Spatio-temporal weather and airspace graph network (SWAG-Net) for trajectory-based estimated time of arrival prediction

Accurate prediction of flight estimated time of arrival (ETA) is crucial for efficient air traffic management (ATM) and operational planning. Several factors that affect arrival time, such as convective weather, traffic density and air traffic controllers’ instructions, pose significant challenges to ETA prediction due to their impact on airspace capacity and flight plans. This paper proposes a novel Spatio-temporal Weather and Airspace Graph Network (SWAG-Net) to predict flight ETA under various operational and meteorological scenarios. The proposed model integrates diverse data sources, including real-time multi-layer constant altitude plan position indicator (MCAPPI) weather radar, Automatic Dependent Surveillance-Broadcast (ADS-B) and the geographical structure of the airspace’s arrival routes. By leveraging the dynamic adaptive spatio-temporal graph neural network, the model can capture both spatial and temporal dependencies, providing robust predictions in dynamic airspace and weather conditions. Experimental results demonstrate substantial enhancements in flight time estimation accuracy across diverse operational scenarios. It indicates a marked reduction in mean absolute error (MAE) to 1.28 min for a 30 min prediction horizon, representing a 41.6% improvement over traditional baseline models (MAE: 2.19 min). Meanwhile, the model indicates robust performance under challenging conditions, with pronounced accuracy gains observed in adverse weather and variable traffic density environments, and offers valuable insights into the impact of meteorological factors on flight operations. Further analysis shows that regional precipitation patterns within airspace greatly affect flight durations and the performance of predictive models. These findings suggest that the SWAG-Net framework has the potential to support weather-aware strategies for next-generation aviation systems.