In a partnership with Johns Hopkins University, Baltimore is borrowing a trick from stargazers to predict housing abandonment.
Almost 17,000 houses sit boarded-up and vacant throughout Baltimore. These are the ones deemed officially unlivable by the city, some with rooftops or walls missing. But those structures represent just a fraction of a larger problem. Estimates from the Census and other community surveys suggest anywhere between 30,000 and 54,000 other homes are currently unoccupied. The question is: Which ones?
It’s a similar story in other cities that have experienced severe population drops, such as Detroit and Cleveland. Keeping track of the exact locations of vacancies can prove difficult as the only occupancy data available is often out of date or incomplete. This information gap represents a challenge for housing authorities trying to stabilize shaky neighborhoods.
So what’s a city to do? Baltimore is taking an unorthodox approach to the problem by enlisting some heavenly assistance. Roughly two years ago, through Johns Hopkins University’s 21st Century Cities initiative, the Baltimore Housing and then-deputy commissioner Michael Braverman reached out to Tamas Budavari, an astrophysicist and mathematician at the Hopkins’s Whiting School of Engineering,* who also researches the statistical challenges of mapping the universe. His task, among many others, is to use big data to help the city find unoccupied buildings before they reach a state of terminal disrepair. To accomplish that, he does what astronomers do when they study distant stars: Look into the past to predict the present.
Budavari and Phil Garboden, a doctoral student in sociology and applied math, are working on a statistical tool to predict abandonment. They’re combining publicly available data with GIS technology to create a database of the city’s housing stock. This will serve as a base to do high-level statistical analyses that can help officials make better, data-driven evaluations of current and future interventions. It could help Baltimore study, among other things, when and why homes are abandoned, and at what point a vacant home starts affecting nearby properties.
Once abandoned, a home is more likely to attract crime and lower the property value of surrounding houses, in turn driving more neighbors away. If cities can predict where clusters of vacant homes are likely to form, they can intervene before the entire neighborhood empties. They can, for example, consider lower-cost alternatives to demolition. Getting rid of all 17,000 homes in Baltimore would take $500 million and half a century—money and time the city doesn’t have on hand.
Taking a cue from stargazers
On the surface, Budavari and Braverman seem like an unlikely pair. But astronomy and urban analysis actually have a lot in common, Budavari says. “Just like how galaxies cluster in the universe, houses also cluster in the city,” he says. “So if you have a vacant house in a given place, there's a higher probability of finding other ones next to it.”
Astronomers rely on a wealth of studies and massive databases compiled over decades to find those galaxy clusters. Cities, on the other hand, often lack detailed and real-time data. “Whether a property is occupied is fairly invisible,” says Garboden. The U.S. Census comes around every 10 years and tracks housing occupancy as a five-year average, but only on the tract level. “What the city needs to know is, are there neighborhoods that are suddenly incredibly unoccupied?”
That’s hard to detect; cities can’t tell which homes are only temporarily unoccupied as renters move in and out, and which ones are on the path of long-term abandonment as residents flee their neighborhoods for good. The statistical tool he and Budavari are developing will hopefully be able to find these empty homes and figure out if they’re about to be abandoned, which will help officials monitor when neighborhood begins showing signs of distress. Such a model would be based on a variety of data, including water, gas, and electricity usage, postal deliveries, and possibly even cellphone use. Essentially, the team is going back in time—as astronomers and mathematicians* often do—mining years worth of data to detect abnormal patterns that predict the future.
Consider, for example, hourly water use. In an occupied home, it may be normal to see low usage during the day when people are at work, and high usage in the mornings and evenings. Deviations from that pattern could signal leaky pipes somewhere in a home that’s not being maintained, or that the water is turned on only when someone has broken in to use it. “In places like Detroit and elsewhere, where a lot of properties are vacant, they’re nonetheless being used by local residents for a number of things,” Garboden says. “Sometimes that's using water to wash their car; sometimes that's stealing electricity from that house, or sleeping in it.”
The data might also help researchers determine whether a house might soon become occupied, though Garboden says it’s still too early to say which patterns are predictive. Still, that is one of the many questions the team is trying to answer. As more data come in, from third parties and on-ground investigations, the team hopes to integrate them into sophisticated algorithms that will eventually refine the tool’s predictive capabilities.
Last March, a 69-year-old West Baltimore resident named Thomas Lemmon was sitting in his Cadillac parked next to an abandoned rowhouse when the building collapsed in high winds. The home was one of five to come down, igniting anger among residents who say they should have been torn down long ago.
The question of what to do with Baltimore’s most-decayed structures has flummoxed city leaders for decades. Some 500 buildings are so dilapidated that, according to The Baltimore Sun, they have to be manually inspected every 10 days.
In response to Lemmon’s death, acting housing commissioner Braverman asked Budavari to conduct a one-time emergency investigation using his database to narrow down the number of vacant houses the city should inspect for signs of imminent danger. The researchers came back with a list of 5,000 most likely to be unstable; they were either built as end-of-row houses or had become untethered due to previous mid-row demolitions. Comparing that information with aerial photography, the city identified 300 that were missing structural components like rooftops or floor joists. Upon further inspection, he says, some 200 met the criteria for emergency demolition (which allows the city to bypass the process of obtaining permits) and were torn down by the end of 2016, says Braverman. He adds that another 74 have been flagged for immediate removal.
The city has a limited budget for demolitions: an annual $10 million from the mayor and $75 million in state funding over four years as part of Project C.O.R.E., which aims to demolish vacant buildings and replace them with new development. Razing a two- and three-story rowhouse in Baltimore can cost upwards of $14,000 and $25,000, respectively, and that doesn’t include the cost of rebuilding walls to stabilize adjacent homes or relocating residents.
Part of the partnership between the housing department and Hopkins is to develop a strategy in choosing which houses truly need to be demolished. One way is to target blocks that are entirely uninhabited. “We wanted to know what the dataset look like of all of the demolitions that we could do without a single relocation,” says Braverman. “The researchers gave us this analysis of all of the vacant buildings in Baltimore where we have no occupied properties in between, which helps inform the process.”
That’s the kind of detailed information that housing advocate Shana Roth-Gormley hopes the city will eventually make available to the public. “The more that communities have access to that data, the better it's going to be, because while the city does important work collecting the data, they can’t do it alone,” says Roth-Gormley, pro bono coordinator at the Baltimore housing nonprofit Community Law Center. “The data allows neighborhoods to craft their own plans and say, ‘Here are the things we are facing’—not just anecdotally but with data to back it up.”
Budavari recently submitted a grant proposal to the National Science Foundation’s Smart & Connected Communities initiative to expand the partnership to New Orleans and Kansas City, Kansas. Both are part of Bloomberg Philanthropies What Works Cities initiative, aimed at getting mid-sized American cities to use open data for decision-making.
Well before Hurricane Katrina hit in 2005, New Orleans had been facing a vacant housing crisis, with over 26,000 uninhabitable properties. That number rose to 43,000 by 2010, as Katrina forced many homeowners to abandon their flood-damaged homes. That same year, Mayor Mitch Landrieu moved to streamline the process of identifying blighted properties and pushed for rigorous data collection—even setting up a online map to publicly track vacant homes.
Kansas City’s problems are on a smaller scale: Records show that around 900 buildings are deemed vacant. Just last year, the city launched an open data platform to make housing data easier to access. It also has a team called SOAR, for Stabilization, Occupation and Revitalization, dedicated to data research and analysis.
The ability to understand occupancy and predict abandonment is a common goal among all cities facing a vacant housing crisis. “People can get this information in these large aggregate levels from the American Community Survey and the Census, but it doesn’t happen fast enough” Garboden says. “Theres a lag, and the city wants that information quickly.”
“It's like looking at the heavens and only seeing visible light,” says Braverman. Just as the latest infrared and ultraviolet observatories can peer beyond the visible spectrum and detect the faint signatures of distant galaxies, he hopes this tool can perform similar feats of detection here on Earth. “It will allow us to see the full spectrum of vacant buildings.”
*CORRECTION: A previous version of this post listed Tamas Budavari astrophysicist. The post has been updated to reflect his current title and role as a mathematician.