Utilizing Reanalysis Datasets to Improve the Performance of Low-Cost Air Sensors in the Global South

Abstract

A Low-cost sensors for particulate matter can provide high spatiotemporal resolution monitoring of air quality, especially in much of the Global South, and sub-Saharan Africa (SSA) in particular, where reference-grade instrumentation is often not available. However, ensuring high-quality data from low-cost sensor (LCS) platforms is essential. Until now, LCS required calibration by collocation with a reference-grade monitor to be used for more than qualitative studies of air quality, but reference-grade monitors are not available in many countries of the Global South. Since a key artifact in optical PM sensors is aerosol hygroscopic growth, we explore the viability of an alternative LCS calibration method: a hygroscopic growth correction factor using particle composition data from the MERRA-2 reanalysis dataset. We compare 3 different LCS located in 3 different areas of SSA – Kenya, Ghana, and South Africa - with 3 different calibration techniques: traditional linear calibrations with a reference-grade monitor, a κ-Köhler-derived correction with MERRA-2 data, and a random forest machine learning regression utilizing MERRA-2 and the regulatory-grade monitor. Random forest regressions using MERRA-2 particle composition data and collocation with a reference-grade monitor improve sensor performance to near that of regulatory-grade monitors. But even without collocation, a hygroscopic growth correction based on MERRA-2 particle composition alone can improve LCS PM2.5 performance by reducing mean-normalized bias to near-zero and reducing error by up to 40%.