A National-Scale Historical Assessment of Nitrate in Public Drinking Water Supplies in New Zealand: Data Integration and Machine Learning Imputation Approaches

Abstract

Nitrate in drinking water is a known health hazard for infants, although a growing body of epidemiological evidence suggests an increased risk of adverse pregnancy outcomes and some cancers. A major constraint of epidemiological research is the ability to quantify nitrate concentrations in public drinking water supplies over time. Data on nitrate concentrations in public drinking water supplies were retrieved by information requests, linked to a national dataset on the spatial extent of water distribution zones (WDZs) and linked with census information. We applied a number of data cleaning and imputation processes to address complexities in the raw data as well as missingness. In total, 599 WDZs (95.4%) had at least one nitrate measurement between 2000 and 2024 (n = 20,875 raw observations). After applying a set of imputation methods, the final dataset covered 89.8% of all person-years (n = 92,800,000) of the population on a public drinking water supply during the most recent period from 2000 to 2024. Overall, XGBoost imputation outperformed a range of other imputation methods when synthetic missingness was added to the original data. The large majority (95.3%) of the population was estimated to be on drinking water supplies of less than 1 mg/L nitrate-nitrogen. The population-weighted median nitrate concentration was 0.05 mg/L (IQR 0.04–0.36). This extensive assessment provides the foundation for epidemiological research into the health effects of nitrate contamination of drinking water in New Zealand. The effectiveness of the system for drinking water nitrate surveillance could be enhanced in several ways that would improve its ability to meet its intended purpose.