Abstract: To address problems including underestimation of heavy precipitation, rapid decay of synoptic systems and low computational efficiency in operational forecast of CMA-GFS V3.3, some key technologies related to physics and dynamics of the model are developed and applied.A suite of graupel-related microphysical processes is adopted in the cloud microphysics scheme to improve the forecast performance of heavy precipitation. These processes include graupel colliding with cloud water, ice crystals and snow, automatic conversions of ice crystals to graupel and snow to graupel, melting process of graupel to raindrop and sublimation process of graupel. In addition, the evaporation rate of cloud and rainwater is restricted, which can increase the liquid water content in warm areas and improve precipitation efficiency.In the convection parameterization scheme, the role of the sub-cloud environmental relative humidity to convection triggers is considered, and the unreasonable occurrence of convections in dry environment is suppressed. Also, the sensitivity of the entrainment rate of the convective updraft to the relative humidity outside the cloud is enhanced to weaken the convections in dry environment. At the same time, the quasi-equilibrium closure scheme is optimized to improve the accuracy in calculating cloud-base mass flux which is related to the convection intensity.To solve the problem of the mass loss in long time integration, a mass conservation correction method is introduced to the model dynamic framework. The method is developed to ensure the mass conservation by adjusting mass in each grid box according to different weight coefficients which are determined by the change of total atmospheric mass of the current time step relative to the previous step.In terms of computational efficiency, the two-dimensional reference profile algorithm is developed. Without losing calculation accuracy, the model integration time step is extended from 240 s to 300 s using the new profile instead of the original three-dimensional reference profile. Meanwhile, the PCSI method is adopted instead of the GCR method, which reduces the time consuming of solving Helmholtz equation. In addition, the radiation scheme and predictor-corrector algorithms are also optimized to improve the computational efficiency.Through the application of the above key technologies, the forecasting skills for weather pattern and precipitation of CMA-GFS are significantly improved. And its computational efficiency is increased by about 1/3, which meets operational time requirement for the model with 0.125° horizontal resolution. Based on the improved model and the research achievements in other aspects of the forecast system, CMA-GFS is upgraded to V4.0 with a significantly improved comprehensive performance.