Fire spread simulations using Cell2Fire on synthetic and real landscapes

Area of Science:

• Wildfire modeling
• Computational science
• Environmental science

Background:

• Wildfires are a global concern impacting natural and built environments.
• Fire spread models (FSMs) are crucial for understanding and predicting wildfire behavior.
• Existing FSMs may not always produce perfectly realistic simulations.

Purpose of the Study:

• To evaluate Cell2Fire, a cellular automata-based FSM, against established models.
• To enhance Cell2Fire's simulation accuracy using optimization techniques.
• To assess Cell2Fire's computational efficiency compared to other FSMs.

Main Methods:

• Tested Cell2Fire on synthetic landscapes, including U.S. terrains.
• Applied multi-objective optimization and blackbox optimization to adjust fire spread parameters.
• Validated optimized Cell2Fire against the 2001 Dogrib Fire data.

Main Results:

• Cell2Fire showed high agreement with existing FSMs on synthetic landscapes.
• Optimized Cell2Fire significantly improved prediction accuracy (F1-score from 0.74 to 0.83) for the Dogrib Fire.
• Cell2Fire demonstrated superior computational efficiency with linear runtime scaling versus exponential for Prometheus.

Conclusions:

• Optimized Cell2Fire provides more realistic wildfire simulations than benchmark FSMs.
• Cell2Fire offers enhanced accuracy and computational efficiency for wildfire modeling.
• The model can be adapted with custom data for broader applications.