Search

Modeling spatially continuous variables from point measurements is integral to environmental scientific research in many fields. This is especially the case for groundwater, which is accessible only at boreholes and springs. Often, further management decisions are dependent on spatially continuous values of groundwater level or quality parameters.For this task, deterministic or geostatistical interpolation methods are traditionally used, involving the spatial structure of the point locations as a set of XY-coordinates. In this case, the spatial model is usually created based solely on the geographic location of the measurement and the spatial autocorrelation of the target values. With few exceptions (e.g. co-kriging), classical interpolation techniques do not support the incorporation of covariates that are spatially correlated to improve spatial prediction accuracy.Spatial predictions using machine learning (ML) models are an attractive and increasingly prevalent alternative, which use correlated, spatially continuous covariates (e.g. meteorological data, land-use, or geological maps) as predictors for groundwater level or quality parameters. They are trained on the nonlinear relationship between these predictors and the target values with the available point measurement data. However, spatial autocorrelations of the target values are usually not considered, as the points are treated independently of their location. Therefore, the machine learning models cannot exploit and represent the spatial dependence structures in the target data, without any further information about the geographic locations. The incorporation of locational information into ML models is, however, not trivial. From other fields, diverse approaches (e.g. using coordinates directly, distances to certain locations in space, Euclidean distance matrices, transformed coordinates) exist, and also different assessments of their suitability. For example, it is often stated that XY-coordinates are very well suited for the use with decision trees, but not for neural networks.We systematically investigate the impact of the most commonly applied methods for the integration of spatial information, such as XY-coordinates, transformed coordinate-based input features, Euclidean distance matrix or distances to corners or center, Wendland transformed coordinates, and combinations of the aforementioned, on the interpolation results for selected hydrogeological parameters and different test sites in two ML models (Random Forest and Multi-Layer-Perceptron). We compare the results by cross-validation and with kriging reference models as well as visual assessment for plausibility.The results show that the incorporation of spatial covariates can significantly improve model performance, especially when the data have high spatial autocorrelation (and the data set is sufficient to capture this). In particular, the Euclidean distance matrix and Euclidean distances to defined locations proved to be efficient approaches to provide the spatial data structure for the model, while the application of XY-coordinates often resulted in significant artifacts in the resulting prediction surfaces.

At the general assembly of the European Geophysical Union in 2022, the “Groundwater Time Series Modeling Challenge” was launched (Haaf et al., 2022). We challenged our colleagues in the field to model five time series of hydraulic heads measured in groundwater observations wells around Europe and North America. Part of the head data was not made available to the participants and held back as independent evaluation data. In this presentation, we will share and summarize the results from the challenge. The challenge attracted submissions from 17 teams using a variety of modeling techniques (https://github.com/gwmodeling/challenge). The models used in the submissions ranged from machine and deep learning models to empirical and bucket-type models. Many of the participants devoted notable attention to the uncertainty quantification, providing not only the results of the best-fit model but also uncertainty intervals for their models. The time to set up each model ranged from a couple of minutes to a couple of hours, indicating that models are generally set up in a limited amount of time. For most models, the majority of the time was used for the training and/or uncertainty quantification. The time spent on training showed larger differences between the models, ranging from a few minutes to more than a day for a single model. The data used to model the time series varied per model, with empirical models using less information than the machine learning models. A preliminary analysis of the modeling results showed that most of the models performed well (as measured by goodness-of-fit metrics such as NSE, MAE, KGE) in both the training and evaluation period. For one of the time series, none of the models showed a good fit with the data in the evaluation period and we suspect that a systematic change in the groundwater system may have occurred. The best-performing model differed between observation wells; none of the models outperformed all other models for all time series. In the coming months up to the EGU 2023 General Assembly we will further analyze and synthesize the results.Haaf, E., Collenteur, R., Liesch, T., & Bakker, M. (2022). Presenting the Groundwater Time Series Modeling Challenge(No. EGU22-12580). Copernicus Meetings.

ZusammenfassungMit zunehmender Anzahl an Grundwasserwärmepumpenanlagen steigt auch das Potenzial an Nutzungsüberlagerungen. Im Rahmen des wasserrechtlichen Genehmigungsverfahrens ist es daher erforderlich, u. a. die von einer Anlage ausgehenden Temperaturfahnen im Grundwasser zu berechnen. In Baden-Württemberg wurde als Erweiterung des für Anlagen ≤ 45.000 kWh/a gültigen Leitfadens des Umweltministeriums Baden-Württemberg das Thermische Online-Modell (TOM) entwickelt. Hierbei handelt es sich um ein vereinfachtes, browserbasiertes, numerisches Grundwassermodell. Ziel dieser Studie ist, die mit TOM berechneten Ausdehnungen von Temperaturfahnen durch mit analytisch und einem kalibrierten numerischen Modell berechnete Temperaturfahnen zu evaluieren. Der Vergleich mit den analytischen Berechnungsmethoden zeigt, dass diese die Ausbreitung der Temperaturfahnen in Relation zu den numerischen Modellen grundsätzlich überschätzen. Für kleine und mittlere Anlagen ( 10 l/s) ist aufgrund der weiträumigeren Ausbreitung der Temperaturfelder die Simulation mit einem kalibrierten numerischen Planungsmodell zu empfehlen.

Hydrological models are widely used to characterize, understand and manage hydrosystems. Lumped parameter models are of particular interest in karst environments given the complexity and heterogeneity of these systems. There is a multitude of lumped parameter modelling approaches, which can make it difficult for a manager or researcher to choose. We therefore conducted a comparison of two lumped parameter modelling approaches: artificial neural networks (ANNs) and reservoir models. We investigate five karst systems in the Mediterranean and Alpine regions with different characteristics in terms of climatic conditions, hydrogeological properties and data availability. We compare the results of ANN and reservoir modelling approaches using several performance criteria over different hydrological periods. The results show that both ANNs and reservoir models can accurately simulate karst spring discharge but also that they have different advantages and drawbacks: (i) ANN models are very flexible regarding the format and amount of input data, (ii) reservoir models can provide good results even with a few years of relevant discharge in the calibration period and (iii) ANN models seem robust for reproducing high-flow conditions, while reservoir models are superior in reproducing low-flow conditions. However, both modelling approaches struggle to reproduce extreme events (droughts, floods), which is a known problem in hydrological modelling. For research purposes, ANN models have been shown to be useful for identifying recharge areas and delineating catchments, based on insights into the input data. Reservoir models are adapted to understand the hydrological functioning of a system by studying model structure and parameters.

Performance criteria play a key role in the calibration and evaluation of hydrological models and have been extensively developed and studied, but some of the most used criteria still have unknown pitfalls. This study set out to examine counterbalancing errors, which are inherent to the Kling–Gupta efficiency (KGE) and its variants. A total of nine performance criteria – including the KGE and its variants, as well as the Nash–Sutcliffe efficiency (NSE) and the modified index of agreement (d1) – were analysed using synthetic time series and a real case study. Results showed that, when assessing a simulation, the score of the KGE and some of its variants can be increased by concurrent overestimation and underestimation of discharge. These counterbalancing errors may favour bias and variability parameters, therefore preserving an overall high score of the performance criteria. As bias and variability parameters generally account for two-thirds of the weight in the equation of performance criteria such as the KGE, this can lead to an overall higher criterion score without being associated with an increase in model relevance. We recommend using (i) performance criteria that are not or less prone to counterbalancing errors (d1, modified KGE, non-parametric KGE, diagnostic efficiency) and/or (ii) scaling factors in the equation to reduce the influence of relative parameters.

Karst springs are the natural outflow of karst water to the surface. These springs occur where the water table can reach the surface unimpeded. This study examines the effect of alluvial deposits with varying thickness and permeability, covering the main outlet (karst spring) of a karst network on karst drainage (e.g., development of the karst water, drainage patterns, conduit-matrix interaction) as a result of a positive base level shift. This was realized with a numerical conceptual model (FEFLOW) of a hypothetical karst aquifer with 6 model configurations (inactive vs. active conduit flow, free vs. confined spring conditions with 20 m and 50 m sediment cover, respectively, with low and high hydraulic conductivity). Conduit flow and coupled conduit-matrix interactions were incorporated into the model with one-dimensional discrete feature elements. The results show that the permeability of the sediments has a more distinctive effect on conduit discharge than their thickness. The conduit network significantly contributes to the drainage even with a fully confined spring outlet. The conduit system acts as a water collector from the matrix in the recharge zone. The buried outlet increases the hydrostatic pressure farther along the conduit, and water is pushed upwards back into the matrix in the vicinity of the stratigraphic contact. Depending on the depositional setting, this results in the evolution of one to multiple new flow systems towards new potential spring sites. The results obtained here provide insight into the likely responses of natural karst systems.

Groundwater level time series are the most common source of information on the dynamics of subsurface water resources. The modeling of such time series is of crucial importance to increase our understanding of the system and make predictions on future changes. This becomes ever more important with global challenges such as climate change and human over-exploitation of groundwater resources. Different types of models can be applied to model groundwater level time series, ranging from purely statistical models (black-box), through lumped conceptual models (grey-box), to physically based models (white-box). Traditionally, physically based, distributed models are predominantly used to solve groundwater problems. In recent years, the use of grey- and black-box models has been receiving increased attention. In this poster we will present the “Groundwater Time Series Modeling Challenge”, which aims to systematically compare different methodologies to model groundwater level time series. We challenge researchers to model a predefined set of groundwater time series observed in unconsolidated aquifers worldwide, set in a variety of physiographic and climatic conditions. Participating groups are free to use the model of their choice, but are required to use the same forcing data and periods for calibration. Model performance will be centrally assessed by the organizers using non-public validation data set, which will be made public after the challenge. A more detailed description of the rules for this challenge and all groundwater and forcing data is available in a GitHub repository at https://github.com/gwmodeling/challenge. The results of the challenge will be presented at the General Assembly of the EGU in 2023 and reported in a peer-reviewed paper. With this challenge, we hope to increase the awareness in the groundwater community about all the different types of models that are available to model groundwater level time series.

Performance criteria such as the mean squared error (MSE), the Nash-Sutcliffe efficiency (NSE) and the Kling-Glupta efficiency (KGE) are extensively used to calibrate hydrological models. In recent years, numerous authors have stressed the inherent limitations of squared-error based criteria such as MSE and NSE. As a result, KGE criterion is gaining in popularity and is being widely used for calibration and for assessment. KGE has been initially proposed to address the poor consideration of discharge variability by NSE, but it also helps to lower the impact of squared errors in highly variable time series. KGE is a combination of (i) the Pearson correlation coefficient (r), (ii) the ratio between simulated and observed means (β), and (iii) the ratio between simulated and observed variances (α). In this study, we used KGE to compare the performance of two karst hydrological models (ANN and LP) over different flow regimes (dry, intermediate, wet) of two karst springs. The considered karst systems exhibit high contrasts in geometrical and hydrodynamic properties, inducing a high variability of the discharge at the springs. The discharge time series were divided into three sub-time series (dry, intermediate, and wet flows) according to fixed thresholds of discharge values. KGE values were higher for LP model for each sub-time series of both karst systems, thus indicating a better performance of LP over ANN at dry, intermediate and wet flows. However, KGE of the whole discharge time series were higher for ANN model, thus indicating a better overall performance of ANN over LP. The analysis of the decomposition of KGE (r, β, α) alongside a visual assessment of the simulated discharges of both models revealed that a compensation bias may be induced by β and α parameters. Simultaneous and equal overestimations and underestimations of multiple parts of the discharge time series seem to favour β and α values, leading to an overall better KGE coefficient without being associated to an increased model relevance.

Groundwater monitoring and specific collection of data on the spatio-temporal dynamics of the aquifer are prerequisites for effective groundwater management and determine nearly all downstream management decisions. An optimally designed groundwater monitoring network will provide the maximum information content at the minimum cost (Pareto optimum). In this study, PySensors, a Python package containing scalable, data-driven algorithms for sparse sensor selection and signal reconstruction with dimensionality reduction is applied to an existing groundwater monitoring network (GMN) in 1D (hydrographs) and 2D (gridded groundwater contour maps). The algorithm first fits a basis object to the training data, then applies a computationally efficient QR algorithm that ranks existing monitoring wells (for 1D) or suitable sites for additional monitoring (for 2D) in order of "importance" based on the state reconstruction to this tailored basis. This procedure enables a network to be reduced or extended along the Pareto front. Moreover, we investigate the effect of basis choice on reconstruction performance by comparing three types typically used for sparse sensor selection (identity, random projection, and singular value decomposition resp. principal component analysis). We define a gridded cost function for the extension case penalizes unsuitable locations. Our results show that this approach is generally better than the best randomly selected wells. The optimized reduction makes it possible to adequately reconstruct the removed hydrographs with a highly reduced subset with low loss. An average absolute reconstruction accuracy of 0.1 m is achieved with a subset of 6 % wells, 0.05 m with 31 %, and 0.01 m with 82 % wells.

International audience; Despite many existing approaches, modeling karst water resources remains challenging as conventional approaches usually heavily rely on distinct system knowledge. Artificial neural networks (ANNs), however, require only little prior knowledge to automatically establish an input-output relationship. For ANN modeling in karst, the temporal and spatial data availability is often an important constraint, as usually no or few climate stations are located within or near karst spring catchments. Hence, spatial coverage is often not satisfactory and can result in substantial uncertainties about the true conditions in the catchment, leading to lower model performance. To overcome these problems, we apply convolutional neural networks (CNNs) to simulate karst spring discharge and to directly learn from spatially distributed climate input data (combined 2D-1D CNNs). We investigate three karst spring catchments in the Alpine and Mediterranean region with different meteorological-hydrological characteristics and hydrodynamic system properties. We compare the proposed approach both to existing modeling studies in these regions and to our own 1D CNN models that are conventionally trained with climate station input data. Our results show that all the models are excellently suited to modeling karst spring discharge (NSE: 0.73-0.87, KGE: 0.63-0.86) and can compete with the simulation results of existing approaches in the respective areas. The 2D models show a better fit than the 1D models in two of three cases and automatically learn to focus on the relevant areas of the input domain. By performing a spatial input sensitivity analysis, we can further show their usefulness in localizing the position of karst catchments.

International audience; The complexity of karst groundwater flow modelling is reflected by the amount of simulation approaches. The goal of the Karst Modelling Challenge (KMC) is comparing different approaches on one single system using the same data set. Thirteen teams with different computational models for simulating discharge variations at karst springs have applied their respective models on one single data set coming from the Milandre Karst Hydrogeological System (MKHS). The approaches include neural networks, reservoir models, semi-distributed models and fully distributed groundwater models. Four and a half years of hourly or daily meteorological input and hourly discharge data were provided for model calibration. The validation comprised forecasting one year of discharge, without the observed discharge data. The model performance was evaluated using the volume conservation, Nash-Sutcliffe efficiency (NSE) and the Kling-Gupta efficiency (KGE) applied on the total discharge and individual flow components. As a result, the comparison of model performances is a challenging task due to the differences in the model architecture but also required time steps: some of the models require aggregated daily steps while others could be run using hourly data, which provided some interesting differences depending on how the data was transformed. The use of instantaneous data (e.g. value at noon) produces less bias that averaging hourly data over one day. The transformation of hourly into daily data produces a decrease of Nash and KGE of 0.05 to 0.08 (i.e. from 1 to ~0.93). The resulting simulations (forecasted values for year 2016) produced KGEs ranging between 0.83 and 0.37 (0.83 to −0.24 for NSE). Although the simulations matched the monitored flows reasonably well, most models struggled to simulate baseflow conditions accurately. In general, the models that performed the best for this exercise were the global ones (Gardenia and Varkarst), with a limited number of parameters, which can be calibrated using automatic calibration procedures. The neural network models also showed a fair potential, with one providing reasonable results despite the relatively short dataset available for warming-up (4.5 years). Semi-and fully distributed models also suggested that with some more effort they could perform well. The accuracy of model predictions does not seem to increase by using models with more than 9–12 calibration parameters. An evaluation of the relative errors between the forecasted and the observed values revealed that for most models, 50% of the forecasted values contained more than 50% of difference against the observed discharge rate, with 25% having a difference larger than 100%. A significant part of the poorly forecasted values corresponded to base-flow which was surprising given that as base-flow is generally much easier to predict than peak flow. Hence, this shows that modelling approaches and criteria for the calibration are too oriented towards peak-flow sections of the hydrographs, and that improvements could be gained by more focus on the base-flow.

Improving water availability and distribution is key to combatting water stress in semi-arid regions. This study presents an approach that integrates hydrogeological surveys and economic analysis at a pilot site in the eastern Lower Jordan Valley to assess the potential for managed aquifer recharge (MAR). Based on exploratory drillings, hydraulic tests, water-quality analyses and previous studies, an appropriate storage space in the alluvial fans along the eastern margin of the valley was found. Average measured infiltration rates of 9.8 × 10−5 m/s in a gravel pit and average hydraulic conductivity values of 6.8 × 10−5 m/s in the aquifer indicate that up to 1 Mm3/year can be infiltrated and stored underground. It is assumed that water is trapped towards the center of the valley where the alluvial sediments inter-finger with fine-grained layers. An increase in electrical conductivity values in the same direction probably indicates irrigation return flow and the presence of evaporitic layers and suggests a water recovery before reaching these saline zones. Depending on whether the water provided by MAR is used for irrigation or domestic water purpose, an average incremental profit of 0.81 and 2.44 US$/m3 can be achieved, respectively. These calculations result in a net present value of the MAR plant of 10.6 or 31.9 million US$, respectively, over a lifetime of 30 years. The presented study serves as a basis for further investigations and planning of a MAR plant, and can be transferred and upscaled to other sites in the Jordan Valley.

Groundwater is an important resource for drinking water supply, and is subject to natural variation as well as climate change effects. It has been shown that long-term natural variations of groundwater levels can often be attributed to climatic oscillations. Long-term groundwater level records are rare, but of special importance for the detection of longer, decadal to multi-decadal periodicities, which are vital for predictions of future development of groundwater levels and the distinction between natural variation and climate change effects. We have examined periodicities of nine groundwater level time-series with records of more than 100 years, as well as possible impacts of climatic teleconnections (NAO, AMO and ENSO) with wavelet analysis. The monitoring wells are located in Germany, Netherlands, UK and Denmark and cover different depths to groundwater, different aquifer types and hydraulic conditions. Our results show that all evaluated monitoring wells exhibit significant relations to long-term climatic periodicities of NAO, AMO and ENSO. Among the wells in phreatic porous aquifers, there is a signal damping, which can be related to the thickness of the unsaturated zone. Further, the damping is higher in the lower permeable aquifers and there is less damping in the karstic aquifers compared to the porous aquifers, in spite of a much thicker unsaturated zone.

In this study we investigate how climate change will directly influence the groundwater resources in Germany during the 21st century. We apply a machine learning groundwater level prediction framework, based on convolutional neural networks to 118 sites well distributed over Germany to assess the groundwater level development under the RCP8.5 scenario, based on six selected climate projections, which represent 80% of the bandwidth of the possible future climate signal for Germany. We consider only direct meteorological inputs, while highly uncertain anthropogenic factors such as groundwater extractions are excluded. We detected significant declining trends of groundwater levels for most of the sites, revealing a spatial pattern of stronger decreases especially in the northern and eastern part of Germany, emphasizing already existing decreasing trends in these regions. We can further show an increased variability and longer periods of low groundwater levels during the annual cycle towards the end of the century.

Clear signs of climate stress on groundwater resources have been observed in recent years even in generally water-rich regions such as Germany. Severe droughts, resulting in decreased groundwater recharge, led to declining groundwater levels in many regions and even local drinking water shortages have occurred in past summers. We investigate how climate change will directly influence the groundwater resources in Germany until the year 2100. For this purpose, we use a machine learning groundwater level forecasting framework, based on Convolutional Neural Networks, which has already proven its suitability in modelling groundwater levels. We predict groundwater levels on more than 120 wells distributed over the entire area of Germany that showed strong reactions to meteorological signals in the past. The inputs are derived from the RCP8.5 scenario of six climate models, pre-selected and pre-processed by the German Meteorological Service, thus representing large parts of the range of the expected change in the next 80 years. Our models are based on precipitation and temperature and are carefully evaluated in the past and only wells with models reaching high forecasting skill scores are included in our study. We only consider natural climate change effects based on meteorological changes, while highly uncertain human factors, such as increased groundwater abstraction or irrigation effects, remain unconsidered due to a lack of reliable input data. We can show significant (p

Hydrograph clustering helps to identify dynamic patterns within aquifers systems, an important foundation of characterizing groundwater systems and their influences, which is necessary to effectively manage groundwater resources. We develope an unsupervised modeling approach to characterize and cluster hydrographs on regional scale according to their dynamics. We apply feature-based clustering to improve the exploitation of heterogeneous datasets, explore the usefulness of existing features and propose new features specifically useful to describe groundwater hydrographs. The clustering itself is based on a powerful combination of Self-Organizing Maps with a modified DS2L-Algorithm, which automatically derives the cluster number but also allows to influence the level of detail of the clustering. We further develop a framework that combines these methods with ensemble modeling, internal cluster validation indices, resampling and consensus voting to finally obtain a robust clustering result and remove arbitrariness from the feature selection process. Further we propose a measure to sort hydrographs within clusters, useful for both interpretability and visualization. We test the framework with weekly data from the Upper Rhine Graben System, using more than 1800 hydrographs from a period of 30 years (1986-2016). The results show that our approach is adaptively capable of identifying homogeneous groups of hydrograph dynamics. The resulting clusters show both spatially known and unknown patterns, some of which correspond clearly to external controlling factors, such as intensive groundwater management in the northern part of the test area. This framework is easily transferable to other regions and, by adapting the describing features, also to other time series-clustering applications.

Karst systems are challenging to model due to their heterogeneous hydraulic properties resulting in highly variable discharge behavior. Distributed models can be applied to karst aquifers but require detailed system knowledge and extensive hydraulic parameter datasets; lumped-parameter models are less complex, but still require parametrization. In this work, we demonstrate the application of a data-driven approach to model the discharge behavior of the Aubach spring in the Gottesacker karst system in the northern Alps, a well-investigated study site for which previous models are available for comparison (Chen et al. 2018; Fandel et al. 2020). Our approach is based on convolutional neural networks (CNN), which have proved to be well suited for time series forecasting in water-related contexts like runoff modelling or groundwater level prediction (Wunsch et al.). The approach is comparably simple in terms of data requirements as we rely mainly on widely available and easy-to-measure parameters such as precipitation and temperature. By implementing Bayesian techniques (Monte-Carlo dropout) we are able to report the predictive uncertainty of the CNN based forecasts. Our results challenge existing modelling results based on lumped-parameter models in terms of common error measures such as Nash-Sutcliffe efficiency. Furthermore, we explore the important role of snow accumulation and melting by coupling our model with a snow-routine to better represent their influence on spring discharge and further improve model performance. Our results demonstrate that the presented machine-learning approach can be applied to simulate karst spring discharge and has certain advantages in comparison with conventional karst modelling approaches, which require hydraulic parameters that are often not available.Chen, Z.; Hartmann, A.; Wagener, T.; Goldscheider, N. (2018) Dynamics of water fluxes and storages in an Alpine karst catchment under current and potential future climate conditions. Hydrology and earth system sciences, 22 (7), 3807–3823.Fandel, C.; Ferré, T.; Chen, Z.; Renard, P.; Goldscheider, N. (2020) A model ensemble generator to explore structural uncertainty in karst systems with unmapped conduits. Hydrogeology journal, published online.Wunsch, A.; Liesch, T.; Broda, S. (2020) Groundwater Level Forecasting with Artificial Neural Networks: A Comparison of LSTM, CNN and NARX. Hydrology and Earth System Sciences Discussions 2020:1–23. https://doi.org/10/ghtcz3

It is now well established to use shallow artificial neural networks (ANN) to obtain accurate and reliable groundwater level forecasts, which are an important tool for sustainable groundwater management. However, we observe an increasing shift from conventional shallow ANNs to state-of-the-art deep learning (DL) techniques, but a direct comparison of the performance is often lacking. Although they have already clearly proven their suitability, especially shallow recurrent networks frequently seem to be excluded from the study design despite the euphoria about new DL techniques and its successes in various disciplines. Therefore, we aim to provide an overview on the predictive ability in terms of groundwater levels of shallow conventional recurrent ANN namely nonlinear autoregressive networks with exogenous inputs (NARX), and popular state-of-the-art DL-techniques such as long short-term memory (LSTM) and convolutional neural networks (CNN). We compare both the performance on sequence-to-value (seq2val) and sequence-to-sequence (seq2seq) forecasting on a 4-year period, while using only few, widely available and easy to measure meteorological input parameters, which makes our approach widely applicable. We observe that for seq2val forecasts NARX models on average perform best, however, CNNs are much faster and only slightly worse in terms of accuracy. For seq2seq forecasts, mostly NARX outperform both DL-models and even almost reach the speed of CNNs. However, NARX are the least robust against initialization effects, which nevertheless can be handled easily using ensemble forecasting. We showed that shallow neural networks, such as NARX, should not be neglected in comparison to DL-techniques; however, LSTMs and CNNs might perform substantially better with a larger data set, where DL really can demonstrate its strengths, which is rarely available in the groundwater domain though.

While the application of neural networks for groundwater level forecasting in general has been investigated by many authors, the use of nonlinear autoregressive networks with exogenous inputs (NARX) is relatively new. For this work NARX were applied to obtain groundwater level forecasts for several wells in southwest Germany. Wells in porous, fractured and karst aquifers were investigated and forecasts of lead times up to half a year were conducted for both influenced (e.g. nearby pumping) and uninfluenced wells. Precipitation and temperature were chosen as predictors, which makes the selected approach easily transferable, since both parameters are widely available and simple to measure. Input and feedback delays were determined by applying STL time series decomposition on the data and using auto- and cross-correlation functions on the remainders to determine significant time lags. Coefficient of determination, (relative) root mean squared error and Nash-Sutcliffe efficiency were used to evaluate forecasts, the model selection was based on an out-of-sample validation on rolling basis. The results are promising and indicate an outstanding suitability of NARX for groundwater level predictions with such a small set of inputs in all three aquifer types.

The selection of the best possible method to interpolate a continuous groundwater surface from point data of groundwater levels is a controversial issue. In the present study four deterministic and five geostatistical interpolation methods (global polynomial interpolation, local polynomial interpolation, inverse distance weighting, radial basis function, simple-, ordinary-, universal-, empirical Bayesian and co-Kriging) and six error statistics (ME, MAE, MAPE, RMSE, RMSSE, Pearson R) were examined for a Jurassic karst aquifer and a Quaternary alluvial aquifer. We investigated the possible propagation of uncertainty of the chosen interpolation method on the calculation of the estimated vertical groundwater exchange between the aquifers. Furthermore, we validated the results with eco-hydrogeological data including the comparison between calculated groundwater depths and geographic locations of karst springs, wetlands and surface waters. These results show, that calculated inter-aquifer exchange rates based on different interpolations of groundwater potentials may vary greatly depending on the chosen interpolation method (by factor >10). Therefore, the choice of an interpolation method should be made with care, taking different error measures as well as additional data for plausibility control into account. The most accurate results have been obtained with co-Kriging incorporating secondary data (e.g. topography, river levels).

To protect karst spring water resources, catchments must be known. We have developed a method for correlating spring hydrographs with newly available, high-resolution, satellite-based Global Precipitation Measurement data to rapidly and remotely locate recharge areas. We verify the method using a synthetic comparison of ground-based rain gage data with the satellite precipitation data set. Application to karst springs is proven by correlating satellite data with hydrographs from well-known springs with published catchments in Europe and North America. Application to an unknown-catchment spring in Pennsylvania suggests distant recharge, requiring a flow path that crosses topographic divides, as well as multiple lithologies, physiographic provinces, and tectonic boundaries. Although surprising, this latter result is consistent with published geologic/geophysical, monitoring well, and stream gage data. We conclude that the method has considerable potential to improve the speed and accuracy of catchment identification and hydrodynamic characterization, with applications to water resource protection and groundwater exploration, among others.

Karst aquifers in semi-arid regions are particularly threatened by surface contamination, especially during winter seasons when extremely variable rainfall of high intensities prevails. An additional challenge is posed when managed recharge of storm water is applied, since karst aquifers display a high spatial variability of hydraulic properties. In these cases, adapted protection concepts are required to address the interaction of surface water and groundwater. In this study a combined protection approach for the surface catchment of the managed aquifer recharge site at the Wala reservoir in Jordan and the downstream Hidan wellfield, which are both subject to frequent bacteriological contamination, is developed. The variability of groundwater quality was evaluated by correlating contamination events to rainfall, and to recharge from the reservoir. Both trigger increased wadi flow downstream of the reservoir by surface runoff generation and groundwater seepage, respectively. A tracer test verified the major pathway of the surface flow into the underground by infiltrating from pools along Wadi Wala. An intrinsic karst vulnerability and risk map was adapted to the regional characteristics and developed to account for the catchment separation by the Wala Dam and the interaction of surface water and groundwater. Implementation of the proposed protection zones for the wellfield and the reservoir is highly recommended, since the results suggest an extreme contamination risk resulting from livestock farming, arable agriculture and human occupation along the wadi. The applied methods can be transferred to other managed aquifer recharge sites in similar karstic environments of semi-arid regions.

To deal with the challenge of groundwater over-extraction in arid and semi-arid environments, it is necessary to establish management strategies based on the knowledge of hydrogeological conditions, which can be difficult in places where hydrogeological data are dispersed, scarce or present potential misinformation. Groundwater levels in the southern Jordan Valley (Jordan) have decreased drastically in the last three decades, caused by over-extraction of groundwater for irrigation purposes. This study presents a local, two-dimensional and transient numerical groundwater model, using MODFLOW, to characterise the groundwater system and the water balance in the southern Jordan Valley. Furthermore, scenarios are simulated regarding hydrological conditions and management options, like extension of arable land and closure of illegal wells, influencing the projection of groundwater extraction. A limited dataset, literature values, field surveys, and the ‘crop water-requirement method’ are combined to determine boundary conditions, aquifer parameters, and sources and sinks. The model results show good agreement between predicted and observed values; groundwater-level contours agree with the conceptual model and expected flow direction, and, in terms of water balance, flow volumes are in accordance with literature values. Average annual water consumption for irrigation is estimated to be 29 million m3 and simulation results show that a reduction of groundwater pumping by 40% could recover groundwater heads, reducing the water taken from storage. This study presents an example of how to develop a local numerical groundwater model to support management strategies under the condition of data scarcity.

Effective groundwater monitoring networks are important, as systematic data collected at observation wells provide a crucial understanding of the dynamics of hydrogeological systems as well as the basis for many other applications. This study investigates the influence of six groundwater level monitoring network (GLMN) sampling designs (random, grid, spatial coverage, and geostatistical) with varying densities on the accuracy of spatially interpolated groundwater surfaces. To obtain spatially continuous prediction errors (in contrast to point cross‐validation errors), we used nine potentiometric groundwater surfaces from three regional MODFLOW groundwater flow models with different resolutions as a priori references. To assess the suitability of frequently‐used cross‐validation error statistics (MAE, RMSE, RMSSE, ASE, and NSE), we compared them with the actual prediction errors (APE). Additionally, we defined upper and lower thresholds for an appropriate spatial density of monitoring wells. Below the lower threshold, the observation density appears insufficient, and additional wells lead to a significant improvement of the results. Above the upper threshold, additional wells lead to only minor and inefficient improvements. According to the APE, systematic sampling lead to the best results but is often not suited for GLMN due to its nonprogressive characteristic. Geostatistical and spatial coverage sampling are considerable alternatives, which are in contrast progressive and allow evenly spaced and, in the case of spatial coverage sampling, yet reproducible coverage with accurate results. We found that the global cross‐validation error statistics are not suitable to compare the performance of different sampling designs, although they allow rough conclusions about the quality of the GLMN.

Gravity Recovery and Climate Experiment (GRACE) derived groundwater storage (GWS) data are compared with in-situ groundwater levels from five groundwater basins in Jordan, using newly gridded GRACE GRCTellus land data. It is shown that (1) the time series for GRACE-derived GWS data and in-situ groundwater-level measurements can be correlated, with R 2 from 0.55 to 0.74, (2) the correlation can be widely ascribed to the seasonal and trend component, since the detrended and deseasonalized time series show no significant correlation for most cases, implying that anomalous signals that deviate from the trend or seasonal behaviour are overlaid by noise, (3) estimates for water losses in Jordan based on the trend of GRACE data from 2003 to 2013 could be up to four times higher than previously assumed using estimated recharge and abstraction rates, and (4) a significant time-lagged cross correlation of the monthly changes in GRACE-derived groundwater storage and precipitation data was found, suggesting that the conventional method for deriving GWS from GRACE data probably does not account for the typical conditions in the study basins. Furthermore, a new method for deriving plausible specific yields from GRACE data and groundwater levels is demonstrated.

Numerous damages at structures and failures of dams can be ascribed to solution processes and attendant subsidence. Leaching in gypsum-bearing layers of the Middle Triassic has induced subsidence underneath the structure of the weir Hessigheim, River Neckar. The goals of this study were to gain a better understanding of the prevailing groundwater flow processes as well as the interaction between surface water and groundwater in the area of the weir lock and to verify the success of former remediation measures. In order to attain these objectives a combined groundwater and surface water tracer test in combination with hydraulic borehole tests were performed. In addition, a study of the dissolved sulfate distribution was carried out using sampling data from 1987 to 2013. Based on the results of these methods a strongly heterogeneous aquifer with main runoff occurring in few major flow paths could be characterized. Former remediation measures showed to be successful.

a b s t r a c t The feasibility of combined extraction–injection one-well systems for geothermal groundwater utilization was evaluated by means of a numerical parameter study. One-well systems require little space, but it is critically important to avoid circulation flow resulting in a thermal short circuit. Numerical models of groundwater flow and heat transport were performed to check the influence of various hydrogeological and technical parameters. Results show that circulation flow can be avoided in most cases, mainly when hydraulic gradients and conductivities are high and when the distance between the two well screens is sufficient. One-well systems are thus favourable options in urban sand-and-gravel aquifers.

Storm-water harvesting and storage via managed aquifer recharge (MAR) is a promising approach to combat water scarcity in semi-arid regions, but poses a challenge for karst aquifers and regions with highly variable water availability. The infiltration of low-mineralized surface water and its impact on highly mineralized groundwater of a karst aquifer was investigated at Wala reservoir in Jordan over a period of approximately 10 years. The results show significant groundwater-level rise in a wellfield, in response to the yearly average infiltration of about 6.7 million m3. This corresponds to about 60 % of the yearly average abstraction of about 11.7 million m3, confirmed by mixing calculations with tritium. A decreasing trend in infiltration due to sedimentation is observed. Mean groundwater residence times of several thousand years, derived from carbon-14 dating, indicate a large storage capacity of the aquifer. The heterogeneous distribution of the residence times is caused by strong groundwater withdrawals and artificial recharge along with karst-specific aquifer characteristics. Temporal groundwater salinity fluctuations in the wellfield are observed after the first MAR infiltration. Enhanced groundwater flow along the wadi course was demonstrated, which is an important aspect with regards to future MAR projects in similar wadis of the region.

The construction of the combined ship lock weir in Hessigheim, Germany, created a river stage difference of more than 6 m and intensified subsequently the surface–subsurface water exchange and modified the natural subsurface flow field. In the sulfate-bearing hydrogeological setting, increased infiltration and intensified subsurface flow triggered related dissolution processes. Soon after construction, the structure started suffering from subsidence due to progressive leaching and demanded extensive grouting. In the course of approximately 25 years, groundwater heads and subsurface sulfate concentrations responded distinctively to grouting campaigns. The analysis confirms that as a result of its high solubility and altered hydrogeological conditions, gypsum karst develops on a human rather than on a geological time scale. Grouting measures considerably slowed down these processes, however, at the cost of subsurface flow field variations which will ultimately transfer the zone of enhanced dissolution.

Water supply in developing countries is prone to large water losses due to leaky distribution networks and defective sewers, which may affect groundwater quality and quantity in urban areas and result in complex subsurface mixing dynamics. In this study, a multi-stable isotope approach was used to investigate spatiotemporal fluctuations of surface and sub-surface water source partitioning and mixing, and to assess nitrogen (N) contamination in the urban water cycle of As-Salt, Jordan. Water import from the King Abdullah Canal (KAC), mains waters from the network, and wastewater are characterized by distinct isotopic signatures, which allowed us to quantify city effluents into the groundwater. Temporal variations in isotopic signatures of polluted groundwater are explained by seasonally fluctuating inflow, and dilution by water that originates from Lake Tiberias and enters the urban water cycle via the KAC. Isotopic analysis (N and O) and comparison between groundwater nitrate and nitrate from mains water, water imports and wastewater confirmed that septic waste from leaky sewers is the main contributor of nitrate contamination. The nitrate of strongly contaminated groundwater was characterized by highest δ15NNO3 values (13.3 ± 1.8‰), whereas lowest δ15NNO3 values were measured in unpolluted groundwater (6.9‰). Analogously, nitrate concentration and isotopic ratios were used for source partitioning and qualitatively confirmed δDH2O and δ18OH2O-based estimates. Dual water isotope endmember mixing calculations suggest that city effluents from leaky networks and sewers contribute 30–64% to the heavily polluted groundwater. Ternary mixing calculations including also chloride revealed that 5–18% of the polluted groundwater is wastewater. Up to two thirds of the groundwater originates from mains, indicating excessive water loss from the network, and calling for improved water supply management.

The Lower Jordan Valley in the Middle East comprises parts of Israel, Jordan and the Palestinian Territories and is amongst the water scarcest regions worldwide. As the average water availability per person is below 150 m³/year, this region is facing absolute water scarcity. The water resource systems are overexploited and subject to further stress through high population growth, climate change and the recent migration of refugees from Syria. Declining water levels of the Dead Sea and within the regional aquifers are visible evidence of overexploitation. Further, salinization and contamination including pharmaceutical residues in surface and groundwater indicate the strong deterioration of water quality by anthropogenic activities. Integrated Water Resource Management (IWRM) is a concept to address these water related issues but needs to be incorporated into the water strategies of the riparian states. SMART (Sustainable Management of Available Resources with Innovative Technologies) is a research and development project within the framework of IWRM and has the overall goal to develop a transferable concept for a sustainable water resource management in the Lower Jordan Valley. Besides natural available freshwater, unconventional sources such as treated waste water, artificially recharged groundwater, and desalinated brackish groundwater are considered. Based on detailed water balances, management concepts have been developed and applied to allow the assessment of the current state of the water resources, as well as the future development, taking the actions of the respective water strategy into account. The SMART approach is transferable to other regions in the world suffering under natural and/or manmade water scarcity.

International audience; In semi-arid regions with high seasonal variability of water availability, adaptive management strategies and technical measures are required to ensure the sustainable use of water resources. In this study, managed recharge of storm water into a karst aquifer and the water level fluctuations related to pumping in a nearby wellfield were simulated at Wadi Wala, Jordan. We used a numerical equivalent porous medium (EPM) approach with specific adaptations to account for the heterogeneity and anisotropy of the karst aquifer. The model domain was vertically projected along the wadi course, resulting in a 2-dimensional model, and subdivided into hydraulic zones representing the karst-specific flow pattern of fast flow and slow depletion. Results show satisfying agreement of measured and simulated groundwater tables from 2002 to 2012 and predict a lowering of the average groundwater table until 2022 of around 2.7m in the immediate surroundings of the reservoir and an increased depletion towards the wellfield, mainly caused by sedimentation in the reservoir and an associated decrease in infiltration. Abstraction at the wellfield changed considerably over the regarded time period and strongly influences the groundwater fluctuations, which shows the need of improved pumping management and monitoring. The results can serve as a basis for decision makers regarding an optimization of water management at the reservoir and wellfield. Furthermore, the presented numerical approach can be transferred to karst regions with similar physio-geographical conditions to assess managed aquifer recharge.

Due to its radiological and toxicological properties even at low concentration levels, uranium is increasingly recognized as relevant contaminant in drinking water from aquifers. Uranium originates from different sources, including natural or geogenic, mining and industrial activities, and fertilizers in agriculture. The goal of this study was to obtain insights into the origin of uranium in groundwater while differentiating between geogenic sources and fertilizers. A literature review concerning the sources and geochemical processes affecting the occurrence and distribution of uranium in the lithosphere, pedosphere and hydrosphere provided the background for the evaluation of data on uranium in groundwater at regional scale. The state of Baden-Wurttemberg, Germany, was selected for this study, because of its hydrogeological and land-use diversity, and for reasons of data availability. Uranium and other parameters from N=1935 groundwater monitoring sites were analyzed statistically and geospatially. Results show that (i) 1.6% of all water samples exceed the German legal limit for drinking water (10 μg/L); (ii) The range and spatial distribution of uranium and occasional peak values seem to be related to geogenic sources; (iii) There is a clear relation between agricultural land-use and low-level uranium concentrations, indicating that fertilizers generate a measurable but low background of uranium in groundwater.