Machine-learning detection and variability of mesospheric frontal waves observed by VIIRS day/night band

Abstract Frontal waves, characterized by sharp boundaries of airglow jump accompanied by following undulations, were detected using machine learning techniques, and their variability was examined. Frontal waves are thought to be manifestations of ducted waves called mesospheric bores or “wall” waves (large-amplitude gravity waves). The YOLOv3 machine learning model, short for “You Only Look Once version 3,” was trained to detect frontal wave events in Day/Night Band (DNB) data from the Visible/Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (Suomi NPP) satellite. The YOLOv3 detector was trained with DNB images, including manually labeled objects of 756 unique frontal waves. The model achieved 83.19% of average precision (AP) for frontal wave event detection during the testing phase. Utilizing the trained model, 1,150 frontal wave events were identified out of all available 515,187 moonless images from Suomi NPP VIIRS/DNB from January 2012 to June 2023. Over the past eleven years, the monthly occurrence of frontal wave events has gradually decreased from approximately 15 in 2012 to around 5 in 2022. Frontal waves exhibit a high occurrence peak at equatorial latitudes and weaker occurrence peaks at winter mid-latitudes. In these regions, the migrating diurnal and semidiurnal tides exhibit large temperature amplitudes, which could create a favorable environment for ducted waves or mesospheric bores, such as a temperature inversion layer. Frontal waves detected in this study show higher occurrences in regions where conditions favor the formation of ducted waves or mesospheric bores. Graphical Abstract