Diurnal, seasonal, and spatial patterns of weather influences on PM2.5 concentrations in Poland: An explainable machine learning approach

Understanding the interplay between weather conditions and fine particulate matter (PM2.5) is crucial for improving air quality management and public health. This study investigates the diurnal, seasonal, and spatial variability in the influence of weather factors on PM2.5 concentrations across Poland during 2015–2024. Hourly PM2.5 data from a nationwide network of monitoring stations were analyzed alongside meteorological parameters derived from the Weather Research and Forecasting (WRF) model. Key weather variables included wind speed and direction, precipitation, temperature, atmospheric pressure, relative humidity, solar radiation, and planetary boundary layer (PBL) height.

The machine learning model, XGBoost, was employed to predict PM2.5 concentrations, and the SHAP (SHapley Additive exPlanations) method, optimized using TreeSHAP, was used to assess variable importance and uncover patterns in the data. The results reveal complex, nonlinear relationships between meteorological factors and PM2.5, with notable variability across time and space. Seasonal and diurnal analyses highlight that weather influences are context-dependent: for instance, precipitation has negative impact on PM2.5 concentrations in winter, but its effect diminishes during summer months. Additionally, the role of PBL height was found to vary diurnally, reflecting their influence on pollutant dispersion. Spatial differences in weather-PM2.5 relationships were also observed, emphasizing the role of local topography, urbanization, and emission sources.

The comprehensive analysis provides a decade-long perspective on how weather factors influenced PM2.5 concentrations in Poland, offering valuable insights for regional air quality modeling and policy interventions aimed at mitigating air pollution under varying climatic conditions.