Exploring the accuracy of data products commonly used to identify private well locations

About 43 million residents in the United States rely on private wells, but there are no comprehensive data on their locations. To address these knowledge gaps, agencies are developing models to predict private well locations or are using various available well datasets. However, these data products have not been validated. In this study, we examined the accuracy of commonly used data products (well permits, property tax records, and a national prediction model) by comparing estimates to a well location census. In 2019, Wake County, NC identified 14442 private wells in the eastern part of the county to notify users of radiological contaminants in groundwater. This census was developed using municipal boundaries, parcel records, and public water supply information. We observed high accuracy in all data products (80-89%) when predicting absence or presence of wells per census block. Additionally, we observed high accuracy in predicting number of wells (78-87%) per block, but this was driven largely by accuracy in blocks within city corporate limits (91-98%), in which only 5% of wells in Wake County reside. This was further evident when fully municipal water reliant blocks were removed – overall accuracy decreased by as much as 40% (Mann Whitney U-Test, p<0.01), highlighting reduced prediction accuracy in non-municipally reliant areas. Furthermore, we found trends between accuracy and sociodemographic variables. Tax records and the prediction model both over and underpredicted BIPOC well users, while permits underpredicted BIPOC well users. With the growing impacts of climate change such as contamination from flooding and depletion of water resources, accurately modeling at-risk private well communities is increasingly crucial. To do so, it is necessary to understand the nuances of currently utilized data products. This research thus provides an assessment of accuracy of well location data methods, data discrepancies, and insights for outreach and policy.