Abstract: A series of data assimilation and forecast test has been carried out with a hemispheric spectral model. It was found that the numerical scheme for determing hemispheric initial wind is important to data assimilation and forecast. An inappropriate scheme may cause the computational divergence sources near the equator, which are responsible for the spurious strong precipitation and corresponding latent heat release. Obviously, this problem differs from either the hemispheric/global domain effect or the tropical data effect pointed by Daley et al. Based on the previous studies. the new divergence and vorticity correction schemes is presented. and the difference with other schemes and its effects on the data assimilation are discussed against the control test. Preliminary tests have shown that the new divergence correction scheme proposed in this paper may be a preferable choice to overcome the initial computational errors in a hemispheric data assimilation.