Speaker
Description
Cloud detection is essential for operational satellite work, particularly in data analysis and atmospheric reconstruction. This study focuses on identifying clouds using machine learning techniques with only satellite radiance observations, specifically utilizing Support Vector Machines (SVM) on spectral radiances from the Infrared Atmospheric Sounding Interferometer (IASI) - Level 1C dataset, covering the wavenumber range from 645 cm−1 to 1600 cm−1.
The study conducted three different analyses, each dividing the training and test sets in different ways. All analyses used the Soil Type index from the ERA5 database, the fifth generation of ECMWF reanalysis, for division. However, they differed in the global divisions used: the first considered satellite observations of the entire Earth’s surface without further subdivisions, the second divided it into five climatic zones, and the third additionally divided it into three longitudinal zones.
The findings of this study are particularly noteworthy. They demonstrate a remarkably high cloud detection accuracy using only radiance information, without the need for incorporating physical models or prior domain-specific knowledge. This result underscores the potential of machine learning algorithms, particularly SVM, in simplifying and improving cloud detection processes in satellite meteorology. Although the research is still ongoing, the preliminary results are promising and represent a significant step forward in using machine learning for atmospheric studies. The current presentation reflects the progress made thus far and lays the groundwork for further refining and applying these techniques.
