A Satellite-Based Spatio-Temporal Machine Learning Model to Reconstruct Daily PM_2.5 Concentrations across Great...

Epidemiological studies on the health effects of air pollution usually rely on measurements from fixed ground monitors, which provide limited spatio-temporal coverage. Data from satellites, reanalysis, and chemical transport models offer additional information used to reconstruct pollution concentrations at high spatio-temporal resolutions. This study aims to develop a multi-stage satellite-based machine learning model to estimate daily fine particulate matter (PM_2.5) levels across Great Britain between 2008-2018. This high-resolution model consists of random forest (RF) algorithms applied in four stages. Stage-1 augments monitor-PM_2.5 series using co-located PM₁₀ measures. Stage-2 imputes missing satellite aerosol optical depth observations using atmospheric reanalysis models. Stage-3 integrates the output from previous stages with spatial and spatio-temporal variables to build a prediction model for PM_2.5. Stage-4 applies Stage-3 models to estimate daily PM_2.5 concentrations over a 1 km grid. The RF architecture performed well in all stages, with results from Stage-3 showing an average cross-validated R² of 0.767 and minimal bias. The model performed better over the temporal scale when compared to the spatial component, but both presented good accuracy with an R² of 0.795 and 0.658, respectively. These findings indicate that direct satellite observations must be integrated with other satellite-based products and geospatial variables to derive reliable estimates of air pollution exposure. The high spatio-temporal resolution and the relatively high precision allow these estimates (approximately 950 million points) to be used in epidemiological analyses to assess health risks associated with both short- and long-term exposure to PM_2.5.