Related Experiment Video
Updated: Nov 8, 2025
09:27
Wind Tunnel Experiments to Study Chaparral Crown Fires
Published on: November 14, 2017
9.8K
Emulation of wildland fire spread simulation using deep learning
Frédéric Allaire1, Vivien Mallet1, Jean-Baptiste Filippi2
1Institut national de recherche en informatique et en automatique (INRIA), 2 rue Simone Iff, Paris, France; Sorbonne Université, Laboratoire Jacques-Louis Lions, France.
Summary
This study introduces a deep neural network to rapidly predict wildland fire spread, significantly reducing computational time for fire danger mapping. The developed emulator achieves thousands of times speed-up, enabling near real-time predictions for large areas.
Area of Science:
- Environmental Science
- Computational Science
- Artificial Intelligence
Background:
- Wildland fire spread simulation is crucial for operational decision-making and planning.
- Traditional simulators face computational time limitations for large-scale, short-term applications like fire danger mapping.
Purpose of the Study:
- To develop a computationally efficient emulator for wildland fire spread prediction.
- To enable rapid generation of burned area maps for short-term fire danger assessment.
Main Methods:
- Utilized a deep neural network with a hybrid architecture to emulate fire spread.
- Input data included spatial landscape fields and scalar environmental conditions.
- Trained the network on a large dataset of fire simulations.
Main Results:
- The emulator achieved a Mean Absolute Percentage Error (MAPE) of 32.8% on a test dataset.
- Demonstrated a speed-up factor of several thousands compared to traditional simulators.
- Enabled one-hour burned area map computation for an entire island in under a minute.
Conclusions:
- The developed deep learning emulator significantly reduces computational time for wildland fire spread prediction.
- This approach facilitates new applications in short-term fire danger mapping and crisis management.
