Jasper 2024: The Monster That Standard Models Couldn't See
The July 2024 Jasper Fire created winds equivalent to an EF-4 tornado, uprooting mature trees and generating flames 100 meters high. Traditional fire models predicted 6.7x less than actual burn. SimFire was within 3.1 percentage points.
SimFire Prediction
31.9%
Actual: 28.8%
Cell2Fire (FBP)
8.0%
6.7x underestimate
Fire-Induced Winds
EF-4
Tornado equivalent
Structures Lost
358
of 1,113 total
Fire Progression Timeline
Interactive timeline based on Parks Canada official timeline and CBC investigation
What Made Jasper Different
The Fire
- Location: Jasper National Park, Alberta
- Ignition: Lightning, July 22, 2024 at 7:02 PM
- Final Size: 32,722 hectares (80,860 acres)
- Duration: Declared extinguished April 1, 2025
- Personnel: 3,000+ firefighters engaged
Extreme Behavior
- Flame Height: 100+ meters
- Fire-Induced Winds: EF-4 tornado equivalent
- Spotting: Embers 500m ahead of fire front
- Pyroconvection: Created its own weather system
- Tree Damage: Uprooted mature trees with roots
"Fire-induced winds caused tree damage and ground scouring comparable to an EF-4 to EF-5 tornado."
25,000
People Evacuated
Largest evacuation in park history
358
Structures Destroyed
32% of town's buildings
70%
Town Saved
All critical infrastructure preserved
Hour-by-Hour: How the Fire Became a Monster
July 22, 2024 — Ignition
July 23, 2024 — Rapid Growth
July 24, 2024 — Catastrophe
The physics: The fire became so intense it created its own convective column, lifting hot air and embers thousands of meters. When this column collapsed, it dumped burning debris across the southwest corner of town. Standard fire models have no mechanism to predict this behavior.
Why Standard Models Failed
| Model | Predicted | Actual | Error | Limitation |
|---|---|---|---|---|
| SimFire | 31.9% | 28.8% | 3.1pp | Models fire-generated winds & pyroconvection |
| Cell2Fire (CFFDRS) | 8.0% | 28.8% | 6.7x under | Uses static wind from weather forecast |
| ELMFIRE | N/A | - | - | Requires LANDFIRE (US-only data) |
What Cell2Fire Assumes
- ✗Wind comes from weather forecast (87 km/h max)
- ✗Fire spread follows FBP equations linearly
- ✗No feedback between fire intensity and atmosphere
- ✗Spotting limited to standard ember physics
What SimFire Captures
- ✓Fire-generated winds calculated from intensity
- ✓Catastrophe mode triggers above thresholds
- ✓Neural network trained on extreme fires
- ✓Pyroconvection and column collapse effects
Cell2Fire used the correct fuel data (CFFDRS). It ran successfully.
The problem is fundamental: standard fire physics cannot model extreme fire behavior.
100m
Flame Height
Standard models assume 30-50m max
EF-4
Fire-Induced Winds
267-322 km/h. Standard models use forecast winds.
6.7x
Model Error
Cell2Fire missed 72% of the burn area.
Benchmark Methodology
Apples-to-Apples Comparison
To ensure a fair comparison, we ran both models on identical conditions:
- Same grid: 64x64 cells covering the fire bounding box + 2km buffer
- Same weather: 87 km/h winds from west (270°), 38°C, 8% humidity
- Same terrain: DEM from Jasper area at 100m effective resolution
- Same ignition: Western edge of actual fire perimeter
- Same ground truth: Official fire perimeter from Parks Canada
SimFire Model
- Architecture: Neural network + differentiable fire physics
- Training: 100 historical fires across North America
- Validation Dice: 73.6% on held-out fires
- Inference: 30 simulation timesteps, ~50ms on GPU
Cell2Fire Baseline
- Algorithm: Canadian FBP (Fire Behavior Prediction) System
- Physics: Rate of spread from fuel type + wind + slope
- Limitation: Cannot model fire-atmosphere feedback
- Expected: ~28% of actual for extreme pyroconvective fires
Why the "Burn Percentage" Metric?
We report burn area as a percentage of the analysis grid (fire bounding box + buffer) rather than absolute hectares. This is more meaningful because the actual burn percentage depends on how large a grid you analyze. A fire that burns 10,000 ha is 100% of a 10,000 ha grid but only 10% of a 100,000 ha grid. By using the fire-appropriate bounding box, we measure prediction accuracy within the relevant area where the fire actually occurred.
Sources & Documentation
CBC: "The Monster of Jasper"
In-depth investigation with hour-by-hour timeline, firefighter accounts, and damage assessment.
Parks Canada: Official Timeline
Official government timeline with fire behavior details and response actions.
Canadian Forest Service Report
Scientific analysis documenting EF-4 tornado equivalent fire-induced winds.
Wikipedia: 2024 Jasper Wildfire
Comprehensive overview with statistics, impact assessment, and references.
Need Accurate Predictions for Extreme Fires?
SimFire is the only model that correctly predicted the Jasper fire's extreme behavior. See how it performs on your specific use case.