On the hunt for cheap and fast strategies to prevent deadly wildfires, utilities across the US and Europeare cont
racting with a handful of artificial intelligence startups to map wildfi
re risk along thousands of miles of power lines, picking out individual trees to cut and poles to replace.
The most obvious way to prevent power equipment from sparking a blaze is to bury lines underground. But at a cost of upwards of $3 million a mil
e, many investor-owned utilities are limiting themselves to just a few hundred miles of buried lines per year. T
hat’s proved to be a major business opportunity for several tech companies that use machine learning to deliver
custom recommendations for targeted, relatively low-cost fixes — including some that can be paid for from maintenance budgets.
Take Massachusetts-based Overstory, which analyzes high-definition satellite images to spot trees that are m
ost likely to topple or drop a limb on power distribution lines —
those most in need of trimming. The company started out making tools to find and discourage deforestation, said Chief Executive Officer Fiona Spruill. Overstory
pivoted after a realization: “We could have a huge impact from a climate standpoint, in that we could help prevent wild
fires,” she said. The company recently closed an oversubscribed Series B funding round that pulled in more than $43 million.
Watch More Image Part 2 >>>
Overstory’s clients now include American Electric Power Co.’s Texas utility andCalifornia’s Pacific Gas & Electric Co., al
ong with other major US utilities, Spruill said. UK-based National Grid P
lc, meanwhile, has signed a deal with Rhizome, a San Francisco-based climate tech firm, for custom modeling to suggestinvestments in fire safety.
The stakes for utilities are high: Edison International’s Southern California utility is facing lawsuits for its po
ssible link to the Eaton fire in Altadena, which killed 19 people in January. Wildfire liability from deadly blazes in Norther
n California pushed PG&E to declare bankruptcy in 2019.
In the years since, PG&E has worked with Overstory’s staff to map the precise location and health of individual
trees across its vast service area, which stretches for more than 500 miles (805 kilometers) in California. The premise
is that it’s cheaper and more comprehensive than sending utility workers
to inspect trees from the ground — or deploying helicopters to spot problem areas from the air.
To train its AI, Overstory relied on a large team of arborists who contributed on-the-ground measurements a
nd observations of diseased or dying trees. The company’s data an
d risk predictions are loaded into a custom web platform, with suggestions for what to trim and how often.
Rhizome co-founder and CEO Mishal Thadani said his company set out to help utilities decide which fire risk interventions — from cutting trees to replacing flammab
le wooden poles and installing new safety equipment — will deliver the most benefit for the cost. Rhizome’s gridF
IRM model can parse an array of site-specific information about
a given grid, including local weather trends and maintenance records, and factor in information about nearby properties and populations that would be impacted in a fire.
National Grid plans to start using the model soon in the UK and already employs it on grids in New York and Massa
chusetts. Those states may not traditionally be thought of as fire prone, but the amount of land area burned within their borders more than tripled from 2023 to 202
4, according to data from the US National Interagency Coordination Center, a federal hub for tracking and managing wildfire response.
Rhizome’s technology has already rewired some of the utility’s assumptions about its fire risk, according to Casey Kirkpatrick, National Grid’s director of group strategic engineering. Previously, National Grid assumed it was worse in areas near bigger cities with more fire-ignition sources, along with sections of the grid near overgrown forests. Some of the biggest risks, however, lurked in suburbs.