Particulate matter variability; a space-time study of pm2.5 pollution in an urban environment

By Marnic Baars

Abstract
Poor air quality is a global health concern, with PM2.5 particles (particulate matter with size of <2.5 µm) linked to cancer and respiratory diseases. In Lent, the Netherlands, a citizen-deployed monitoring network collected PM2.5 data that remained largely unanalysed. This study aimed to (1) analyse spatio-temporal patterns, (2) assess the effect of environmental factors on PM2.5 concentrations, and (3) develop a Random Forest model for an average hourly PM2.5 prediction. The results showed a seasonal trend, with higher concentrations in winter and lower levels in summer, no weekly pattern and an unanticipated diurnal pattern with elevated levels during the night. The spatio-temporal analysis revealed minimal variation across the network. Weak-to-moderate correlations were observed between PM2.5 and meteorological variables, including air pressure (ρ = 0.23) and temperature (ρ = −0.18), while strong correlations were found with pollutants such as black carbon (ρ = 0.60) and NO2 (ρ = 0.47). PM2.5 and ozone showed a moderate correlation (ρ = −0.38) due to opposing weather dependencies. Feature engineering identified key predictors for prediction: 1-hour lag, 1-hour rolling average, black carbon, and ozone. The Random Forest model achieved high accuracy on a 10-fold chronological cross-validation (R² = 0.95, RMSE = 1.57 µg/m³). The findings highlight spatio-temporal patterns and give valuable insights in the state of the air quality in the Lent neighbourhood.