The Application Of Adjoint Model In The Parameter Estimation Of Melt Pond Parameterization Scheme

Numerical models in marine sciences usually have some important sub-grid physical processes which need to be parameterized in the models.Parameterization scheme of these processes often has physical basis,but some of their key parameters have great uncertainty,which will affect the long-term integration results of the model.Such uncertain parameters exist in the melt pond scheme of sea ice model.Due to the limitation of observation,determination of these key parameters often requires a lot of sensitivity tests and is limited by the huge computation cost.The adjoint model method provides another way for parameter estimation,which has been widely used in the estimation and optimization of sea ice model parameters.However,it is a time-consuming and difficult work to develop the adjoint model by the code conversion method.At present,the adjoint model has not been used in parameter estimation of the melt pond parameters in the sea ice model.In this study,the adjoint model of the CESM melt pond parameterization scheme in CICE6 sea ice model was developed,and the pond aspect ratio parameter was estimated using the adjoint model.The estimated parameter was used in simulation to evaluate the impact of the estimated parameters on the melt pond and sea ice simulation.The main conclusions are:(1)The tangent linear model and adjoint model of the CESM melt pond parameterization scheme in CICE6 sea ice model were developed correctly through forward model code translation.(2)The melt pond forward model,adjoint model and L-BFGS minimization algorithm were used to construct the melt pond parameter estimation algorithm.The parameter estimation algorithm was verified using the output of model forward integration using the default parameter value of CICE6 as the "observation" constraint.The results show that the parameter estimation algorithm was correct and effective.(3)The pond aspect ratio of CESM was estimated in several time periods during summer time and two different regions.One region is for first-year ice and the other multi-year ice.MODIS observation data was applied as constraints.The results show that,the estimated parameter changed over time.This is different from the CICE6 default configuration.There were significant differences between the parameter estimated on multi-year ice and first-year ice.The sea ice simulation experiment was carried out using the estimated melt pond parameters.The results show that the simulated pond fraction using the estimated parameters was closer to the observation than that using the default parameters.The adjoint model provides an efficient way for estimating melt pond parameter in a sea ice model and can be useful for improving sea ice models.