Abstract: The results of Part I of this study show that the imposition of the Linear Balance Equation(LBE) in the tropics is not correct and additional steps should be taken to decouple the analysis of mass and wind fields. Based on the results in Part I, this paper developed a LBE-regression hybrid balance constraint in GRAPES(global/regional assimilation and prediction system) global variational data assimilation system. In the new scheme, after the calculation of LBE on each level, a vertical regression whose coefficients could vary at different latitudes and model levels was introduced. By analyzing and comparing the implied background error covariance of different schemes, the authors found that the new scheme could efficiently reduce the unreasonable correlations of control variables. Additionally, the forecast error variance of the balanced pressure and its ratio to the whole pressure were also both dramatically reduced in the tropics. Results of the single-observation experiments indicated that, the new scheme had little impacts on the mid- and high-latitude compared to the LBE scheme, but it did successfully decouple the mass/wind analyses in the tropics. Although equatorial wave modes were not explicitly considered in the new scheme, the structure of covariance was consist with the theory analysis results based on those modes in Part I. The results of cycle analysis and forecast experiments in one month showed that, the new scheme could bring slightly positive results in the extratropics and significantly improve the wind analysis and forecast accuracy in the tropics, the abnormally large tropical wind errors in the LBE scheme were dramatically suppressed.