The fires went on for three relentless days in the summer of 2021, scorching over 13,000 hectares of western Sardinia. Residents “saw their whole world go up in flames around them,” says Carlo Poddi, a forestry expert on the island. Although they began with a roadside car accident that was immediately reported to firefighters, the conditions—temperatures over 40°C; an ongoing drought; and strong, hot sirocco winds blowing from Africa—made the blaze difficult to stop.

Fires like those are bound to become more common and vicious in a warming world, says Mr Poddi, walking through the forest of Santu Lussurgiu, one of the areas hit that summer. So the island is preparing.

As part of those preparations, Mr Poddi’s team at MEDSEA, a Sardinia-based environmental non-profit organisation, has installed 20 fire-detection sensors in the Santu Lussurgiu forest. These are part of a pilot programme by Dryad Networks, a German forest-monitoring company, that began in 2022 and is sponsored by Vodafone, a telecoms firm.

The sensors, which hang from branches three to four metres off the ground like green Christmas-tree ornaments, collect information about everything from concentrations of carbon monoxide and hydrogen to temperature, humidity, and air pressure. These data are then sent off to be analysed by bespoke artificial-intelligence (AI) models trained on data collected from forests around the world. If any anomalies are spotted, a call for action is sent to the emergency services.

Until now, fire detection on Sardinia has been carried out the old-fashioned way: by sight. During fire-risk season, from June 1st to October 31st, observers in lookout towers work to spot telltale smoke plumes snaking out of the greenery. But in most of these cases, says Carsten Brinkschulte, Dryad’s chief executive, the fire is noticed only once it has started to spread. Dryad’s sensors work more like a sense of smell, identifying the airborne chemicals caused by smouldering before flames and smoke are visible. This allows firefighters to intervene before the blaze is too large to control, says Mr Brinkschulte, and to have a clearer sense of where to tackle it.

The company’s sensors cost $100 per unit. They are designed to be installed at intervals of a few hundred metres in strategic locations where fires are most likely to occur. Since around 85% of wildfires are accidentally caused by humans, these high-risk positions include along power lines, hiking paths, roads and railways. This sort of strategic deployment can make sensors tens of times cheaper than alternative wildfire suppression methods, says Mr Brinkschulte. Once installed, the sensors can last for up to 15 years in the field and can have their firmware updated remotely.

Since every environment is different, and what passes for an anomaly in one region may represent business as usual in another, the AI model has to be tailored to incorporate factors such as local variations in wind speed, humidity, and temperatures. A system designed for an Italian forest would not work equally well for one in Canada, says Bogdan Diaconu at the Constantin Brâncusi University in Romania, who is not involved with Dryad. Mr Poddi says the ultimate test for Sardinia will be if the sensor networks can be deployed with equal effectiveness along the coastal pine forests near the island’s beaches. These environments—though equally susceptible to fires—are very different from those in Santu Lussurgiu.

Dryad has thus far run 50 pilot programmes to test its technology, from Spain to Indonesia, with positive preliminary results. In a pilot programme in Lebanon, for example, the sensors detected a small illegal fire caused by a farmer burning dry grapevines within 30 minutes of when it was started, whereas traditional sighting techniques would have taken several hours.

A handful of other companies around the world have had similar successes with analogous technology. Together with N5 Sensors, an American firm, Hamburg-based Breeze Technologies has installed sensors across forests in California. Their highly sensitive detectors are deployed at intervals of between two and five kilometres and can smell wildfires from within five to 15 minutes after the initial burn, says Robert Heinecke, the company’s boss. These sensors also measure polluting particulate matter (PM2.5 and PM10) in the air. Their results have been confirmed in projects throughout America, Canada, Germany and Peru.

To make a real difference in the effective early detection of fires, however, sensor networks like these will need to be installed in large numbers, says Mr Brinkschulte. Even so, the sensors will only be able to do so much on their own. Their real impact will depend on how well they are integrated with other existing firefighting measures, from mapping and data-analysis tools to the ability to promptly deploy firefighters. What’s more, says Dr Diaconu, the risks of time-wasting false positives from such sensitive systems are very real.

Overcoming such challenges will yield substantial rewards. Ankita Mohapatra runs a research laboratory at California State University in Fullerton that has also filed a patent application for similar fire-detecting sensor networks. “All the various technologies can and should work together to build a robust solution,” says Dr Mohapatra. Early fire detection will allow for quicker decision-making and communication with nearby towns that may need to be evacuated. This could save lives.

For now, Dryad has also received a European grant of €3.8m ($4m) and invested an additional €1.2m of its own to build an autonomous system capable of deploying camera-carrying drones to the site of sensor-spotted smouldering. The drones will deliver live feeds from above so as to help firefighters decide how best to intervene. “Eventually”, says Mr Brinkschulte, “we want to digitise the forest.”