Solving Data Challenges While Fulfilling Lead and Copper Rule Requirements

HDR supported the City of St. Petersburg, Florida, with advanced data analytics to prioritize field investigation efforts used to identify pipe material and develop the City’s lead and copper inventory.

Because of the risk lead services lines pose to public health, the U.S. Environmental Protection Agency recently revised the Lead and Copper Rule requirements. All utilities are required to:

• Identify all lead service lines within their system
• Make an inventory of lead service line status, which must be submitted to the EPA by October 2024 and be made available for the public
• Prioritize the replacement of lead service lines
• Develop a plan to replace lead service lines 

Addressing St. Petersburg's Challenge

The City has approximately 96,000 service lines in its system; only approximately 1% of which had a documented known material type before this project. While the City did not have a robust inventory of known material types, it did have a rich database of work activities performed on its assets in a work order management system. Much of the valuable information about the work activities is found within free-form text fields, which, due to the unstructured nature of those fields, made it difficult to identify necessary information through manual review.

Developing a GeoAI Solution

Our team has assisted the City with the creation of its service line inventory using Esri’s GeoAI Solution.

To address the problem of unstructured data, we used geospatial artificial intelligence to train a text classification model that reads through the 500,000-record dataset and identifies:

• Whether a service line has been replaced and when
• If there is a known material type for the original or replaced service line 
• If the service line was replaced on the City side, the homeowner side, or both

This toolbox incorporates the known material types from the existing inventory, the newly created dataset from GeoAI and other key variables (such as building construction date and connected mainline installation date) into a Presence Prediction model. The model outputs a probability distribution for each material type, at each service line location, to assist with the prioritization of field investigation for the City.

When Iteration Increases Accuracy — Simplify

The model that we used helps prioritize where field crews should be dispatched for field investigation. Because of this, the model was required to be rerun intermittently as new data was collected. The larger the sample size for each known material type, the more accurate the model in its presence prediction.

Leveraging Geospatial Technology

Lead and Copper Rule Inventory projects inherently have a spatial component — each service line exists at a specific location and many utilities have this information in their geographic information systems. However, there are often critical pieces of information that may not be stored in a manner that makes them easily attributable to a spatial object.

By incorporating Esri solutions and our technical expertise, the model allowed disparate datasets to be aggregated, analyzed and visualized in ArcGIS Online — all of which increases the efficiency and accuracy of the data so that experts could make informed decisions and comply with federal requirements.

Advancing Technology

The GeoAI text classification model is technically doing nothing that couldn’t be done with brute force; however, with 500,000 records, an analyst would spend countless hours classifying these unstructured text fields into categorical domains and would have a high potential for error. By using GeoAI, we increased the efficiency and the accuracy of the text classification and provided a repeatable model that can be quickly used when new records need to be assessed.

This project is a highly collaborative effort between drinking water industry experts, our geospatial practice, spatial statistics analysts, and strategic communications specialists. The need for subject matter experts from all disciplines cannot be stressed enough — the model provides a structured, repeatable process that gives industry experts the resources they need to make data-driven decisions.