Combining statistical methods for detecting potential outliers in groundwater quality time series

Area of Science:

• Environmental Science
• Hydrogeology
• Statistical Modeling

Background:

• Quality control of large-scale monitoring networks necessitates automated outlier detection.
• Groundwater quality data present unique challenges: varied dynamics, few measurements, non-normal distribution, outliers, trends, and values below detection limits.
• Time-varying detection limits add complexity to data analysis.

Purpose of the Study:

• To develop and present a robust statistical methodology for unambiguous and reproducible outlier detection in groundwater quality monitoring data.
• To address the specific characteristics of groundwater measurement series, including data below detection limits and temporal trends.
• To provide a tunable method adaptable to different datasets and user requirements.

Main Methods:

• Robust regression on order statistics (ROS) to handle measurements below detection limits.
• Biweight location estimator to filter temporal trends from measurement series.
• Outlier detection in z-score space with tunable parameters for robustness and accuracy.

Main Results:

• The methodology was successfully applied to the Dutch national groundwater quality monitoring network (approx. 350 wells).
• The method effectively detected outliers at the extremes of the measurement range and near detection limits.
• Potential outliers identified by the method require further expert assessment and validation.

Conclusions:

• The developed statistical methodology offers a reliable approach for quality control in groundwater monitoring.
• The method's adaptability through tuning parameters enhances its utility across diverse datasets.
• While effective, the method complements, rather than replaces, expert judgment in outlier validation.