Abstract： Spaceborne infrared hyperspectral data have emerged as a cornerstone of modern Numerical Weather Prediction (NWP) systems, enabling high-resolution atmospheric profiling and improved forecast accuracy. However, the utility of these data is significantly constrained by cloud interference, as infrared spectral radiation is strongly attenuated or scattered by cloud particles. Consequently, clear-sky identification— specifically the discrimination of cloud-free pixels and channels—has become an indispensable preprocessing step in data assimilation, ensuring that only reliable observations are integrated into NWP models.This review provides a systematic overview of the evolutionary landscape of clear-sky identification methods for spaceborne infrared atmospheric sounding data, spanning both foreign and domestic sensor systems. It critically evaluates techniques applied to datasets from iconic foreign missions, such as the High-Resolution Infrared Sounder (HIRS), Atmospheric Infrared Sounder (AIRS), Infrared Atmospheric Sounding Interferometer (IASI), and Crosstrack Infrared Sounder (CrIS), alongside domestic advancements using the Hyperspectral Infrared Radiation Sounder (HIRAS) and Global Infrared Imager and Interferometer Sounder (GIIRS) aboard China’s Fengyun satellites.The methodologies are categorized into three distinct technological frameworks: Spectral Feature-Based Approaches: Early techniques rely on single-spectral thresholding, where channels are flagged as clear-sky based on predefined radiance thresholds sensitive to cloud absorption or emission. Advanced variants employ crossspectral consistency checks, leveraging the spectral dependence of cloud properties across multiple wavelength bands to enhance discrimination accuracy. For example, IASI’s cloud-clearing algorithm uses a combination of shortwave and longwave infrared channels to identify consistent clear-sky signatures.Data-Driven and Machine Learning Techniques: Principal component analysis (PCA) has been widely used to reduce the dimensionality of hyperspectral datasets, enabling the extraction of latent variables that distinguish clear-sky from cloudy conditions. More recently, machine learning models—including random forests, support vector machines, and deep neural networks—have demonstrated superior performance in pixel-level clear-sky classification. These models learn complex nonlinear relationships between spectral features and cloud states, achieving higher precision in heterogeneous cloud environments. For instance, AIRS has adopted neural networks to improve clearsky identification in regions with thin cirrus clouds. Domestic Innovations in Assimilation Systems: China’s Fengyun satellite program has developed bespoke clear-sky identification schemes tailored to the HIRAS and GIIRS instruments. These methods integrate physical constraints from radiative transfer models with statistical learning, optimizing clear-sky channel selection for regional NWP models over the Tibetan Plateau and monsoonaffected areas. Such innovations have significantly enhanced the utilization of domestic hyperspectral data in operational assimilation systems. The review highlights two transformative technologies: Three-Dimensional (3D) Clear-Sky Identification: By incorporating vertical atmospheric structure from NWP model forecasts, 3D methods enable the assimilation of clear-sky data above cloud tops, extending the utility of hyperspectral observations in partially cloudy conditions. This approach has been shown to improve upper-tropospheric humidity analysis in Arctic NWP systems. Cross-Spectral Matching with Cloud Parameter Inversion: At the pixel scale, matching hyperspectral observations with cloud properties derived from complementary sensors (e. g., microwave radiometers or visible imagers) has proven particularly effective in multi-phase cloud environments. Compared to standalone 3D methods, this hybrid approach achieves a 15%~20% improvement in clear-sky identification accuracy over ice-water mixed clouds, as demonstrated in CrIS data applications. Looking forward, the review identifies key challenges in all-sky and full-spectrum assimilation, including the handling of sub-pixel cloud heterogeneity, spectral bias in multi-sensor datasets, and computational scalability for real-time operations. To address these, a novel framework is proposed: a machine learning-enhanced 3D clear-sky identification model trained on cross-spectral matching datasets. By fusing physical radiative transfer principles with data-driven learning, this approach promises to unlock the full potential of spaceborne infrared hyperspectral data, offering robust technical support for next-generation NWP systems and advancing global weather forecasting capabilities.