Improving Atmospheric Waves Experiment (AWE) Airglow Measurements Using Machine Learning Identification Methods

The Advanced Mesospheric Temperature Mapper (AMTM) is an infrared imaging system with a 20-year legacy of observing atmospheric gravity waves (AGWs) near the mesopause. Since November 2024, an AMTM has been operating aboard the International Space Station (ISS) as part of NASA’s Atmospheric Waves Experiment (AWE), capturing wide-field (~90°, ~600 km at mesopause altitude) and narrow spectral band measurements of OH (3,1) airglow emissions near 1.5 μm. Using the ratio method, it derives rotational temperatures from intensity measurements of P1(2), P1(4), and a background emission line, with the Q1(1) measurements of the (2,0) band providing improved accuracy and cross-calibration.

A major challenge in processing AMTM data is contamination from clouds, which affects more than 20% of the images, particularly in the Q1(1) channel. These clouds interfere with the retrieval of accurate mesospheric temperatures, which are essential for AGW characterization. To address this, we developed a machine learning approach that automatically detects clouds in AMTM images, both identifying whether clouds are present and locating the affected areas. This integrated method improves precision in filtering out contaminated regions while preserving usable data, leading to more reliable temperature retrievals.

We have extended this machine learning framework to identify additional unwanted features in AWE imagery. These include solar panels occasionally blocking part of the field of view, which can cause a blurry streak in the averaged images. Ground-based light sources can have a similar effect as clouds, though they are smaller, appearing as tiny bright dots, and tend to have fixed geolocations with some seasonal variation. Automatically detecting and excluding these non-atmospheric features is crucial for maintaining the quality of derived temperatures and improving the consistency of wave analyses.

By applying machine learning across multiple levels of the AWE data pipeline, we aim to improve data quality, reduce manual intervention, and enhance our ability to observe and understand the global behavior of atmospheric gravity waves from space.