A Graph Neural Network Based Model to Predict Flight Delay Due to Weather-Induced TMI Activation

Weather induces delays in air transportation at different phases of flight, that is, departure, arrival, and en-route, causing flight delays, cancellations, and diversions. In addition, adverse weather conditions such as thunderstorms, strong winds, fog, low visibility, snow/ice, etc. significantly impact airport and airspace operations. Traffic Management Initiatives (TMIs) are critical for managing demand and capacity in the National Airspace System (NAS) to ensure safe and efficient operations. A preliminary analysis of Traffic Flow Management System (TFMS) data shows that weather impacts can result in TMIs such as ground delay programs (GDP), ground stops (GS), flight rerouting, airspace flow programs (AFP), and deicing. Improved and accurate classification of various weather conditions that could potentially disrupt airport and airspace operations along with the prediction of subsequent flight delays play a critical role in achieving greater aviation safety and efficiency. In this paper, we propose a graph neural network (GNN)-based multi-task model to classify weather observations that could potentially lead to TMIs and predict flight delays under such conditions.