Exploring PM2.5 dynamics in Santiago, Chile: how satellite data and machine learning could inform environmental policy

Rapid urbanization and high population densities in Latin American cities pose significant challenges for air quality monitoring, particularly for fine particulate matter (PM2.5), recognized as the foremost health risk for urban populations. This study leverages satellite data and machine learning to estimate the spatial-temporal variability of PM2.5 in the Santiago City Metropolitan Area (Chile), providing critical data to enhance public health policies and address environmental injustices.

Our methodology encompassed four stages: data preprocessing, model architecture selection and construction, model validation, and spatial mapping of PM2.5 concentrations. Initially, we processed PM2.5 data from 12 ground monitoring stations (period 2015 to 2024), integrating it with meteorological data (ERA-5), land cover and NDVI products (MODIS), as well as aerosol products from MERRA-2. We employed a variety of modeling techniques, such as Decision Trees, Treebag, Random Forest, and Extreme Gradient Boosting (XGB). The XGB model was ultimately selected for its superior performance metrics (i.e., higher R2 and lower RMSE).

The XGB model captured a significant range of PM2.5 concentrations across Santiago for the test period (2023-2024), with winter months showing the highest levels, peaking at 75 µg/m³ in June 2023. In contrast, the lowest concentrations occurred from February to April and from October to December, with a minimum of 10 µg/m³ in November. The 1 km² resolution maps revealed a pronounced gradient of PM2.5 concentrations from the west (the Coastal Mountain range) to the east (the Andes Mountain range), negatively correlated with elevation. Densely populated communes such as Quinta Normal and Lo Espejo, which have lower socioeconomic standings, recorded the highest average PM2.5 concentrations (67 to 63 µg/m³). In contrast, wealthier and less densely populated areas like Lo Barnechea and Vitacura exhibited lower concentrations (21 to 23 µg/m³). By identifying how socioeconomic disparities intersect with environmental risks, this study provides a solid foundation for policymakers to formulate interventions that not only improve air quality but also promote social equity.