An evaluation of the U.S. EPA's correction equation for PurpleAir sensor data in smoke, dust, and wintertime urban pollution events

Abstract

PurpleAir sensors (PASs) are low-cost tools to measure fine particulate matter (PM) concentrations and are now widely used, especially in regions with few regulatory monitors. However, the raw PAS data have significant biases, so the sensors must be calibrated to generate accurate data. The U.S. EPA recently developed a national correction equation and has integrated corrected PAS data onto its AirNow website. This integration results in much better spatial coverage for PM2.5 (particulate matter with diameters less than 2.5 μm) across the US. The goal of our study is to evaluate the EPA correction equation for three different types of aerosols: typical urban wintertime aerosol, smoke from biomass burning, and mineral dust. We identified 50 individual pollution events, each having a peak hourly PM2.5 concentration of at least 47 μg m-3 and a minimum of 3 h over 40 μg m-3 and characterized the primary aerosol type as either typical urban, smoke, or long-range transported dust. For each event, we paired a PAS sampling outside air with a nearby regulatory PM2.5 monitor to evaluate the agreement. All 50 events show statistically significant correlations (R values between 0.71-1.00) between the hourly PAS and regulatory data but with varying slopes. We then corrected the PAS data using either the correction equation from Barkjohn et al. (2021) or a new equation that is now being used by the U.S. EPA for the AirNow Fire and Smoke Map (U.S. EPA, 2022b). Both equations do a good job at correcting the data for smoke and typical pollution events but with some differences. Using the Barkjohn et al. (2021) equation, we find mean slopes of 1.00 and 0.99 for urban and smoke aerosol events, respectively, for the corrected data versus the regulatory data. For heavy smoke events, we find a small change in the slope at very high PM2.5 concentrations (> 600 μg m-3), suggesting a ∼20 % underestimate in the corrected PAS data at these extremely high concentrations. Using the new EPA equation, we find slopes of 0.95 and 0.88 for urban and smoke events, respectively, indicating a slight underestimate in PM2.5 using this equation, especially for smoke events. For dust events, while the PAS and regulatory data still show significant correlations, the PAS data using either correction equation underestimate the true PM2.5 by a factor of 5-6. We also examined several years of co-located regulatory and PAS data from a site near Owens Lake, California (CA), which experiences high concentrations of PM2.5 due to both smoke and locally emitted dust. For this site, we find similar results as above; the corrected PAS data are accurate in smoke but are too low by a factor of 5-6 in dust. Using these data, we also find that the ratios of PAS-measured PM10 / PM1 mass and 0.3 μm / 5 μm particle counts are significantly different for dust compared to smoke. Using this difference, we propose a modified correction equation that improves the PAS data for some dust events, but further work is needed to improve this algorithm.