. In this paper, the BOGUS data assimilation (BDA) scheme and the TC vortex relocation scheme are combined, and using MC-3DVAR technique, a new vortex initialization scheme called cycling data assimilation (CDA) scheme is developed. In this scheme, the profiles around the TC center are extracted from the previous 6-h ours forecast fields of a regional model. Then these profile data are displaced to an accurate location determined based on the TC location from the observations and assimilated into the first guest fields which are from the global analyses. The CDA scheme insures that the TC vortex developed in the regional model will be kept in the next run. Case study shows that the CDA scheme has a positive impact on the initial TC intensity and structure. Compared to that of the control experiment, the intensity of the initial TC is closer to the observations with the structure more compact. Therefore, the experiment using the CDA scheme has a better performance on the track and intensity forecast than the control experiment and the experiment using the BDA scheme. The overall results for thesuper typhoon “CHANGMI” show that compared to the control experiments, the CDA sch eme improves the track forecast evidently. The 24-h distance error is reduced from 121 to 89 km and 48-h distance error is from 187 to 179 km. At the same time, the CDA scheme also has a positive impact on the intensity forecast. The RMSE of 24-h minimum sea level pressure is less than 20 hPa which is much less than that of control experiment. The RMSE of 48h minimum sea level pressure is a little larger thanthat of control experiment. In 2009, GRAPES-TCM based on the CDA scheme was put into quasi-operational run. So far, the performance of the quasi operational system on track and intensity forecasts is better than the present GRAPESTCM opera tional system.