Using AI to help prevent environmentally devastating forest fires

On 13 July 2021 a wildfire was ignited in northern California. The Dixie Fire, as it was subsequently named, burned about 390,000 hectares of woodland and countryside in a region spanning five counties, north of the state capital Sacramento. It was only extinguished in October 2021, by which time it had become the second most devastating wildfire in the state’s history. A subsequent investigation by the California fire department, Cal Fire, found that it had been started by a tree hitting a power line.

Wildfires are a significant contributor to global warming: the EU’s Copernicus Atmosphere Monitoring Service reports that they contribute about 1.76 billion tonnes of C02 into the atmosphere – which is more than double the annual CO2 emissions of the whole of Germany. They also cause significant but incalculable damage to wildlife and biodiversity. Wildfires also have other economic effects, including power cuts, destruction of communities, and a negative effect on the balance sheets of the utility operators who own the lines and the insurance sector.

Trees hitting power lines are the single most common cause of wildfires; but when a utility builds or operates thousands of kilometres of overhead cables, often running through forests in wild and unpopulated territories, it is a significant challenge to anticipate where the next tree-cable interface event will be.

That is where Roelof Pieters comes in. Pieters is co-founder and CTO of Overstory, an AI-based company that uses satellite data to analyse all vegetation on Earth, to prevent power outages and wildfires. “Our mission is to get a better overview of the world’s natural resources in real time, to enable anyone in the world who has access to that information to make better decisions about the problems we’re facing today, including the climate crisis and environmental problems related to that,” says Pieters.

It's quite a mission, and Pieters and his colleagues in Amsterdam and Boston, where Overstory has offices, use a combination of machine learning and human expertise to get to their goal on behalf of their clients, which are predominantly companies in the utilities sector.

Large utilities’ infrastructure runs through cities, forests, and numerous other kinds of vegetation. Historically it has been impossible to keep an eye on changes and risks throughout the length of this infrastructure. “With our satellite imagery, we can give a very detailed overview of the situation on the ground,” says Pieters. “That helps provide intelligence and scale, because we can do it from space, looking at these places on Earth. We’re not limited by accessibility as with drones or helicopters: we can go anywhere in the world and get that analysis.”

Machine learning can only provide 80 per cent of the solution. The remaining 20 per cent is provided by their own expert team of arborists and field experts, who further analyse data then visit the at-risk sites to make a human analysis. Pieters says that this hybrid AI-human model comes from his own training. Although he is now a leading developer of AI and machine learning in Europe, and a frequent keynote speaker at industry conferences, he studied social and cultural anthropology for his undergraduate and graduate masters degrees, before undertaking a doctorate in theoretical computer science.

He says the human expertise input into AI to produce a more effective solution is a relatively new and growing field, one which he is trying to help pioneer. Overstory’s clients are spread across the world. One utility, based in the US, has testified that as well as decreasing power cuts by 15 per cent, working together with Overstory has reduced the number of trees they have trimmed by 18 per cent, due to the precision of the company’s analysis.

As well as the utilities sector, clients and potential clients include the infrastructure sector: railway and road builders and operators who want to know how vegetation changes are affecting their routes, and where healthy trees in close proximity to infrastructure can be beneficial as they can prevent landslides and rockfalls.

“We also worked a lot with the Nordics in sustainable logging, understanding the health of the forest better,” says Pieters. “This not only allows your forests to be more healthy and for you to trace your impact, but also to have bigger timber gains for large, mono-yield timber forests.”

The insurance market is also a significant potential client, as is the fast-growing market for carbon credits. This field, which has been the subject of controversy because of its lack of transparency and question marks about the effectiveness of individual products, has great potential for the kind of hybrid AI-human analysis model pioneered by Overstory.

Pieters explains that corporates are understanding that it is not just a case of planting trees and counting carbon credits. “It can take up to 10 years just for all the inputs that are needed to decide how to start planting the trees, and then it takes (depending on the trees and also the area where you are planting) between 20 and 30 years for the trees to be grown up, and start to really have a good carbon draw down potential,” he says. Trees have other benefits, like contributing to biodiversity, holding water, and protecting against erosion. “But maybe we should just focus more on not cutting down trees, or stopping forest fires, rather than planting new ones.” The AI-led approach can help verification companies working with the growing carbon credit industry to track developments in existing forests within carbon credit schemes, he points out.

What drives Pieters? It is, he says, “about being motivated to make a positive impact. That can mean so many different things to different people”. Fortunately, he says, it is becoming an increasingly important driving force for a new generation that is both pioneering and driven.

Biography / key highlights

2012 Completes a masters degree in social anthropology at Stockholm University

2012 Becomes co-founder and director of R&D, Vionlabs

2014 Begins as a PhD candidate in theoretical computer science at KTH in Stockholm

2014 Founds Graph Technologies, the AI and Deep Tech consultancy

2016 Founds the Swedish AI network Stockholm AI

2016 Co-founds creative.ai as Chief Science Officer

2019 Founds the sustainable solutions consultancy Sunshine Lab, and later that year joins Overstory as co-founder and chief technology officer