Now, after four years of punishing drought, the city badly needs a fix on the condition of its urban forest.
But doing that the old way — by sending out tree counters with clipboards — would cost about $3 million, money that’s already committed elsewhere.
So the city Bureau of Street Services has called on Caltech for help. And Caltech has called on Google.
Pietro Perona, an academic who hopes to become one of the world's most prolific urban tree counters, has developed a method using Google Earth and Google Street View to let a computer do the counting.
For Mayor Eric Garcetti, a partnership with Caltech offers a potential example of his campaign promise to apply smart technology to the city’s nagging problems.
There’s just one hitch: Google’s terms of service allow academics free use of the company’s vast library of images, but not cities. The city has been in negotiations with Google for more than two months. In an email, Google said it had no comment on the negotiations.
In the meantime, Perona, whose specialty as Caltech’s Allen E. Puckett Professor of Electrical Engineering is making machines see, is busy teaching his computers to see all sorts of things.
Perona works with global communities that post digital images of things online. He renders the images into geometric algorithms stored in what he calls a “visipedia.”
Theoretically, once he has trained a computer to see birds, anyone could upload a picture to a website that would determine its species.
One day, everything under the sun will be in a visipedia, reptiles and mushrooms included.
Perona’s tree project is a panoramic version of that technology.
Using the downloaded satellite view, Perona is teaching computers to record the latitudes and longitudes of shades and shapes that look like trees.
He then matches those coordinates to an image on Street View.
The computer examines the closeup to confirm the presence of a tree and judge its variety.
It takes a human years to learn the taxonomy of trees, Perona said. And it would take dozens of humans months to apply that knowledge to the city’s trees.
But a bank of computers could do it overnight, Perona said.
“Thirty-two CPUs. Sixteen square miles for one CPU,” Perona said.
Perona has tested the methodology in a section of Pasadena where the city recently commissioned a sidewalk survey. By comparing the results to the known inventory, he determined that the computer was about 80% accurate.
As an academic, Perona wants his algorithms to recognize every tree variety. But that may never be practical.
That’s because tree varieties are not spread evenly. There are about 600 varieties in Los Angeles, and a third of them have fewer than 100 specimens.
The corollary is that only 85 varieties make up 90% of all trees. Teaching the computer to identify those could be a big help to the city.
The initial benefit will be the macro picture, said Greg Good, Garcetti’s executive officer for city services.
“It’s important to understand where the gaps are in terms of canopy,” Good said. “We want to use this inventory to identify storm water capture opportunities, ameliorate urban heat island impacts.”
It’s also valuable that, unlike a human count, the survey can be easily replicated and will get better with time.
“We will have inspectors going out for the next 20 years validating and correcting the data from this model,” he said.
The hard part is training the first computer.
The instruction is a very hands-on human affair. Someone has to pore over samples of the Google images and grade the computer. Is it actually a tree, or a lamppost? Is it oak, as the computer judged, or something else?
Perona is getting help from the TreePeople, whose founder, Andy Lipkis, introduced him to the city. TreePeople volunteers are examining samples to find any that are misidentified. Perona then tweaks the algorithm to make the computer see better.
The roadblock with Google prevents him from giving the city what he has done so far.
Good is optimistic there will be a deal.
“Google from the outset has been exceedingly supporting,” Good said. “We are very excited about the idea of this tripartite partnership.”
But just in case, the mayor’s office is also pursuing other options, he said.
One would be to piggyback trees on the old-fashioned human count of all the city’s sidewalks that will be required by the settlement of a lawsuit.
Either way, Perona will keep counting. He’s also seeking federal funding to continue his research. If a deal is worked out, Los Angeles will pay Caltech only for programming done specifically for the city. The research would be for the benefit of all.
“If it’s successful, it benefits all the cities in the U.S. and the world,” he said.
©2016 the Los Angeles Times Distributed by Tribune Content Agency, LLC.
