In our previous post, we discussed using predictive modeling for service line inventory classification and how that process works. We reviewed thresholding and how it works to classify lines as lead, non-lead, or unknown. It’s these thresholds, set by water systems and state regulators (where applicable, more on that in a minute) that help water systems use the predictions generated by the models to classify their lines as lead, GRR or non-lead.
The Lead and Copper Rule Improvements (LCRI) provides federal guidance from the EPA and several deadlines all water systems must meet, including inventory, classification, validation, and replacement deadlines. However, it has left the enforcement of these deadlines, along with the methodologies used to meet these deadlines up to the states. It’s these different state guidelines (or lack thereof) that help determine how predictive modeling is used to classify unknown service lines as lead, GRR or non-lead. Let’s look at a few different examples:
Massachusetts
The Massachusetts Department of Environmental Protection (Mass DEP) is one of two states that pre-determines thresholding limits for predictive modeling. The state has set the following thresholds:
- Service Lines with an 80% or higher likelihood of lead may be classified as “lead” in the inventory.
- Service lines with a 15% or lower likelihood of lead may be classified as “Unknown, definitely does not contain lead or galvanized.” This will exempt the line from any compliance related communication for unknown service lines.
- Service lines with a likelihood of lead between 16% and 79% must be categorized as “Unknown, may contain lead and/or galvanized.”
North Carolina
North Carolina Department of Environmental Quality (NC DEQ) is the other state that has set specific thresholds for water systems using predictive modeling to classify unknown service lines. These thresholds are explained below.
It’s important to note that if no lead or GRR lines are found during the records review and representative field investigations, all unknowns can be reclassified as non-lead.
Ohio
Ohio is one of many states that don’t have any specific guidance on using predictive modeling for classification and inventory purposes. This leaves water systems with a lot of flexibility (and uncertainty, quite honestly) in how they classify any unknowns.
The pros of not having any state-specific rules around predictive modeling for classification: Water systems can set their own thresholds for classifying unknowns based on their unique risk factors and preferences, allowing them to be as stringent as they would like. The cons: Water system operators typically aren’t experts in the fields of data science and regulatory policy and likely need a lot of guidance and recommendations for understanding how thresholding works and what it means for them; without this guidance, water systems risk making decisions that aren’t in compliance with federal LCRI requirements (or the best interests of their customers).
That’s where BlueConduit can help. Our data scientists make thresholding recommendations to clients tasked with setting their own thresholds in an effort to help guide water systems who otherwise might not know how to proceed. These recommendations can be based on existing state guidance examples (such as the ones above) as well as best practices in statistics and what we’ve seen in the field over the past 10 years. BlueConduit will never recommend thresholds it cannot defend should the state ask for a methodology report or metrics to support the water system’s decision.
As these examples show, state guidance can heavily influence how predictions are used to classify unknown service lines. For the majority of systems in the US, it is up to them to set their own thresholds for classifying unknown service lines, but they may need to defend their decisions to their state regulator.
Navigating LCRI compliance requirements is complex, especially at a state level. With limited federal funding, annual customer notifications, and the first LCRI compliance deadline in 2027, water systems can’t afford to waste time or resources classifying unknown lines. Contact us to learn more about how we can help you efficiently classify your unknowns with high accuracy and minimal digging.
Learn more about how predictive modeling is used for inventory classification at our webinar, From Unknown to Known: Using predictive modeling to classify unknown service lines on Feb. 12 at 1 pm ET. BlueConduit’s data science, policy and product experts will talk through the process and tools needed to classify service line materials based on statistical models. Then, Jenny Puffer, Director of Water Distribution from Des Moines Water Works will discuss how the utility used predictive modeling to reduce their unknowns by more than 50%.
