Learning Terminal Airspace Traffic Models from Flight Tracks and Procedures

Constructing robust aircraft trajectory models is important when developing modern air traffic management systems. In real operations, aircraft are often guided to follow the standard flight procedures. However, there is some variability in their actual flown paths due to different factors. Therefore, a trajectory model should be able to capture the variability as well as the general tendency of the aircraft behavior in relation to the corresponding procedure. This paper proposes a generative probabilistic model that can learn this directly from radar surveillance data and procedure data. We fit a Gaussian mixture model for each flight stage to learn how much the aircraft trajectories deviate from standard procedures. The performance of the proposed model is validated using Jensen-Shannon divergence between distributions of original and generated trajectory data.