A novel Contrail Detection Algorithm for the Meteosat Second Generation satellite

Contrail cirrus is estimated to be the largest contributor to global effective radiative forcing from aviation, surpassing even aviation CO₂ emissions in their impact. One promising mitigation strategy is contrail avoidance by rerouting flights to avoid regions where warming contrails can persist. These regions are forecast with numerical contrail models fed with weather prediction model output. Satellite imagery presents a good opportunity to evaluate both the models and the success of the mitigation strategy.

An automatic contrail detection algorithm was implemented by Mannstein et al. (1999) for a polar orbiting satellite with high spatial resolution (≈ 1 km and used two thermal channels in the atmospheric window). Since then, it has been adapted to the Meteosat Second Generation (MSG) satellite because geostationary satellites have the big advantage of high temporal coverage of a large area. However, its lower spatial resolution (≥ 3 km) is a challenge for contrail detection. In recent years AI algorithms have been presented for the American geostationary GOES-R/S satellites (spatial resolution ≥ 2 km). In this study, a new improved contrail detection algorithm for MSG is proposed based on image processing.

To establish a new detection algorithm a labeled dataset was compiled. This labeled dataset contains 140 MSG images with data from the years 2013-2018 and 2023-2024. This data volume is very suitable to develop and evaluate an algorithm with a classical approach, would however not be sufficient to train an AI based algorithm. The data covers the whole MSG disk with a wide distribution of satellite viewing angles, cloud cover, number of contrails in the image and other properties. Each image was labeled by three individuals, and a common contrail mask was established as the consensus of the labelers. Based on this dataset, a new detection algorithm was developed. Making use of the spectral information of MSG, an image is created as input for the algorithm where contrail visibility is enhanced. The algorithm takes advantage of several new techniques compared to Mannstein et al. (1999). In addition to the contrail mask, uncertainty information is provided.

This new algorithm for contrail detection in MSG images demonstrates superior performance compared to the previous algorithm for MSG based on the Mannstein approach with a probability of detection higher than 70%. The contrail detection algorithm can now be employed for generating datasets to evaluate contrail models as well as to assess the success of contrail mitigation strategies such as flight path alteration. Thanks to this classical image processing approach, in the future the algorithm can be adapted to other satellites like Meteosat Third Generation.