Flight Diversions: Understanding Air Traffic Disruptions

A comprehensive data science analysis of U.S. flight diversions (2021-2024)

1 Welcome

This project analyzes flight diversions across the United States from July 2021 through December 2024, using data from the U.S. Bureau of Transportation Statistics (BTS) and Federal Aviation Administration (FAA). By combining temporal clustering analysis with interactive geospatial visualizations, we uncover patterns in when, where, and how flights are diverted from their intended destinations—and reveal significant differences in how airlines manage disruptions.

2 Why This Matters

Flight diversions represent a significant disruption to both airlines and passengers. Understanding diversion patterns helps:

Airlines: Optimize route planning and operational resilience. By identifying which airports, time periods, and routes are prone to diversions, airlines can pre-position aircraft, adjust crew schedules, and reroute flights proactively. American Airlines’ 29% higher diversion rate reveals opportunities for operational improvement through benchmarking against Delta and United.
Airports: Plan capacity for overflow demand. The analysis identifies consistent diversion hub airports (DEN, IAD, LAX) that naturally absorb system disruptions, informing infrastructure and ground service planning.
Policymakers: Identify systemic bottlenecks in air traffic management. Diversion clusters reveal capacity constraints at specific airports and demonstrate how disruptions cascade through the network, informing infrastructure investment and ATC policy decisions.
Researchers: Better understand the vulnerabilities and resilience patterns in the aviation system. Flight diversions expose how the network responds to disruptions and reveal dependencies between airports.

3 Project Overview

3.1 Key Findings

64,815 total diversions documented from July 2021 through December 2024
25,386,632 total flights analyzed, with 0.26% experiencing diversions
3 major diversion clusters identified: San Diego (system-wide), Dallas/Fort Worth (regional), Chicago O’Hare (regional)
American Airlines diverts 29% more frequently than Delta despite similar network size
System-wide events cascade across 60+ destination airports; regional events stay concentrated in geography zones
Top diversion hubs: Denver (1,935), Washington Dulles (1,680), Los Angeles (1,608)
Average arrival delay for diverted flights: 278.8 minutes (4.6 hours)

3.2 Methodology

This project follows a complete data science pipeline:

Data Collection: Aggregating 45 months of BTS OTMC-OTP data (25.4M flights) with FAA airport coordinates
Data Cleaning: Filtering for diversions, validation, removing duplicates, geocoding
Analysis: Temporal clustering (12-hour threshold) to identify disruption events
Visualization: Interactive Plotly maps showing individual and system-wide impacts
Communication: Clear documentation and actionable insights

4 The Three Clusters

4.1 Cluster 1: San Diego (December 2024) - System-Wide Disruption

140 diversions to SAN; 290 system-wide across 60 destinations
Only 48.3% diverted to primary airport (cascaded across network)
Affected 73 origin airports and 60 destination airports
Finding: Coastal hub disruptions have broader ripple effects through the system

4.2 Cluster 2: Dallas/Fort Worth (November 2024) - Regional Disruption

110 diversions to DFW; 186 system-wide with regional concentration
59.1% diverted to primary airport (more localized recovery)
Texas corridor airports (IAH, SAT, AUS) absorbed overflow
Finding: Regional hubs keep diversions contained in geographic area

4.3 Cluster 3: Chicago O’Hare (August 2021) - Regional Disruption

108 diversions to ORD; 187 system-wide with Midwest focus
57.8% diverted to primary airport (regional coordination)
Midwest corridor (IND, STL, MKE) natural overflow network
Finding: Hub location determines scope of cascade effects

5 American Airlines: Why the Higher Diversion Rate?

American Airlines accounts for 27.7% of all diversions despite operating roughly 18-22% of U.S. flights:

Airline	Diversions	% of Total	Status
American (AA)	18,004	27.7%	Higher than market share
United (UA)	13,921	21.4%	Expected
Delta (DL)	10,886	16.8%	Better than expected
Southwest (WN)	10,285	15.8%	In line
Alaska (AS)	3,922	6.0%	In line

Possible explanations: - More aggressive scheduling relative to airport capacity - Less efficient operational recovery procedures - Different network vulnerabilities (route design, hub exposure) - Older fleet with higher mechanical diversion rates

This represents both a competitive disadvantage and an opportunity for operational improvement.

6 Interactive Dashboard

Explore the data yourself with the interactive dashboard:

Running Locally:

cd MUSA5500-finalProject
conda activate geospatial
python app.py

Then visit: http://localhost:5006

Features: - Filter by airline (multi-select) - Adjust date range (July 2021 - December 2024) - Real-time map and statistics updates - Explore top 15 diversion airports - Compare airlines side-by-side

7 Documentation Structure

This site contains complete documentation:

Data Collection: Data sources, cleaning pipeline, quality assessment
Analysis Methodology: Clustering approach, statistical methods, airline comparison
Visualizations: Interactive maps with explanations of each cluster
Results & Conclusions: Key findings, operational implications, recommendations
Dashboard Documentation: How to use the interactive Panel dashboard

8 Quick Stats

Metric	Value
Analysis Period	July 2021 - December 2024 (45 months)
Total Flights	25,386,632
Total Diversions	64,815
Diversion Rate	0.26%
Airlines	10 major carriers
Airports	377 diversion airports, 373 destination airports
Avg Departure Delay	32.2 minutes
Avg Arrival Delay	278.8 minutes
Major Clusters	3 (SAN, DFW, ORD)

9 About This Project

This analysis was conducted as a final project for MUSA 5500: Data Science for Planning and Policy at the University of Pennsylvania, emphasizing:

Multi-source data integration (BTS + FAA databases)
Complex geospatial and temporal clustering analysis
Interactive web-based visualization and exploration
Clear communication of technical findings to multiple audiences
Actionable recommendations for industry stakeholders

9.1 Technologies Used

Data Processing: Pandas (2.3.1), NumPy (2.3.1), GeoPandas (1.0.1)
Analysis: Temporal clustering (custom 12-hour threshold algorithm)
Visualization: Plotly (6.5.0), Panel (1.7.5), Matplotlib
Documentation: Quarto, Jupyter Notebook
Dashboard: Panel (local deployment)
Code: Python 3.13, GitHub

10 Key Insights for Stakeholders

10.1 For Airlines:

Pre-position crews and aircraft at hub diversion airports (DEN, IAD, LAX)
Implement tiered disruption protocols (regional vs. system-wide)
Benchmark operational practices against DL and UA
Monitor routes with historically high diversion risk

10.2 For Airports:

Plan for surge capacity at consistent diversion hubs
Establish regional coordination agreements (Texas, Midwest corridors)
Position ground crews to handle unexpected diversion traffic

10.3 For Policy:

Infrastructure investment should target secondary hubs that absorb diversions
ATC coordination between regional clusters can mitigate cascade effects
Monitor whether system is near saturation during peak periods

11 Get Started

Start here: Read the overview
Understand the data: Check Data Collection
Learn the approach: Review Analysis Methodology
Explore visually: See the Visualizations
Understand findings: Read Results & Conclusions

12 Technical Details

Data Sources: - Bureau of Transportation Statistics (BTS): https://www.transtats.bts.gov/ - Federal Aviation Administration (FAA): https://www.faa.gov/ - OpenFlights Database: https://openflights.org/

Code Repository: - GitHub: MUSA5500-finalProject (private - available by request)

Reproducibility: - All analysis code included in Jupyter notebooks - Data pipeline documented and reproducible - Environment specifications provided (Python 3.13, Pandas 2.3.1, etc.)

13 Questions?

This project demonstrates: - Sound data science methodology applied to real-world aviation data - Complex analysis combining temporal, geospatial, and statistical methods - Clear communication of technical findings to diverse audiences - Effective use of interactive visualization for exploration and discovery - Actionable insights that drive operational and policy decisions

For questions about the methodology, data sources, analysis approach, or findings, see the detailed documentation pages or review the code in the Jupyter notebooks.

Project Completed: December 2024
Analysis Period: July 2021 - December 2024
Author: Jun
Course: MUSA 5500: Data Science for Planning and Policy