When the city of Flint, Michigan, started to replace lead service lines in 2016—two years after mismanagement of the city’s water system caused pipes to corrode and lead in drinking water spiked to dangerous levels—it faced a basic challenge. It didn’t know which homes had lead lines and which didn’t. The same problem exists across the country, where there may be as many as 10 million lead service lines, but no map exists showing their locations.
“The records are poor, and many of them have just been handwritten over decades,” says Eric Schwartz, a professor at the University of Michigan’s Ross School of Business, who worked with the city of Flint on the problem. “Sometimes we were looking at records that had been manually entered in a spreadsheet after reading an atlas of hand-drawn maps, or index cards that have been in file cabinets in the basements of City Halls for decades. Even if you do your best to digitize those, they’re old and incomplete.”
That means that cities typically have to dig to figure out whether a pipe buried underground is lead or copper—and each dig can cost thousands of dollars. Schwartz, along with other researchers at the University of Michigan, worked on a machine-learning tool that could help the city predict where lead pipes were most likely to be, eventually forming a company called BlueConduit that started working with other cities on the same issue. Now, as the Biden administration is pushing to remove lead pipes nationally, the team is using a grant from Google.org to build an open-source tool that other cities can use to understand the scope of their lead pipe problem.
The software works by pulling in data about the age of the home, the neighborhood, whether any records exist about the service line or others nearby, the size of the property, and other factors, and then uses machine learning to estimate the likelihood that the service line on the property is made from lead. When the researchers first used it in Flint in 2016 and 2017, the tool found lead pipes around 80% of the time. In 2018, the city temporarily stopped using the algorithm; residents who had lost faith in the government wanted to physically see that their own pipes were okay, rather than trusting software. But as the city dug more randomly, the hit rate dropped to 15%. The next year, after a court order, the city went back to the software.
It’s not a complete solution, since it can’t say with certainty that a house isn’t connected to a lead service line. But it helps more people who need replacements get them more quickly. “If there were enough resources to inspect every service line and replace the hazardous ones tomorrow, or next month, or even next year, then that would be fantastic,” says Schwartz. “No need to worry about optimizing. But in reality, there isn’t going to be enough funding or available labor to do that everywhere. Some needs to go first, so the order matters. The best a city can do is use its best information available, like the probability each home has lead, to aim to cut down the total time all residents are living with lead.” At a later point, he says, when a city has almost all of the lead out, more affordable new technology may emerge for physically inspecting the remaining lines without having to dig each one up.
In the cities where BlueConduit works, the tool helps save money by making it more likely that a dig will discover a pipe that needs replacement. “You can get the lead out sooner, so people are living with lead for less time, and there’s the money avoided,” says Schwartz. “Thousands of dollars per dig that can be saved. And that money doesn’t just sit, that’s money going to active replacement of lead at a neighbor’s house.” Cities typically need to remove thousands of service lines. In Trenton, New Jersey, for example, where BlueConduit is working now, there’s an estimated 37,000 lead service lines.
The tool, which is open-source, will help cities estimate how many lead service lines exist and get a basic map that can be used to understand how much funding is needed to replace them. The company will also continue doing more detailed work with specific cities, pulling in more data to calculate the probability house by house. (The grant will cover this type of work with three yet-to-be-determined cities, others will pay for the service.) It also plans to create the first national map of lead service lines—a critical step if the country is actually going to succeed in removing them.