In this study, we investigate how different physical parameterizations impact the simulation of surface (2 m) air temperature over China using the Model for Prediction Across Scales-Atmosphere (MPAS-A) with a customized 92-25 km variable resolution. All simulations use ERA-Interim reanalysis data as the initial conditions for three selected typical years (1987, 1998 and 2011), and seven sub-regions is divided to assess the performance of different parameterization schemes on varying climate types. We performed 12 simulations using combinations of two shortwave–longwave (SW-LW) radiation, three cumulus convective (CUM) and two microphysics (MP) schemes. Model performance is evaluated using CN05 observation. The simulation shows significant sensitivity to SW-LW schemes, with the large difference of RSME scores. And RRTMG exhibits lower bias over more than half of the sub-regions. The influence of CUM schemes is weaker than that of SW-LW schemes on surface air temperature simulation. GF shows best model BIAS over more sub-regions. Followed is KF. For the MP schemes, the difference between the simulation results of Thompson and WSM6 is not significant for the same time and sub-region. Further we analysis the influence of different parameterization combinations. RRTMG_Thompson_NTDK combination performs best relative to other combinations under all conditions, with higher spatial correlation coefficient, lower root-mean-square error and bias. The annual cycles of surface air temperature could also be reasonably reproduced. Our results indicate a better parametrization scheme configuration, evaluate the simulation performance of the MPAS model for surface air temperature over China, and provide a meaningful reference for future researches.