Not since March 2024. We found this by querying the public record ourselves, and we keep checking every hour.
South Africa has 138 government air quality monitoring locations that report fine particulate matter to the main open data aggregator. Every one of them stopped on the same day: 12 March 2024. Stations do not fail simultaneously across five provinces. A publishing feed did.
This is checkable. Anyone with a free API key can query the open record and see the same date on all 138 records. We are not asking you to take our word for it.
A network of low-cost community sensors, roughly 43 locations, is now the only live public source of fine particulate readings in South Africa. Most of them sit in two coastal clusters. They are real instruments producing real data, and they are not reference-grade.
The Highveld industrial belt has the worst air and the heaviest exposure burden in the country. It currently has no live public measurement of any kind. Not a dead station that could be restarted. Nothing.
The people breathing the worst air live where nobody is counting.
Pick a town. We will tell you how far away the nearest live public measurement is. Not a reading, a distance.
Distances are straight-line to the nearest community sensor reporting as of 7 August 2026. These are low-cost sensors, not reference instruments, and a sensor ten kilometres away across different terrain may say very little about the air on your street. Proximity is not the same as coverage. If you use the location button, your coordinates are used inside this page to work out a distance and are not sent anywhere or stored.
Every hour, at ten locations spanning the range of South African conditions, we compare the modelled estimate we display against the nearest live reference sensor, and log the difference. Five of those locations have no live reference at all, and we record that absence rather than filling it in.
The model has never once produced a value above 80.87 µg/m³. The sensors crossed 75 on 174 occasions, crossed 100 on 89, and peaked at 379. When the ground reads 75 or above, the model’s typical answer is 30. We are not claiming the model is wrong, and this stays an observation rather than a finding.
The modelled estimate and the sensors on the ground disagree, and the disagreement is not random. It widens as the air gets dirtier, and it turns over as the air gets cleaner.
The model has never once produced a value above 80.87 µg/m³. In the whole record it has crossed 75 exactly twice, and it has never reached 100. The sensors crossed 75 on 174 occasions, crossed 100 on 89, and peaked at 379. When the ground reads 75 or above, the model's typical answer is 30.
At the clean end it inverts. Where sensors read below 15 µg/m³, the model reads roughly twice as high.
Too high at the bottom, far too low at the top. That is the shape averaging leaves behind when it is stretched across a field with sharp local peaks, and it matters most in the places that need it most. An estimate that cannot see above roughly 80 will look calmest exactly where the air is worst.
We are still not claiming the model is wrong. Low-cost sensors are themselves unreliable at high concentrations, and that explanation stays open. Two things make it harder to hold. The pattern appears against two different kinds of reference instrument, differently sited and maintained by different people. And the official stations, which are not low-cost optical units, show the gap more strongly than the optical ones do, not less.
What we have not done is rule out humidity, and an earlier version of this page wrongly said we had. The humidity readings we hold come from only one of our two reference networks, they are taken inside a sensor housing that runs warm and therefore reads drier than the air around it, and few of them fall in genuinely humid conditions. That comparison cannot settle the question. Settling it needs a calibrated reference instrument standing beside one of these sensors for a season, and we do not have one yet. So this stays an observation rather than a finding.
Figures counted 10 September 2026 across ten locations. The three counters above update themselves hourly. This paragraph does not, and is rewritten by hand whenever the picture changes.
Every air quality product displays a confident local number. Producing one is easy. Knowing whether it is right is not, and in this country right now, nobody can check it against an official source, because there isn't one publishing.
So we would rather show you the gap, with a date on it, than a number we cannot yet defend.