Machine learning resurrects post-WWII neighborhoods in 3D using old maps

Reviving lost towns with 90% accuracy: "A game changer"

The research started with Sanborn Fire Insurance maps, initially developed to aid fire insurance companies in evaluating their risks across approximately 12,000 cities and towns in the United States throughout the 19th and 20th centuries.

"The story here is we now have the ability to unlock the wealth of data that is embedded in these Sanborn fire atlases," said co-author Harvey Miller in a press release, a professor of geography at The Ohio State University.

"It enables a whole new approach to urban historical research that we could never have imagined before machine learning. It is a game changer."

Initially, researchers faced the challenge of manually gathering data from these maps, which proved laborious and time-intensive. Thanks to the Library of Congress, digital versions are now available. 

Study co-author Yue Lin, a geography doctoral student at Ohio State, then created machine learning tools to extract specific building details from the maps. These details encompassed building locations, footprints, floor count, construction materials, and primary residential or commercial usage.

"We are able to get a very good idea of what the buildings look like from data we get from the Sanborn maps," Lin said.

The researchers applied their machine learning technique to two neighborhoods in Columbus, Ohio, largely demolished in the 1960s, to construct the U.S. highway I-70.

The first neighborhood, Hanford Village, was established in 1946 to accommodate returning Black veterans from World War II. The second neighborhood, Driving Park, also housed a vibrant Black community until it was divided by the construction of I-70.

A comparison of Sanford maps to present data revealed 380 demolished buildings in the neighborhoods, comprising 286 houses, 86 garages, five apartments, and three stores. 

The analysis indicated that the machine learning model achieved high accuracy in replicating the map information, with approximately 90 percent precision for building footprints and construction materials.

"We want to get to the point in this project where we can give people virtual reality headsets and let them walk down the street as it was in 1960 or 1940 or perhaps even 1881," Miller said. 

The researchers argue that this approach can be used to re-create neighborhoods lost to natural disasters like floods, as well as urban renewal, depopulation, and other types of change.

The complete study was published in PLoS ONE and can be found here.