Every firefighter knows the equation: every minute saved in detecting a fire means several hectares of forest spared and far less water used to extinguish it. To illustrate, putting out a fire covering one hectare requires around 22,000 litres of water — a resource that has become precious in the face of recurring droughts.
This is precisely the issue that the SDIS 77 is working on in Seine-et-Marne. “The aim is to detect the outbreak of a fire as early as possible, so that we can deploy as few firefighters as possible and use as little water as possible,” explains Commander Paul Laurain, the departmental forest fire officer at the Fontainebleau Fire and Rescue Centre.
The idea: turning telecoms masts into 2.0-style watchtowers
The system relies on the infrastructure of TDF (formerly Télédiffusion de France), a telecoms infrastructure operator in France. The idea is as simple as it is ingenious: to install smart cameras at the top of existing pylons that continuously scan the tree canopy.
The trial began on the TDF pylon at La Croix d'Augas, which rises to 45 metres above the Fontainebleau forest. To put the algorithms to the test, Two controlled burns were set at 5 and 10 kilometres of the site in order to assess the actual performance of the detection systems.
The choice of telecoms masts is not insignificant. Their height provides an unobstructed view for several kilometres in an environment where there are numerous visual obstacles. An additional advantage for manufacturers is that this infrastructure already has a reliable power supply and secure access procedures, which significantly speeds up deployment.
Today, the system is being expanded: eight cameras, spread across four TDF pylons, now monitor the forest, covering a significant part of the Fontainebleau massif, which spans nearly 25,000 hectares.
Pyronear: the open-source AI that can detect smoke
The driving force behind the system is the French association Pyronear, the project’s technical partner. Founded in 2019 as part of a ‘Data for Good’ programme, this collective of volunteers made a radical decision: to develop a solution entirely open source, which can be reused free of charge by any SDIS.
The system is deliberately simple. The detection towers consist of four to five high-resolution cameras and a microcomputer which captures one image per camera at regular intervals — every 30 seconds — and then analyses it locally using a forest fire detection model. If a fire is detected, alert mode is activated, and all images from the camera that detected the fire are sent to the database via an API.
The AI model — trained on thousands of images of smoke plumes — works locally on a standard Raspberry Pi, making it a solution energy-efficient and can be deployed even in remote areas. Furthermore, a triangulation method makes it possible to pinpoint the source of the fire as soon as at least two cameras detect the same smoke from different angles.
A paradigm shift: humans as sentinels, AI as lookouts
The system marks a methodological shift. Historically, the forest, covering nearly 25,000 hectares and known for its centuries-old oak trees and rock formations popular with climbers, was monitored during the summer months by students stationed at the top of wooden watchtowers, from sunrise to sunset. This seasonal surveillance was limited to daylight hours and relied on human vigilance.
AI is a game-changer. “Humans will take on a passive role in the future: artificial intelligence means they won’t have to constantly monitor the footage,” explains Commander Laurain. In practical terms, when a suspicious event is detected, an alert is automatically generated and verified by the SDIS teams before any response is initiated. The algorithm does not replace the firefighter — it provides them with a preliminary assessment to be confirmed.
A roll-out that forms part of a disastrous year in 2025
This trial is no coincidence. According to the National Forestry Office (ONF), France experienced a particularly severe fire season in 2025, with around 20,000 hectares destroyed, 15,000 fire outbreaks, of which 1,800 were classified as major forest fires. The most devastating incident occurred in the Corbières mountains at Ribaute (Aude) in August 2025, resulting in the loss of more than 11,300 hectares of vegetation, one fatality, 42 people injured and 52 homes destroyed.
Across Europe, 2025 was a record-breaking year: more than one million hectares went up in smoke in the European Union, mainly in Spain and Portugal. In this context, every minute saved through early detection can make the difference between a fire being brought under control and a disaster.
A new role for telecoms infrastructure
Beyond firefighting, the Fontainebleau experiment opens up broader prospects. The high points operated by TDF — historically used for digital terrestrial television (DTT), radio and mobile telephony — could eventually host other environmental monitoring devices: air quality monitoring, landslide detection, wildlife monitoring and the measurement of urban heat islands. Telecoms masts are becoming multi-purpose sensors for the climate transition.
Other organisations are already active in this field: the start-up ADCIS (FireTracking), the European MidGard project, and the trials being carried out by the ONF in the Île-de-France region. But Pyronear remains unique in one respect: its code remains open-source, free of charge and adaptable by any local authority.