High-fidelity CNN-based surrogate modeling for wildfire spread forecasting.

Area of Science:

• Environmental Science
• Computer Science
• Artificial Intelligence

Background:

• Wildfires increasingly threaten ecosystems, infrastructure, and lives, necessitating advanced mitigation strategies.
• Effective wildfire management hinges on accurate and rapid fire spread prediction for timely interventions.
• Current high-resolution wildfire modeling faces computational challenges, limiting its practical application in time-sensitive scenarios.

Purpose of the Study:

• To develop a computationally efficient deep learning surrogate model for high-fidelity wildfire spread prediction.
• To address the bottleneck of computational burden in wildfire modeling for improved risk assessment.
• To provide a tool supporting public safety power shutoff planning and firefighting/evacuation efforts.

Main Methods:

• Development and training of a convolutional neural network (CNN) model.
• Utilizing deep learning to create a surrogate model for predicting wildfire propagation.
• Focusing on achieving high-resolution predictions with significantly reduced computational time.

Main Results:

• The proposed CNN model demonstrates high efficiency in predicting wildfire spread.
• Achieved a significant F1 score of 0.92, indicating strong predictive performance.
• The model provides predictions in a fraction of a second, drastically improving response times.

Conclusions:

• The deep learning-based surrogate model offers a fast and accurate solution for wildfire spread forecasting.
• The model serves as a valuable tool for electric utilities in planning public safety power shutoffs.
• Provides critical support for firefighting and evacuation teams, enabling data-informed risk assessment and rapid response.