AI Models Show Promise in Predicting Wildfire Spread

Wildfires are growing more deadly each year, but a team of researchers just found a powerful new ally in protecting communities: artificial intelligence.

Scientists at the University at Buffalo completed one of the most comprehensive tests ever of AI models designed to predict how wildfires spread. They trained five different deep learning systems using over a decade of wildfire data from Hawaii, then put them to the ultimate test against the devastating 2023 Maui fires.

"Wildfires are becoming more frequent and more destructive, and communities need tools that can anticipate how these events will unfold," said Yingjie Hu, associate professor of geography at UB. His team wanted to know if these new AI tools could match or beat FARSITE, the traditional physics-based model firefighters have relied on since the 1990s.

The results revealed both exciting potential and important limitations. Two AI models, called ConvLSTM and its attention-enhanced version, performed best among the five tested. While FARSITE still showed better overall accuracy, the AI models brought something equally valuable to the table: flexibility.

Traditional models like FARSITE need detailed, structured data that takes significant time and labor to produce. By the time it's ready, that information is often outdated. The AI models can work with diverse data sources, including real-time satellite imagery, adapting quickly as new information becomes available.

The Bright Side

The research revealed something remarkable about how AI understands fire behavior. By analyzing the 2023 Maui fires, the models identified exactly which factors drove the extreme spread: temperature, humidity, precipitation, wind and vegetation. This kind of insight helps firefighters understand not just where a fire might go, but why.

The team isn't stopping here. They're now working to create hybrid models that combine AI's adaptability with traditional fire science's precision. They're also incorporating higher-resolution environmental data to capture on-the-ground conditions with even greater detail.

This approach could transform how communities prepare for and respond to wildfires. Instead of choosing between old and new technology, firefighters might soon have the best of both worlds: AI that learns and adapts quickly, guided by decades of proven fire science.

The work matters now more than ever, as climate change pushes wildfire season longer and blazes burn hotter across the American West and beyond.

With these smarter prediction tools, communities threatened by wildfires won't just react faster—they'll see danger coming before it arrives.