Machine learning might be able to tally trees that human labor simply can’t
In 2017 the San Francisco Planning Department finished a long-anticipated census of trees in San Francisco, and even launched an interactive map for identifying nearby leafy neighbors.
It’s a great tool, but the city only counted street trees on its Open Tree Map. The count didn’t include the likes of public parks, which meant that the majority of trees in San Francisco remained unaccounted for.
A more complete census would take longer—longer perhaps than anyone has time and resources to dedicate to the process of tallying trunks.
Why would anyone really care that much?
Geographer Tim Wallace explains in a Medium article: “Urban trees reduce crime and help stormwater management. […] They make oxygen for breathing, suck up CO₂, provide shade, reduce noise pollution, and just look at them — they’re beautiful.”
And as CityLab points out, “Planting trees has long been a low-tech strategy to fight the effects of climate change and the urban heat island effect. [Trees] help reduce stress, they’ve been linked to the lower obesity rates, and may even curb pedestrian deaths.”
Wallace says that manually counting trees across a city is labor intensive. NASA uses satellite data to try to estimate vegetation canopy at specific locales:
That’s a useful tool, but still not something that can easily provide an actual tree count. So Wallace is pushing Santa Fe, New Mexico-based satellite data company Descartes Labs’ new approach instead, trying to use artificial intelligence and machine learning as an arboreal abacus.
Supposedly, Descartes’ tech can read satellite images and other high-res scans and pick out which green bits are individual trees and which aren’t.
Whether it actually works is hard to suss out at first glance. However, the graphical results are pretty striking; first we see a map made using the city’s street tree census, and then one using Descartes tech to fill in the blanks:
For a look at the complete canopies of other major U.S. cities, go here.