Simultaneous retrieval of temperature and humidity profiles from ground‐based infrared hyperspectral spectrometer by combined machine learning and physical iterative method

Abstract The temperature and humidity profiles are the basic parameters to describe the vertical structure of the atmosphere. Efficiently and accurately obtaining atmospheric profiles has always been an important premise in the study of the thermal and dynamic characteristics of the boundary layer. Aiming at this problem, this paper proposes a retrieval method by combining the physical iterative method and the Convolutional Neural Network (CNN), by which the atmospheric temperature and humidity profiles can be simultaneously retrieved based on ground‐based infrared hyperspectral observation. From the results of retrieval experiments, it can be found that the hybrid retrieval method can not only improve the retrieval accuracy of temperature and humidity profiles, but also significantly reduce time consumption compared with the physical iterative method. For the monthly statistical results of temperature, the mean bias (BIAS), and root mean‐squared error (RMSE) of the hybrid retrieval method are improved by at least 29% and 28% compared with the physical iterative method, and also show an improvement of 7% and 6% over the CNN method. The retrieval accuracy has also improved notably for humidity profiles, where compared with the physical iterative method, the BIAS is reduced by nearly 38% in October and the RMSE is reduced by nearly 35%. However, in May, both BIAS and RMSE are reduced by more than 35%. Compared with the CNN algorithm, the BIAS and RMSE of humidity decreased by 6% and 5% in September, respectively.