Localization and Classification of Gravity Wave Events From VIIRS Day/Night Band Satellite Imagery Using Machine Learning Techniques

Abstract

Gravity waves in the upper mesosphere produce diverse spatial patterns in nighttime airglow, but their global morphology has been difficult to characterize because manual identification is impractical for the massive satellite image archive. We developed a machine learning framework to detect and classify gravity wave events in imagery from the Visible Infrared Imaging Radiometer Suite Day/Night Band (VIIRS DNB) on three satellites. A machine learning based object detector called YOLOv8 (You Only Look Once, version 8) was trained to identify four airglow morphological classes: concentric gravity waves, frontal waves, ripples, and other gravity wave events. On an independent test set, the model achieved a mean average precision of 0.832 at an intersection over union threshold of 0.3. At the confidence threshold adopted for catalog generation, the mean precision and recall were 0.828 and 0.778, respectively. The trained detector was applied to 12 years of moon-free VIIRS DNB observations from Suomi National Polar-orbiting Partnership, NOAA-20, and NOAA-21, resulting in an event catalog containing more than 100,000 detected wave event instances from more than 65,000 images with positive detections, which was archived in a public repository. The catalog captures physically plausible large-scale gravity wave occurrence patterns while also showing that detectability is reduced under elevated sensor noise and strong artificial light contamination. These results demonstrate that machine learning based object detection can systematically extract diverse mesospheric gravity wave events from the long-term VIIRS DNB nightglow record.