Gendrome Editors' Note: The article below provides the raw material for a proof and is not the proof itself. In addition, the raw material may contain one or more false statements and/or some offensive, outside content.
How many trees are in your city?It might seem like a straightforward question, but finding the answer can be a monumental task. New York City’s 2015-2016 tree census, for example, took nearly two years (12,000 hours total) and more than 2,200 volunteers. Seattle’s tree inventory won’t be complete until at least 2024. Such efforts aren’t done in vain; in the short term, they allow cities to better maintain their urban trees. And over the long run, they lay out the foundation for various initiatives that address everything from climate change to public health.So to make the task of counting trees easier, a team of cartographers and applied scientists at geospatial analytics startup Descartes Labs is turning to artificial intelligence. In their quest to leave no tree uncounted, they built a machine learning model that can map an entire city’s canopy, even subtracting other greenery that might look like trees in satellite imagery. The resulting maps reveal a green thumbprint of each city—like this one of Baltimore and its surrounding leafy suburbs.A map of Baltimore’s tree canopy. (Tim Wallace/Descartes Labs)The challenge of mapping trees comes from several factors. On the ground, the human eye can easily distinguish a tree from the rest of the urban landscape. But the inability to access private areas, or places guarded by tall fences, means some trees don’t get counted. Mapping trees from above should solve that problem; the Normalized Difference Vegetation Index (NDVI) derived from satellite imagery has long been a reliable survey of a city’s greenery. Even so, there are limitations.“Often when [the New York Times] needed to map things like trees, they get lumped in with other types of vegetation like grass or crops,” says Tim Wallace, a former cartographer for the newspaper who now works at Descartes Labs. The NDVI detects vegetation by measuring the distinct wavelengths and near-infrared light reflected by all plants, which means it can’t tell the difference between trees, grass, bushes, and other kinds of greenery.What is notably different among those types of greenery are their heights; trees are obviously taller than shrubs and grass. And that can be measured using LIDAR data—essentially shooting a light from a drone or airplane at those plants, and recording the length of the light that bounces back up. Kyle Story, an applied scientist at Descartes Labs, says this “third dimension” is crucial. But collecting LIDAR data for any city is expensive because of the costly equipment involved. Luckily for his team, there are plenty of publicly available datasets that can be used to train their machine learning model.The tree canopy of Boston. The map made using machine learning is lighter as grassy areas, like Fenway Park, disappear. (Tim Wallace/Descartes Labs)“Using the NDVI and the LIDAR, those two datasets can tell us where trees are in an area. If there are satellite pictures, we can train an algorithm to say, ‘Okay, looking at that imagery, I can learn what trees look like,’” Story says. “Once you’ve trained that algorithm, you can run it anywhere you have satellite imagery, because you’ve taught your machine to differentiate them from bushes and grass.”Wallace says the team has run the algorithm in over 2,000 cities so far. And according to chief marketing officer Julie Crabill, the company is hoping to talk city planners, as well as businesses and nonprofits, about implementing the technology in tree counts and other projects.Cities’ tree counts are more than just a good bit of trivia. Urban development in the U.S. means more cities are losing tree cover—often where and when it’s needed most. Planting trees have long been a low-tech strategy to fight the effects of climate change and the urban heat island effect. Aside from that, trees are a boon for public health. They help reduce stress, they’ve been linked to the lower obesity rates, and may even curb pedestrian deaths.A map of tree cover in New York City reveals where the “tree deserts” are. (Tim Wallace/Descartes Labs)Yet lower-income and minority neighborhoods that are most vulnerable to such environmental and health stresses tend to have the least tree cover. So by having an accurate map of where the leafy and barren neighborhoods, and in a timely manner, allow local government to better target tree-planting initiatives.That’s not to devalue the work of researchers, tree experts, and volunteers who are still ultimately needed to paint an accurate picture of a city’s urban canopy, though. Like most algorithms, this one isn’t perfect—it has picked up shadows cast onto buildings as trees, for instance. It can provide a broad overview of the tree population, but gathering more granular data will still require more work.“It takes time—and humans—to go out and classify these trees based on fields like how tall they are, what their diameters are, what species they are, and whether they're healthy or not,” says Wallace. “All of those details are not grabbable from this machine learning technique.”What it does do is allow the researchers and volunteers to jump into that deeper data collection faster, Wallace says, by automatically answering the most basic question: “Where the heck are the trees?”You can see more maps on Wallace’s Medium post.