SIParCS 2026 - Maximus Jessey

Firecracker: Wildfire Modeling and Leveraging AI for Strategic Wildfire Suppression

Recorded Talk

In the U.S. alone, hundreds of billions of dollars are spent every year on wildfire management. As climate change increases the frequency and severity of wildfires, the need for more effective strategies to both suppress active fires and reduce their long-term impacts has become increasingly urgent. In this work, we present Firecracker, a research environment featuring a simplified, yet expressive, fire model that captures many of the key complexities of real-world wildfire behavior, together with a suite of tools for machine learning. Inspired by successes in game-playing artificial intelligence, we use this environment to train a deep learning model that employs an internal learned world model for planning, enabling it to identify effective locations for placing fire blockers within the simulation. We highlight this as a proof-of-concept that planning-based machine learning agents can learn strategies that successfully contain large-scale simulated wildfires, providing a foundation for future research on AI-assisted wildfire management, planning, and decision-making.

Mentors: John Schreck

Slides and poster