High-Resolution PM2.5 Exposure Modelling for Nationwide Assessment of Respiratory Mortality Risks in South Africa

Fine particulate matter (PM2.5) is a major environmental health risk, yet long-term, high-resolution exposure assessments remain limited across sub-Saharan Africa. Robust exposure estimates are essential for quantifying health impacts and informing mitigation policies. This study focuses on developing a high-resolution, machine-learning-based PM2.5 dataset for South Africa and demonstrates its application for assessing short-term mortality impacts using country-wide daily respiratory mortality data.

We developed a daily PM2.5 exposure dataset for South Africa using an XGBoost regression framework, trained on ground-based PM2.5 measurements from 2007–2021. Predictors include satellite aerosol optical depth (AOD), meteorological variables (temperature, relative humidity, precipitation, wind speed), soil moisture, road density, population, carbon monoxide (CO), nitrogen dioxide (NO2), emission data from EDGAR, and cyclic temporal predictors (sine and cosine of day-of-year and month). Model performance is strong, with R = 0.95, R² = 0.86, RMSE = 10.9 µg m⁻³, and MAE = 4.15 µg m⁻³, demonstrating high skill in capturing spatial and temporal variability. Using the resulting exposure dataset, we assess population exposure patterns across South Africa and apply a Distributed Lag Non-Linear Model (DLNM) to link district-level daily PM2.5 exposure to all-cause mortality over 1997–2018. Models control for temperature, relative humidity, precipitation, co-pollutants, day of week, and seasonal trends, following established epidemiological approaches. Effect modification by demographic and socio-economic characteristics is explored through stratified analyses.

The high-resolution PM2.5 dataset reveals widespread and persistent exceedances of the South African daily air quality guideline (40 µg m^-3). In the highly populated Johannesburg–Pretoria region, PM2.5 exceeds this threshold on more than 50% of days, while elevated concentrations are also common in coastal cities such as Cape Town, Durban, and East London, particularly during winter. Population-weighted PM2.5 exposure has increased by more than 5% nationally between 2000 and 2023, indicating a growing public health concern. Preliminary epidemiological analyses are consistent with existing evidence from comparable settings, suggesting increased mortality risks associated with short-term PM2.5 exposure, with ongoing work to quantify effect sizes and vulnerable sub-populations.

This study provides the first nationwide, high-resolution PM2.5 exposure dataset for South Africa based on machine learning, offering substantial improvements over existing products. The results highlight widespread guideline exceedances, rising population exposure, and the potential for significant health impacts. The framework enables robust future assessments of air pollution - health relationships and supports evidence-based air quality management and health equity policies in South Africa.