Twenty Years of High Spatiotemporal Resolution Estimates of Daily PM_2.5 in West Africa Using Satellite Data, Surface Monitors, and Machine Learning

Estimates of air pollution mortality in sub-Saharan Africa are limited by a lack of observations of fine particulate matter (PM_2.5). Satellite data represents a promising solution with near-complete spatial coverage and high temporal coverage, but representativeness of surface conditions is a critical issue. Here we estimate surface PM_2.5 concentrations over West Africa at a daily, 1 km² spatiotemporal resolution based on satellite-derived and reanalysis inputs trained against surface PM_2.5 observations using several machine learning algorithms. Among machine learning models tested, Extreme Gradient Boosting (XGBoost) demonstrated the highest accuracy, with a 0.91 r², mean absolute error of 9.1 μg m^–3, and a CvMAE of 0.1, indicating about a 10% error across all sites on aggregate. Seasonal and annual PM_2.5 patterns were well captured, revealing severe air quality challenges via near-universal exceedances of World Health Organization air quality guidelines and interim targets. The data set’s long-term perspective (2005–2024) highlights worsening air quality trends in both rural and urban areas. Our findings provide actionable data to support air quality management and policy, public health, and environmental justice initiatives in a critically underserved region of the world.