Deep learning predicts wildfire spread faster using fuel load data
Interpretable Patch-Based Deep Learning for Wildfire Spread Prediction from Ensemble Simulations
Machine LearningArtificial Intelligence
Summary
Predicting where wildfires will spread helps communities prepare and respond. Traditional methods use physics-based simulations that are slow when running many scenarios. The authors trained deep learning models on thousands of simulated fires to predict fire spread more quickly. They found that surface fuel load was the most useful environmental factor for prediction and that their models could generalize fairly well to new regions. Different model types focus on different features like distance to the fire or terrain.
What this means in practice
- •For wildfire management teams: Predict wildfire spread in specific regions quickly with deep learning using surface fuel load and high-resolution simulations.
- •For environmental data engineers: Integrate patch-based convolutional models to accelerate large-scale fire spread simulations at a fraction of traditional computational cost.
Tested on simulated data.
Authors
Marcin Lawenda, Aleksandra Krasicka, David Caballero, Luis Torres, Łukasz Szustak
Abstract
Wildfire spread is traditionally predicted using physics-based simulators, which are physically interpretable but whose cost increases with each additional ensemble member. We ask how well deep learning surrogates can reproduce these simulations at a fraction of this cost, training them on 10,584 fire spread simulations at 2m resolution for the Rectoret region in Catalonia, Spain. Four architectures are compared: a patch-based U-Net, a transfer-learned ResNet-50, a physics-informed network constrained by the wind-driven advection equation and a Swin-Unet transformer. Among the terrain and vegetation variables, only surface fuel load predicts burn probability with any strength (r = 0.27) and including it lowers prediction error by 21%. The remaining variables correlate weakly and are highly duplicative. Next, an experiment with saliency, occlusion and rotation demonstrates the models' learning. Convolutional models rely primarily on distance from the current fire front, while Swin-Unet assigns more weight to fuel and terrain, a finding also noted in an unrelated wildfire dataset. When applied without retraining to the second region, Pedriza, all three convolutional models still predict fire spread, losing accuracy by a small but systematic margin.