Detection of Contrail Formation Using Machine Vision

As air travel demand grows, contrail-related environmental impacts are expected to rise. This leads to an ever-increasing need for monitoring of contrail formation and persistence, especially as their radiative forcing is a function of the latter. However, predicting where contrails will form and persist in real-time remains a challenging problem. This work investigates the use of aircraft-mounted cameras and machine vision for real-time contrail detection and prediction. With the help of atmospheric measurements and visual footage from NASA’s Airborne Sciences program, a framework that leverages machine-learning algorithms was developed to autonomously classify contrail formation from visual input. The Schmidt-Appleman criterion identified conditions favorable for contrails. Visual data was segmented, filtered, and reconstructed through an autoencoder. A secondary neural network, trained on the Schmidt-Appleman results, was used for the classification of the visual data. Preliminary results show the system can identify contrail absence, short-lived presence, and persistence with approximately 93%, 81%, and 77% accuracy, respectively, indicating strong potential for real-time application. With further development, this approach could enable fleet-wide monitoring, helping airlines optimize flight paths to avoid harmful contrails. Unlike sensor-heavy alternatives, this solution relies only on onboard cameras and software, making it a cost-effective and scalable approach for reducing aviation’s climate impact.