The spatial distribution of seepage at a flow‐through lake in western Denmark was investigated at multiple scales with integrated use of a seepage meter, lake–groundwater gradients, stable isotope fractionation (δ ¹⁸ O), chlorofluorocarbon (CFC) apparent ages, land‐based and off‐shore geophysical surveys, and lake bed coring. Results were compared with a three‐dimensional catchment‐scale groundwater flow model using the MODFLOW and LAK3 codes for simulating lake–groundwater interaction. Seepage meter and model results of discharging groundwater to the lake compared well, if direct seepage measurements from near shore were combined with measurements from deeper parts of the lake. Discharge rates up to 9.1 × 10 ⁻⁷ m s ⁻¹ were found. Ground‐penetrating radar used to map the lake bed sediments proved very effective in recognizing low‐ and high‐permeability areas but also in understanding the complex recharge pattern of the lake and relating these to the geologic history of the lake. Recharge of the surrounding aquifer by lake water occurs offshore in a narrow zone, as measured from lake–groundwater gradients. A 33‐m‐deep δ ¹⁸ O profile at the recharge side shows a lake δ ¹⁸ O plume at depths that corroborates the interpretation of lake water recharging offshore and moving down gradient. Inclusion of lake bed heterogeneity in the model improved the comparison of simulated and observed discharge to the lake. The apparent age of the discharging groundwater to the lake was determined by CFCs, resulting in ages between 3 and 36 yr with an average of 16 yr. The simulated average groundwater age was 13.2 yr.

Observed groundwater head and soil moisture profiles are assimilated into an integrated hydrological model. The study uses the ensemble transform Kalman filter (ETKF) data assimilation method with the MIKE SHE hydrological model code. The method was firstly tested on synthetic data in a catchment of less complexity (the Karup catchment in Denmark), and later implemented using data from real observations in a larger and more complex catchment (the Ahlergaarde catchment in Denmark). In the Karup model, several experiments were designed with respect to different observation types, ensemble sizes and localization schemes, to investigate the assimilation performance. The results showed the necessity of using localization, especially when assimilating both groundwater head and soil moisture. The proposed scheme with both distance localization and variable localization was shown to be more robust and provide better results. Using the same assimilation scheme in the Ahlergaarde model, groundwater head and soil moisture were successfully assimilated into the model. The hydrological model with assimilation showed an overall improved performance compared to the model without assimilation.

This paper presents a first attempt to estimate future groundwater levels by applying extreme value statistics on predictions from a hydrological model. Climate scenarios for the future period, 2081–2100, are represented by projections from nine combinations of three global climate models and six regional climate models, and downscaled (including bias correction) with two different methods. An integrated surface water/groundwater model is forced with precipitation, temperature, and potential evapotranspiration from the 18 models and downscaling combinations. Extreme value analyses are performed on the hydraulic head changes from a control period (1991–2010) to the future period for the 18 combinations. Hydraulic heads for return periods of 21, 50 and 100 yr (T_21–100) are estimated. Three uncertainty sources are evaluated: climate models, downscaling and extreme value statistics. Of these sources, extreme value statistics dominates for return periods higher than 50 yr, whereas uncertainty from climate models and extreme value statistics are similar for lower return periods. Uncertainty from downscaling only contributes to around 10% of the uncertainty from the three sources.