Ask a Scientist: How can researchers use AI to spot a wildfire?

Google Research has spent over a decade investigating how artificial intelligence and machine learning can identify early-stage wildfires and predict their spread patterns. The team collaborated with Muon Space to support the Earth Fire Alliance’s FireSat program, which involves a new satellite constellation designed specifically for wildfire detection and monitoring. These satellites will use AI to distinguish small fires from false alarms such as backyard grills or industrial heat sources, providing critical early data to firefighting teams.

Traditional satellite imagery often lacks the resolution or timeliness needed to spot small, fast-growing wildfires, with data sometimes being hours old or too coarse. To overcome this, the FireSat satellites will operate in low-Earth orbit, capturing images every 20 minutes to enable near real-time monitoring. The system aims to detect fires as small as 25 square meters—about the size of a single car—by combining lower-cost, smaller satellites with advanced AI models trained on controlled burns and real-world fire scenarios.

The AI models are designed to differentiate between controlled burns, agricultural fires, and actual wildfires by analyzing how fire scenes evolve over time. This ground-truth data collection helps refine detection algorithms and prioritize alerts that matter most to first responders. The satellites’ frequent imaging allows firefighters to forecast fire behavior and respond more effectively, potentially preventing rapid spread before it becomes unmanageable.

Researchers emphasize that the primary benefit of this system is time—giving firefighters the earliest possible warning to contain fires before they escalate. The initiative also aims to create a comprehensive dataset on wildfire activity, which will be shared with the scientific community and authorities to improve long-term fire management strategies and understanding of local fire risks.