Quantification and optimization of parameter uncertainty in the grid-point atmospheric model GAMIL2

Physical parameterization is one of the most important sources of uncertainties in the current climate system models. With the increasing complexity of models and the diverse requirements for climate studies, the priori and manual model tuning method for physical parameterization has become a bottleneck to further improve the climate system model. In this study, we propose a “two-step” parameter optimization approach. In the first step, an improved full factor sampling scheme is presented to determine the area where the optimal solutions are likely to be found. In the second step, the simplex downhill algorithm is used to perform the search with low computational costs. When applying this “two-step” method to GAMIL2, the grid-point atmospheric model of LASG (State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics), IAP (Institute of Atmospheric Physics), three important parameters from deep convection scheme and cloud fraction scheme are tuned to improve the model performance measured by a comprehensive metrics based on precipitation, wind, temperature, humidity, potential height as well as radiation flux fields. Results show that the proposed metrics is improved by 7.5% compared with the standard GAMIL2 version using our proposed optimization method. The optimal parameters improve the condensation efficiency, leading to reducing the simulated bias of moisture and cloud fraction. Meanwhile, the adjustment of condensation further affects the simulation of temperature, geopotential height, and wind.