Abstract: By continuously developing and optimizing the linearized physical process of the tangent linear model and keeping it consistent with the nonlinear model, the analysis and forecasting performance of four-dimensional variational data assimilation (4DVar) of China Meteorological Administration Global Forecast System (CMA-GFS) can be effectively improved. Currently, the planetary boundary layer (PBL) parameterization scheme adopted by CMA-GFS model is based on Charney-Phillips (C-P) grid, while the linearization PBL parameterization scheme used in CMA-GFS 4DVar is on the basis of Lorenz grid. The zigzag noise of the temperature and moisture in the boundary layer is removed and the correspondent profiles appear to be smooth with C-P PBL parameterization scheme, and the forecast errors of CMA-GFS model are effectively reduced. To improve the analysis and prediction effects of 4DVar on the boundary layer, a new linearized PBL parameterization scheme based on the C-P grid (TL_NMRF_CP) is developed. By making more refined regularization constraints on the disturbances of the surface heat flux, the surface water vapor flux, Richardson coefficient of the free atmosphere, the heat and momentum exchange coefficient of the boundary layer, the influence of the linearization process on the accuracy of the tangent linear approximation is reduced while ensuring stable operation of the tangent linear and adjoint models. Two tests are designed, one is TL_MRF test based on the original TL_MRF scheme and the other is TL_NMRF_CP test based on the TL_NMRF_CP scheme. The tangent linear approximation tests show that, compared with the original scheme, the TL_NMRF_CP scheme can improve the tangent linear model forecast accuracy on the boundary layer, reduce the relative error of the potential temperature of the boundary layer by up to 10%, reduce the relative error of the specific humidity by 5% at the most, and reduce the relative error of zonal wind by up to 12%. In the minimization process of CMA-GFS 4DVar, the TL_NMRF_CP scheme can reduce the relative difference of the cost functions caused by different resolutions between inner and outer loops of the 4DVar system, and improve the convergence efficiency of 4DVar. The batch cycle assimilation forecast experiments of CMA-GFS 4DVar further verify that the TL_NMRF_CP scheme can provide better analysis and prediction results. The TL_NMRF_CP scheme has also been applied to CMA-GFS 4DVar operation system.