Air quality · 5 min
Smoke travels further than you think
The cities with the worst air during a fire season are frequently nowhere near a fire. Dispersion forecasting is a different problem from fire forecasting.
Ensemble Weather AI Research ·
Wildfire smoke routinely crosses a continent. Fine particulate matter lofted into the free troposphere can travel thousands of kilometres before subsiding, which is how a city with no fire within a thousand kilometres records its worst air quality on record.
This makes smoke a forecasting problem rather than an observation problem. By the time a monitoring station registers elevated PM2.5, the exposure has already happened. A public health advisory issued on observed data is issued too late to change behaviour.
Dispersion forecasting couples fire emissions estimates with the same wind fields that drive the weather forecast. That coupling is why it sits naturally alongside the rest of our stack: the transport is just atmospheric dynamics, and we are already running it.
We forecast PM2.5 and PM10 out to five days at 0.1° resolution, with AQI conversion for the local standard rather than assuming the US scale. A public health team in Europe or Australia should not have to convert our numbers before it can use them.
Every response carries the spread
Ensemble members, model weights and p10/p50/p90 on every value — including on the free tier.