Graduation Year


Document Type




Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Civil and Environmental Engineering

Major Professor

Yu Zhang, Ph.D.

Committee Member

Fred Mannering, Ph.D.

Committee Member

Xiaopeng Li, Ph.D.

Committee Member

Mingyang Li, Ph.D.

Committee Member

Gabriel Picone, Ph.D.


Convolutional Neural Network, Deep Learning, Ensembled Gridded Weather Forecast, Integrated Weather Information, Multi-Airport System


Automated prediction of runway configuration and airport capacity is critical for the future generation of air traffic management. In the future aviation industry, multi-sources weather forecast information will be available for air traffic decision-making units; how to use these data efficiently is key for overall efficiency of air traffic management. Currently, air traffic management personnel lack tools to assist them to translate weather forecast data into real-time airport capacity. Runway configurations and AARs of airports in a multi-airport system are determined by different air traffic controller personnel. The lack of synchronization may lead to the loss of efficiency of the sharing airspace and airfield capacity. Thus, a decision-making tool that can better use weather forecast information and translate it into real-time runway configuration and airport acceptance rate is an urgent need.

The objectives of this study were to investigate significant factors that have impact on determination of AARs; to provide data-driven neural network methods to predict AAR for a single airport scenario by applying fusing different aviation weather forecasts; and to develop a multi-Convolutional Neural Network model to predict runway configuration and AAR simultaneously for multi-airport systems by utilizing high-precision ensembled gridded weather forecasts.

To achieve the objectives, first, airport operation, flight, weather observation, and weather forecast data were collected from different sources. Second, simultaneous equation models were applied to understand the cross effect of airport departure call rate on AAR and significant factors that have effect on determining AAR. Several toolboxes were developed to automatically process, decode, reshape, and fuse weather forecast data into the desired data format and particular time frame for different look-ahead horizons. Two models were proposed—a multi-layer neural network and a time-dependent model, a stacked LSTM model to predict single airport AAR where integrated aviation weather forecast information is used. Results are compared for these two models. In addition, a feature importance analysis was conducted to identify critical variables that have effects on the prediction of AAR. Finally, for multi-airport systems, a multi-Convolutional Neural Network was developed to predict runway configurations and AARs simultaneously by employing high-precision gridded weather forecast data. Comparisons with outcomes form previous studies demonstrated the advantages of the proposed method. The outcomes from this research can be embedded into automated air traffic management tools and will potentially improve the performance of air transportation.