Global contrail coverage from machine learning-based detection of line-shaped features in Terra-MODIS imagery
Öhlert, Maurice; Seelig, Torsten; Tesche, Matthias
Meteorologische Zeitschrift (2026)
publication en ligne: Sep 7, 2026
manuscrit accepté: May 5, 2026
révision final du manuscrit reçu: May 4, 2026
révision du manuscrit demandée: Apr 7, 2026
manuscrit reçu: Mar 9, 2026
DOI: 10.1127/metz/1303
Open Access (article peut être télechargé gratuitement)
Abstract
Contrails produced by aircraft induce a local short-term warming effect that adds to the long-term warming caused by the accumulated CO2 emissions of aviation. However, the exact magnitude of the climate impact of contrails still remains highly uncertain. Current attempts to reduce contrail formation, such as flight rerouting, prove difficult because the evaluation of the success of such strategies remains challenging. A reliable method for contrail detection in satellite images can help to assess the magnitude of their climate impact and to evaluate the effectiveness of flight rerouting for contrail avoidance. In the present paper, a machine learning approach is used for contrail detection in global MODIS satellite imagery. Previous studies relying on machine learning for contrail detection used geostationary satellite data, providing an overview of contrail coverage of specific regions. The model is tested with different masks for contrail width and with different wavelength channels as input. The best performing model is applied on satellite data from four entire months in order to gain a global view on contrail coverage. A comparison of global coverage obtained from a contrail detection algorithm generally shows good agreement between the algorithm-based and the machine learning approach. Considering these results, this contrail detection model could be a foundation for future analyses of global contrail coverage using MODIS data.