Cloud detection using machine learning techniques with application to IASI measurements

Cloud detection plays a crucial role in satellite-based atmospheric analysis, and this study explores the use of machine learning techniques for identifying clouds from satellite radiance observations. Specifically, Support Vector Machines (SVM) were applied to spectral radiances from the Infrared Atmospheric Sounding Interferometer (IASI) - Level 1C dataset to different range of IASI spectral wavenumbers. The study included three distinct analyses, each dividing the training and test sets in different ways using the Soil Type index from the ERA5 database, part of the ECMWF reanalysis. The analyses varied in the scope of their divisions: the first considered satellite observations across the entire Earth's surface, the second divided the data into five climatic zones, and the third further subdivided the climatic zone data into three additional longitudinal zones. Results showed a high recognition accuracy using only radiance data, without the need for physical models or prior domain-specific knowledge. Despite the ongoing nature of the study, the preliminary findings highlight the effectiveness of SVM in performing cloud detection across diverse climatic regions, underscoring the method's scalability and adaptability. Looking ahead, integrating advanced AI techniques, such as deep learning models, could further enhance detection accuracy and generalizability. Additionally, the combination of multimodal data and transfer learning may offer new opportunities to capture complex atmospheric patterns, providing valuable advancements for both operational meteorology and climate research.