S. Kröger, S. Lück, J. Goeing

Contrails significantly contribute to aviation-induced global warming. Accurate prediction and simulation models of contrails are essential both to better estimate their climate impact, which remains subject to significant uncertainties, and to predict when and where flights will produce contrails, with the long-term goal of rerouting them to reduce their impact. This study investigates whether automated detection of contrails in geostationary satellite imagery, combined with the attribution to the flights that produced them, can create a database of real-world observations capable of validating contrail prediction models. We used an open-source neural network to detect contrails in geostationary satellite images. To match these contrails with flights, we developed a two-stage matching algorithm, building on and extending previous work in this field. The first stage applies geometric filters to narrow down the number of candidate flights per contrail, while the second stage uses wind data to calculate the predicted contrail track for each previously matched flight trajectory and compares it with the observed contrail locations. Additionally, a scoring method was introduced to quantify match quality, distinguishing well-aligned from poorly aligned associations. The system also enables the retrieval of environmental conditions present at the time of contrail formation for each identified match. Compared to an existing matching algorithm, our approach produced one-third fewer matches per contrail on the same dataset, and the individual results appear more plausible. A statistical evaluation shows that while a single contrail cannot always be attributed to one specific flight, the method consistently narrows the results to a small set of candidate flights per contrail. A case study shows that manually reviewing the results, particularly by combining multiple observations of the same contrail over time, helps to further narrow down the list of potential source flights. This demonstrates that although the current version cannot yet automatically identify a contrail’s source flight, further improvements could make this possible. In particular, automatic contrail tracking combined with dedicated filtering logic and refined contrail detection may enable the creation of a reliable large dataset of contrail–flight associations usable for model validation.

Deutscher Luft- und Raumfahrtkongress 2025, Augsburg

Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V., Bonn, 2025

Conference Paper

englisch

21,0 x 29,7 cm, 12 Seiten

urn:nbn:de:101:1-2512221116069.109139779049

10.25967/650223

contrails, matching contrails with aircraft, contrail detection, aviation climate impact, satellite imagery

Download - Bitte beachten Sie die Nutzungsbedingungen dieses Dokuments: CC BY 4.0  OPEN ACCESS

Kröger, S.; Lück, S.; Goeing, J. (2025): AI- and Data-driven Identification of Contrail Sources. Deutsche Gesellschaft für Luft- und Raumfahrt - Lilienthal-Oberth e.V.. (Text). https://doi.org/10.25967/650223. urn:nbn:de:101:1-2512221116069.109139779049.

22.12.2025