A system of tower mounted cameras called ALERTWildfire gives early warning alerts of smoke and flames.

The 37,000-acre Tubbs Fire, in 2017, was the most destructive wildfire in California’s history. According to Sam Wallis, a community alert and warning manager in Sonoma County, California, it was “an absolutely massive and fast-moving fire that we had no way of tracking.” But he now has help from ALERTWildfire, a system of artificial intelligence and sophisticated cameras to help him tackle fires before they rage out of control. (SCIENTIFIC AMERICAN)

ALERTWildfire

Sonoma County has in place 21 high-powered devices that send images every 10 seconds from areas vulnerable to wildfires. These images are displayed on a bank of video monitors in the county’s fire emergency centre, also fields 911 calls from citizens.

Since it is humanly impossible to cope with the sheer volume of imagery that is coming through, and because of the risk of error, the county has now deployed AI software that is linked to the cameras.

Effective May 1, AI software now analyses all the incoming images and compares them almost instantaneously with historical photographs of the same location. The system alerts the emergency centre in the event it finds a disparity.

Graham Kent, Director, Nevada Seismological Laboratory at the University of Nevada, Reno, developed the ALERTWildfire system. The objective is to flag potential fires much earlier for appropriate firefighting action.

It is observed that the AI system is ahead of a 911 call on a fire warning by as much as 10 minutes. This is not a small margin given how fast wildfires can spread, and how much damage they can inflict if not stamped out immediately.

Though the system is improving steadily as emergency dispatchers become more familiar with it, the AI algorithms still have considerable learning ahead of them, including experience from at least 70 real-world blazes.

Sometimes, the system is still vulnerable to false positives – it once sounded the alarm on steam from the Geysers, a local geothermal field.

However, Wallis is confident that ALERTWildfire will become consistently dependable by November.

Related Story:   Satellites and AI Could Together Predict Wildfires Accurately