SimFire
Universal Wildfire Spread Prediction
Point to any location on Earth. Get accurate fire spread predictions in seconds. No fuel maps. No calibration. No regional experts required.
0.74
Dice Score
Validated on 100 fires
0
Calibration Required
Works out of the box
Global
Coverage
Any ecosystem
<10s
Prediction Time
Real-time capable
The Problem with Current Fire Models
Require Detailed Fuel Maps
Traditional models need expensive fuel surveys that take months to complete and become outdated quickly. Most of the world has no fuel data.
Need Regional Calibration
Each ecosystem requires expert parameter tuning. A model calibrated for California chaparral won't work in Canadian boreal forest.
Fail on Extreme Fires
Standard physics models underpredict catastrophic fires by 6x or more. They don't account for fire-generated winds and extreme behavior.
Too Slow for Emergencies
Many operational models take hours to run. During active fires, incident commanders need predictions in minutes.
SimFire: A Different Approach
Trained on 500+ real wildfires with real satellite data. We combine physics-based fire spread equations with machine learning that learns regional fuel characteristics directly from coordinates. VLA 512-bit precision eliminates numerical error accumulation.
Global Coverage
Works from Canadian boreal to Australian eucalyptus to Mediterranean scrublands. No regional setup needed.
Trained on Real Fires
500+ real wildfires from USA and Canada. Real satellite perimeters, real terrain, real weather. No synthetic data.
Handles Extreme Fires
Trained on catastrophic fires. Accounts for fire-generated winds, crown fire runs, and pyroconvection.
How It Works
INPUT
Coordinates + Weather
STEP 1
Region Encoder
Predicts fuel from location
STEP 2
Physics Core
Rothermel + CFFDRS + VLA
STEP 3
ML Correction
Handles extreme cases
OUTPUT
Burn Probability Map
Proven Performance
Validated against real fires across North America. Outperforms traditional models on extreme events.
Jasper 2024 Fire — Extreme Fire Test
The Jasper fire burned 32,700+ hectares with documented 200 km/h fire-generated winds. Traditional models failed catastrophically. We benchmarked both models on the same 64x64 grid.
| Model | Predicted | Actual | Error | Notes |
|---|---|---|---|---|
| SimFire | 31.9% | 28.8% | 3.1pp | Correctly modeled extreme behavior |
| Cell2Fire (FBP) | 8.0% | 28.8% | 6.7x under | No fire-generated winds |
| ELMFIRE | N/A | - | - | No data for Canada (LANDFIRE US-only) |
California Benchmark (Oak, Caldor, McKinney)
SimFire
0.573
Avg Dice Score
ELMFIRE
0.344
SimFire +67% better
Cell2Fire
0.298
SimFire +92% better
Applications
Emergency Response
Real-time spread predictions during active fires. Help incident commanders position resources and time evacuations.
- 6-72 hour predictions
- API integration available
Insurance Risk Modeling
Property-level fire risk assessment with proper tail risk quantification using Mixture LDT.
- Catastrophe scenario modeling
- Heavy-tailed loss distributions
Utility PSPS Planning
Better public safety power shutoff decisions. Reduce liability while minimizing unnecessary outages.
- Forecast-based scenarios
- $30B+ utility liability context
Land Management
Prescribed burn planning, WUI assessment, and climate change projections for long-term planning.
- Safe burn window identification
- Future climate scenarios
Why Traditional Models Fail on Extreme Fires
The Jasper 2024 fire revealed a critical gap: standard fire physics can't model catastrophic behavior. Cell2Fire uses the same validated CFFDRS equations as Canadian operational models, yet it predicted 8% burn when the actual was 28.8% on the same grid.
The problem isn't bad data or poor calibration. It's that extreme fires operate in a different regime:
Fire-Generated Winds
Extreme fires create their own weather. Jasper had documented 200 km/h winds generated by the fire itself. Standard models use static wind input.
Pyroconvection
Large fires develop convection columns that loft embers 10+ km. This creates mass spot fires that traditional models don't anticipate.
Runaway Feedback
Above certain thresholds, fires enter "catastrophe mode" with exponential intensification. Standard equations are calibrated for typical fires.
Trained on 500+ Real Fires with Real Data
Including Fort McMurray, Camp Fire, Black Saturday, Jasper, and hundreds more from NIFC and CWFIS. Real satellite perimeters, real 30m terrain, real weather. VLA 512-bit precision ensures numerical accuracy.
How We Compare
The fire modeling market is projected to grow from $400M (2024) to $2.5B by 2032. Here's why existing solutions leave gaps that SimFire fills.
| Feature | SimFire | FlamMap/FARSITE | Technosylva | Pano AI / AlertWest |
|---|---|---|---|---|
| Primary Function | Spread Prediction | Spread Prediction | Spread Prediction | Detection Only |
| Requires Fuel Maps | No | Yes | Yes | N/A |
| Regional Calibration | None | Extensive | Per-region | N/A |
| Global Coverage | Yes | US Only | Select Regions | Camera Network |
| Prediction Speed | <10 seconds | Hours | Minutes | N/A |
| Extreme Fire Accuracy | 500+ real fires trained | 6x underestimate | Limited | N/A |
| Numerical Precision | VLA 512-bit | float64 | float64 | N/A |
| API Access | REST API | No | Enterprise only | Enterprise only |
| Pricing | Affordable SaaS | Free (DIY setup) | $500K+/year | $200K+/year |
Why Detection Companies Don't Predict
Companies like Pano AI and AlertWest focus on finding fires early using camera networks and satellite imagery. This is valuable but fundamentally different from predicting how a fire will spread.
Fire spread prediction requires physics modeling, terrain analysis, and weather integration. Detection companies optimize for fast alerts, not fire behavior simulation.
Why Traditional Models Need Calibration
FARSITE and FlamMap require detailed fuel maps that take months to survey and cost millions to produce. They also need regional experts to tune parameters for each ecosystem.
SimFire's neural encoder learns fuel characteristics directly from coordinates and satellite data, eliminating both requirements while maintaining physics interpretability.
Growing Market, Underserved Customers
Fire modeling market growing 30% annually. Yet most agencies still use 1990s tools or pay $500K+ for enterprise solutions that require months of setup.
$400M
2024 Market
$2.5B
2032 Projected
Ready for Better Fire Predictions?
We're seeking partners in fire agencies, insurance, and utilities. See SimFire in action on your specific use case.