Circumferential Background Field Temperature Inversion Prediction and Correction Based on Ground-Based Microwave Remote Sensing Data

Microwave radiometers are passive remote sensing devices that provide important observational data on the state of the oceanic and terrestrial atmosphere. Temperature retrieval accuracy is crucial for radiometer performance. However, inversions during strong convective weather or seasonal phenomena are short-lived and spatially limited, making it challenging for neural network algorithms trained on historical data to invert accurately, leading to significant errors. This paper proposes a long short-term memory (LSTM) network forecast correction model based on the temperature inversion phenomenon to resolve these large temperature inversion errors. The proposed model leverages the seasonal periodicity of atmospheric temperature profiles in historical data to form a circumferential background field, enabling the prediction of expected background profiles for the forecast day based on temporal and spatial continuity. The atmospheric profiles obtained using the radiometer retrieval are compensated with the forecast temperature inversion vector on the forecast day to obtain the final data. In this study, the accuracy of the forecast correction model was verified utilizing meteorological records for the Taizhou area from 2013 to 2017. Using a hierarchical backpropagation network based on the residual module for comparison, which had a forecast accuracy error of 0.0675 K, the error of our new model was reduced by 34% under the temperature inversion phenomenon. Meanwhile, error fluctuations were reduced by 33% compared with the residual network algorithm, improving the retrieval results’ stability in the temperature inversion state. Our results provide insights to improve radiometer remote sensing accuracy.