Groundwater level monitoring using exploited domestic wells: outlier removal and imputation of missing values

Groundwater-level monitoring networks provide vital information for hydrogeological studies. Including exploited domestic wells in these monitoring networks can provide a low-cost means of obtaining a broader set of data; however, the use of these sites is limited because the frequent pumping of these wells generates outliers in the recorded time series. Here a slope criterion is applied to identify and remove outliers from groundwater-level time series from exploited domestic wells. Nonetheless, eliminating outliers creates a problem of missing values, which biases the subsequent time series analysis. Thus, 14 imputation methods were used to replace the missing values. The proposed approach is applied to groundwater-level time series from a monitoring network of 20 wells in the Lanaudière region, Québec, Canada. The slope criterion proves very effective in identifying outliers in exploited domestic wells. Missing values generated by outlier removal can reach up to 99% of the recorded data. Among the characteristics of the missing value pattern, the gap size and the position of the gaps along the time series are the most important parameters that affect the performance of the 14 imputation methods. Of the imputation methods tested, linear interpolation and Stineman interpolation, and then Kalman filtering, were the most effective. The present study demonstrates that exploited domestic wells can be used for groundwater monitoring by removing the outliers and imputing the missing values.