Kriging-Based Metamodeling for Accurate and Explainable Real-Time Taxi-Time Prediction in Airport Surface Networks

This paper proposes a Kriging-based metamodel for real-time taxiing time prediction on complex airport surface networks. By combining the strengths of analytical and machine learning methods, the model offers superior adaptability and interpretability. It produces not only accurate point predictions but also confidence intervals to quantify predictive uncertainty, with each prediction traceable to specific training samples. The model relies on simple and readily available inputs, including flight schedules and observed runway-queue information, which facilitates adaptation to other airports after local calibration. Trained on historical operational records and evaluated through high-fidelity simulations of Beijing Capital International Airport and benchmarked against current operational methods and artificial intelligence-based alternatives, the results demonstrate significant improvements over existing taxi-time estimation techniques. These advantages highlight the model’s potential as a robust solution for future automated air traffic management systems.