Estimation of Air Temperature from FY-4A AGRI Data: A Comparison of Different Machine Learning Algorithm

Abstract

Air Temperature(Tair),a basic meteorological observation element, is an essential meteorological parameter in physiology, hydrology, meteorology, environment, etc. The Tair data ,which is characterized by high precision, is of great significance for the greenhouse effect, land surface processes and so on. With the advent of high performing imaging instrument onboard geostationary satellites such as Advanced Geostationary Radiation Imager(AGRI) onboard FY-4A of China, it provides high spatial and temporal resolution data. To estimate Tair from such high-resolution data, this paper presents an effective method for estimation Tair based on AGRI data. Different machine learning algorithms–-random forest (RF), k-nearest neighbors(KNN) and extreme gradient boosting(XGB)–-are evaluated for estimation of Tair under clear sky conditions in the Southwest of China. For the training dataset, the two infrared brightness temperatures of AGRI (BT12 and BT13), digital elevation model(DEM), latitude and longitude, surface pressure, time and relative humidity(RH) are selected. The Tair data obtained by National Centers for Environmental Information(NCEI), evaluates different machine learning algorithm performance in the Southwest of China. The results show that the performance of the XGB model is better than RF and KNN with a correlation coefficient (R) of 0.977, a mean bias of -0.036□,and the root mean square error (RMSE) of 1.266□.