A Fast Atmospheric Correction Model for L-band Spaceborne Microwave Radiometer Based on Neural Network

Atmospheric correction is critical for the application of spaceborne microwave radiometers. After the successful launch of China's HY-4 satellite, the sea surface salinity detection accuracy is expected to be further improved, which puts higher demands on the accuracy of atmospheric correction. In this context, the neural network is used for the atmospheric correction of L-band spaceborne microwave radiometers for the first time. Firstly, the traditional top-of-atmosphere brightness temperature model was reformulated, yielding a novel linear atmospheric correction equation with respect to the Earth's surface brightness temperature. The slope and intercept of this equation can serve directly as the atmospheric correction coefficients for modeling purposes, thus improving efficiency and accuracy. Secondly, based on the MPM93 atmospheric radiative transfer model and ERA5 hourly reanalysis data, the sensitivity of surface water vapor density and total column water vapor was analyzed for the purpose of optimizing the model input parameter. Thirdly, the A-B coefficients atmospheric correction model was developed using the neural network method, which can greatly simplify the atmospheric correction process. Finally, comprehensive comparative tests were performed using the Peng model and SMAP L1B data. The results demonstrate that the A-B model has good consistency with the Peng model, and has an average error of about 0.03K compared to the SMAP L1B data. This proves the accuracy and reliability of the A-B model, and provide a reliable basis for its future atmospheric correction applications in China's ocean salinity satellite mission.