Adaptive metamodeling approach for real-time taxiing time estimation on airport surface

Surrogate models enable efficient approximations and real-time decision-making in complex scenarios by utilizing analytical and tractable mathematical structures. While these methods have proven effective in real-time optimization, traffic control, and flow prediction for road traffic, they have not yet been applied to real-time taxiing time estimation for airport surface operations. This paper proposes a metamodeling framework, named Kriging, designed to predict taxiing time in real-time, providing flexibility to balance model granularity, complexity, and accuracy. To address the computational challenges, we introduce two algorithms: (1) an adaptive sampling and infill strategy, and (2) an adaptive selection of primary hyperparameters algorithm within the Kriging framework. These methods significantly improve the model’s computational speed without compromising its accuracy. The proposed Adaptive Kriging model is tested using a high-fidelity simulation of Beijing Capital International Airport (PEK) and compared with current operational methods and AI-based alternatives. The results demonstrate that the proposed approach significantly outperforms existing taxiing time estimation methods, achieving an RMSE of 1.59 min, MAE of 0.72 min, MAPE of 4.53 %, and 69.35 %, 91.86 %, and 96.91 % of predictions within ±1, ±3, and ±5 min, respectively. The potential impact of this work extends to accommodating distinct traffic flow characteristics, with the capability for real-time updates, providing more reliable taxiing time predictions and route optimization decisions for future automated Air Traffic Management systems.