Gap-filling groundwater level time series of irregular temporal resolution using physical modeling (Pastas) and simple statistical scaling

Groundwater level (GWL) monitoring datasets are essential for effective groundwater resource management and understanding the potential impacts of climate change. However, these datasets frequently contain gaps and irregular measurement intervals, posing challenges for time series analyses that depend on consistent sampling. As a result, GWL datasets with substantial gaps are frequently excluded from further analysis, leading to a loss of temporal and spatial coverage, regional representativity, and potentially valuable insights. Addressing this issue requires effective and interpretable imputation techniques to fill missing values while preserving the physical realism of the reconstructed data. Traditional statistical imputation methods and advanced machine learning algorithms, such as missForest, have been used to address data gaps. While these approaches often yield effective imputations, they lack physical interpretability, particularly for extreme events, which is crucial for understanding the variability and resilience of groundwater systems under changing environmental conditions. This study proposes a novel hybrid imputation approach that combines physical modeling with statistical adjustments. First, GWL data are simulated using Pastas, an open-source framework that leverages hydrometeorological variables and impulse response functions to model GWL time series. These simulations serve as a physically consistent basis for imputing missing values. In the second step, a linear scaling approach is applied to scale the simulated GWL to match the observed start and end point of each gap, ensuring consistency with observations. The hybrid method was tested on data from 213 monitoring wells across Sweden, encompassing diverse temporal resolution and gap characteristics. This process generated continuous daily time series spanning 34 years (1990–2023), enabling the evaluation of long-term groundwater dynamics across Sweden (future work). Validation focused on the ability to capture extreme GWL events. While Pastas-only simulations performed well in reproducing seasonal GWL variability, they failed to accurately capture extremes. The hybrid technique demonstrated significant improvements in representing extreme variability, offering a robust solution for handling irregular and incomplete datasets. Additionally, the study provides insights into regional data characteristics, such as variations in gap patterns and hydrometeorological drivers, offering valuable information for groundwater modeling and analysis. The proposed method not only enhances the reliability of GWL datasets but also supports better decision-making in groundwater resource management. The work is a contribution to the Water4All GroundedExtremes project.