Satellite-Based Validation of Contrail Prediction Models for Sustainable Aviation

Abstract

The aviation industry significantly contributes to global warming through the formation of contrails, which trap heat in the atmosphere and exacerbate climate change. To mitigate this effect, sophisticated models have been developed to predict contrail formation and its associated warming effects, but these require empirical validation for accuracy. This project leverages satellite imagery to validate contrail prediction models, enabling effective contrail avoidance strategies for airlines.  U-Net variants, a convolutional neural network architecture, is utilized for image segmentation to identify contrails in satellite imagery. By optimizing the threshold for the softmax layer, contrail detection accuracy, and validating model predictions with real-world data had been enhanced. This enables pilots to minimize contrail formation during flights, aiming to reduce the aviation industry's environmental impact. The research offers a scalable and cost-effective solution for enhancing aviation sustainability and aligns with global efforts to combat climate change.