Can We Predict The Next City with Unsafe Drinking Water?

When Newark started handing out bottled water to residents in mid-August after finding unsafe levels of lead in their drinking water, comparisons to Flint quickly followed. But while Newark and Flint might be the most prominent examples in recent years of cities with troubled water systems, they are by no means the only municipalities that have struggled—and they won’t be the last, either. 

“There are thousands and thousands of water utilities in the United States, and many of them have similar issues with Safe Drinking Water Act compliance,” said David Switzer, a professor who studies water policy at the University of Missouri.

To be precise, there are 51,988 water utilities in the U.S. Fifty-six percent of those systems serve fewer than 500 people, and an additional 27% serve less than 3,300. Compare that to around 3,000 electric utilities covering the entire country, and Switzer says it’s obvious how water problems can occur: a fragmented market means most of them are underfunded, and therefore unequipped to do the massive system overhauls need to replace aging infrastructure, one of the primary causes of SDWA violations. 

When the SDWA was approved by Congress in 1974, lots of money was funneled into municipalities to update their systems. But Switzer said that funding dried up right when local governments need it most. “Take Flint for example—white flight caused the tax base to shrink, so they can’t pay for needed updates,” he said. “Often, water utilities that were built for much bigger populations end up serving a small group of homes. The utilities could raise the rates, but then you risk making water unaffordable.”

Without sufficient funding to maintain or replace entire systems, cities may begin to look to technology as a way to grapple with water problems before they rise to the level of a crisis. A new project in Massachusetts that uses data to try to predict the systems that could face violations in the future looks promising—and if successful, could help both state and local government leaders prioritize investments, especially for the communities with the fewest resources.

Only Some Communities Can Adapt 

Old, corroding infrastructure (particularly lead service lines from water treatment centers to individual houses) are the main cause of lead issues, and high lead levels are considered one of the most dangerous scenarios for a water system. But lead isn’t the only concern, and therefore, neither are the pipes in many communities. Heavy metals, pesticides, and E. coli are some of the other potentially harmful contaminants in water systems, and can appear from a variety of sources. 

Problems can be caused by drenching rains, droughts, chemicals dumped upstream of water sources, and agricultural or ranching runoff that leaches into groundwater. Some utilities are well-suited to adapt to these contaminants, usually the largest ones that have advanced detection technology and a full staff of chemists and engineers who can devise solutions. But small, rural, and tribal utilities struggle, and those serving populations made up predominantly of minorities and low-income folks are often hit the hardest.

“Communities that are underrepresented equivalent to their population in the state and federal government are let down in a lot of ways, and water is no different,” Switzer said.

On the other hand, well-funded utilities can afford to fortify their systems with the latest technology, such as in-pipe contaminant monitoring and leak testing systems.

Moving Towards Prediction

A project in Massachusetts aims to arm communities with information before they decide to make investments, creating a logistic regression model to predict potential sites of unsafe drinking water using real time data. Connor Johnson, a high school student working on the Safe Drinking Water Project, an initiative run by Code for Boston, said the group of volunteers is using data modeling and machine learning to study E. coli contamination in the state.

Johnson said that Massachusetts is uniquely well-suited for testing this kind of prediction model because there are so many water sources and individual systems providing a rich data set for researchers. But that doesn’t mean the work isn’t challenging. “There’s a lot of literature on what could be predictors for water violations, but a model has never existed that could predict violations as they happen, and it’s extremely difficult to do on a large scale,” he said. “The work is cutting edge right now, and there’s a limit to how strong the predictions can be because of the availability of data.”

The Code for Boston team is primarily relying on the Safe Drinking Water Information System for data, but they’re looking to expand to secondary sources so that the model can become more accurate. To explain why they need real time data to create more complex indicators, Johnson used the example of E. coli contamination caused by livestock runoff.

A simple secondary source for predicting E. coli would be where farmland and livestock are concentrated, he explained. So the team took USDA data on livestock, and mapped that information over where water systems are located by latitude and longitude, giving them a better sense as to which water systems would be vulnerable to E. coli outbreaks from farm animals. The team can keep adding datasets—like rainfall and flood zones, for example, to map where and when manure might be washed into water sources—until the model gets more and more accurate.

“We’re narrowing the scope of what were looking for and broadening the scope of how we look for it,” Johnson said. “Right now our model is running at about 64% accuracy, but it’s hugely consequential to predict violations, so we need to keep making it more reliable before we scale to looking at larger areas and more violations beyond E. coli.”

Identifying strong predictors and finding data sets for them is difficult—and even the SDWIS data isn’t the most reliable, because the latitude and longitude markers for water systems are sometimes placed by utility headquarters, not the water source. Even after the team acquires accurate data (sometimes after months of searching), they have to keep updating it, and if the model isn’t run continuously, its predictions will be out of date. Johnson said he’s hopeful that they’ll get to a model with a 90% accuracy rate for E. coli violations in Massachusetts soon, and then they’ll be able to expand to other violations and other states.

But Johnson, who does project management for the group of around eight other volunteers, said that he was surprised that no municipalities have reached out so far looking for help. “We haven’t run into a high demand for this work,” he said. “We want to help districts know in advance when something might happen, so they can act preemptively. But it feels like no one is searching for something like this, even though it’s a really pressing issue.”

Bringing Trust Back

With a nationwide prediction model not yet realized, Switzer said municipalities are really choosing between two options: trying to repair aging infrastructure or replacing it outright. 

Not acting fast enough can lead to real criticism, as happened in Newark, where residents and advocates say the city was slow to act after a few years of warnings. After the Environmental Protection Agency ordered the city to begin giving out bottled water, citing lead problems, Newark and county officials began moving forward with a $120 million plan to replace significant chunks of the infrastructure. 

The bond proposal calls for replacing all lead pipes in the city with copper ones over the next few years, free of cost to residents. Frank Baraff, Newark’s director of communications, said that the city is working on a campaign now to let residents know about the plan, because they have to give consent before the pipes in their homes can be replaced. “Once the lead pipes are removed, the problem will be gone,” he said. “It’s a permanent solution.”

But Switzer warned that government leaders should be thinking beyond immediate crises, so they don’t just create the same infrastructure issues for the people in charge 40 years in the future. Instead, he thinks that local governments need to consider more radical solutions, like consolidating smaller utilities under broader districts. Though they’d give up some local control, municipalities banded together might find it easier to comply with SDWA regulations and would have more capacity to invest in upgrades, he said.

And then, Switzer said, those local governments have to work to regain the public trust. “People who don’t trust the government drink bottled water at higher rates, and I don’t think that is unrelated to situations like those in Flint and Newark,” he said. “That lack of trust is warranted. We don’t need campaigns telling us to trust the water now. The only way to rebuild trust is to fix the problems for good.”