Predictive Modeling of Aircraft Arrival Times in the Terminal Maneuvering Area Through Data-Driven Techniques
Author(s)
Choi, Hyungu
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Recently, the consistent increase in domestic air travel within the United States has exerted significant pressure on the National Airspace System (NAS). The arrival phase is especially impacted, as operations in the Terminal Maneuvering Area (TMA) face limited airspace, high traffic, and changing weather conditions. Although aircraft systems and navigation have improved, delays still pose a major challenge. Approximately 20 percent of commercial flights are delayed annually, highlighting the continued difficulty of handling busy terminal areas. A large part of these delays happens within the TMA, where operational coordination is heavily influenced by traffic congestion and environmental
conditions.
This dissertation aims to improve the accuracy of arrival time predictions in terminal airspace, where operations are often limited by congestion, restricted airspace, and unpredictable weather. To achieve this goal, the research is divided into three interconnected studies. Each study employs a distinct methodological approach to support the shared goal of enhancing ETA prediction in terminal airspace, with a focus on traffic patterns, trajectory behavior, and key factors such as weather and congestion. Specifically, the three studies focus on (1) Spatiotemporal Modeling for Airport Traffic Forecasting, (2) Bi-Level Clustering of Arrival Trajectories, and (3) ETA Prediction with Traffic and Weather Features.
Our first study examined short-term forecasting of airport traffic volume at both the individual airport and network levels. A Long Short-Term Memory (LSTM) model was trained to forecast traffic at individual airports using two years of 30-minute interval data. The model was trained with various feature combinations, including traffic volume, dely indicators, weather conditions, and aircraft composition. Chicago O’Hare International Airport (ORD) was used as a case study to assess the model's performance after hyperparameter tuning.
To extend the forecasting to the airport network, a Graph Convolutional Network (GCN) was integrated into the LSTM model. The LSTM-GCN model effectively captured spatial dependencies among airports. Adjacency matrices were created to represent the connectivity between airports. Seven types of adjacency were analyzed to illustrate spatial relationships among airports, encompassing both static and dynamic forms. Static adjacency matrices were constructed using geographic distance, total flight volume, and airport size. Dynamic matrices used monthly flight volumes to reflect temporal changes in airport connectivity. The results showed that how spatial features were defined impacted the prediction accuracy.
The second study focused on identifying flow patterns of arrival trajectories within the TMA. Arrival paths differed depending on traffic levels, weather, runway use, and aircraft type. A bi-level clustering framework was created to categorize these trajectories by operational conditions and shape. In the upper-level clustering, arrival trajectories were grouped based on operational factors present at the time of TMA entry. These factors included traffic volume, weather conditions, aircraft composition, and delay statistics, which collectively represented the broader arrival environment. Using K-means clustering for this categorization, trajectories were classified within shared flow environments, effectively capturing how traffic and environmental conditions influenced the structure of approach paths.
In the lower-level clustering, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm was used within each cluster to identify different trajectory onfigurations. The Dynamic Time Warping (DTW) technique was applied to measure the similarity of trajectories. To assess how data preprocessing affected clustering results, three resampling methods were compared: distance-based, time-based, and key point-based. This analysis helped examine how different representations of trajectory shapes influenced cluster formation.
The third study explored how adding traffic and weather features can improve arrival time prediction accuracy within the TMA. Two main types of features were examined. The first included traffic conditions and aircraft characteristics when entering the TMA. The second consisted of weather data collected from different parts of the terminal airspace. Weather information was sourced from multiple locations, such as the airport surface, TMA entry points, airspace grid areas above the airport, and sectors defined based on historical trajectory patterns. These weather features describe conditions that influence aircraft movement within the TMA. Linear regression, Lasso regression, and Random Forest were used, each with different combinations of these features. The results indicated that using traffic and weather data relevant to the aircraft’s TMA entry time improved ETA
prediction.
Together, the three studies help build a practical framework for predicting arrival times in complex terminal airspace. Each study addresses a different part of the problem: forecasting short-term traffic, identifying arrival flow patterns, and improving ETA prediction with weather and traffic features. Together, they provide essential insights for the development of tools that facilitate real-time traffic management as well as long-term planning.
Sponsor
Date
2025-12
Extent
Resource Type
Text
Resource Subtype
Dissertation (PhD)