Global Meteor Network: Large Scale Ground-Based Camera Validation of Contrail Model Predictions using Machine Learning

Contrail cirrus contributes an estimated 1-2% of all anthropogenic radiative forcing, but this estimate carries significant uncertainty (~70%). A proposed mitigation strategy involves redirecting aircraft to avoid contrail-producing regions, which requires accurate predictions of atmospheric states.  To improve these predictions, direct observations of aircraft forming contrails can validate and constrain atmospheric models. Ground-based cameras bridge the spatial resolution gap left by satellite observations, allowing us to observe contrail formation and attribute contrails to specific flights.

We present first results from a large-scale dataset of flight-attributed contrails observed by the Global Meteor Network (GMN) across two continents over several months. The GMN operates 1,600 calibrated ground-based video cameras in 45 countries which have been modified for 24-hour observations to monitor contrails. Contrails were detected and segmented from camera timelapses using machine learning algorithms, automatically associated with flights from Automatic Dependent Surveillance–Broadcast (ADS-B) flight data (then manually validated), and compared to the CoCiP model predictions.

Our analysis highlights the limitations of current prediction models, which early results suggest stem from insufficient vertical resolution to capture vertically thin ISSRs and a limited number of measurements of humidity in the upper troposphere. While errors in model wind data affect our flight associations, the discrepancy between predicted and observed contrail advection offers a new avenue to quantify this wind error and use the derived measurements to improve associations. Finally, we provide statistics of contrail properties observed by the GMN such as altitude, width, and lifetime.