All-sky temperature and humidity retrieval from the MWRI-RM onboard the FY-3G satellite

<p>To investigate the application of deep learning in satellite remote sensing, this study employs brightness temperature observations from the remapped Micro-Wave Radiation Imager-Rainfall Mission (MWRI-RM) onboard the Fengyun-3G (FY-3G) satellite as input data, while temperature and relative humidity profiles (ranging from 1000 to 100 hPa) obtained from ERA5 reanalysis data are used as label data. An Advanced Residual Convolutional Neural Network (AR-CNN) model was developed to retrieve atmospheric temperature and relative humidity profile data. The results show that: (1) The retrieval of temperature profiles achieves a root-mean-square error (RMSE) of approximately 1.24 K, and the RMSE for relative humidity profiles is 12.98 %. (2) A comparison between retrieved and labeled samples reveals consistent results for temperature retrieval but some discrepancies in extreme high and low humidity regions, suggesting the need for further refinement. (3) Gradient-based analyses and perturbation experiments confirm that 118 GHz oxygen channels are critical for mid-to-upper tropospheric temperature (500–200 hPa), indirectly impacting upper-level humidity (200–100 hPa) through thermal coupling, while 183 GHz water vapor channels dominate lower-to-mid tropospheric humidity retrievals (1000–500 hPa) and constrain temperature via moisture-radiation feedbacks. (4) Additional channel ablation experiments demonstrate that channels with smaller frequency offsets mainly affect upper atmospheric layers, whereas larger-offset channels have stronger impacts on lower layers, supporting the spectral contribution patterns identified in previous studies. These findings highlight the model's ability to capture temperature-humidity coupling and confirm the complementary roles of 118 and 183 GHz channels in improving vertical profile retrievals.</p>