Machine learning-enhanced high-resolution exposure assessment of ultrafine particles

Area of Science:

• Environmental Health
• Air Quality Science
• Computational Toxicology

Background:

• Ultrafine particles (UFPs) <100 nm present health risks not captured by mass-based metrics.
• Particle number concentration (PNC) is key for UFP exposure assessment, yet large-scale data are limited.
• Existing WHO exposure guidelines may not be interchangeable.

Purpose of the Study:

• To develop a national-scale UFP exposure assessment framework.
• To evaluate UFP exposure levels and spatial heterogeneity across Switzerland.
• To inform future UFP air quality standards.

Main Methods:

• Developed a stacking machine learning model integrating data-driven and physical-chemical approaches.
• Utilized long-term, standardized PNC measurements for a national assessment.
• Achieved 1 km spatial and 1-hour temporal resolutions.

Main Results:

• Approximately 20% of the Swiss population faces high UFP exposure (>10^4 particles/cm³).
• National average UFP exposure is (9.3 ± 4.7)×10³ particles/cm³, with urban areas higher than rural.
• UFP spatial heterogeneity is significantly greater than PM2.5, especially in rural regions.

Conclusions:

• The study provides the first national-scale UFP exposure assessment using PNC.
• Identified a nonlinear relationship between WHO 1-hour and 24-hour exposure levels.
• Findings highlight the need for refined UFP standards and targeted public health interventions.