For more than a decade, Google Research has been working to use artificial intelligence and machine learning to detect wildfires in their early stages and model how they spread. This work is currently focused within the FireSat program, which Google supports in cooperation with Muon Space for Earth Fire Alliance, with the goal of building a satellite constellation that monitors fires from space before they turn into widespread disasters.
The company says that, once complete, the system will be able to detect a fire covering 25 square meters, roughly equivalent to a fire covering an area of 5×5 meters or the size of a single car. The role of artificial intelligence is not limited to detecting a heat source; it is also expected to help distinguish an actual wildfire from sources that may look similar from orbit, such as light-reflecting clouds, factory chimneys, barbecues, and controlled agricultural burns.
Why Was Space-Based Monitoring Difficult?
Firefighting authorities need fast information when a fire is small and containable, but some currently available satellite imagery may arrive after about 11 hours or have low resolution. This makes the data less useful when confronting a rapidly spreading fire.
The traditional alternative was to use a large, high-resolution satellite that monitors Earth for a long period, an expensive option with limited coverage. FireSat, by contrast, relies on a larger number of smaller, lower-cost satellites, while using machine-learning models to compensate for part of the resolution gap and analyze changes in scenes over time.
How Does the Model Learn to Distinguish Between Fires?
Google developed a dedicated camera to improve wildfire monitoring and trained its initial models using aerial flights over controlled fires. It also collected images from California of scenes that look like fires from above, then tracked their changes over time to identify cases that develop into real fires and those that represent false alarms.
Google aims to build a reference record showing which small fires develop into large fires and which do not. These data would improve the algorithms and reduce unnecessary alerts that could overwhelm response teams, while keeping the focus on the most dangerous fires.
What Changes in Practice for Response Teams?
The project's main practical value is reducing the time between a fire's start and its detection. Repeated imaging will help firefighting teams spot a fire while it is still small, and may also provide better information about its rate and direction of spread. The data could also support long-term planning, such as identifying suitable locations for creating firebreaks.
However, this benefit depends on completing the constellation, the accuracy of the models, and their ability to handle differences in local conditions and similar heat sources. The announcement also describes future operational goals, not complete global coverage at present.
Timeline and Next Goals
The system launched its first batch of operational satellites, while the next batch is expected to arrive during the coming year. Google says reaching full operational capability will take several years. The goal of imaging the world approximately every 20 minutes will require about 50 satellites, with that target expected to be reached around 2030. The nearer-term goal, over the next two years, is to image the entire surface of Earth once every hour.