Abstract
Temperature profiles are important meteorological parameters of the atmosphere that can determine atmospheric thermal processes. Detecting global spatial and temporal continuous atmospheric temperature profiles is crucial for weather protection work. Atmospheric datasets such as ERA5 (fifth generation ECMWF reanalysis) provide global and continuous temperature profile datasets with good resolution. RAOB (radiosonde) sounding data have high confidence and representativeness and are commonly used for data accuracy validation. In this paper, we use the RAOB sounding data of 2017 as the true value and revise the ERA5 reanalysis data based on machine learning methods to optimize the data. The algorithm not only improves the problem of RAOB distribution discontinuity but also improves the accuracy of ERA5 itself. In order to verify the results of the algorithm, the RAOB sounding data are compared with it, and it is found that the accuracy of the revised data is reduced by about 3K compared to the preprocessing RMSE, which is closer to the RAOB data. The algorithm proposed in this paper can provide important data support for subsequent meteorological studies.