Using crowd-sourced low-cost sensors in a land use regression of PM_2.5 in 6 US cities

Assessing exposure to ambient fine particulate matter (PM_2.5) is important for improving human health. With rapidly expanding low-cost sensor networks globally, it is possible for monitoring networks to be located by a variety of users (i.e., crowd sourcing) to increase measurement density and coverage for use in exposure assessment, e.g., national land use regression (LUR) models. Few studies have integrated low-cost sensors into LUR models across multiple cities, limiting the ability of modelers to fully utilize growing low-cost sensor networks worldwide. We developed five LUR models to predict annual average PM_2.5 concentrations using combinations of regulatory (six cities: n = 68; national: n = 757) and low-cost monitors (n = 149) from six US cities. We found that developing Hybrid LURs that include the low-cost (i.e., PurpleAir) network may better capture within-city variation. LURs with the PurpleAir data only (tenfold CV R² = 0.66, MAE = 2.01 µg/m³) performed slightly worse than a conventional LUR based on regulatory data only (tenfold CV R² = 0.67, MAE = 0.99 µg/m³). Hybrid models that included both low-cost and regulatory data performed similarly to existing national models that rely on regulatory data (hybrid models: tenfold CV R² = 0.85, MAE = 1.02 µg/m³; regulatory monitor models: R² = 0.83, MAE = 0.72 µg/m³). Integrating crowd-sourced low-cost sensor networks in LUR models has promising applications to help identify intra-city exposure patterns especially for regions with limited regulatory networks internationally.