Collecting at-home water quality observations through a citizen science project to characterize risk of lead in drinking water

Lead is a toxic metal that can be harmful to human health even at low exposure levels. For homes that draw water from community water systems, the most significant source of lead in the water comes is typically lead service lines, or lead pipes that connect homes to the water main. Households that are exposed to high water lead levels (WLLs) should be identified rapidly to mitigate exposure through preventative measures. Laboratory testing of water samples across a community is expensive, and at-home water chemistry test kits are affordable and report water quality parameters including copper, iron, pH, and lead. At-home chemistry tests, however, are subject to high levels of error and do not reliably identify WLLs. This research explores the use of machine learning approaches to improve the use of at-home water chemistry test results and classify WLLs at households. Water chemistry parameters, such as iron, copper, and pH, and observations of tap water quality, including odor, taste, and discoloration, are correlated with the presence of lead and can be used to predict WLLs in a statistical or machine learning approach. Data were collected by citizen scientists, who used at-home water chemistry test kits and reported household characteristics, plumbing type, build year, water quality observations, and at-home water chemistry reports through a citizen science project, Crowd the Tap. Regression models and machine learning methods, including Bayesian Belief Networks models, were applied to use variables reported by participants to classify WLLs above lead thresholds. This research explores and couples machine learning and citizen science methods to predict WLLs in tap water and prioritize laboratory testing of households in a community with high risk of lead.