Machine Learning for Lead Line Replacement: Detroit, MI
Introduction
As of October 2024, $480 billion of federal funding from the $1.2 trillion Bipartisan Infrastructure Law (BIL) has been announced to support more than 60,000 projects nationally, from traditional bridge repair to demonstrating emerging technologies. This unprecedented funding will help bridge the infrastructure investment gap, strengthen America’s workforce, and mitigate the impacts of climate change. With more than $700 billion remaining, local governments continue to compete for funding and deliver projects for their communities that embrace innovation, including the implementation of emerging technologies, adaptive policies, and community-driven solutions to improve mobility and advance more equitable infrastructure investment.
The Innovative Infrastructure Initiative (i3), a project of Accelerator for America and our partners, is releasing a series of publications that highlight innovative projects across the country with the goal of supporting local governments, community leaders, and private sector partners in project ideation and delivery. These projects exemplify how innovative approaches to planning, financing, and the use of transformative technologies can change the status quo of how local and state governments deliver infrastructure.
The BIL has provided an overdue and historic funding opportunity to advance local infrastructure priorities in communities of all sizes across the country. However, these federal funds are not designed for, nor sufficient to, cover the entirety of a given project’s costs. To complete project capital stacks, cities must explore a larger range of funding and financing options, including public-private partnerships and other alternative delivery models. This i3 local project case story is the first in a series that will showcase how emerging technologies are funded and deployed by local and state governments to solve complex problems.
Machine Learning: An Innovative Tool for Local Governments
Across the country, local governments are leveraging machine learning to solve complex challenges. In cities like Detroit, Kansas City, and Pittsburgh, this cutting-edge technology is being used to optimize critical infrastructure projects, from replacing lead water lines to predicting fire risks. These early adopters are showcasing the power of machine learning to improve efficiency, reduce costs, and deliver smarter, more equitable public services.
What is Machine Learning?
As a subfield of artificial intelligence, machine learning can simply be described as computer systems programming themselves based on the data they’re supplied, including numbers, photos, video, and text. With “training data” and a machine learning algorithm in place, machine learning systems can find patterns and build a model that makes predictions or decisions.
Functionally, machine learning can be:
Descriptive - the system uses the data to explain what happened.
Predictive - the system uses the data to predict what will happen.
Prescriptive - the system will use the data to make suggestions about what action to take.
Source: MIT Sloan School of Management
Problem to Solve
In 2018, following the Flint water crisis, Michigan strengthened its Lead and Copper Rule (LCR) under the state’s Safe Drinking Water Act, requiring that 100% of lead service lines be replaced by 2040. Additionally, the LCR requires all Michigan water supplies to build, maintain, and submit a comprehensive inventory of service line material to the state by 2025. In 2018, before the state revised the LCR, the City of Detroit launched its Lead Service Line Replacement Program with the intent to identify and replace all lead service lines in the city. At the time of the program launch, Detroit estimated there to be 120,000 lead service lines in the city, about 40% of the city’s water service lines. While some areas of the city were known to have lead service lines, many of the lead service line locations were unknown due to missing, inaccurate, or outdated records. With 300,000 total water service lines, outdated location data, and state law to consider, Detroit was looking at an estimated $165 million to excavate and verify the material of each of its service lines.
In 2021, utilizing philanthropic support from the Rockefeller Foundation and the Kresge Foundation, the City of Detroit’s Water and Sewerage Department (DWSD) procured a solution from tech startup BlueConduit, an Ann Arbor based water analytics company, to use their predictive machine learning technology to map the probable location of lead service lines into a comprehensive inventory.
Timeline: Lead Service Lines and Detroit
Impact
The technology developed by BlueConduit offers a reliable and cost-effective solution to locate lead service lines. Instead of excavating all 300,000 water service lines to check their material makeup, BlueConduit’s software only required input data from 384 locations. In 2021, using parcel and permit data, including the age of a property, BlueConduit found that Detroit had approximately 80,000 lead service lines, a 33% decrease from the city’s initial estimate of 120,000. By utilizing this new technology, the Detroit Water and Sewerage Department significantly minimized the number of costly diagnostic excavations required and is expected to save hundreds of millions of dollars. In addition to the cost savings, the City of Detroit has built an updated and accurate lead service line inventory while complying with state regulations.
Per the US Environmental Protection Agency (EPA) and the Michigan Department of Environment, Great Lakes, and Energy (EGLE), predictive machine learning has been identified as an acceptable tool for building out a comprehensive service line inventory. In fact, the EPA specifically references Detroit’s use of machine learning to identify lead service lines in their Guidance for Developing and Maintaining a Service Line Inventory. BlueConduit’s technology, which was first piloted in Flint, MI, has also been deployed in Summerville, SC and South Bend, IN. Similar predictive modeling technology has been used for planning and lead service line inventories in Denver, CO and San Antonio, TX.
How does the technology work?
BlueConduit’s machine learning technology combines and analyzes geographical data, age of the housing stock, property values, and other data points from a variety of sources, including site visits, to predict the probable location of lead service lines.
To simulate real world conditions, some data points will be purposefully left out when training the model. The model will then be trained on the withheld data to see how the model performs across several different metrics such as accuracy, recall, and precision.
Is it accurate?
BlueConduit predictive models routinely identify 80-90% of lead total lines (recall) and these positive predictions are 80-90% accurate (precision). Recall and precision are two key metrics in machine learning models.
Recall (80-90%+): This means that out of all the actual lead service lines present (the total number of lead lines), the model is able to correctly identify 80-90% of them. In other words, it successfully "recalls" most of the true positive cases.
Precision (80-90%+): Precision refers to the accuracy of the positive predictions made by the model. Of all the lines that the model predicts as lead lines, 80-90% are actually correct. This means that only a small portion of its positive predictions turn out to be false positives.
In simpler terms:
Recall tells you how good the model is at finding all the lead lines.
Precision tells you how accurate the model’s predictions are when it says a service line is a lead line.
Impact (continued)
To supplement the predictive modeling, DWSD has internal processes in place to identify the remaining 10-20% of lead service lines that may not be identified by machine learning. DWSD utilizes its own GIS analysts along with the Esri ArcGIS platform to identify and track assets across the city. This is in addition to the boots on the ground staff that identify lead service lines through their daily duties.
The use of predictive machine learning positions Detroit to more equitably prioritize lead service line replacement throughout the city. In 2023, with state and federal grant funds in hand, Detroit began approaching lead service line replacement neighborhood-by-neighborhood. According to DWSD, neighborhoods are prioritized based on density of housing built prior to 1945, significant number of children and seniors in the area, and likely high number of low-income households based on census tracts.
In addition to the equitable prioritization of lead service line replacement and significant cost savings, this project has a wide array of impacts, including:
Time Savings: The City of Detroit’s initial goal was to replace all lead service lines in the city by 2040. Use of the new technology, along with state and federal funds, have accelerated efforts, and the work is now expected to be completed in 2034.
Community Engagement: Detroit’s Lead Service Line Replacement Program has a robust website that offers several details on its approach and work to date. On the ground efforts include neighborhood meetings and informational packets.
Workforce Development: The DWSD Workforce Development program emphasizes the hiring of Detroit residents to support the lead service line replacement efforts. DWSD hosts several recruitment events each year to increase employment opportunities and diversify staff. The program also provides employees the opportunity to oversee and manage the contractual aspects of the lead service line replacement. Utilizing city employees, it costs the city only $3,700 per line replacement. This dynamic and expanding contractor capacity and competition has since driven down the cost per house by contractors to $9,300 per line.
Funding Sources
Rather than relying solely on limited local funds, Detroit has assembled and leveraged approximately $100 million in grants and financing from philanthropic, state, and federal sources.
*Additionally, the Rockefeller Foundation provided $50,000 directly to BlueConduit to fund their work with the City of Detroit.
Conclusion
This case story highlights Detroit's innovative use of machine learning to tackle the complex challenge of lead service line replacement. By leveraging technology, Detroit has been able to identify and prioritize lead line replacements more accurately and cost-effectively, reducing the number of unnecessary excavations and accelerating project timelines. With significant cost savings and a focus on equitable distribution of resources, Detroit’s approach serves as a model for other cities facing similar challenges. This initiative not only demonstrates the potential of emerging technologies in public infrastructure projects but also emphasizes the importance of collaborative efforts and strategic funding in transforming local government services.