Water quality monitoring with purpose: Using a novel framework and leveraging long-term data

Water quality monitoring programs are developed to meet goals including attaining regulatory compliance, evaluating long-term environmental changes, or quantifying the impact of an emergency event. Methods for developing these programs often fail to address multiple aspects of development (hazard identification, parameter selection, monitoring locations/frequency) simultaneously. We develop a framework for monitoring program development that is both versatile and systematic, the Hazard Based Water Quality Monitoring Planning framework, and apply it to the Quabbin watershed in Massachusetts, USA. We use a novel application of dataset deconstruction of long-term water quality datasets and the Seasonal Kendall test for trends to evaluate the effects of sampling frequency on long-term trend detection at several watershed sites. Results showed that when sampling frequency is decreased, ability to detect statistically significant trends often decreases. Absolute error in trend slopes between biweekly (twice monthly) and reduced sampling frequencies was relatively small for specific conductance and turbidity but was high for total coliform, likely due to interannual variation in rainfall and temperature We found that no one sampling reduction method resulted in a consistently lower absolute error compared to the truth (biweekly sampling), highlighting the importance of evaluating conditions that may affect water quality at sites in different parts of a watershed. We demonstrate the framework's usefulness, particularly for parameter and sampling frequency selection, using methods that can be readily applied to other watershed systems.

Automated summary

Water quality monitoring programs often fail to address multiple aspects simultaneously. This study develops a versatile and systematic framework for monitoring program development and applies it to the Quabbin watershed in Massachusetts, USA. The results show that decreasing sampling frequency often reduces the ability to detect significant trends. No sampling reduction method consistently resulted in a lower absolute error.