Low-Cost Sensor Networks and Science

Until about 2015, air quality data in the United States came almost entirely from a network of around 1,000 federal reference monitors run by state and tribal agencies. They cost $30,000 to $100,000 each, need climate-controlled shelters, and are sparse: most U.S. counties have none. The rise of consumer-grade optical sensors (PurpleAir, Clarity, AirGradient, IQAir AirVisual, and others) has changed that picture, but the research community spent the better part of a decade figuring out how trustworthy the new data actually is.

Background

Most low-cost PM2.5 sensors use a laser and a photodetector (a Plantower PMS5003 in PurpleAir's case) to count particles flowing through a small chamber. Federal reference monitors weigh collected particles or use a beta-ray attenuation method that is traceable to mass. The cheap sensors infer mass from optical counts, which means they can drift with humidity, particle size, and particle composition.

Two research programs anchored the credibility work. The Air Quality Sensor Performance Evaluation Center (AQ-SPEC) at South Coast AQMD in California has run side-by-side comparisons of dozens of consumer sensors against reference monitors since 2014, posting results publicly. EPA's Office of Research and Development, led on this topic by Karoline Barkjohn and Andrea Clements, has done parallel work focused on national-scale correction.

What the studies found

AQ-SPEC's 2016 PurpleAir evaluation found correlations (R squared) of 0.78 to 0.90 against reference monitors, with a tendency for the raw PurpleAir to over-report PM2.5 by roughly 40 percent at typical concentrations and to drift further off in high humidity.

In 2021, Barkjohn and colleagues published the national EPA correction equation in Atmospheric Measurement Techniques. The formula uses PurpleAir's two internal sensor channels plus an onboard relative humidity reading to bring corrected PM2.5 close to federal reference values. EPA integrated the corrected feed into AirNow's Fire and Smoke Map, which now displays PurpleAir data alongside regulatory monitors during smoke events.

A 2023 evaluation by Daniel Jaffe (University of Washington) and NOAA collaborators tested the EPA correction across 50 high-pollution events. Corrected PurpleAir tracked regulatory data with mean slopes near 1.0 for both urban winter pollution and smoke. The exception was extreme smoke above roughly 600 micrograms per cubic meter, where the correction undershoots by about 20 percent. A 2024 paper added a humidity- and temperature-aware multilinear correction that cut error another 16 to 23 percent.

Why it matters

Low-cost networks fill in the map. PurpleAir alone has more than 30,000 active outdoor sensors worldwide, roughly 30 times the size of the U.S. regulatory network. Researchers have used the dense data to map intra-urban PM2.5 gradients, track wildfire plume movement in real time, and identify environmental justice disparities at neighborhood scale (work by Joshua Apte at UC Berkeley and Sarah Chambliss at UT Austin).

EPA accepts low-cost sensor data for some non-regulatory uses, including community air monitoring grants funded under the Inflation Reduction Act and the Fire and Smoke Map. Several state agencies use corrected sensor data to issue local advisories.

Open questions

The hard problems are sensor drift over time (Plantower units degrade after 18 to 24 months of continuous use), differential bias for different particle types (corrections for smoke and dust are not the same as for urban aerosol), and how to credit volunteer-hosted data for regulatory decisions. Research on next-generation low-cost gas sensors for NO2, ozone, and VOCs is less mature, with reported correlations against reference instruments still in the 0.4 to 0.8 range.

Sources

• AQ-SPEC Sensor Evaluation Center, South Coast AQMD
• Development and application of a United States-wide correction for PM2.5 data collected with the PurpleAir sensor (Barkjohn et al., AMT 2021)
• An evaluation of the U.S. EPA's correction equation for PurpleAir sensor data in smoke, dust, and wintertime urban pollution events (Jaffe et al., AMT 2023)
• Correction and Accuracy of PurpleAir PM2.5 Measurements for Extreme Wildfire Smoke (Barkjohn et al., Sensors 2022)
• EPA AirNow Fire and Smoke Map
• High-resolution air pollution mapping with Google Street View cars (Apte et al., ES&T 2017)
• Calibration of PurpleAir sensors under high relative humidity conditions (PMC 2024)