Cheap sensor networks are slowly reshaping air quality monitoring

A small article in a trade magazine

I have a copy of a trade magazine called Asian Environmental Technology on hand.

It’s a specialist environmental-measurement publication packed with ads and articles for exhaust gas analyzers and water quality meters. Flipping through it, a small article caught my eye: the Acid Deposition Monitoring Network in East Asia (EANET) had adopted a new medium-term plan.

The plan expands what’s being monitored from acid rain to PM2.5, ozone, and VOCs, and one line noted that the network had adopted “Guidelines for a Hybrid Air Quality Monitoring Network.”

It’s a framework for combining reference-grade monitoring stations with low-cost sensors, expanding the scope of observation while maintaining data quality through calibration and validation.

Article on EANET adopting a hybrid monitoring strategy (Source: Asian Environmental Technology, August/September 2026 issue (ILM Media))

Reading it left me with a feeling I can’t quite put into words.

A box of sensors in San Francisco

Over ten years ago, I was trying to start a business in air quality measurement in San Francisco.

Fortunately, things took some shape: I won a business contest, exhibited at trade shows, and even managed to raise funding.

What I was building at the time was, as far as I could think of and as far as I could source, a box crammed with pollutant sensors.

PM sensors, gas sensors, whatever I could get my hands on, I put it in.

I always carried that box hanging from my bag, and when I went into the consulate to renew my passport, the security screener gave me a thoroughly puzzled look.

Of course he did — there was this unidentifiable clump of sensors dangling off me.

That they let me through anyway still makes me laugh a little when I think back on it.

The sensor box I used to carry hanging from my bag back then

While I walked around taking measurements like that, I already had another picture in my head.

Scatter a lot of cheap sensors around, keep correcting them against the values from expensive reference stations placed here and there, and maybe you could measure the air not as points but as a surface.

I believed it was technically feasible, and given that it’s now become a framework at the intergovernmental level, I think the reasoning wasn’t wrong.

I just couldn’t marshal the people, the time, or the funding, and it stayed a picture on paper.

Still — ten years.

The idea of correction itself already existed back then.

You place your device next to a reference instrument for a while, fit the coefficients by regression, and then take it out into the field.

Anyone who’s ever done measurement work would think of this; it’s no special invention.

So why did it take this long for it to make it into a government document as a “guideline”?

What’s frustrating is that the purpose of measuring the air hasn’t changed at all between then and now.

Long-term inhalation of fine particles like PM2.5 raises the risk of heart disease, stroke, and lung cancer.

When a leak or fire happens at a chemical plant, toxic gas drifts past the fence line into residential areas.

The same issue of this trade magazine carried a story about an accident at a pharmaceutical plant in India, where a gas believed to be sulfur dioxide leaked and killed three workers, along with a story about a chlorine leak at a chemical plant in Punjab where nearby residents complained of a strong odor.

Article on a chlorine leak at a chemical plant in Punjab (Source: Asian Environmental Technology, August/September 2026 issue (ILM Media))

Article on a sulfur dioxide leak at a pharmaceutical plant in India (Source: Asian Environmental Technology, August/September 2026 issue (ILM Media))

To protect citizens from health damage and disasters like gas leaks, you have no choice but to measure what’s happening where.

The tools for that were already becoming cheap ten years ago, and yet it took this long before it became acceptable to use them officially.

What actually changed

The first thing that comes to mind is the performance of the sensors themselves.

Among the cheap PM sensors commonly used ten years ago was the Shinyei PPD42NS, popular enough that researchers wrote evaluation papers on it in the lab. After that, compact light-scattering PM sensors spread rapidly, and both price and accuracy improved considerably.

But the premise was always to use them with correction applied, so there was never really a need to wait for the accuracy of the sensor itself to improve. Indeed, low-cost sensors still can’t match reference instruments on their own even today.

It’s hard to believe this was the deciding factor.

Photo of a PPD42NS unit (Source: Austin et al. (2015), PLoS ONE, https://doi.org/10.1371/journal.pone.0137789（CC BY 4.0)

Next, the correction methods.

Machine-learning-based correction using temperature and humidity as explanatory variables did substantially improve accuracy.

But the weakness that coefficients drift once you move away from the location where they were trained still remains today, so it’s hard to say that perfecting this method is what set the world in motion.

What about communications and the cloud?

LPWA and cellular modules got cheaper, and the cost of collecting data dropped dramatically.

I think this had an effect, but even ten years ago you could upload data over a 3G module, so it’s hard to call this the decisive factor either.

Lining up the timeline, what was actually moving was less the technology than the “yardstick,” and the people who demonstrated how to use it.

The US Environmental Protection Agency (EPA) held workshops with experts in 2018 and 2019, and in February 2021 released a report laying out testing methods and target values for PM2.5 and ozone sensors, then extended it to PM10, NO2, CO, and SO2 in February 2024.

In Europe, the technical specification CEN/TS 17660, which classifies the performance of low-cost sensors, was established for gas sensors first and then, in 2024, for PM sensors as well.

And then there’s this EANET guideline.

Air Sensor Performance Targets and Testing Protocols Testing methods and performance targets for air sensors compiled by the US Environmental Protection Agency (EPA) epa.gov

Makers and users became able to talk about how much trust to place in corrected data using the same yardstick.

What PurpleAir and the EPA demonstrated

Talking about yardsticks alone stays abstract, so let me name the players who actually put this into practice.

The clearest example, I think, is PurpleAir.

Adrian Dybwad, who lived in Draper, Utah, wanted to know in 2015 how much dust was being kicked up every time the wind blew across a large gravel pit near his house.

The nearest government monitoring station was 20 miles away, and he couldn’t find an affordable instrument to buy.

So he drew on his experience in electronics and programming to build his own sensor, and started making and handing them out to his neighbors too.

By early 2016 there were still only about seven sensors, but he sent nearly half of them, three, to the Air Quality Sensor Performance Evaluation Center (AQ-SPEC), a low-cost sensor evaluation program run by a Southern California air pollution control agency, to have their performance tested.

By his own account, the results were the most accurate the program had tested up to that point. From there universities and research institutions began testing them one after another, and the network grew to over 30,000 units by 2021.

That’s roughly the same period when I was getting puzzled looks at the consulate.

The PurpleAir Story The story of PurpleAir’s founding, starting from dust at a gravel pit purpleair.com

What’s interesting is what happened next.

Karoline Barkjohn, Andrea Clements, and others in the EPA’s research division cross-checked PurpleAir data against public monitoring stations across the country and built a nationwide correction formula.

In 2020, the EPA added this corrected PurpleAir data to its official wildfire smoke map, the “AirNow Fire and Smoke Map.”

During the 2020 wildfires, readings exceeded 1,000µg/m³ in some places, revealing that the existing correction formula underestimated at high concentrations. The following year the EPA updated it with a formula extended to cover the high-concentration range.

Scatter plot comparing PurpleAir’s raw and corrected data against reference instruments (Source: Barkjohn, Gantt & Clements (2021), Atmospheric Measurement Techniques, https://doi.org/10.5194/amt-14-4617-2021（CC BY 4.0)

Cheap sensors placed by citizens at their own homes, corrected by the government, and put on an official map.

Here is something quite close to the finished version of the picture I had in my head ten years ago.

At the same time, reading through the substance of these yardsticks, gaps remain.

The EPA’s target values are voluntary, and testing against them doesn’t earn you EPA certification.

CEN/TS 17660 also goes out of its way to note that it evaluates individual sensor units, not networks, and doesn’t guarantee performance outside the tested conditions or over the long term.

How to evaluate, as a standard, the part where a network of cheap sensors is continuously corrected against reference stations seems to still be a matter for the future.

The people who built the yardstick

Looking back like this, I find myself feeling a bit nostalgic for the person who used to walk around with a sensor box hanging off his bag.

I had nothing but a conviction that measuring things must reveal something, and honestly, I didn’t have much of a grasp on how to get anyone to trust that, or who those people would even be.

Over these ten years, there have been people steadily building up exactly that: how to earn trust.

Someone who grew a sensor built to measure dust from a gravel pit into a network of 30,000 units.

Someone who spent years cross-checking citizen data against public data to build a correction formula.

Someone who worked through discussions with officials from various countries to compile it into a guideline.

It wasn’t any single person’s great invention that moved things forward. It was that kind of accumulation that got corrected data onto an official map.

Coming from someone who was sketching a similar picture around the same time, I genuinely find how far this has come remarkable.

I’m simply impressed, nothing more.

References

• Asian Environmental Technology, August/September 2026 (Full EANET article: ilmt.co/TL/EKWj, India SO2 leak article: ilmt.co/TL/2QeJ, Punjab chlorine leak article: ilmt.co/TL/wkyR)
• WHO: Ambient (outdoor) air pollution fact sheet
• EPA: Air Sensor Performance Targets and Testing Protocols
• CEN/TS 17660-2:2024 Air quality – Performance evaluation of air quality sensor systems – Part 2
• Austin et al. (2015) Laboratory Evaluation of the Shinyei PPD42NS Low-Cost Particulate Matter Sensor, PLoS ONE
• PurpleAir: The PurpleAir Story
• SFGATE: The story behind PurpleAir, which has become a necessity for Bay Area summers (2022)
• Business Wire: PurpleAir’s 30,000 Citizen Scientists Across the World (2021)
• EPA: AirNow Fire and Smoke Map: Extension of the U.S.-Wide Correction for PurpleAir PM2.5 Sensors
• EPA Science Inventory: Sensor data cleaning and correction: Application on the AirNow Fire and Smoke Map

This piece was conceived and directed by Kuzuryu, with the writing done by AI.

Originally published in Japanese at https://clazytech.com/2026/09/1817/. Translated with LLM assistance and reviewed before publication.