Extraction of spatially confined small-scale waves from high-resolution all-sky airglow images based on machine learning

Abstract. Since June 2019, a scanning airglow camera is operated operationally every night at DLR Oberpfaffenhofen (48.09° N, 11.28° E), Germany. It provides nearly all-sky images (diameter 500 km) of the OH* airglow layer (height ca. 85–87 km) with an average spatial resolution of ca. 150 m and a temporal resolution of ca. 2 min. We analyse about three years (941 nights between October 2020 and September 2023) of OH* airglow all-sky images for spatially confined wave structures with horizontal wavelengths of ca. 20 km and less. Such structures are often referred to as ripples and are considered to be instability structures. However, Li et al. (2017) showed that they could also be secondary waves. While ripples move with the background wind, secondary waves do not. To identify small-scale and spatially confined structures, we adapt and train YOLOv7 (You Only Look Once, version 7), a machine learning approach, to determine their position and extent on the sky as well as their horizontal wavelength. Those wavelengths are compared to two-dimensional FFT (Fast Fourier Transform) results. We analyse the seasonal variations in the propagation direction and horizontal wavelengths of these structures and deduce that instability signatures are observed especially in summer. Finally, we introduce a concept for “operating-on-demand” in order to derive energy dissipation rates from our measurements.