How these numbers are calculated
This run is scored against NIFC fire perimeters — hotspots are not used.
- Actual: each NIFC perimeter in the 7 days after the run, scored at its own time:
every one with a mapped time, the latest one published on a day without one (timed by its
last edit, shown in italics), and the perimeters scored before November 2026.
- Predicted: where the forecast's time of arrival is before that perimeter's time.
- Growth only: what had burned when the run began is removed from both, so only new spread counts: the latest perimeter mapped before the run, and where the forecast itself started burning (Pyrecast adds satellite detections to the perimeter).
Precision = overlap ÷ predicted growth
Recall = overlap ÷ actual growth
IoU = overlap ÷ area either one covers
Every perimeter is shown: no minimum growth, timing window or off-grid
filter. When the fire didn't grow, recall is blank and precision and IoU are 0 if the
forecast predicted growth.
A worked example →