To facilitate the increasing amount of air traffic, current and future decision support tools for air traffic management require efficient and accurate trajectory predictors. With uncertainty inherent to almost all inputs of a trajectory predictor, accurate predictions are not a simple task. In this study, Monte Carlo simulations of a trajectory predictor are performed to estimate the prediction uncertainty up to 20 minutes look-ahead time and to assess the correlation between inputs and prediction errors. Selected inputs are aircraft bank angle, constant calibrated airspeed (CAS) and Mach number speed settings, vertical speed, temporary level-offs, air temperature, lapse rate, wind, and air traffic control intent. These inputs are provided in the form of their distribution functions obtained from observed data. Simulations are performed for heavy, medium and light wake turbulence category (WTC) aircraft. Results indicate that at 20 minutes look-ahead time, when outliers are not considered, along-track errors can reach up to 50 nautical miles for heavy and medium WTC and 100 nautical miles for light WTC aircraft. Cross-track errors are mostly within 3 nautical miles for heavy and medium WTC and 8 nautical miles for light WTC. Altitude errors can reach up to 25,000 feet. Wind conditions, vertical speed and CAS/Mach number speed settings are determined to be the most influential inputs.

Trajectory predictor performance is dependent on input data quality, which is usually far from perfect. Whereas previous research has mostly focused on mathematical models used for the trajectory prediction and errors present in the prediction process, the comparison of common assumptions made about input values with observed (real) data has not received much attention. In this paper, probability distribution functions are obtained per aircraft wake turbulence category for a set of inputs to a trajectory predictor based on observed data. Considered inputs include aircraft bank angle, indicated airspeed and Mach number speed profiles, vertical speed, temporary level-offs in climb and descent, air temperature, wind speed and wind direction, and air traffic control instructions. Surveillance data, weather forecasts, and air traffic controllers' inputs are used as the data source. Obtained distributions are compared with common simplified assumptions and their characteristics are addressed. The results of this study could be used to improve trajectory prediction.