- Gentrification is shifting the socioeconomic landscape of cities around the world
- Previous method of using census data doesn’t accurately capture the process
- New model combines Google Street View images and machine learning to map gentrification
When neighborhoods gentrify, the new, higher cost of living drives working people out of the central parts of cities and into more affordable locations, causing prices there to also rise and forcing long-time residents out.
Displacement is closely tied to factors like race and socioeconomic status and can lead to further inequalities. Gentrification affects who has access to things like quality healthcare, reliable public transportation, and healthy food.
Sawada, who is also co-lead of the Ottawa Neighborhood Study, says another reason people move to a certain area is because they want to be associated with its way of life or its perceived vibrancy.
“In the end, what happens is the community that built that vibrancy, that culture, disappears because of gentrification,” says Sawada.
“The process is going on in almost every major city,” says Lazar Ilic, a graduate student at the University of Ottawa. “I feel like there is injustice when people get pushed out against their will.”
The sheer size of the problem means it’s not only difficult to solve, but also difficult to track or monitor in a meaningful way.
“Deep-mapping” the problem
That’s beginning to change thanks to a recent study carried out by Ilic, Sawada, and their colleague Amaury Zarzelli. Combining a machine learning model and images from Google Street View, they created a map that highlights instances of gentrification in Ottawa.
Previously, gentrification has been studied using census data. However, that data is only collected every five years in Canada and every ten years in the US. Gentrification doesn’t keep to that schedule.
“Prior to what we've done, gentrification had no good way to be measured in a timescale that would be appropriate to the actual process,” says Sawada. “It can completely happen within a five year period. Or, more often, it's not detectable until a ten-year census, for example. And so it's completely missed.”
In an attempt to better capture the gentrification process, the researchers accumulated hundreds of thousands of images from different Ottawa neighborhoods taken between 2007-2016. Google Street View updates images every two to three years, so they had multiple images of the same properties over this nine-year period.
They then fed all of these images into their model, and the artificial intelligence told them if it picked up visual changes correlated with gentrification.
“Gentrification is one of those things that you know it when you see it,” says Sawada. “If, suddenly, your neighbor is building a new addition, changing the siding, upgrading the landscaping significantly and then that starts to happen in a few other places near you, you know what’s happening. It's not just painting the front door or putting up a new fence. It's these significant changes that the model was trained to look for.”
All of these instances were represented in a process called “deep-mapping,” allowing researchers to better visualize what’s happening, where, and when. It also helped weed out outliers like odd household maintenance or unclear images.
One city at a time
In the end, the model produced a map of Ottawa with clusters, or “hot spots," where many gentrification-like changes took place within the nine-year period. After verifying their outcomes with permit data, the researchers determined that their model had a 95 percent accuracy reading.
“It's almost as good as a human at capturing these things,” says Sawada. “So it's a very good result.” Sawada and Ilic believe they would probably achieve a similar accuracy in other cities in Canada or the US with the current model.
“The model has learned a set of what we call weights from our imagery. This could be transferred to another image set and save a lot of time and effort in training a new model,” says Sawada.
However, thanks to a pay wall implemented by Google Street View in July of 2018, looking at gentrification in this way may no longer be so easy. Before the change, the team could access 25,000 images a day. Now it takes them a month to acquire the same number.
“That puts these types of tools now in the hands of people who have money, like developers who want to get ahead of this process or profit off of it before advocate groups can do anything to mitigate it,” says Sawada. “Even local governments can't afford to access Google Street View imagery now.”
Shedding a new light
To the researcher’s surprise, their model also captured something most people don’t consider: a phenomenon Ilic calls “super-gentrification” where the wealthy displace people in neighborhoods that are already middle or upper-middle class.
“It's not a process necessarily whereby you have people at a disadvantage being taken advantage of, it's people that are middle-class now being taken advantage of, too,” says Sawada. For example, in Florida, wealthier people are relocating to higher elevations to escape destructive storms and rising oceans, displacing the middle class people already living there.
Sawada says that finding these gentrified areas that weren’t previously being looked at means Ottawa’s policymakers can start thinking of solutions.
“Gentrification always starts under the radar,” says Sawada. “It's not until it becomes talked about and you start seeing changes in the commercial spaces that gentrification is then looked at by the city itself. And so the city is very interested in what we've done as a means by which to look at future planning as well as zoning in order to perhaps mitigate negative consequences.”
One possible solution proposed by Sawada would be to have the property developers who are causing the rising prices contribute to building affordable housing.
While it’s a big step forward, Sawada and Ilic concede that their model has a ways to go before it can accurately capture all aspects of gentrification.
“If anything, it's a conservative measure of gentrification than what may actually be happening in the geographic space,” says Sawada. “But I think the overall message is that this type of an approach now gives us the ability to measure gentrification at a fine spatial and temporal scale over large areas. That's something that nobody has been able to address before.”