In San Francisco, it took a team of arborists a year to map every street tree in the city–124,795 in total. In New York, a similar tree count with volunteers took nearly two years. A new algorithm, by contrast, uses AI and satellite data to map a city’s tree canopy in just hours.
“You can use either aerial imagery or satellite imagery to do basically the same task, but a lot faster,” says Aidan Swope, a Caltech undergrad who created the algorithm as an intern at the tech startup Descartes Labs. Because taking a census by hand takes months or years, some trees are inevitably cut down before it’s complete, so the final map won’t be completely accurate. And these censuses typically also only include street trees, not trees in parks or on private property, while the algorithm includes everything.
The tool uses a convolutional neural network, similar to those used for facial recognition. While it’s not hard for a machine to find green areas in an aerial image, Swope also trained the model with lidar data, a type of remote sensing data that shows height, making it possible to distinguish trees from grass or other plants.
In a city map, the result shows dense sections of green in some areas and treeless gaps in others. The tool can’t replace every aspect of counting trees by hand–in San Francisco, for example, the team of arborists carefully noted the species and condition of each tree, which the city is using as it plans how to maintain them efficiently. But it could help with planning efforts. “If you’re looking at trying to identify things like canopy coverage goals, or simply where do we need more trees in our city, you don’t need to know the condition of every tree to do that,” says Carla Short, superintendent of the Bureau of Urban Forestry at San Francisco Public Works. “That’s where having that kind of 30,000-foot view is really, really helpful.”
It could also help cities prioritize where to send humans for a tree census, especially in places without the budget for a detailed count like San Francisco’s. Though the Descartes Labs algorithm was made simply as a demonstration of its technical capabilities, and isn’t commercially available, the company would be interested in working directly with other cities. It’s one of a handful of tools designed to speed up the process of counting trees; another tool uses Google Street View.
Cities have increasingly ambitious plans for tree planting. In Italy, Milan plans to plant 3 million trees over the next 12 years, something that can fight climate change by absorbing an extra 5 million tons of CO2 each year, clean the city’s polluted air to improve health, and lower temperatures by 2 degrees Celsius. In the U.K., Manchester plans to plant a tree for every resident. Melbourne plans to double its canopy cover by 2040. New York City is planting 1 million trees. Large-scale tree planting is also happening at national levels–Pakistan, for example, recently met a goal to plant a billion trees, and plans to plant 100 million more. The U.K. plans to plant a new forest with 50 million trees. And China deployed 60,000 soldiers in 2018 to plant trees on more than 30,000 square miles of land.
Urban trees have a long list of benefits: Living near more trees makes you happier, improves health, and even makes people feel as healthy as if they were seven years younger (or $10,000 richer). In Dallas, tree planting is helping fight rising heat. Urban trees can store almost as much carbon as tropical forests, making them an important tool in the race against climate change.