PCRF: An Independent Cloud Detection Framework for FY-4B GIIRS Hyperspectral Infrared Sounder Using PCA Reconstruction and Random Forest

Accurate cloud detection is essential for the assimilation of hyperspectral infrared observations from the Geostationary Interferometric Infrared Sounder (GIIRS) aboard Fengyun-4B (FY-4B). This study introduces a novel, independent cloud detection framework termed PCRF, which integrates Principal Component Analysis (PCA)-based reconstruction of clear-sky spectra with an optimized Random Forest (RF) classifier. In contrast to operational methods that depend on external imager products e.g., Advanced Geostationary Radiation Imager (AGRI), PCRF exploits the intrinsic spectral characteristics of GIIRS longwave channels to identify cloud-contaminated pixels through reconstruction error analysis. Fourier-based spectral features are further incorporated to better represent cloud microphysical properties. When validated against the Moderate Resolution Imaging Spectroradiometer (MODIS) cloud masks, PCRF achieves overall accuracies of 91.3% over ocean and 86.7% over land, outperforming the operational AGRI cloud product by 2.1% and 6.6%, respectively. An independent validation using July 2024 data confirms the model’s temporal stability and spatial generalizability. Spatial comparisons reveal that PCRF delineates sharper cloud boundaries and more effectively detects fragmented clear-sky regions, particularly over complex surfaces. These results highlight PCRF’s potential as a robust cloud detection solution that can operate without external imager cloud products during routine application, offering direct benefits for numerical weather prediction (NWP) data assimilation and satellite product quality control.