Optimized sea ice simulation in MITgcm-ECCO2 forced by ERA5

Abstract

Coupled ice–ocean models require an accurate performance of sea ice dynamics and thermodynamics to realistically simulate sea ice circulation in the polar regions. However, atmospheric surface variables from reanalysis products that force the models differ in spatial and temporal resolutions, and there are also substantial inconsistencies in the distributions and magnitudes of these variables between the products. Therefore, several key parameters in the models need to be adjusted for the models forced by different atmospheric reanalysis products. This study compares the sea ice properties simulated by the previously optimized MITgcm-ECCO2 model forced by JRA25 and forced by the recently released ERA5. The comparisons with high-resolution satellite sea ice data reveal that the simulations forced by these two datasets significantly overestimate the Arctic sea ice concentration in summer, whereas the simulation forced by ERA5 underestimates the winter Antarctic sea ice. It is likely that the downward longwave radiation and wind stress regime in ERA5 is responsible for the underestimated Antarctic sea ice cover. We then optimize MITgcm-ECCO2 model forced by ERA5 by using Green functions. We consider six key parameters: sea ice strength, ice-atmospheric drag coefficient, ice–ocean drag coefficient, ice and snow albedos, and sea ice salinity. The optimized simulation significantly improves the sea ice distribution in both regions using a more realistic set of sea ice parameters. The most significant improvement is in the Antarctic, where winter sea ice extent misfit decreases by 48%. The optimized simulation reveals that ERA5 reproduces a closer representation of the Arctic sea ice extent and thickness than JRA25. In the southern hemisphere, ERA5 simulates the sea ice drift better, whereas the JRA25 have the closest match to the sea ice extent. Despite the improvements, further work is necessary for the Antarctic, where the lack of reliable observational data and limitations of the ocean model hinders the optimization.