Instrument networks
Air quality sensors: the citizen network that outgrew the official one
Air quality monitoring split into two worlds about a decade ago. One is regulatory, expensive and sparse. The other is amateur, cheap and everywhere. The interesting question is not which is better — it is what each is actually for.
What gets measured
Air quality is not one measurement. The pollutants that matter have different sources, different health effects and completely different sensing techniques.
- PM2.5 and PM10 — particulate matter below 2.5 and 10 micrometres. PM2.5 is most strongly tied to mortality, because particles that small reach deep lung tissue. Sources: combustion, brake and tyre wear, wood burning, wildfire.
- NO₂ — nitrogen dioxide, overwhelmingly from vehicle exhaust and combustion. A good proxy for traffic exposure, and it varies sharply over tens of metres.
- O₃ — ground-level ozone, not emitted directly but formed photochemically from NOx and volatile organics in sunlight. It peaks on hot afternoons and is often higher downwind of a city than in it.
- SO₂ and CO — industrial and combustion markers, now low across much of Europe and North America but still central elsewhere.
How the two tiers actually work
Reference monitors
A regulatory-grade particulate monitor draws a metered air sample through a filter or an oscillating microbalance, under controlled temperature and humidity, with scheduled calibration against known standards. It costs as much as a car, needs a serviced enclosure and mains power, and produces data admissible for legal compliance.
Because of that cost, they are sparse. A large European city may have a dozen. Enough to establish whether a country meets a limit value; nowhere near enough to tell you about your street.
Low-cost optical sensors
The cheap sensors work by optical scattering: a small fan pulls air past a laser, and a photodetector counts the flashes as particles cross the beam. From pulse sizes and rates, the device infers a mass concentration.
That inference is where the error lives. Optical counters measure particle number and size, then convert to mass using assumptions about density and refractive index. When the real aerosol differs from the assumption — as in fog, sea spray or wildfire smoke — the conversion drifts. High humidity is the classic failure: water condenses onto particles, they scatter more light, and the sensor reports a pollution spike that is really weather.
The community networks
sensor.community, which began in Stuttgart as a citizen response to diesel pollution, is now the largest air quality network in the world by station count. The design is deliberately simple: an off-the-shelf optical particulate sensor, a temperature and humidity probe, a microcontroller, a housing made from two plumbing elbows, and open firmware that publishes readings publicly by default.
The result has properties no agency network has — extreme density, full transparency, near-zero marginal cost per station — and weaknesses no agency would accept: no calibration regime, no siting standard, and stations that vanish when their owner moves house.
Why you need both
The productive framing is not accuracy but purpose.
- Reference monitors answer legal questions. Did this city exceed the annual limit value? That needs traceable, defensible measurement, and only the expensive tier provides it.
- Dense networks answer spatial questions. Which side of the road is worse? Does the school playground differ from the main road 200 m away? No twelve-station network can answer that; a thousand-station one can.
- Dense networks answer temporal questions cheaply. Wildfire smoke arriving overnight, a nearby bonfire, an inversion trapping morning traffic — these appear within minutes across many stations.
The strongest deployments use both: reference monitors as anchor points, and low-cost networks calibrated against them. Correcting a cheap sensor against a nearby reference instrument — particularly for humidity — recovers much of the accuracy gap at almost none of the cost.
Reading the data honestly
A few habits separate useful analysis from noise.
- Treat a single low-cost station as an indicator, not a measurement. Agreement across neighbours is the real signal.
- Expect humidity artefacts. A PM2.5 spike at dawn that resolves by mid-morning is very often fog, not pollution.
- Check what the value actually is. Feeds differ on whether they publish raw or humidity-corrected mass, and mixing the two produces a step change that looks like an event.
- Watch for stale stations. A sensor that stopped reporting three months ago still has a last value, and plotting it as current is a common and invisible error.
A thousand imperfect sensors, understood, beat a dozen perfect ones misread.