ABSTRACT This study addresses the challenge of radio‐frequency interference (RFI) in brightness temperature (BT) observations from China's FY‐3G satellite, a low‐inclination‐orbit platform equipped with a Microwave Radiation Imager (MWRI) designed for precipitation monitoring. To mitigate RFI contamination, we propose a deep neural network (DNN) framework that integrates multidimensional features, including interchannel correlations, topographic variables, and principal components, with particular emphasis on the complex conditions near land–sea boundaries. Validation using FY‐3G seasonal orbital data demonstrates outstanding performance, with simulated BTs achieving correlation coefficients greater than 0.999 and standard deviations below 0.5 K. An RFI index derived from residual analysis enables effective identification and correction of contamination, and results further show that in snow‐ and ice‐covered regions during winter, the proposed method significantly outperforms conventional techniques. The corrected BT dataset improves spatial and temporal consistency and provides reliable input for geophysical retrievals, including land surface temperature estimation, thereby confirming the practical applicability of the framework. Compared with existing approaches, the proposed method enhances generalization capability, improves detection accuracy, and reduces false alarms, ultimately establishing a robust quality control mechanism for the low‐frequency channels of FY‐3G MWRI.
Xiao et al. (Wed,) studied this question.