SmogRank / Methodology / Low-cost sensor accuracy
How accurate are low-cost PM2.5 sensors?
Mostly close, with real exceptions. Over the 30 days to October 6, 2026, 248 low-cost sensors within 5 km of a government-grade reference monitor typically read 0.95× what the monitor read in the same hours (5% below), and 57% of them were within a third of it.
That average hides what matters when you buy one or read one: the maker and the sensor inside, the humidity, and how dirty the air is. Sensor.Community's home-built sensors, most of them the Nova Fitness SDS011, typically read 0.56× (44% below) over the 30 days to October 12, 2026, 231 of them. These are our own measurements; how we made them is below.
By maker and sensor
Each row is one kind of sensor, and each tick one sensor, placed at its median ratio to the reference monitor beside it. Clarity's units read closest. AirGradient's read 0.85× on the whole, but that counts units reading under 35% of the monitor next to them, most likely indoors or under cover: the 101 of 116 that pass our checks read 0.96×. Sensor.Community's SDS011 units read lowest, 0.34× on the whole and 0.54× for those that pass.
| Sensor | Typical | Each sensor beside its reference monitor |
|---|---|---|
| Clarity via OpenAQ | 0.97× | |
| AirGradient via OpenAQ | 0.85× | |
| Plantower PMS5003 Sensor.Community, home-built | 0.72× | |
| Plantower PMS7003 Sensor.Community, home-built | 0.65× | |
| Sensirion SPS30 Sensor.Community, home-built | 0.65× | |
| Nova Fitness SDS011 Sensor.Community, home-built | 0.34× | |
0.1×0.25×0.5×reference2×4×10× |
Ratio to the reference monitor, on a log scale. Dashed line: the reference monitor's reading. Shaded: within a third of it (0.75 to 1.33 times). Bold mark: the median.
Humidity
Most low-cost sensors count particles by the light they scatter, and damp particles swell and scatter more. In AirGradient's units, which report their own humidity, the ratio rose from 0.78× under 50% relative humidity to 1.25× at 85% and over; at 70–85% it was 1.16×.
| Relative humidity | Hours | Ratio | Compared with the reference monitor |
|---|---|---|---|
| <50% | 19,131 | 0.78× | 22% below |
| 50–70% | 20,117 | 1.00× | the same |
| 70–85% | 2,188 | 1.16× | 16% above |
| 85%+ | 299 | 1.25× | 25% above |
Sensor.Community's sensors read low in dry and damp air alike: 0.49× under 50% humidity, 0.60× at 85% and over. Its SDS011 units went from 0.28× to 0.44×.
Clean air and dirty air
Across all the pairs, the ratio held near 1 in clean and moderate air and fell in dirty air: low-cost sensors under-read the worst hours. Few hours were over 55 µg/m³, so the higher rows rest on less.
| Reference monitor reading | Hours | Ratio | Compared with the reference monitor |
|---|---|---|---|
| <12 µg/m³ | 63,893 | 0.96× | 4% below |
| 12–35 µg/m³ | 18,560 | 0.93× | 7% below |
| 35–55 µg/m³ | 770 | 0.92× | 8% below |
| 55–150 µg/m³ | 595 | 0.78× | 22% below |
| 150+ µg/m³ | 41 | – | too few hours to say |
Correcting them
The US EPA publishes a correction for Plantower-based sensors (Barkjohn and others, 2021, with its 2022 extension for smoke), fitted to US sensors beside US monitors. On our 115 AirGradient pairs, which report the humidity it needs, it moved the median from 0.85× to 0.63× (37% below): further from the monitors, not nearer. So we don't apply it: a correction fitted to one place's sensors and air can move another's the wrong way.
What it means if you buy one
- A little over half the outdoor units we paired landed within about a third of a reference monitor in their typical hour (57%).
- Put it outside in open air, away from a kitchen, a chimney or a car park. The units we found reading under 35% of the monitor beside them were most likely indoors or under cover.
- Read humid days with care: in damp air some makers' units read high. And if yours is an SDS011, expect it to read low, about 0.54× a reference monitor even among the units that pass our checks.
- Compare it with the nearest government monitor for a week or two. We leave out any sensor averaging under 35% of the nearest reference monitor: one that far off is telling you about its placement or its health, not the air.
- Look at several sensors, not one: a city's figure here needs three or more low-cost sensors, because one can be well off.
- If yours also shows CO₂ or VOCs: what those numbers mean.
Paid link: PM2.5 monitors on Amazon.ca (Canada store) · PM2.5 monitors on Amazon.com (United States store)
How SmogRank uses them
Government-grade reference monitors set a city's value wherever it has one. Low-cost sensors count only in a city with none, and only three or more of them, each checked against the nearest reference monitor and our model estimate; those cities carry a one-bar mark on every board (what the bars mean). Sensor.Community's sensors, which read lowest, count only where no other low-cost network can. We don't correct any sensor's readings.
Cities with only one or two low-cost sensors are worked out apart and not ranked. Each hour we compare their 7-day values with the nearest reference-monitor city within 25 km: across 496 such pairs the typical ratio is 1.13×, and 87% are within half and double.
How we measured
We took low-cost sensors within 5 km of a reference monitor we count (1,162 such sensors on the networks we read through OpenAQ, 2,517 of Sensor.Community's), picked across makers and countries, and matched their hourly readings to the monitor's for the same UTC hours over 30 days: September 6, 2026 to October 6, 2026 for AirGradient and Clarity, September 12, 2026 to October 12, 2026 for Sensor.Community, whose hours come from its daily archive. Only hours when the monitor read 5 µg/m³ or more count, and a sensor needs 48 of them. Each sensor's figure is the median of its hourly ratios, and each group's the median of its sensors'. The sensors are those near our monitors, not a random sample of every unit sold, and 30 days is one season in each place. The rules that use these results are on the methodology page.
As an Amazon Associate, SmogRank earns from qualifying purchases.
Not health or buying advice: for decisions about your health, follow your local air quality authority.