Sending Smoke Signals

With many high-profile incidents occurring this season, wildfires are at the forefront of our minds lately. From Canada to Hawaii, numerous blazes have burned out of control, leaving widespread devastation and affecting the landscapes, ecosystems, and human communities in their paths. The damages they cause are extensive and multifaceted, impacting both immediate and long-term aspects of affected regions.

Human communities often bear the brunt of wildfire damages. Homes, infrastructure, and livelihoods are at risk as flames spread rapidly through wooded areas and urban interfaces. Evacuations become necessary, disrupting lives and causing emotional distress. In the worst-case scenarios, lives can be lost, and firefighters face dangerous conditions in their efforts to contain the flames.

The economic costs of wildfires are substantial. In the United States alone, the 2020 wildfire season burned around 10.3 million acres, with firefighting costs exceeding $2 billion. Immediate expenses include firefighting efforts, evacuation and emergency response, and property damage. Long-term costs encompass reforestation, habitat restoration, and infrastructure repair. In addition, the loss of tourism and outdoor recreational activities due to fire-damaged landscapes can impact local economies that rely on these industries.

Unfortunately, preventing all wildfires is not a realistic goal. From droughts to lightning strikes and volcanic activity, we simply cannot control all of the factors that lead to these blazes. This means we must be reactive, but there is still a lot of opportunity to do better. Presently, most wildfires are spotted visually. By the time a fire is detected in this way, it will already be sizable and difficult to extinguish, greatly increasing the odds that it will burn out of control.

Experts widely agree that bringing down response times could significantly improve outcomes. And that is the goal that Berlin-based Dryad Networks is working to achieve with their networks of AI-based sensors that can sniff out a forest fire in minutes that otherwise might not be noticed for hours. Once a wildfire has been recognized, the sensors can wirelessly communicate with firefighters in real-time, even in remote regions of forests by using a clever networking technique.

The roughly human hand-sized sensors are hung on trees, with each sensor being able to monitor an area about the size of a football field. The devices leverage gas sensors to look for miniscule quantities of gasses like hydrogen and carbon monoxide that could indicate the presence of fire. Each unit is charged by a solar cell so that it can operate off the grid. To avoid the risk of starting fires themselves, the sensors store power in capacitors rather than the more common lithium-ion batteries found in most portable electronics.

Exactly what chemical signature indicates the presence of a forest fire is difficult to define, so the sensors run machine learning algorithms that have been trained to recognize the relevant patterns. Moreover, the indicators vary from location to location. For example, some areas may be frequented by passing trucks that release exhaust fumes, where that would not be the case in other regions. To avoid false positives in such situations, the devices continually update their machine learning model to fine tune it for the location that it is in.

When a wildfire is suspected, the sensors cannot just connect to the nearest cell tower to phone home in most cases due to their remote locations — most cell towers only cover a radius of a few miles. Rather, the sensors form a mesh network in which they communicate with one another, passing the message from one to the next. In this way, the message will quickly make its way to the forest edge, where it can then be transmitted via traditional cellular networks to inform emergency responders of the situation.

Being alerted to the exact location of a suspected fire within minutes gives firefighters the opportunity to extinguish the fire before it has a chance to rage out of control. The technology has recently been rolled out in regions of the redwood forests of northern California, so we will learn how effective it is in the real-world in the years to come.