Non-linear models for black carbon exposure modelling using air pollution datasets.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Public Health

Background:

• Black carbon (BC) is a significant urban aerosol component from incomplete combustion, primarily road traffic.
• Epidemiological studies link BC exposure to adverse cardiovascular and respiratory health outcomes.
• BC is not currently regulated by the EU Air Quality Directive, leading to data gaps in urban monitoring.

Purpose of the Study:

• To develop a machine learning-based proxy for estimating black carbon concentrations in urban areas.
• To utilize readily available air pollution datasets as input for the BC proxy model.
• To address the lack of official BC monitoring data for improved exposure assessment.

Main Methods:

• Employed machine learning models, specifically Support Vector Regression (SVR) and Random Forest (RF).
• Utilized input data including particle mass and number concentrations, gaseous pollutants, and meteorological variables.
• Validated model performance using experimental data from two urban sites in Barcelona over a two-year period.

Main Results:

• The SVR model demonstrated a strong correlation with measured BC (R² = 0.828, RMSE = 0.48 μg/m³).
• Model performance varied with seasonality and time of day, influenced by new particle formation events.
• Validation at a second site showed decreased performance (R² = 0.633, RMSE = 1.19 μg/m³) due to data limitations.

Conclusions:

• The developed BC proxy model shows flexibility and potential for use with EU-regulatory air quality parameters.
• The model can effectively complement experimental measurements for urban black carbon exposure assessment.
• Optimal application is suggested for environments where traffic is the primary source of ultrafine particles.