Weather is a major influence on the performance of the aviation system in the United States. This paper describes models for predicting weather-related aircraft delays and cancellations at the national, regional and airport levels. Federal Aviation Administration (FAA) uses the delay estimates for system status briefings, long-term post-season reviews and future system-wide analyses. The models estimate delay based on the number of aircraft affected by the expected weather in the en-route environment as well as the terminal areas. The estimation and prediction models are developed using both regression methods and neural networks, using two different operational databases maintained by the FAA. The paper compares the performance of traditional linear regression models with several neural network models in the estimation of key airspace metrics such as total aggregate delay, arrival delay, and airborne delay as well as flight cancellations. The performance metrics are predicted at the national, regional and airport levels. The results are based on using the traffic, weather and delay data for the period 2005-2008. Some of the conclusions based on the results of the study are: (a) the metric based on the number of aircraft expected to be impacted by weather and the extent of the impact is a good proxy for delay of various types at all levels, (b) different delay models are preferable for different seasons and the delay estimation accuracy is higher in the convective weather season (April-September) than the non-convective weather season (October-March), (c) the delay estimation accuracy at all levels and for different metrics is about the same, (d) models resulting from the use of either of the FAA databases are complementary and provide similar level of accuracy and (e) the neural network delay models perform slightly better in the sense that they have a higher correlation between model output and airspace metrics than linear regression methods.
Theme: ATM Performance Measurement and Management
/ Other authors: