Healthy Building & Building Performance Consultants | SYNKTECT
Can You Trust Low-Cost Indoor Air Quality Sensors? What 2026 Research Says About Calibration
New 2026 studies show low-cost PM2.5 sensors can be trusted after local calibration, while many gas sensors cannot. Here is what that means for your building.
11/4/20268 min read
A wall-mounted air monitor that costs less than a dinner for four can now report particulate matter, carbon dioxide, volatile organic compounds, and several combustion gases every few seconds. Facility managers are installing them by the dozen, schools are publishing their readings, and homeowners are making ventilation and purifier decisions based on the numbers. The question that matters for anyone acting on those numbers is simple: how much of what the screen shows is real?
Three peer-reviewed studies published in 2026 give a clearer answer than we have had before. Taken together, they show that the reliability of low-cost sensors depends far more on which pollutant is being measured and how the device was calibrated than on the price tag or the polish of the dashboard. For particulate matter, a properly calibrated sensor can be a genuinely useful instrument. For several gases, the same class of device may be producing numbers that look precise but carry very little information. This article explains why, and what building professionals should do about it.
The Problem: More Data Does Not Automatically Mean Better Data
Until about a decade ago, continuous indoor air monitoring required reference-grade instruments that were expensive, bulky, and demanded trained operators. That kept most buildings in the dark. Affordable sensors changed the economics, as we described in our earlier introduction to low-cost indoor air sensors: a complete low-cost monitoring node can now be assembled for well under 1% of the capital cost of a regulatory monitoring station.
The catch is that low-cost sensors do not measure pollutants the way reference instruments do. They infer concentrations indirectly, and those inferences are vulnerable to humidity, temperature, aerosol composition, interfering gases, and gradual changes in the sensor itself. A device that leaves the factory reading correctly in a laboratory may drift, saturate, or simply respond to the wrong thing once it is installed in a real room.
Recognizing this, the U.S. Environmental Protection Agency published voluntary performance testing protocols and target metrics for PM2.5 and ozone sensors in 2021, with supplemental reports for PM10, nitrogen dioxide, carbon monoxide, and sulfur dioxide following in 2024. ASTM International has also issued a standard test method specifically for PM2.5 sensors used indoors (ASTM D8405-21) and a newer one for indoor CO2 sensors (ASTM D8624-25). These frameworks are important, but they are voluntary, they do not certify products, and most consumer and commercial devices are sold without any published results against them.
What the 2026 Evidence Shows
Particulate matter: the strongest case for low-cost sensing
A ten-month field study published in Sensors in October 2026 compared three tiers of equipment side by side at two London roadside sites: low-cost metal oxide and optical sensors, mid-cost electrochemical gas sensors, and government reference stations. The optical PM2.5 sensor was the clear success. After field calibration with a simple linear model that accounted for temperature and relative humidity, it explained roughly 57–78% of the variation measured by the reference monitors, it remained usable through the colder months it had not been trained on, and its calibration transferred reasonably well from one site to the other.
An indoor-focused study in Atmospheric Measurement Techniques (January 2026) reached a compatible conclusion under very different conditions. Researchers at the University of Birmingham exposed 40 low-cost PM2.5 monitors from two manufacturers to cigarette smoke, incense, and cooking emissions inside an instrumented test container alongside a reference-grade aerosol spectrometer. Out of the box, one brand overestimated at low concentrations while the other substantially underestimated, and both compressed their readings during heavy pollution events. A staged calibration approach, built with automated machine learning and separate models for clean and polluted conditions, brought the corrected readings into close agreement with the reference instrument and largely removed systematic bias.
A third study, in Aerosol and Air Quality Research (2026), tested widely used PurpleAir sensors in two very different environments, Raleigh, North Carolina, and Delhi, India. With location-specific calibration, hourly PM2.5 estimates came within about 12% mean absolute bias of regulatory monitors, and daily averages within about 6%. The authors' central practical message was that calibration for the specific location or region is essential before the data are used for analysis.
Gas sensors: a much weaker picture
The London study is equally instructive about what does not work. The low-cost metal oxide sensors for nitrogen dioxide, carbon monoxide, and ozone were essentially unusable with their manufacturer calibration curves, largely because those curves were designed for concentrations far above typical ambient levels. Field calibration rescued some signal for carbon monoxide and a weak signal for nitrogen dioxide, but the ozone sensor failed under every calibration method tested.
Two further findings deserve attention. First, the nitrogen dioxide calibration that worked during spring and summer stopped describing the data in winter at both sites. Second, a mid-cost electrochemical nitrogen dioxide sensor, roughly six times the price of its low-cost counterpart, did not outperform it. In other words, spending more did not buy reliability, and a calibration that looked good for six months could quietly expire.
Why Pollutants Behave So Differently
The divergence comes down to physics.
Optical particle sensors shine a laser through a small air stream and count the light scattered by passing particles. Scattered light is reasonably well related to particle mass, so the main corrections involve humidity, which makes particles swell and appear larger, and the type of particle, since smoke, dust, and cooking aerosols scatter light differently. These effects are systematic, and systematic errors can be modeled and removed.
Metal oxide gas sensors work differently. A heated semiconductor surface changes its electrical resistance when gases react with it. Those reactions are nonlinear, sensitive to temperature and humidity, and not very selective: an ozone sensor may also respond to nitrogen dioxide, for example. The sensor's baseline resistance in clean air also drifts over weeks, so a fixed calibration gradually loses its anchor. At the low concentrations typical of most buildings, the useful signal can be smaller than this background noise.
This is also why machine learning is not a universal fix. In the London study, sophisticated models added nothing over a linear correction for the particle sensor, and the most complex models sometimes learned daily pollution patterns rather than the sensor's actual response. Machine learning helped most where the sensor's response was genuinely nonlinear, but it cannot recover information the sensor never captured.
A Practical Framework for Building Professionals
Based on this evidence, we suggest treating low-cost IEQ data in three tiers of confidence:
Actionable after calibration: PM2.5 from optical sensors that have been co-located with a reference or well-characterized instrument in comparable conditions. These readings can support decisions about filtration upgrades, source control, and cooking or combustion ventilation.
Useful for trends, not thresholds: Uncalibrated optical PM2.5 readings and composite "air quality index" scores. They reliably show when conditions change and which spaces are worse than others, but the absolute numbers should not be compared against health guidelines without correction.
Indicative only: Low-cost metal oxide readings for ozone, nitrogen dioxide, carbon monoxide, and total VOCs. Treat a spike as a prompt to investigate, not as a measurement. Carbon monoxide safety must always rely on certified alarms, not IAQ monitors.
Carbon dioxide deserves a separate note. Most quality CO2 monitors use a different principle, non-dispersive infrared absorption, which is not covered by the studies above. They should be evaluated on their own merits, and the new ASTM test method for indoor CO2 sensors gives buyers a reference point to ask manufacturers about.
For owners and facility teams deploying sensor networks, a few habits make a large difference:
Ask for evidence. Request test results against the EPA performance protocols or the relevant ASTM method, not just a specification sheet.
Calibrate locally. Co-locate a sample of sensors with a reference instrument, or a carefully maintained higher-grade monitor, in representative conditions before trusting absolute values.
Plan for drift. Schedule periodic re-checks, especially across seasonal transitions, and budget for sensor replacement.
Keep the sensor model consistent. Calibrations transferred best between identical devices; mixing brands complicates comparison.
Read readings in context. High humidity, heavy smoke, and very low concentrations are where errors are largest.
Benefits, Limitations, and Open Questions
The benefits are real. Dense, continuous monitoring reveals patterns that a one-day walkthrough cannot: the classroom that exceeds others every afternoon, the kitchen exhaust that underperforms, or the filtration upgrade that actually reduced exposure. At a fraction of the cost of reference equipment, that visibility is accessible to almost any building.
The limitations are equally real, and the current evidence base has gaps. The London study was conducted outdoors at urban roadside sites, so its gas-sensor findings may not translate directly to indoor environments with different pollutant mixtures. The Birmingham chamber study used controlled emission sources and did not evaluate long-term drift. The PurpleAir study involved short calibration deployments. None of these studies settles how often indoor sensors should be recalibrated, how well calibrations transfer between building types or climates, or how reliable low-cost VOC readings are in practice. Each study also tested specific sensor models, and results should not be assumed for others.
Implications for Future Healthy Buildings
Low-cost sensing is likely to become standard infrastructure in healthy buildings, much as occupancy and temperature sensing already are. The 2026 research points toward what a mature version of that infrastructure might look like: a small number of well-maintained reference or higher-grade instruments acting as calibration anchors for larger networks of inexpensive sensors, with automated recalibration routines and transparent data-quality flags.
It also suggests that certification programs and procurement specifications may increasingly need to ask not only whether a building monitors IEQ, but how well the monitoring data have been validated. A dashboard that displays confident numbers for pollutants the hardware cannot reliably measure risks undermining trust in monitoring altogether.
Conclusion
Low-cost indoor air quality sensors are neither a gimmick nor a substitute for professional measurement. For fine particulate matter, the evidence now shows that locally calibrated sensors can deliver data good enough to guide real decisions. For several gases, the same technology remains a rough indicator at best, and even a successful calibration can expire with the season. The practical takeaway is to match your confidence to the pollutant, insist on calibration evidence, and treat sensor data as the start of an IEQ investigation rather than its conclusion.
If you are planning a monitoring program or interpreting data you already collect, SYNKTECT can help you design a measurement strategy, select and validate sensors, and translate readings into building improvements through our post-occupancy evaluation and building performance assessment services. Contact us to discuss your project.
References
Duvall, R., Clements, A., Hagler, G., Kamal, A., Kilaru, V., Goodman, L., Frederick, S., Johnson Barkjohn, K., VonWald, I., Greene, D., & Dye, T. (2021). Performance testing protocols, metrics, and target values for fine particulate matter air sensors: Use in ambient, outdoor, fixed site, non-regulatory supplemental and informational monitoring applications (EPA/600/R-20/280). U.S. Environmental Protection Agency, Office of Research and Development. https://cfpub.epa.gov/si/si_public_record_Report.cfm?dirEntryId=350785&Lab=CEMM
Gupta, P., Doraiswamy, P., Sayeed, A., Baliyan, P., Dey, S., Chartier, R., & Levy, R. C. (2026). Evaluating and calibrating low-cost air quality sensors in contrasting aerosol regimes. Aerosol and Air Quality Research, 26(8), Article 56. https://doi.org/10.1007/s44408-026-00146-x
Mercado-Velazquez, A. A., Escamilla-Ambrosio, P. J., Bodanese, E., Islam, M., Toosy, F., Mosler, G., & Poslad, S. (2026). How accurate are low-cost air quality sensors? A 10-month study of three-tier comparison with three calibration methods and cross-site transferability. Sensors, 26(20), 6441. https://doi.org/10.3390/s26206441
Qian, J., Wynn, T., Liu, B., Shan, Y., Bartington, S. E., Pope, F. D., Dai, Y., & Shi, Z. (2026). Enhancing accuracy of indoor air quality sensors via automated machine learning calibration. Atmospheric Measurement Techniques, 19, 603–615. https://doi.org/10.5194/amt-19-603-2026
ASTM International. (2021). ASTM D8405-21: Standard test method for evaluating PM2.5 sensors or sensor systems used in indoor air applications. https://www.astm.org/d8405-21.html
U.S. Environmental Protection Agency. (n.d.). Air sensor performance targets and testing protocols. Air Sensor Toolbox. Retrieved November 4, 2026, from https://www.epa.gov/air-sensor-toolbox/air-sensor-performance-targets-and-testing-protocols
Image Credits
Featured image: "Air Quality Sensor Provides Big Data for Visualization" by Intel Free Press, via Wikimedia Commons, licensed under CC BY-SA 2.0. Not modified. Diagram: original illustration by SYNKTECT.
