Wildfires are getting worse. How can machine learning help reduce the damage?
Why we invested in Overstory
The world is on fire, and like it or not we’ve all got a front row seat.
Since long before the United States was created, people in North America have been living at the edge of, and often right in the midst of, nature. But things have changed. Wildfires have now increased at pace, the sky is turning Blade-Runner red with growing frequency, ecosystems are being destroyed, and many people are starting to grieve, admitting—as New York Times journalist Elizabeth Weil did of the West Coast—‘this is not the California I married’.
It is, however, the California we now have.
It’s also the Australia we have. It’s the Portugal we have. It’s the Russia, Canada, and Brazil. South Africa, China, Greece, Spain as well. Wildfires impact—directly and indirectly—every part of the Earth. Wildfires not only destroy carbon-sequestering trees, but they also release carbon dioxide, making every forest fire a double loss for the planet. We need to better understand the catastrophic wildfires sweeping the globe, and to find scalable, sustainable ways to prevent them.
But first, what’s causing all the wildfires?
Fires have been a natural part of life for a long time. Indigenous people have lived in fire-prone ecosystems for centuries, and fires have actually played a useful role in the past. Indigenous populations used fires to help release nutrients, aid seed dispersal, and to clear land for agriculture, hunting, and travel. But trees are not built for this rapidly changing climate, and as the world’s temperature increases, so does the impact of uncontrolled wildfires.
Fires today have become more extreme and they are now larger and burn hotter than in the past. The twenty largest wildfires in Colorado have all occurred since 2001. Hotter, drier weather creates conditions conducive to much more damaging fires like the bush fires we saw in Australia. Hot weather also stresses trees, stunts growth and makes them vulnerable to parasites and fungi: branches fall more easily, and more frequent and high intensity fires are allowed to rage.
But what ignites the fires in the first place? It’s not just human error like cigarette butts or gender reveal parties. They’re also started by events like lightning or human-made infrastructure: namely, power lines. Power lines can start fires in several ways: if an electrified cable breaks in the wind and hits a branch, if a branch falls across the line, or if one line collides with another. In California in 2015, data showed that fires caused by electrical lines and equipment burnt through more acres in the state than any other cause.
We need to address this because power lines aren’t going anywhere. In fact, we’re going to need more of them very soon.
A decarbonised future means more power lines
As we start to decarbonise industries, we will need to rely more on electrification and renewable energy sources like solar and wind. This requires infrastructure. To access this energy, we need to move power from where it’s generated to where it is used—which means more power lines. In the US alone, high voltage transmission across the country must increase by roughly 60% in order to meet net-zero targets by 2050, and tree branches falling are a big threat not just to wildfires but to grid reliability and people losing power.
It’s also a huge cost. Vegetation management is one of the largest recurring expenses of the utility industry, costing billions of dollars every year. Falling trees and branches are one of the top causes of power outages, and in the US those power outages cost the economy up to $70 billion every year.
The search for solutions
Though wildfires can be sparked by anything from lightning strikes to arson, those fires sparked by electric utilities tend to be exponentially more devastating, burning larger areas than other fires on average. This is because they tend to occur on remote and challenging terrain, unwitnessed, especially during weather events conducive to fire growth.
Power outages and wildfires cause such tremendous social and economic problems that utility companies are seeking drastic solutions. During extreme weather they might use preventative measures like power shut offs, but this creates other economic and social ramifications. Power shut offs disproportionately affect marginalised communities and individuals with disabilities who depend on electricity for emergency communication, evacuation, and sheltering.
For years, electric utilities have relied on the same techniques to manage vegetation. They use annual trimming cycles, foot patrol, helicopter flights, and reports of concerned citizens to help manage the vegetation across their networks. Some have spent billions of dollars per year on vegetation management, dispatching work crews via email and text to follow up on reports of trees too close to power lines—often only to find that the areas didn’t need immediate action. Utility companies need modern tools to keep up with the increasingly severe weather.
The magic of predictive vegetation intelligence
The problem with trees is that they can’t do much about their predicament. They can’t change the climate, and they can’t cry out for help, so it falls on us to find a better way to stop them from being swept up in smoke and falling on power lines. But we need to do it quickly, not manually, as trees and vegetation are constantly changing and we need relevant data to be able to respond. So far it hasn’t been easy to find the right data and share it with people who need it when they need it.
Overstory, one of our portfolio companies, has a solution. They use different types of satellite imagery—multispectral, hyperspectral and synthetic-aperture radar (SAR)—, process it with their machine learning algorithms, and then turn those actionable insights into trimming recommendations for their customers to help them prevent outages and wildfires.
Their customers, utilities and enterprise organisations based everywhere from the US to Finland to Africa, have projected tremendous results: a 13x return on investment and a 40% improvement in resource efficiency.
Because Overstory uses best-in-class satellite imagery with a resolution of 30cm2, they can detect anything larger than a standard squared ruler. This allows them to capture granular tree-level data on health, height, and proximity to power lines. They also partner with their clients to incorporate third-party data layers and local expertise for things like fast-growing species, high-risk pests, and soil health relevant to their network. Overstory is truly global. Their intelligence has been trained and applied on more than 2 million square miles of forest in 64 countries.
According to the latest IPCC report, vegetation intelligence is one of our best bets—both in economic potential and emission reduction—for fighting the climate crisis. Overstory is building the platform and predictive models to help us do just that.
By partnering with clients, wildfire experts, arborists, and foresters, they have been able to scale hard-won local expertise—all with the scope of satellites and the speed of machine learning. It’s an important step on their journey to providing vegetation intelligence to businesses and communities around the world.
Climate change is the most important challenge of our time, and it demands that we innovate to find better ways to take care of our natural resources. Overstory is now fast scaling their business and tech teams, and as the team at Overstory grows, so does their vision to use technology to protect and restore natural ecosystems. Beyond just responding or reducing damage, Overstory is offering the predictions we need to get ahead of the growing threat of catastrophic storms, drought, and wildfires. In doing so, they’re protecting the California that some of us married, and the world that so many of us have fallen in love with.
Or to check out all the Pale blue dot portfolio roles head here.
Looking to go deeper into the science behind forestry? 🌳 Check out our public research on forestry here.