How We Read Air Quality

The day the number was wrong

Smoke from Canadian wildfires settled over Minnesota. For Anoka, our app showed AQI 62, which reads as Moderate: fine for most runners, maybe ease off if you are sensitive. The EPA monitoring station in Blaine, about 5 km away, was reading 311, Hazardous. The state had an air quality alert out and was telling people to stay inside.

Nothing was broken. We were faithfully showing a global air-quality model, the same kind most weather apps use. Models like this are computed on a grid roughly 25 km wide and do not read the ground. They are good at the shape of a day and can be badly wrong about the amount, especially when smoke arrives faster than the model expects. Which is to say: wrong on exactly the day a runner needs it most.

What we show now

The dashed line is the raw model. The solid line is what RunnersKit now shows: pinned to the station's real reading for the current hour, then handed over to the forecasters' call for the days after, keeping the model's hour-to-hour shape throughout.

Three sources, not one

• The nearest ground station: Real instruments measuring real air, published by the US EPA AirNow program together with state and local agencies and Environment Canada. Where one sits within 100 km of you, it sets the number for right now. Nothing outranks a measurement.
• The human forecast: Forecasters at your state or provincial air agency publish a call for today and the days ahead, in their own words. People who watch smoke plumes for a living call these days early, and their number sets the ceiling for each day.
• The model's shape: The hour-to-hour rise and fall through a day is the part the model gets right, so we keep it and correct the level around it. That comes from Open-Meteo, serving the European CAMS atmospheric model.

A station reading tells you about now, not about tomorrow, so its correction counts fully for the hour it was taken and fades over the following hours. By the next day the forecasters are doing the work.

It runs both ways

The same week, Miami showed the mirror image: the model read higher than the forecasters did. A number you cannot trust in either direction is not much use when you are deciding whether to run outside, so the correction pulls down as readily as it pushes up.

When we cannot check

Ground stations are dense across the United States and Canada, thinner in Mexico, and rare elsewhere. Outside that coverage there is nothing to check the model against, so we show the model estimate as-is and tell you that is what it is, rather than dressing it up as something firmer.

Air quality can also change faster than any forecast. If you can smell smoke, believe your nose over any app, ours included.

Sources

• AirNow (US EPA, with state and local air agencies and Environment Canada): ground-station readings and daily forecasts. airnow.gov
• Open-Meteo, serving the Copernicus Atmosphere Monitoring Service (CAMS) model: the hourly curve. open-meteo.com
• US EPA AQI scale: the 0 to 500 categories behind every label on this page. AQI basics

See it for your run

RunnersKit reads the air for wherever you are about to run, and folds it into the RunReady score alongside heat, wind, and everything else. See how the rest of it works.