Abstract: The selection of the reference profile is critical for the accuracy of the semi-implicit semi-Lagrangian dynamic core of atmospheric model. The GRAPES_GFS (Global Regional Assimilation and PrEdiction System, Global Forecast System) developed in the Numerical Weather Prediction Center of China Meteorological Administration is based on one-dimensional reference profile. While this method is simple and easy to realize, it reduces the accuracy of spatial computation due to large perturbations of dimensionless pressure and potential temperature. Meanwhile, the accuracy of time integration is also low because of the large nonlinear terms. In the present study, with reference to the method of constructing the dynamic core implemented in recent years in several major operational centers, a three dimensional reference profile that does not change with time and satisfies the hydrostatic balance is introduced into the dynamic core of GRAPES_GFS. The reference atmosphere can be as close as possible to the model atmosphere, which improves the accuracy of spatial computation and decreases the nonlinear terms and thus enhances the accuracy of time integration. This paper re-derives the process of solving the dynamic equations after introducing into the dynamic core the three-dimensional reference profile, and verify the results through a number of ideal tests. Results show that the new three-dimensional reference profile can effectively improve the computational accuracy of the dynamic core. The follow-up work will focus on the method of initialization for real-time forecast, and consecutive cycling forecast experiments will be conducted to evaluate the actual prediction performance of the new dynamic core.