By the time that someone discovers that illegal loggers are cutting or burning down trees in a remote forest, it’s usually too late to do anything about it. As much as 58,000 square miles of forest–an about area the size of Illinois–disappears around the world each year.
Now, however, it’s possible to discover threatened areas just before deforestation actually happens. Using artificial intelligence, a startup called Orbital Insight is building a new system that can scan through satellite images of forests to detect suspicious changes–both things that might be visible to the eye, like new roads, and other more subtle changes that humans might otherwise miss.
“The networks just look for patterns,” says James Crawford, CEO of Orbital Insight, who previously led Google Books, using similar AI techniques to scan through 20 million books and make them searchable. “We’ll give the networks a bunch of examples of how the forest changed in look before deforestation, and the neural networks will be looking for patterns that occur,” he says. “When we look at them by eye, what we see is roads going in. But one of the really cool things about deep learning is that the neural network itself will be looking for what kinds of systematic changes occur leading up to major cutting.”
The results will be published on Global Forest Watch, a website from the World Resources Institute that aims to track deforestation in real time. “The method they’ve been using is all about transparency,” says Crawford. “So they don’t do any enforcement. But they put the results up on well-designed, very public websites. So what you find is that local folks, and people around the world who care about this stuff, use it as a primary source.”
For Orbital, which currently uses its technology to help hedge funds and corporations predict global trends, the partnership with Global Forest Watch is a chance to test out using the tech for good. “We realized that there are a number of really interesting humanitarian options,” Crawford says.
Eventually, the company may also begin to analyze things like the health of crops to see how that correlates with global poverty. “That could provide a way to target aid or otherwise improve farming practices,” he says. The tech could also analyze building types–whether roofs are made from wood, or mud, or steel, for example–as another way to potentially create large-scale poverty maps.
“The interesting thing is that you can see so many things in this imagery,” Crawford says.