Search for a command to run...
The development of effective contrail warming mitigation strategies requires the ability to accurately model contrail formation. This is a challenging problem for a number of reasons, including high uncertainty in the humidity data which is a key component of such models. Observational datasets can be used to constrain and improve contrail formation models. Here we present an analysis of existing contrail models on the task of predicting whether a contrail will be observed in a collection of observational datasets (collectively termed 'ContrailBench'). The observational datasets include one based on Ref [1] using automated contrail detections from the GOES-16 satellite and an automated contrail attribution algorithm, another based on Ref [2] which detects contrails using GOES-16 and uses knowledge of the altitude from LIDAR measurements to attribute them to flights, and a third dataset based on Global Meteor Network ground-based camera imagery [3] with automated contrail detection and high-confidence attribution to the flights that formed them. Different downstream applications require different properties from contrail models, so we evaluate the contrail models based on their performance in both ‘high-recall’ mode (which prioritizes identifying all the flights which make contrails) as well as in ‘high-precision’ mode (which prioritizes minimizing the number of flights incorrectly predicted as forming a contrail). We find that models using raw ERA5 weather reanalysis data perform poorly on all metrics, but the use of machine learning to correct the weather data can lead to improvement. [1] A. Sarna et al, “Benchmarking and improving algorithms for attributing satellite-observed contrails to flights”, https://doi.org/10.5194/egusphere-2024-3664[2] V. Meijer, thesis, “Satellite-based Analysis and Forecast Evaluation of Aviation Contrails”[3] D. Vida et al, “The Global Meteor Network – Methodology and first results” https://doi.org/10.1093/mnras/stab2008