Abstract: Melt pond has important influences on sea ice. Due to the lack of field observational data, large uncertainties in melt pond parameterization schemes of current sea ice models still exist, and this is one of the key factors degrading the accuracy of sea ice simulation. This research presents the melt pond parameter estimation based on automatic differentiation technique, and an adjoint model of TOPO melt pond scheme in CICE6 sea ice model was attained for the first time. Through sensitivity test of the adjoint model, the drainage rate parameter with the largest sensitivity was chosen as candidate parameter. A parameter adjustment scheme was constructed using the forward model, the TOPO adjoint model, and the L-BFGS optimization program. The drainage rate parameters for the first-year ice and multi-year ice region were adjusted separately, and the optimized parameters were then used to produce new simulation results. It shows that, compared with the simulation with default model parameters, the root mean square error of the melt pond fraction in the first-year ice region is reduced from 12.97% to 5.29%, a reduction of 59.21%. For the multi-year ice region, the root mean square error of the melt pond fraction is reduced from 11.96% to 7.76%, a reduction of 35.12%. The parameter estimation scheme can effectively adjust the model parameters and improve the melt pond fraction simulation, which lays the foundation to simultaneously achieve parameter optimization and multi-parameter estimation for the entire Arctic.