关键词: CICE6.0, 海冰模式, 融池, 参数化方案, 伴随模式, 参数估计, 北极

Abstract: Melt ponds can significantly affect Arctic sea ice. However, in the widely used sea ice model CICE6.0, large differences still exist between simulation and observation in terms of melt pond fraction. The CESM melt pond parameterization scheme in CICE6.0 can calculate the volume, area fraction and depth of melt ponds.However, the scheme has some uncertain parameters, which can affect melt pond simulation. These parameters could be determined through parameter estimation, which requires a large amount of sensitivity simulations. An adjoint model can efficiently calculate parameter sensitivity. In this work, an adjoint model was developed for the CESM melt pond scheme through a forward model code transformation method. The adjoint model was checked following the standard method of adjoint model development. A melt pond parameter estimation algorithm was also developed, which is composed of the CICE6.0 sea ice model, melt pond adjoint model, and L-BFGS minimization algorithm. The parameter estimation algorithm was further verified under ideal conditions. By using the developed parameter estimation algorithm and MODIS satellite melt pond fraction observation data as the observational constraints, the melt pond aspect ratio parameter in the CESM scheme was estimated in several time periods during summer and for two different regions in the Arctic. One region was for first-year ice, and the other multi-year ice. The estimated parameter values were used in later simulation, and the output values were compared with the simulation using the default parameter. The results show that the estimated parameter values were quite different for multi-year ice and first-year ice and also varied with time. This is quite different from the default constant value of the parameter used in CICE6.0. The estimated parameter values were found to be more appropriate for melt pond simulation. The simulation error of the melt pond fraction was reduced. The simulated melt pond fraction was more consistent with MODIS observations than that using the default parameter.

Key words: CICE6.0, sea ice model, melt pond, parameterization, adjoint model, parameter estimation, Arctic