Presented by: Garima Raheja, Columbia University

Description: Metropolises in sub-Saharan Africa experience high levels of ambient air pollution, yet remain scarcely measured by reference-grade monitors. Combining insights from three years of data collected by newly established and field-calibrated Purple Air low-cost sensor networks, we present the first measurements of PM2.5concentrations in Lomé, Togo. For Accra, Ghana, we utilize an 18-node network of low-cost Clarity low-cost sensors to analyze regional trends, and assess the reductions in PM2.5caused by intentional interventions as well as the COVID-19 pandemic. For both low-cost sensor networks, we apply simple methods such as multiple linear regression, as well as novel more complex methods such as Gaussian mixture regression to develop correction models. We find that the annual average PM2.5in Lomé is 21.1 µg m-3and is heavily influenced by the Harmattan. The network-wide annual mean PM2.5in Accra is 26.3 µg m-3, 5.3 times higher than WHO daily guidelines and 19% higher than neighboring Lomé. Both the PurpleAir network in Lomé and the Clarity network in Accra are validated using results of co-located sensors: 18 Clarity Nodes, 2 PurpleAir, 2 Modulair, 1 Teledyne T640, and 1 Met-One Beta Attenuation Monitor at FEM instrumentation sites in Accra, Ghana. We find that Clarity devices have a mean absolute error (MAE) and R2of 3.36 µg m-3and 0.76 when compared with Teledyne (FEM) PM2.5; Purple Air devices have a MAE and R2of 2.6 µg m-3and 0.88, and Modulair devices have a MAE and R2of 1.66 µg m-3and 0.89. This sensor intercomparison study is contributing to the development of a universal low-cost sensor data correction algorithm. Using these results, we show that low-cost sensors, when combined with data science-based correction techniques, have the potential to close the air pollution data gap in resource-limited areas such as West Africa.

Click here for presentation video!