Global high-resolution ultrafine particle number concentrations through data fusion with machine learning.

Area of Science:

• Environmental Science
• Public Health
• Atmospheric Chemistry

Background:

• Atmospheric pollution, particularly particulate matter (PM), causes millions of deaths annually.
• Ultrafine particles (UFPs, diameter < 100 nm) are a growing concern due to their deep lung penetration and systemic health impacts.
• Particle number concentration (PNC) is a key metric for UFP prevalence, as UFPs dominate particle counts.

Purpose of the Study:

• To develop the first global datasets of PNC and UFPs at 1 km resolution over land.
• To utilize machine learning to predict annual PNC levels from 2010-2019.
• To support global exposure assessments and health impact studies by providing reliable UFP data.

Main Methods:

• Combined ground station measurements with machine learning (XGBoost model).
• Integrated diverse global environmental and anthropogenic variables for prediction.
• Employed spatial and temporal cross-validation and a conformal prediction framework for reliable intervals.

Main Results:

• Achieved R² ≥0.9 and mean relative error of ~30% for polluted urban areas.
• Global annual mean PNCs range from thousands to over 40,000 particles/cm³.
• UFPs constitute approximately 91% of total PNCs.

Conclusions:

• The study provides unprecedented global, high-resolution data on PNC and UFPs.
• The developed model accurately predicts PNC levels and quantifies UFP exposure.
• This data is crucial for understanding and mitigating the health risks associated with UFP air pollution.