Enhancing Atmospheric Gravity Wave Studies through Advanced Cloud Classification and Segmentation in Satellite Infrared Imagery: A Contribution to the AWE Mission

The Atmospheric Waves Experiment (AWE) mission, supported by NASA, focuses on understanding how Earth's weather influences space weather by studying atmospheric gravity waves (AGWs). AWE investigates how these waves are distributed globally, how they change with the seasons, and how they move through the upper atmosphere, aiming to improve our knowledge of their effects on space weather. AWE uses the Advanced Mesospheric Temperature Mapper (AMTM), an infrared imager operating on the International Space Station (ISS), to observe AGWs as they move through the OH airglow layer around 87 km altitude.

The work presented here supports the AWE mission by addressing the task of sorting AMTM images into two categories: 'Cloud' and 'No-cloud'. Accurately determining whether clouds are present in the image data is essential for ensuring the AWE temperature measurements are reliable. The AMTM’s four cameras measure specific emission lines to derive the OH (3,1) band rotational temperature, an excellent proxy for the mesospheric temperature. Clouds and background emissions can interfere with these measurements, making it important to identify and correct for cloud presence. To enhance this classification process, we have developed a machine learning model using transfer learning, achieving over 80% accuracy in categorizing satellite images into 'cloud' and 'clear'.

Building on this success, we are now advancing to image segmentation. This involves dividing each image into 45 small rectangles and identifying which rectangles contain clouds. This refined approach will significantly improve the precision of cloud detection, allowing for more accurate corrections of temperature measurements.