Reconstruction of 2‐m Air Temperature From ERA5 Reanalysis at Dome A, Antarctica

ABSTRACT In this study, we jointly used in situ air temperature from AWS and reanalysis data from ERA5 to make the first‐ever reconstruction of a 42‐year (1978–2020) air temperature time series for Dome A, Antarctica. By analysing the impact of environmental variables, we found that the 10‐m u‐component of wind was the predominant one for air temperature bias between ERA5 and AWS, followed by total cloud cover. Air temperature deviations between ERA5 and AWS during the period of 2005–2020 were successfully reduced by applying a random forest (RF) model, decreasing the bias by 0.52°C, the RMSE by 3.16°C and the MAE by 2.77°C. We next applied the RF model to predict the 2‐m air temperature difference which was added back to correct ERA5 from 1978 to 2004. This yielded an accurate time series of air temperature from 1978 to 2020. Using the innovative trend analysis method to analyse the temperature trend of the corrected data, we found that Dome A has experienced a gradual warming of 0.10°C dec −1 over the 42‐year period. Among the seasonal temperature changes, spring showed a significant warming trend of 0.57°C dec −1 , autumn and winter showed no significant warming, while summer showed a slightly cooling trend. Also, over the 42‐year analysis period, a stable oscillation period of ~28 year was observed. This cycle emerged as the dominant pattern, influencing the overall temperature evolution. The method proposed in this research, which combines machine learning with AWS to correct ERA5 air temperature data, holds the potential to address spurious changes of reanalysis data in long‐time series studies, thus improving the reliability of trend analyses.