Quantifying Aviation-Related Contributions to Ambient Ultrafine Particle Number Concentrations Using Interpretable...

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Machine Learning Applications

Background:

• Ultrafine particles (UFP) from aircraft are a significant air quality concern near airports.
• Quantifying aircraft UFP contributions is difficult due to traffic and meteorological factors.
• Traditional methods struggle to capture complex source interactions.

Purpose of the Study:

• To develop and validate a machine learning model for hourly source attribution of particle number concentrations (PNC).
• To interpret model predictions using SHapley Additive exPlanations (SHAP) for detailed source insights.
• To assess the impact of aircraft activity on near-airport air quality.

Main Methods:

• Applied an ensemble machine learning model to multi-year PNC data near Boston Logan International Airport.
• Incorporated meteorological data, road traffic, and runway-specific aircraft activity.
• Utilized SHAP for model interpretation, identifying feature contributions and interactions.

Main Results:

• The ML model achieved strong performance (R² = 0.66), outperforming typical hourly PNC models.
• SHAP analysis revealed aircraft arrivals on perpendicular runways had the most significant impact.
• Crosswinds were shown to increase ground-level air quality impacts from aircraft not flying overhead.
• Intermediate planetary boundary layer heights correlated with elevated PNC from aircraft.

Conclusions:

• The ML-SHAP framework provides a novel and transferable method for UFP source attribution.
• Aircraft operations, including those not directly overhead, substantially influence near-airport air quality.
• This approach enhances characterization of aviation-related UFP exposure in communities.