Sensitivity of groundwater trend and homogeneity analyses to missing-data imputation methods (Middle Sebaou...

Incomplete groundwater level time series are common in monitoring networks and can bias homogeneity and trend analyses if not properly addressed. This study evaluates how different gap-filling strategies influence homogeneity detection and trend inference in groundwater level records from the Middle Sebaou Basin, northern Algeria. Four imputation approaches (Seasonal Trend decomposition with interpolation (STL), Kalman Smoothing, Moving Average Imputation, and Multiple Imputation by Chained Equations (MICE)) were assessed. Homogeneity and change-point behavior were examined using Pettitt and Buishand tests, while randomness was evaluated with the Von Neumann ratio test. Trend analyses were conducted using autocorrelation-corrected Mann-Kendall, Sen's slope estimator, and Innovative Trend Analysis. Results demonstrate that imputation choice substantially affects homogeneity outcomes and trend significance. MICE preserved temporal variability and reduced artificial change-point detection, whereas smoothing-based methods tended to dampen variability. The study highlights the necessity of selecting imputation methods consistent with groundwater system dynamics and provides a transferable framework for trend analysis in data-scarce aquifers.