FAQ

A low-cost sensor counts particles with a laser. In humid air those particles take on water, swell, and scatter more light — which the sensor faithfully reports as more mass. A raw reading runs about 1.8× the corrected one, which is enough to move a sensor a whole AQI category: 46.3 raw reads “unhealthy for sensitive groups” where 25.4 corrected reads “moderate”.

The US EPA derived a nationwide correction by collocating PurpleAir sensors with reference monitors, and AirNow publishes PurpleAir data through it. Be I/O uses the same equation, so its numbers agree with the official ones.

There’s a second estimate, ALT-CF3.4, computed from the particle counts directly. It needs no humidity reading at all, which is why it’s the default — a failed hygrometer takes the EPA correction with it.

Reference monitors publish more accurate data infrequently. It’s a tradeoff.

Sampled over one area in Chicago, 40 low-cost sensors report data between 1.1 and 4.9 minutes old. In the same area, at the same time, all 16 reference monitors report the previous hour’s data, once an hour: between 103 and 165 minutes old.

Be I/O allows three hours for one and half an hour for a small sensor, because half an hour of silence from a sensor that reports every two minutes really does mean something is wrong.

This matters most during exactly the events you’d want it for. A storm dropped one reporter’s own sensors to nearly zero while the nearest reference monitor, an hour behind, still read 13 µg/m³ — and an hour later the real figure was 5.

Whether to use a reference monitor or a low-cost sensor depends on your needs and priorities, but the air quality changes so rapidly that looking at trends in low-cost sensors will likely give you the most meaningful information.

Meditating in a cloud of dust will have a dramatically different impact on you than doing VO2Max intervals in the same cloud.

Likewise, two people doing the same activity in the same air will likely take in different amounts of particulate matter because they breathe at different rates.

Your heart rate is the best available proxy for how much air you’re moving, and on a watch it’s already being measured. Without it the app has to assume an effort level, which is exactly the assumption that makes the answer generic.

It’s the World Health Organization’s 2021 guideline for PM2.5, averaged over 24 hours. It is not a personal dose budget, and Be I/O using it as one is an extrapolation.

The WHO’s guidance is epidemiological: it is intended to help large populations understand a very complex situation.

Even if the extrapolated guidance isn’t quite right, it is still good for is comparison. If a plan comes out at twice the guideline, that’s a meaningful thing to know about the plan, even though nothing in particular happens at 1.0. How it works sets out the arithmetic and where it’s shaky.

No, and Be I/O will alert you when that’s the case.

Indoor PM2.5 is mostly outdoor PM2.5 that got in. A tight modern house might let a fifth of it through; an old one with the windows open is barely distinguishable from outside. Be I/O assumes a fraction and lets you change it, because the true figure depends on your building and what you’re doing in it.

On a bad enough day the app will tell you to be inside and say that indoors is also above the guideline. That’s not a contradiction — it’s the honest version of “inside is the better of two options”, and saying only “be inside” would imply a safety it can’t offer.

Three primary reasons:

1. PM2.5 varies over surprisingly short distances — a monitor near a road, a grill, or a dryer vent reads its own microclimate rather than the neighbourhood’s. That’s an argument for owning one rather than against it: the air you breathe is the air where you are.
2. Sensors also fail in ways that still produce numbers. A hygrometer stuck at 100% quietly takes 8.6 µg/m³ off every corrected reading.
3. PM2.5 varies over time. Reference monitors report older data, so they can’t be relied upon if a storm rolls in.

Be I/O addresses these variations in a couple ways:

1. When it picks a public monitor for you and that monitor disagrees sharply with its neighbours, it says so.
2. Many sensors have two laser counters that measure the same thing to determine if either can be trusted. If they read close to the same number, it’s very likely accurate. If they read different numbers, neither can be trusted. Be I/O checks for both and refuses to show a figure it doesn’t believe, rather than averaging its way past the problem.
3. You can choose whether to prefer reference monitors (accurate but stale) or the nearest (less accurate but near real-time).