Application of Machine Learning for Simulation of Air Temperature at Dome A

Dome A is the summit of the Antarctic plateau, where the Chinese Kunlun inland station is located. Due to its unique location and high altitude, Dome A provides an important observatory site in analyzing global climate change. However, before the arrival of the Chinese Antarctic expedition in 2005, near-surface air temperatures had not been recorded in the region. In this study, we used meteorological parameters, such as ice surface temperature, radiation, wind speed, and cloud type, to build a reliable model for air temperature estimation. Three models (linear regression, random forest, and deep neural network) were developed based on various input datasets: seasonal factors, skin temperature, shortwave radiation, cloud type, longwave radiation from AVHRR-X products, and wind speed from MERRA-2 reanalysis data. In situ air temperatures from 2010 to 2015 were used for training, while 2005-2009 and 2016-2020 measurements were used for model validation. The results showed that random forest and deep neural network outperformed the linear regression model. In both methods, the 2005-2009 estimates (average bias = 0.86 degrees C and 1 degrees C) were more accurate than the 2016-2020 values (average bias = 1.04 degrees C and 1.26 degrees C). We conclude that the air temperature at Dome A can be accurately estimated (with an average bias less than 1.3 degrees C and RMSE around 3 degrees C) from meteorological parameters using random forest or a deep neural network.