Direction Accuracy Across Four Fires
The same wind+slope heuristic applied to four different fires. The bar chart shows what fraction of satellite detections (FIRMS) fell inside the predicted cone in the first 21–34 hours.
| Fire | Wind at Ignition | Predicted Direction | FIRMS Detections | In Cone | Accuracy | Why |
|---|---|---|---|---|---|---|
| Kincade | 13 km/h from 85° | W (265°) | 330 | 132 | 40% | Steady wind, flat terrain |
| Camp Fire | 26 km/h from 335° | SE (148°) | 706 | 3 | 0.4% | Strong slope interaction, rapid shifts |
| Dixie | 4 km/h from 65° | SW (223°) | 8 | 0 | 0% | Light wind, terrain-dominated spread |
| Marshall | 24 km/h from 260° | E (104°) | 79 | 28 | 35% | Strong steady wind, grassland |
The failures are total, not gradual. Camp Fire's 26 km/h wind predicted SE spread, but steep terrain channeled the fire into canyons the heuristic can't represent. 3 of 706 detections in the predicted cone. In complex terrain, a wind-only direction estimate isn't just inaccurate — it points the wrong way. You need at least a physics model (Q2) to capture topographic channeling.
Where it works, it's all you need. Kincade and Marshall had strong, steady winds over flat terrain — 35–40% of detections fell in the right quadrant. For a first-hour "which way to run" decision, that's enough. The 60% outside the cone arrived later as conditions shifted — a timing problem for Q2, not a direction problem for Q1.
Direction is a solved problem. Timing is not. The heuristic answers "which way to evacuate" in the first hour. Good enough. But "who exactly is threatened and when" requires terrain-based fire routing (Q2) — not because more fidelity is always better, but because the question demands it.
Model: wind direction (from ASOS at ignition) + terrain slope gradient. Cone: ±45° from predicted center bearing. Validated against NASA FIRMS VIIRS detections in first satellite pass window (21–34 hours depending on fire).