Original title: WFDroneBench: A Benchmark for Sensor Placement and Drone Routing for Wildfire Detection
Romain Puech, Danique de Moor, Ana Trisovic, Dimitris Bertsimas · September 2026
Wildfires are becoming more frequent and severe, but spotting them early is hard with current tools. Drones can cover large areas in real time, yet figuring out where to place charging stations and how to route flying drones across shifting fire zones is difficult. To tackle this, researchers created WFDroneBench, an open-source testing platform. It uses realistic fire simulations to test how well different methods place ground sensors and guide drones using fire-risk maps. The system tests plans across thousands of situations to see what works best. The researchers found that drones detect fires better than ground sensors alone. When fire-risk maps are accurate, an approach called Max-Coverage spots the toughest fires fastest. Setting up risk-aware ground equipment also helps, even if the risk map is imperfect. Finally, two big challenges remain: spotting small fires quickly and reliably, and creating better machine learning models to improve risk-map predictions.
Imagine you are a lifeguard at a huge beach and must decide where to patrol. You cannot see everywhere at once, and people sometimes get into trouble in unexpected spots. In this paper, the lifeguard is a drone and the beach is a forest; the benchmark tests whether patrolling near predicted high-risk areas catches fires faster than random wandering, and finds it does—but only if the predictions are accurate.
Wildfires are getting worse, and every minute of delay in detection costs lives and property. This benchmark helps researchers test which sensor placement and drone routing strategies actually work before deploying them in the field. It also reveals that current risk maps are not accurate enough, pointing machine learning straight at the most urgent problem—predicting where fires will start.
The benchmark is simulation-based and does not account for real-world factors like wind, drone battery constraints beyond charging stations, or communication delays. The risk maps used are from historical data; the abstract explicitly states a gap between these and true ignition patterns. The study does not test actual drone hardware or real-time deployment. The two open challenges it identifies—small fire detection and risk map accuracy—are acknowledged as unsolved.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
WFDroneBench: A Benchmark for Sensor Placement and Drone Routing for Wildfire Detection — Romain Puech, Danique de Moor, Ana Trisovic et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.