A probabilistic airport capacity model for improved ground delay program planning

Abstract

Weather uncertainty is a major cause of unnecessary delays within the National Airspace System (NAS). In particular, the high uncertainty in terminal area weather forecasts combined with the uncertain correlation between weather and airport capacity makes the task of planning strategic traffic management initiatives such as ground delay programs (GDPs) very difficult. At the twelve airports in the NAS with the most GDPs issued during 2008 and 2009, weather-related GDPs were canceled an average of 95 minutes earlier than the initially scheduled GDP end time and, when GDP revisions were issued, nearly two hours earlier than the revised GDP end time. As part of NextGen, the FAA and NASA are researching methods for reducing the negative impact of weather on the NAS and exploring ways to improve collaborative air traffic management (CATM) processes through increased automation and improved decision support tools (DSTs). This paper introduces the Weather Translation Model for GDP Planning (WTMG), a statistical model for translating weather forecasts into probabilistic arrival capacity predictions over a strategic time horizon of up to twelve hours. The model is self-training and independent of forecast product. With a sufficiently large historical data set, the model is able to build probability distributions for the airport arrival rate (AAR) in future time intervals conditioned on weather forecasts and the current state of the airport. These distributions are sampled to build probabilistic capacity scenarios with the end goal of providing inputs for CATM DSTs for GDP planning. Two versions of WTMG are presented: static WTMG samples independently at time interval; dynamic WTMG draws sample capacity vectors that are dependent across time intervals. Each version of the model is demonstrated using each of two distinct forecast products: the Terminal Area Forecast (TAF) and the Localized Aviation Model Output Statistical Program (LAMP).