Comparing Surface Mass Balance and Surface Temperatures From Regional Climate Models and Reanalyses to Observations Over the Antarctic Ice Sheet

ABSTRACT This article assesses the performance of surface mass balance (SMB) and surface temperature products from seven climate models (ERA5, MERRA2, HIRHAM5_ERA‐Interim, HIRHAM5_ERA5, MARv3.11, RACMO2.3p1 and RACMO2.3p2) to be used as inputs to firn densification models (FDMs), crucial for the whole Antarctic mass balance estimation. The evaluation compiles and utilises a new dataset combining existing observations with 184 stake SMB measurements in the Vostok region and two SMB radar transects in Blåskimen Island and Nivlisen Ice Shelf, significantly enhancing the AntSMB dataset. For temperatures at 10 m (T10), 992 observations are included, offering vastly improved spatial coverage. Compared with multi‐year SMB observations, RACMO2.3p1 and ERA5 align closely with observed SMB over the inland and the whole Antarctic, with ERA5 exhibiting superior alignment in coastal areas. However, in terms of temporal SMB variations, ERA5, RACMO2.3p2 and HIRHAM5_ERA‐Interim demonstrate the best consistency with observed SMB, with ERA5 excelling in regression slope comparisons. In our surface temperature evaluation, RACMO2.3p2 and ERA5 have a minimal mean bias from T10 over the whole Antarctic. RACMO2.3p2 performs well with few occurrences of deviations beyond ±3°C in coastal areas, while RACMO2.3p2 and HIRHAM5_ERA5 are both close to T10 in inland areas. In inter‐model SMB comparisons, ERA5, HIRHAM5_Interim, HIRHAM5_ERA5, RACMO2.3p1 and RACMO2.3p2 exhibit similar deviations from the multi‐model mean (MMM), while MARv3.11 and MERRA2 show larger deviations in some coastal areas. For surface temperature, MARv3.11 and ERA5 display larger deviations from the MMM in some coastal areas where ERA5 actually matches the observations well, illustrating that MMM should not be taken to represent the true climate conditions. In conclusion, ERA5 demonstrates the best overall consistency with SMB and T10 observations compared with the other models.