Wind Forecast Error Usually Dominates Fire Spread Prediction Error
When a spread forecast turns out wrong, the instinct is to blame the spread equation. Usually the wind is to blame. Rate of spread climbs steeply with wind speed, so a modest speed error scales the whole run, and a direction error is worse than that: turn the wind by thirty degrees and the head of the fire is aimed at a different slope, different fuel and a different community, so the perimeter is not merely bigger, it is somewhere else. Forecast winds are also the input you know least about, because they are themselves model output, averaged over grid cells kilometres across and hours long, while the fire responds to the gust in this drainage. The rule: before improving the model, perturb the wind by the forecast's own stated uncertainty, re-run, and compare that spread of outcomes with the difference between two candidate models. If the wind spread is larger, model choice is not your problem. The confusion this resolves is treating model error and input error as one budget; they are separate, and on most fires the input side is bigger. After this Concept you can attribute a wrong perimeter to its inputs or its equations instead of guessing.
This Concept is waiting for its first lesson!
When a spread forecast turns out wrong, the instinct is to blame the spread equation. Usually the wind is to blame. Rate of spread climbs steeply with wind speed, so a modest speed error scales the whole run, and a direction error is worse than that: turn the wind by thirty degrees and the head of the fire is aimed at a different slope, different fuel and a different community, so the perimeter is not merely bigger, it is somewhere else. Forecast winds are also the input you know least about, because they are themselves model output, averaged over grid cells kilometres across and hours long, while the fire responds to the gust in this drainage. The rule: before improving the model, perturb the wind by the forecast's own stated uncertainty, re-run, and compare that spread of outcomes with the difference between two candidate models. If the wind spread is larger, model choice is not your problem. The confusion this resolves is treating model error and input error as one budget; they are separate, and on most fires the input side is bigger. After this Concept you can attribute a wrong perimeter to its inputs or its equations instead of guessing.
Are you a teacher? Sign in to start contributing.
Sign In