Detection of Mesospheric Bores Using Machine Learning and Eleven Years of Mesospheric Bores Variability as Observed by Day-Night-Band

A machine learning model, YOLOv3 was trained to develop a mesospheric bore detector. As a result, mesospheric bores variability was examined by applying the YOLOv3 detector model to 11-year data of the Day/Night Band (DNB) from the Visible/Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-orbiting Partnership (Suomi NPP) satellite. YOLOv3 was trained with DNB images, including manually labeled 696 unique bore images.  The trained model achieved 83.19% of average precision (AP) for mesospheric bore detection. With the developed model, 1,198 mesospheric bore events were found out of all the available moonless images of Suomi NPP VIIRS/DNB from January 2012 to June 2023. The monthly occurrence of bores gradually decreased over the past eleven years from around 15 in 2012 to around 5 in 2022. Mesospheric bores have a high occurrence peak at equatorial latitudes and weak occurrence peaks at winter mid‐latitudes. This result suggests that DE1 and SW2, which have a large amplitude of temperature at equatorial latitudes and winter mid-latitudes, respectively, generate a preferable background for bores, such as a temperature inversion layer.