Prediction of Arrival Runway Occupancy Time and Exit Taxiway Using ADS-B Trajectories

Abstract

The expected future growth in aircraft movements will require airports to increase runway capacity, which is often constrained, among other factors, by aircraft arrival runway occupancy time (AROT) and by the rapid exit taxiway (RET) selected by pilots. Existing prediction approaches for AROT and RET selection rely mostly on proprietary Advanced Surface Movement Guidance and Control System (A-SMGCS) or radar data and often ignore temporal context in trajectory patterns, leaving gaps for operationally relevant applications. In this study, we used Automatic Dependent Surveillance–Broadcast (ADS-B) trajectory data sourced from the OpenSky Network for flights arriving at Zurich Airport in the years 2024–2025 to train two machine learning models: a LightGBM model for RET prediction and a neural network combining time-invariant features with time-variant ADS-B trajectory snippets for AROT prediction. The results of both models were within the range reported in the literature. The RET prediction model achieved a weighted accuracy of 79.4%, while the AROT prediction model yielded a mean absolute error of 3.95 s and a root mean square error of 5.01 s. These findings demonstrate that ADS-B-based models can support air traffic controllers in reducing separation between arriving aircraft, thereby potentially enhancing runway arrival throughput at aerodromes.