‹ Back to the map
How accurate is SmokeCast?
Every night, SmokeCast checks its recent forecasts against what EPA air monitors actually measured. This page is the scorecard.
Loading the latest scorecard…
What do these numbers mean?
SmokeCast starts with NOAA's federal smoke model. Before you see the map, it checks that model against about 1,200 EPA air monitors and corrects the forecast toward what they measured. The headline numbers grade forecasts one to two hours after that correction. The lines about the uncorrected model show how the starting point scores without the correction, graded on exactly the same monitors and hours.
Where is it weakest?
- No monitor, no correction. About one town in ten has no air monitor within 30 miles. There you see the uncorrected forecast, and the app labels it an estimate.
- It is still a forecast. Tomorrow afternoon has no reading yet to correct against. The further ahead you look, the more you are trusting the model on its own.
How is it scored?
- "How far off" is the typical miss (the median). Half of our forecasts missed by less than that number, and half missed by more.
- Every forecast is graded right at a monitor. We save what the forecast said for each monitor's exact location, then compare it to what that monitor measured.
- The readings are EPA AirNow's. They are preliminary and subject to change.