Bridging the cloud gap: AHI/ATMS synergy through CNN feature fusion for all-weather SST retrieval

Infrared-based Sea Surface Temperature (SST) retrieval methods face persistent challenges from cloud-induced data gaps and accuracy degradation. This study bridges this critical limitation through multisensor satellite synergy, integrating geostationary Advanced Himawari Imager (AHI) with Advanced Technology Microwave Sounder (ATMS) data via a Convolutional Neural Network (CNN) for all weather SST retrieval. The CNN model adaptively extracts features from multi-band AHI/ATMS imagery, effectively predicting SST under varying cloud conditions. Evaluation results demonstrate a root mean square error (RMSE) of 2.07 °C, a mean absolute error (MAE) of 1.22 °C, and a coefficient of determination (R²) of 0.88 on the test dataset. Under the same CNN framework, unimodal retrievals from AHI and ATMS alone yield substantially lower performance (R² = 0.51, RMSE = 3.45 °C; and R² = 0.63, RMSE = 2.64 °C, respectively), confirming the complementary benefits of multisensor fusion. Comparisons with a Transformer-based model, the daily OSTIA product, and the official AHI SST product (clear-sky conditions) further indicate that the proposed CNN achieves the highest accuracy. Although RMSE exceeds 1 °C for certain cloud types, the method substantially mitigates cloud-induced data loss and provides a reliable, high-accuracy, all-weather SST retrieval strategy for satellite ocean remote sensing.