Aquitard hydraulic conductivity: Investigating heterogeneity with lab measurements and pumping test data

Aquitards are low‑permeability layers that separate or confine aquifers. Whereas groundwater in aquifers flows mainly horizontally, flow in aquitards is predominantly vertical. Aquitards protect groundwater from surface contamination and shield surface activities from subsurface interventions such as pumping or geothermal systems. In the Netherlands, aquifers typically contain sand and gravel, while aquitards consist of clay, silt, or peat. These materials have a wide range of hydraulic conductivities (K), are sensitive to disturbances such as compaction, and respond slowly to hydraulic signals. As a result, parameterizing aquitards in groundwater models is challenging. This thesis aims to improve characterization of aquitard hydraulic conductivity across scales, using core measurements and pumping tests. Chapter 2 presents a machine‑learning model trained on a comprehensive Dutch dataset. The model predicts hydraulic conductivity with a log10(K) RMSE of 0.7 and R² of 0.55, indicating either missing explanatory variables or natural variability due to pore‑scale differences or measurement error. Five features contributed most to predictive skill: stratigraphic unit, clay fraction, depth, geographic x‑coordinate, and lithofacies. Stratigraphic unit and lithofacies act as proxies for additional properties such as mineralogy and grain sorting. Chapter 3 describes the calibration of a groundwater flow model using a pumping test in Zeeland. The test used an extraction well and multiple injection wells switched in various configurations to induce different flow paths. After calibrating a homogeneous model, the overlying aquitard was replaced with heterogeneous realizations featuring various horizontal and vertical correlation lengths. Many combinations reproduced observed drawdowns well, though ranges of optimal correlation lengths were identified. Chapter 4 adapts this method to a long‑operating drinking‑water production site, where pre‑pumping heads could not be defined. Instead, temporal fluctuations in drawdown due to pumping variations over several months were modeled. The aquitard was represented with heterogeneous realizations, incorporating uncertainties in lithology fractions and mean core‑scale K. Many parameter combinations yielded acceptable fits, though large uncertainty remained in horizontal correlation length. Calibration was highly sensitive to the mean of the core‑scale K distribution. Analyses showed that hydraulic resistance varied by about one order of magnitude between fine (25 m) and coarse (800 m) grid cells. Heterogeneous models produced shorter groundwater travel times, highlighting the role of aquitard heterogeneity in protecting well fields. Chapter 5 investigates how heterogeneity in different aquitard locations affects drawdowns, using a three‑layer model and six pumping‑test scenarios. Sensitivity maps revealed consistent patterns in the pumped aquifer, with maximum sensitivity near the pumping and observation wells. In the overlying aquifer, sensitivity patterns shifted depending on transmissivity contrasts and the upper boundary condition. Chapter 6 explores how the number and placement of observation wells influence the ability to infer aquitard heterogeneity. Using heterogeneous realizations and pilot‑point calibration, results showed that few wells underestimate variance and misrepresent spatial patterns, especially for short correlation lengths. Approximately 8–10 wells were sufficient to estimate correlation length; more wells were needed to accurately reconstruct K fields. Overall, this thesis demonstrates that pumping tests can reveal aquitard heterogeneity and that aquitard parameterization strongly depends on scale, flow patterns, and modeling objectives.