Key points are not available for this paper at this time.
Assessing the health burden of air pollution in African cities is challenging due to limited monitoring and significant data gaps. Cities like Kampala, Uganda, lack dense reference networks, making long-term health impact assessments difficult. In this study, we use fine particulate matter (PM2.5) measurements from a reference grade monitor, a network of low-cost air quality sensor nodes (LCAQSN) and satellite-derived PM2.5 to assess air quality trends in Kampala (2019–2023). To address the missing data problem in the PM2.5 time-series, we applied a two-stage machine learning (ML) imputation technique. The ML results were assessed using 5-fold cross-validation, having good predictions with an R2 = 0.78. This approach significantly improved data coverage, uncovering an additional 7–928 days of PM2.5 exceeding the 24 h World Health Organisation Interim Target 1 (WHO IT1) across traffic, sub-urban, residential and urban background sites. The comparison of the complete reference and LCAQSN PM2.5 time-series to satellite-derived PM2.5 showed good agreement with relative differences within 17%. Using an updated health-risk assessment model, we calculated an annual mortality burden between 660 and 1010 adult premature deaths. Reference-grade data had the highest PM2.5 mortality burden, which was 6% greater than satellite-derived data and 20% greater than LCAQSN data. Our results show that a two-stage ML imputation method can effectively fill long-term data gaps in ground-level measurements and reveal actual PM2.5 levels and related health impacts. In the absence of ground monitoring, satellite data provide a useful alternative for capturing city-wide PM2.5 trends and related health impacts.
Kassandros et al. (Tue,) studied this question.