Document Type

Article

Publication Date

7-30-2025

Department

Department of Civil, Environmental, and Geospatial Engineering; College of Engineering

Abstract

There is an urgent need for rapid, cost-effective approaches to identify residences with lead service lines (LSLs). We evaluated whether analyzing water for corrosion-related metals could accurately identify residences with LSLs without relying on potentially inaccurate property records. We applied principal component analysis logistic regression (PCA-LR) and classification tree models using 28 analytes per bottle (including Pb, Cu, Zn, Fe, Al, and others) measured in 216 water samples collected in Flint, Michigan, in August 2015. The PCA-LR model achieved 87% accuracy (AUROC = 0.93) with 81% sensitivity and 90% specificity, while the classification tree model achieved 80% accuracy (AUROC = 0.77) with 74% sensitivity and 84% specificity. The classification tree provided interpretable decision rules identifying key predictive metals, primarily relying on 1 min flush Pb concentrations with Zn and Al as secondary predictors. It also revealed distinct metal co-occurrence patterns between LSLs and premise plumbing, offering insights into Pb source identification. The tree’s interpretable structure makes it particularly valuable for practical implementation by utilities. Although additional work is needed to extend these models to other water systems, our results suggest that metal analysis provides an accurate, cost-effective, and minimally invasive tool that complements existing approaches for predicting the presence of an LSL.

Publisher's Statement

© 2025 The Authors. Published by American Chemical Society. Publisher’s version of record: https://doi.org/10.1021/acs.estlett.5c00552

Publication Title

Environmental Science and Technology Letters

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Version

Publisher's PDF

Download

Share

COinS