Assessment of Antarctic Sea Ice Cover in CMIP6 Prediction with Comparison to AMSR2 during 2015–2021

Abstract

A comprehensive assessment of Antarctic sea ice cover prediction is conducted for twelve CMIP6 models under the scenario of SSP2-4.5, with a comparison to the observed data from the Advanced Microwave Scanning Radiometer 2 (AMSR2) during 2015–2021. In the quantitative evaluation of sea ice extent (SIE) and sea ice area (SIA), most CMIP6 models show reasonable variation and relatively small differences compared to AMSR2. CMCC-CM4-SR5 shows the highest correlation coefficient (0.98 and 0.98) and the lowest RMSD (0.98 × 106 km2 and 1.07 × 106 km2) for SIE and SIA, respectively. In the subregions, the models with the highest correlation coefficient and the lowest RMSD for SIE and SIA are inconsistent. Most models tend to predict smaller SIE and SIA compared to the observational data. GFDL-CM4 and FGOALS-g3 show the smallest mean bias (−4.50 and −1.21 × 105 km2) and the most reasonable interannual agreement of SIE and SIA with AMSR2, respectively. In the assessment of sea ice concentration (SIC), while most models can accurately predict the distribution of large SIC surrounding the Antarctic coastal regions, they tend to underestimate SIC and are unable to replicate the major patterns in the sea ice edge region. GFDL-CM4 and FIO-ESM-2-0 exhibit superior performance, with less bias (less than −5%) and RMSD (less than 23%) for SIC in the Antarctic. GFDL-CM4, FIO-ESM-2-0, and CESM2 exhibit relatively high positive correlation coefficients exceeding 0.60 with the observational data, while few models achieve satisfactory linear trend prediction of SIC. Through the comparison with RMSD, Taylor score (TS) consistently evaluates the Antarctic sea ice cover and proves to be a representative statistical indicator and applicable for its assessment. Based on comprehensive assessments of sea ice cover, CESM2, CMCC-CM4-SR5, FGOALS-g3, FIO-ESM-2-0, and GFDL-CM4 demonstrate more reasonable prediction performance. The assessment findings enhance the understanding of the uncertainties associated with sea ice in the CMIP6 models and highlighting the need for a meticulous selection of the multimodel ensemble.