Scientists Built AI to Design Smarter Wildfire-Fighting Aircraft Fleets

Designing a fleet of firefighting aircraft is a nightmare of choices. How many planes? Which types? Which carries water, which carries scouting cameras, which flies at night? Every combination creates new knock-on decisions, and testing each one in a detailed flight simulation eats hours of supercomputer time — and sometimes the simulation simply crashes. It's like designing a whole restaurant menu by cooking every possible dish before deciding what to serve.

The researchers' fix is a technique called hierarchical Bayesian optimization. In plain terms: instead of testing every design, the AI tests a handful, spots patterns, and then makes educated guesses about the ones it hasn't tried — getting smarter with each round. The "hierarchical" part matters because aircraft choices come in layers. You can't pick a water-tank size until you've decided to include a water-dropping plane at all. The method handles those stacked decisions naturally.

They tested it on a real scenario from COLOSSUS, an EU-funded project on wildfire response. The setup involved mixed aircraft — some spotting fires, others dropping retardant — each with different speeds, costs and abilities. Working together, they cover far more ground than any single plane could. The new method found strong fleet designs using far fewer simulations than standard approaches, and it held up even when some individual simulations failed.

The catch: this is still computer modeling. No drones were built, and real wildfires — smoke, wind, rough terrain, tight budgets — are messier than any simulation. The team also notes this is among the first practical uses of this method on a real-world problem, so it's early days. Still, the same trick could help design delivery networks, disaster-response fleets and greener air travel.