We present a decentralized Model Predictive Control scheme for hierarchical systems to tackle the collision avoidance problem for autonomous aircraft in an air traffic control setting. Using a low level controller, the aircraft dynamic equations are abstracted to simpler unicycle kinematic equations. The navigation function methodology is then used to generate conflict free trajectories for all aircraft. In order to ensure that the resulting trajectories respect the aerodynamic constraints of the aircraft, a decentralized model predictive controller is added at a higher level, to provide preview to the otherwise myopic navigation functions. The overall hierarchical, distributed control scheme has the same feasibility properties as the corresponding centralized scheme. Its performance is demonstrated by simulations of dense air traffic scenarios.