Predicting & Quantifying Risk in Airport Capacity Profile Selection for Air Traffic Management*

There is currently no data-driven approach widely used by air traffic managers and controllers to predict the capacity at airports. Instead controllers rely on rules-of-thumb to define the airport acceptance rate (AAR). As the approach is inherently subjective, it can lead to poor definition of Traffic Management Initiatives (TMIs) which rely on accurate airport capacity estimates and can lead to under-delivery or overdelivery of flights to airports. In this paper we propose a methodology for estimating airport capacity and capacity uncertainty based on the environmental conditions within the terminal and airport arrival routes and the projected arrival demand and aircraft spacing. To make these predictions we used a gradient tree boosting model in which the prediction model estimates are time-lagged and conditioned on the previous states. Additionally, estimates from previously predicted states are also used to condition the model based on the history of the predictor variables. The concept was validated against observations from historical data recorded at Newark Liberty Airport (EWR). The proposed method provides accurate prediction of airport capacity and produces a strong quantification of uncertainty in the form of a prediction interval. To explore the implications of applying information about the capacity uncertainty into planning in ground delay programs (GDPs), a stochastic integer programming model for GDP planning was created using the specific quantiles to define a constraint on airport capacity. This model allows the decision maker to make trades based on quantified levels of capacity deviation uncertainty. The results of a sensitivity analysis suggest that the decision maker may benefit from adopting a modest risk premium when planning GDPs. Keywords-Airport Capacity, Capacity Prediction, Ground Delay Programs, Capacity Uncertainty, Stochastic Programming