The impact of data quality and outlier detection in high-frequency water quality data on water management and process understanding

Abstract

Detecting outliers in environmental data is a common challenge in assessing river water quality using high-frequency monitoring. While the importance of addressing outliers on data quality is generally well recognized, their quantitative impact on water quality metrics guiding water management and process understanding has not been evaluated. To address this gap, a four-year water quality dataset of turbidity, nitrate nitrogen, total organic carbon, and total phosphorus measurements was analysed using quality control and advanced outlier detection methods. The outlier detection methods included univariate methods such as the Z-test, naïve prediction, and ARIMA prediction, as well as multivariate methods, including the local outlier factor machine learning approach and Stray’s feature-based approach. The methods were applied to both raw and transformed data to enhance detection performance. The detected outliers were validated through expert evaluation of outliers and a detailed evaluation of the concentration-discharge relationships. These results showed that tailored quality control effectively identified and removed the most obvious outliers, enabling the derivation of reliable and valid descriptive statistics. Therefore, effective quality control is crucial in projects collecting large amounts of high-frequency water quality data to make informed decisions based on these data. This study revealed that neither a single outlier detection method nor a single transformation was universally superior; often, a trade-off exists between overall performance and precision. In this context, the aggregation of results across multiple transformations demonstrated promising potential. Careful application of the methods is essential, as using a poorly performing outlier detection method can degrade data quality even more. The concentration-discharge relationship provided valuable insights to validate the results, which is a new application of this tool as a post-outlier detection step. While setting up quality control measures is relatively straightforward, their implementation plays a critical role in ensuring data integrity and supporting informed decision-making in water management and system understanding.