HS8.2.9 | Data-driven groundwater modeling: methods, applications & challenges
EDI
Data-driven groundwater modeling: methods, applications & challenges
Convener: Ezra Haaf | Co-conveners: Tanja Liesch, Raoul Collenteur, Inga Retike, Mark Bakker
Orals
| Wed, 26 Apr, 14:00–17:35 (CEST)
 
Room 2.31
Posters on site
| Attendance Wed, 26 Apr, 10:45–12:30 (CEST)
 
Hall A
Posters virtual
| Attendance Wed, 26 Apr, 10:45–12:30 (CEST)
 
vHall HS
Orals |
Wed, 14:00
Wed, 10:45
Wed, 10:45
Data-driven models are increasingly used to solve groundwater problems. These types of models require less knowledge about the subsurface and depend more on input and output data. Most common groundwater measurements are groundwater levels and, to a lesser extent, groundwater quality. The overarching question is how to obtain as much information as possible from these measurements. Data-driven models include, but are not limited to, time series models, machine learning models, AI models, statistical models, and lumped groundwater models. Models can, for example, be used to predict future groundwater levels or groundwater quality parameters, determine the effect of anthropogenic activity, or analyze data to support more traditional groundwater modeling methods. In this session, we seek especially contributions on the development of:
- new and improved data-driven methods to model groundwater time series and point data,
- applications and comparative studies of existing methods to solve groundwater problems with data-driven models,
- approaches to typical challenges, such as non-stationary time series, irregular time steps and data scarcity in general,
- concepts and approaches for regionalization, e,g., transfer of model data to unmonitored sites using similarity-, regression- or signature-based methods,
- approaches to improve hydrogeological system understanding from data-driven models and their parameters.

Orals: Wed, 26 Apr | Room 2.31

Chairpersons: Ezra Haaf, Raoul Collenteur, Mark Bakker
14:00–14:05
Improving groundwater assessments with auxiliary information
14:05–14:15
|
EGU23-5322
|
ECS
|
On-site presentation
Moritz Gosses and Thomas Wöhling

With the increasing threat of climate change, projections of its impact on the availability of natural resources, such as groundwater, are becoming significantly more important. While extrapolation or prediction is a valid application for environmental system models, their assumptions and limitations are often not clearly communicated or investigated. This is especially true for data-driven models, which have been applied more frequently, and often with great success, to groundwater problems in the last decade. But are these techniques applicable to long-term future predictions of the impact of climate scenarios on groundwater resources, and if so, under which conditions and limitations?
Within the context of KlimaKonform (Technische Universität Dresden, 2023), a research project studying the impact of climate change on the Central German Uplands, ensembles of multi-layer perceptron (MLP) models have been derived to estimate groundwater levels for a variety of wells in a region in Saxony-Anhalt in Germany. Once trained to replicate historical time series with climatological input data such as precipitation and temperature, these model ensembles were then tasked to predict the groundwater levels up until the end of the current century under different climate scenarios.
We analyse the plausibility of the model ensemble predictions to shed light on the above-proposed question: what are factors (and metrics) of success, as well as limitations and possible failures, of data-driven methods (MLPs in this case) for long-term prediction? First, we propose that using ensembles of models, rather than “the single-best” trained model, is a necessity for such applications. We identify different methods of pre-processing of input and target data, structural model setup as well as target-oriented post-processing of ensemble simulations as vital factors for the successful application of MLPs to long-term prediction of groundwater levels under climate change scenarios. We further highlight remaining limitations and pose the question of how they could potentially be overcome.

 

Technische Universität Dresden, 2023. KlimaKonform: Forschungsprojekt KlimaKonform. https://klimakonform.uw.tu-dresden.de/

How to cite: Gosses, M. and Wöhling, T.: Potentials and limitations of MLPs for predicting groundwater heads under climate change, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-5322, https://doi.org/10.5194/egusphere-egu23-5322, 2023.

14:15–14:25
|
EGU23-2696
|
ECS
|
On-site presentation
Mariana Gomez, Maximilian Noelscher, Stefan Broda, and Andreas Hartmann

Groundwater level forecasting with machine learning has been widely studied due to its generally accurate results and little input data requirements. Furthermore, machine learning models for this purpose are set up and trained in a short time when compared to the effort required for process-based, numerical models. Despite the high performance of models obtained at specific locations, applying the same model architecture to multiple sites across a regional area might lead to contrasting accuracies. Likely causalities of this discrepancy in model performance have been barely examined in previous studies. Here, we investigate the link between model performance and the effects of geospatial site characteristics and time series features. Using precipitation and temperature as predictors, we model groundwater levels at approximately 500 observation wells in Lower Saxony, Germany, using a 1-D convolutional neural network with a fixed architecture and hyperparameters tuned for each time series individually. The performances are evaluated against geospatial and time series features using correlation coefficients. Model performance is negatively influenced at sites near waterworks and densely vegetated areas. Besides, the more complex the time series, the higher the metrics, but autocorrelation reduces the model performance. The new insights evidence that further information is required at certain locations to improve model accuracy due to external impacts.  

How to cite: Gomez, M., Noelscher, M., Broda, S., and Hartmann, A.: Performance assessment of groundwater level forecasting with deep learning: a case study of Lower Saxony, Germany., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2696, https://doi.org/10.5194/egusphere-egu23-2696, 2023.

14:25–14:35
|
EGU23-361
|
ECS
|
On-site presentation
Dolon Banerjee and Sayantan Ganguly

Over last few decades the Punjab region of India has been one of the country's leading contributors to agricultural products. The agricultural farms in the region are supplied with water from a well-established canal system and groundwater reserve in the state. The share of irrigated area in the region fed by canals and groundwater wells are 28 and 72%, respectively. The over and unscientific usage of groundwater over the years has resulted in groundwater depletion at an alarming rate. To help policymakers address the situation and develop effective plans, forecasting groundwater recharge for the future is utmost essential. The recharge process primarily governs the growth or depletion of groundwater reserve. Groundwater recharge is one of the most difficult phenomena to be quantified as it cannot be measured directly and is influenced by several processes varying spatially and temporally. Extensive research work for quantifying the groundwater recharge have been performed in the past. These investigations introduced a number of methodologies, including chemical tracers and physical procedures. These methods, however, being experimental in nature, involve significant time and investment. The use of machine learning algorithms to predict the recharge is promoted as a solution to these problems. These algorithms have proven to be efficient enough to deduce the recharge with very high accuracy. Through a variety of models, ranging from the most basic to one of the more intricate, we have attempted to forecast the recharge scenario in the Punjab region, India. Four machine learning algorithms, namely the Multi-linear regression model, Non-linear regression model (Random Forest), Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) have been employed in this study. The aim was to comprehend the dependence of groundwater recharge on the factors of temperature, precipitation, soil type, LULC, and ground slope. The observed recharge for every subsequent month in a 30-year period is calculated using the observed monthly groundwater level data from observation wells located throughout Punjab. The monthly temperature and precipitation data are used for the study while soil type and ground slope for the location of the observation stations are extracted from digital elevation models (DEMs). At intervals of three years, the LULC maps are created. The models are then used to forecast and compare with the available observation data after the entire data set was split into a training and testing set using the 80/20 method. The models were then assessed for their ability to predict observational data using the Root Mean Squared Error (RMSE) and Coefficient of Determination (R2) in each case. The groundwater recharge prediction is then performed using the model with the highest accuracy.

How to cite: Banerjee, D. and Ganguly, S.: Predicting future groundwater recharge scenario in the Punjab region of India using machine learning techniques, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-361, https://doi.org/10.5194/egusphere-egu23-361, 2023.

14:35–14:45
|
EGU23-12633
|
ECS
|
On-site presentation
Martin Le Mesnil, Jean-Raynald de Dreuzy, Luc Aquilina, Frédéric Gresselin, and Alexandre Gauvain

We determine the sensitivity of groundwater-induced flooding forecast to different modelling strategies (evolving model structure and quantity/nature of data). We test three calibration strategies on shallow coastal aquifers:

  • we use the river network as a proxy of aquifer seepage and determine uniform hydraulic conductivity and porosity. Despite limiting assumptions of surface/groundwater connections, this strategy could be broadly deployed with rapid advances in temporal and spatial river mapping at regional to national scales.
  • we calibrate uniform conductivity and porosity on hourly piezometric data time-series close to the seashore, which are controlled by both tide and inland aquifer recharge fluctuations. We investigate the limitations of the uniform assumptions and demonstrate the interests of coastal and inland forcing as complementary sources of information.
  • we introduce spatially variable conductivities and porosities and use additional inland piezometers. Hydraulic parameters are mapped to the main geological structures of the studied area.

We analyse the results of these 3 competing strategies on the hydraulic parameters and, in a more prospective way, on the spatial distribution of vulnerabilities to groundwater fluctuations. We perform our analysis within the “Rivages Normands 2100” research project, in several sites of Western Normandy (France). In this area, low-lying coastal areas and neighbouring lands are prone to groundwater table increase, leading to flooding of buried networks and building foundations. The site of Saint-Germain-sur-Ay (66 km2 coastal watershed) is equipped with 6 piezometers from which we collected 1.5 year-long hourly time series of groundwater level. It includes a low-elevation coastal area made up of sands and a continental area made up of schists.  The modelling approaches allowed us to simulate groundwater levels up to 2100, and to analyse the associated evolution of vulnerability. Simulations are performed using Modflow on a daily timestep, with a 75 m spatial resolution.

Results obtained from the 3 models consistently show that vulnerable areas are mostly clustered close to the shoreline. We also show that, in the studied watershed, the first calibration strategy using the river network data leads to long-term simulations close to the second calibration strategy using the piezometric data of the coastal area. Integration of extra piezometric data from the continental area in the third strategy provides more reliable simulations. This analysis is progressively deployed to other sites and extended to cost-benefit analyses including the costs of site instrumentation and the benefits of forecast reliability. Integration of flood zones aerial photography during extreme events is being implemented to the first modelling strategy, as well as wetland mapping.

How to cite: Le Mesnil, M., de Dreuzy, J.-R., Aquilina, L., Gresselin, F., and Gauvain, A.: Calibration of coastal hydrogeological models for the analysis of groundwater-induced flooding: from instrumentation to model reliability, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12633, https://doi.org/10.5194/egusphere-egu23-12633, 2023.

14:45–14:55
|
EGU23-12842
|
On-site presentation
Marc Ohmer, Fabienne Doll, and Tanja Liesch

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.

How to cite: Ohmer, M., Doll, F., and Liesch, T.: Incorporating spatial information for regionalization of hydrogeological parameters in machine learning models, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12842, https://doi.org/10.5194/egusphere-egu23-12842, 2023.

14:55–15:05
|
EGU23-16375
|
ECS
|
On-site presentation
Mourad Jadoud, Abderrahim El Achheb, Noureddine Laftouhi, Mustapha Namous, Abdellah Khouz, Jorge Trindade, Fatima El Bchari, Blaid Bougadir, Hasna Eloudi, and Said Rachidi

Estimating the potential of underground water sources has become a top priority for authorities in semi-arid mountain regions, particularly in the context of recent climate change that has made surface water less accessible and available. However, exploratory methods are typically quite expensive, such as geophysical and topographic methods, which take into account the vastness of the terrain and the extremely limited accessibility in the mountains such as the High Atlas. To achieve this, it has become necessary to use indirect methods first to define potential groundwater zone boundaries before turning to direct measures. This study proposes an indirect, straightforward, quick, and minimally costly method for identifying potential areas for groundwater in the Rherhaya watershed, High Atlas-Morocco, using machine learning, geographical information systems, and remote sensing.

Machine learning algorithms have been increasingly applied to define potential groundwater areas mapping. It’s performance to predict the spatial distribution of potential groundwater zones was tested with an inventory-based spring’s geodatabase. 254 spring’s obtained inventory was split into two independent datasets, including 70% of the springs for the training set and the remaining 30% of springs for validation purposes in the test set. 19 layers of landslide-conditioning factors were prepared and checked for collinearity issues, to produce the potential groundwater zones map. The conditioning factors were selected, prepared and classified, in order to determine the contribution of each class of factors to potential groundwater areas, including: Elevation, Aspect, Slope angle, Curvature plan, Curvature profile, Stream Power Index (SPI), Topographic Wetness Index (TWI), Normalized Difference Vegetation Index (NDVI), Distance to rivers, Lithology, Rainfall, Land Use and Land Cover (LULC), Drainage density, Valley Depth, Topographic Position Index (TPI), Terrain Ruggedness Index (TRI), Slope Length (LS), Geomorphons, Distance to faults, Distance to spring and Distance to roads. Both the inventory and the conditioning factors are obtained from the observation and interpretation of different data sources, namely high-resolution satellite images, aerial photographs, topographic maps, and extensive field surveys.

Using RStudio software, three machine-learning algorithms were used in this study: Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM). The model’s evaluation and validation was assessed using a hybrid approach based on k-cross validation, ROC Curve - AUC, and confusion matrix for the estimation of the predictive performance. The three models provided very encouraging results in terms of identifying the areas that are very likely to produce subsurface water in the Rherhaya basin.  The main contributing factors to the potentiality of underground waters, according to the three methods, are the valley depth, TPI, distance to rivers, and curvature plane.  With an AUC of 84.4% in the test data, it was clear that the SVM model outperforms the other models for the 70/30 percent subdivision. These findings may contribute to the development of a significant database that will help in the management of water resources in vulnerable areas by decision-makers.

How to cite: Jadoud, M., El Achheb, A., Laftouhi, N., Namous, M., Khouz, A., Trindade, J., El Bchari, F., Bougadir, B., Eloudi, H., and Rachidi, S.: Machine Learning Models for Spatial Prediction of Groundwater Potentiality in a Large Semi-Arid Mountainous Region: Application to the Rherhaya Watershed, High Atlas, Morocco, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16375, https://doi.org/10.5194/egusphere-egu23-16375, 2023.

15:05–15:15
|
EGU23-2998
|
On-site presentation
Jen-Tsung Kuo, Jet-Chau Wen, and Hong-Ru Lin

Since there are many uncertainties in the state of groundwater, after determining the research direction and objectives, a numerical model should be established by means of simulation to analyze the hydraulic characteristics such as transmissivity and storativity in the research area. Sensitivity analysis is carried out effectively, and the optimal pumping test strategy is designed based on it. In many previous studies, the sensitivity map made by the two-dimensional model was used to discuss the sensitivity. However, after a long period of data accumulation, it was found that although there would be a continuous change in the discharge, the pumping well and the vicinity of the observation well's hydraulic gradient did not change significantly. From this we get an inference, that is, we cannot obtain the exact location of the sensitivity source in space by simply using a two-dimensional sensitivity map, at most, only the average parameter value of the active area can be obtained. In addition, most of the current research is based on the assumption of homogeneous steady-state conditions, but groundwater mostly exists in the form of heterogeneous transient state in nature. Therefore, this research hopes to use the method of inverse calculating the heterogeneous field to plan and use the developed software (VSAFT2 and VSAFT3) for effective 2D and 3D space simulation, so as to use the characteristics of the three-dimensional space concept to confirm various information and extend the discussion of the source. This will have a great contribution to the changes of water bodies in the future analysis area, groundwater recharge patterns and diffusion control of underground pollution.

How to cite: Kuo, J.-T., Wen, J.-C., and Lin, H.-R.: Sensitivity analysis and research of hydraulic characteristic parameter and observation wells during the pumping test, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2998, https://doi.org/10.5194/egusphere-egu23-2998, 2023.

15:15–15:25
|
EGU23-10596
|
ECS
|
Virtual presentation
Yueling Ma, Elena Leonarduzzi, Amy Defnet, Peter Melchior, Laura Condon, and Reed Maxwell

Groundwater is one of the most valuable resources in the US. The United States Geological Survey (USGS) reported that about 26% of the water used in 2015 came from groundwater. Due to the scarcity of groundwater observations, it is still challenging to monitor groundwater resources at the watershed scale (where local decision making occurs). In addition, for long-term high-resolution simulations, physically-based models become very computational demanding and data hungry. With much less computational time and physical knowledge, machine learning (ML) techniques are able to capture complex nonlinear connections between groundwater dynamics and atmospheric and land surface processes from historical data. Recent studies have shown their success in groundwater modeling.

In this study, we develop a ML-based estimator to produce water table depth (WTD) estimates over the Contiguous US (CONUS) at a spatial resolution of 1 km using USGS WTD observations and other hydrometeorological data. The WTD estimator consists of two components. One component captures spatial variations in WTD using random forests and the other component learns temporal variations in WTD by Long Short-Term Memory networks. We combine the results from the two components to obtain WTD estimates. The estimated WTD are compared to USGS WTD observations to show their reliability. Based on the WTD estimates, we study groundwater changes in the Upper Colorado River Basin (UCRB) in recent drought years, which is one of the principal headwater basins in the US. To better interpret the performance of the WTD estimator, we conduct sensitivity analyses on input variables for both components. Moreover, we assess the uncertainty of the WTD estimator in estimating WTD over the CONUS. Our study demonstrates that the WTD estimator can generate reasonable WTD estimates over CONUS, thereby facilitating understanding of groundwater systems in the US. The WTD estimator can also be transferred to other regions in the world that have a similar hydrologic regime to a region in the US.

How to cite: Ma, Y., Leonarduzzi, E., Defnet, A., Melchior, P., Condon, L., and Maxwell, R.: The Development of a ML-based Estimator to Reconstruct Water Table Depth over the Contiguous US, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-10596, https://doi.org/10.5194/egusphere-egu23-10596, 2023.

15:25–15:35
|
EGU23-12629
|
ECS
|
Virtual presentation
Vipul Bhadani, Abhilash Singh, Vaibhav Kumar, and Kumar Gaurav

We applied different machine learning algorithms to predict the groundwater water fluctuation in a semi-arid river basin. Precipitation, temperature, evaporation, relative humidity, soil type, and groundwater lag were modeled as input features. Feature importance analysis of the input features indicate that the groundwater lag is the most relevant whereas the soil type is the least relevant input features. We applied the backward elimination approach to eliminate the less relevant features in mapping groundwater. Using the relevant input features we trained different machine-learning models (random forest, decision tree, neural network, linear regression, ridge regression, support vector regression, k-nearest neighbors, recurrent neural network). All these algorithms predicted the groundwater level with a high correlation of coefficient (R) ranging from 0.83 to 0.91 and a Root Mean Square Error (RMSE) ranging from 1.61 to 2.17 m.  We found that the random forest algorithm outperforms the other algorithms with a RMSE of 1.61 and R = 0.91. 

How to cite: Bhadani, V., Singh, A., Kumar, V., and Gaurav, K.: Machine learning models to predict groundwater level in a Semi-arid river catchment, Central India, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12629, https://doi.org/10.5194/egusphere-egu23-12629, 2023.

15:35–15:45
|
EGU23-6165
|
ECS
|
Virtual presentation
Ying Hu, Nengfang Chao, Jiangyuan Wang, Zheng Liu, and Kaihui Zou

Abstract: Groundwater depletion is a serious threat to agriculture and economic development, with an adverse impact on the ecological environment in Beijing. With the successful implementation of the South-to-North Water Diversion Project (SNWDP), groundwater (GW) depletion is expected to be alleviated. Here, we bridged the gap of the monthly terrestrial storage anomaly (TWSA) observations of the Gravity Recovery and Climate Experiment (GRACE) and its successor GRACE Follow-On (GRACE-FO) missions, then downscaled the spatial resolution of GW storage anomaly (GWSA) from 0.5°×0.5° to 0.25°×0.25° in Beijing using deep learning (DL) method, and precisely quantified the characteristics and causes of GWSA before and after the SNWDP, including that 1)  reconstructed 0.5°×0.5° TWSA in Beijing during 2004 to 2021 using three DL architectures (Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), Multilayer Perceptron (MLP)); 2) selected the optimal performance of DL for downscaling GWSA from 0.5°×0.5° to 0.25°×0.25°; 3) quantitatively analyzed of GWSA characteristics before and after the SNWDP basing on Random Forest (RF). The results indicated that the trend of downscaled GWSA was consistent with in-situ grounder water level measurements, while the seasonal amplitude differed. Before the SNWDP (2004-2014), the GW in Beijing showed a declining trend with a rate of -20.8 mm/yr, of which human factors contributed 75.7% (21.9% and 21.1% for domestic water and agricultural water, respectively), and climatic factors accounted for 24.1%. After the SNWDP (2015-2021), GW gradually increased by 39.3 mm/yr, and GW was significantly restored, of which the human factors accounted for 59.2% (30.7% and 14.2% for domestic water and water diverted to Beijing by the SNWDP, respectively). Our research could provide a new reliable theoretical support and technical reference for GRACE/GRACE-FO dynamic monitoring of groundwater.

Keywords: Beijing GW storage; spatial resolution downscaling; DL; GRACE/GRACE-FO; SNWDP

How to cite: Hu, Y., Chao, N., Wang, J., Liu, Z., and Zou, K.: South-to-North Water Diversion weaken groundwater depletion in Beijing: insight from downscaling GRACE/GRACE Follow-on data, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6165, https://doi.org/10.5194/egusphere-egu23-6165, 2023.

Groundwater Quality
Coffee break
Chairpersons: Tanja Liesch, Inga Retike, Ezra Haaf
16:15–16:25
|
EGU23-12956
|
On-site presentation
Joel Podgorski and Michael Berg

Chronic consumption of elevated concentrations of fluoride in groundwater can cause detrimental health effects including dental mottling and skeletal fluorosis. However, the concentration of fluoride is not known in many aquifers. To help address this, we have used machine learning to create a global fluoride prediction map based on the WHO drinking-water guideline of 1.5 mg/L. Over 400,000 data points of fluoride in groundwater (10% greater than1.5 mg/L) from 77 countries were used along with 12 predictor variables out of an initial set of 62 spatially continuous variables relating to geology, soil, climate and topography. The model performs very well, (e.g. AUC of 0.90) and was used to produce a global prediction map. This helps gauge the scope of the problem and identify potential hotspots that should receive the focus of more groundwater testing, including parts of central Australia, western North America, eastern Brazil and many areas of Africa and Asia. This fluoride hazard model was also used to estimate the global at-risk human population at about 180 million people, most of whom live in Asia and Africa. Another model was created using additional physicochemical parameters measured in situ. Although this model (AUC of 0.95) could not be used to create a map, it helps to better understand the processes related to the dissolution and accumulation of fluoride. For example, both the spatially continuous and in-situ predictor variables confirm that arid conditions promote the dissolution of fluoride in groundwater.

How to cite: Podgorski, J. and Berg, M.: Global machine-learning model of naturally occurring fluoride in groundwater, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-12956, https://doi.org/10.5194/egusphere-egu23-12956, 2023.

16:25–16:35
|
EGU23-4745
|
ECS
|
On-site presentation
Mohamed Yassin, Sani Abba, Abdullahi Usman, and Isam Aljundi

As one of the vulnerable arid regions, Saudi Arabia suffered from an optimal water crisis. Rapid population, industrialization, and urbanization mandated the water stress and agricultural water balance in the region; as such, there is a need for integrated water resources planning and management. To meet sustainable development goal six (SDGs-6), the ground and surface water need to be within the required quality and quantity. Hence there is a need for both the physical, chemical and hydrogeological assessment of groundwater in the Eastern part of Saudi Arabia. This study proposed three scenarios to assess the hydrogeological groundwater quality, namely, experimental laboratory based on fieldwork, geospatial mapping of the parameters (EC, pH, CaCO3, Turbidity, NA, k, Mg, Ca, F, Cl, Br, Li, B, Al, V, Cr, Fe, Mn, Ni, Co, Cu, Zn, As, Se, Sr, Mo, and Ba), and intelligent computational analysis using machine learning (ML). This research is motivated toward intelligent prediction of some heavy metals using neural network (NN), and adaptive neuro-fuzzy inference system (ANFIS). The evaluation criteria of the predictive results were analyzed based on mean absolute percentage error (MAPE), correlation coefficient (CC), and determination coefficient (DC). The outcomes proved the satisfactory ability of the NFIS model over the NN approach, despite its predictive credit. 

How to cite: Yassin, M., Abba, S., Usman, A., and Aljundi, I.: Spatiotemporal and hydrogeological assessment of groundwater supported by soft computing modeling of heavy metal in Al-Hassa, Eastern Province, Saudi Arabia, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-4745, https://doi.org/10.5194/egusphere-egu23-4745, 2023.

Explaining and modelling stresses in groundwater systems
16:35–16:45
|
EGU23-619
|
ECS
|
On-site presentation
Shubham Goswami and Muddu Sekhar

Processes controlling surface-groundwater interaction in terms of recharge and baseflow have been a topic of pursuit in the hydrological research community. The groundwater recharge and baseflow in hard-rock aquifers is significantly impacted by rainfall pattern, aquifer characteristics, weathering/soil condition, topography, land use and land cover. Analysis of the recharge and baseflow process in tropical semi-arid hard-rock aquifer regions of southern India is crucial due to heavily tailed monsoon system prevailing in the region, heterogeneity of aquifers in terms of fractures and lineaments and presence of several man-made irrigation tanks along with the drainage network. Process uncertainties in groundwater recharge simulation due to impact of climate change on vegetation and resulting changes in evapotranspiration can significantly impact groundwater projections. Poor representation of diffused and focused recharge pathways and inadequacy in capturing the feedback among climate, land use and groundwater systems are other factors introducing uncertainties. Finally, the gaps in long-term observational data make it challenging to assess the impact of climate change. A taluk scale study conducted in Karnataka State of India indicated a significant correlation between the rainfall intensity distribution and climatology on groundwater recharge in Hard-rock aquifers. The current study targets to add a baseflow component represented by a 2D groundwater model and extend the previous study to a larger scale. The primary objectives of the study are to estimate the long-term groundwater recharge and baseflow trends and evaluate their association with rainfall and climatological variability. As the projected climate scenarios reflect higher frequency of high-intensity rainfall, it becomes essential to evaluate the impacts of varying rainfall patterns on the surface-subsurface processes.

How to cite: Goswami, S. and Sekhar, M.: Understanding impact of Rainfall Intensity Distribution and Climatology on Recharge and Baseflow in tropical Hard-rock Aquifers of South India, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-619, https://doi.org/10.5194/egusphere-egu23-619, 2023.

16:45–16:55
|
EGU23-11745
|
On-site presentation
Liang Cheng Chang, You Cheng Chen, and Tsai Chen Lin

The groundwater level is a comprehensive response of the external stimuli of the aquifer, such as rainfall and river recharge, and is one of the main factors affecting the formation deformation. The spatiotemporal analysis of stratum elastic deformation identifies different stimulus sources and their influence on groundwater level changes, as well as the influence of groundwater level changes on stratum elastic deformation. In addition to the numerical test to verify the correctness of the analysis process, this study uses the daily water level observation data of each aquifer in Yunlin area of Taiwan to conduct principal component analysis to explore the influence of rainfall and river water level on the variation of groundwater level in each aquifer. Furthermore, based on the magnetic ring layered subsidence observation well data, the iterative principal component analysis was used to explore the influence of the variation of the groundwater level in each layer on the elastic deformation of each layer.

The research results show that the first principal component of the groundwater level in each aquifer in the Yunlin area can fully reflect the impact of rainfall on the groundwater level, and the residual water level affected by rainfall can be deducted to analyze the impact of the river water level. The results show that the shallower water level Compared with the deeper aquifers 2-2 and 3, layers 1 and 2-1 have a more significant effect on the river water level. In the analysis of formation compression deformation, the correlation analysis between the principal component of groundwater level and the principal component of formation deformation shows that there is a high negative correlation between the water level change of each aquifer and the formation deformation, which is in line with the physical mechanism of groundwater level and formation deformation. The above analysis results show that principal component analysis, combined with the analysis process of this study, can indeed identify and isolate the influence of various external stimuli to the groundwater level, and can also obtain the main changes in the elastic deformation of the formation, which can be used as the basis for further water resources planning and analysis. Important reference.

How to cite: Chang, L. C., Chen, Y. C., and Lin, T. C.: An application of iterative principal component analysis on analyzing the regional groundwater level and stratum elastic-deformation, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11745, https://doi.org/10.5194/egusphere-egu23-11745, 2023.

16:55–17:05
|
EGU23-993
|
Virtual presentation
|
Tengfei Feng, Yunzhong Shen, Qiujie Chen, and Fengwei Wang

Groundwater overdraft in North China (NC) has posed adverse threats to sustainable development due to the reduction of freshwater availability. To comprehensively clarify the groundwater change and formulate reasonable control strategies, groundwater storage anomaly (GWSA) is investigated using the high-resolution time-variable gravity field model Tongji-RegGrace2019 together with the hydrological model. The results show that GWS presents a downward trend of -0.87±0.04 cm/yr from January 2004 to December 2015 and the trend aggravates to -3.71±0.49 cm/yr from January 2014 to December 2015, which is basically consistent with those detected by monitoring well. Moreover, by analyzing the spatiotemporal characteristics of GWSA with the independent component analysis (ICA) approach, the driving factors and corresponding mechanisms of groundwater changes are determined. Among the four independent components (ICs) of GWSA, the first two ICs (IC1 and IC2) cooperatively reflect the long-term and intra-annual groundwater changes caused by water consumption of coal mining and agricultural irrigation in Shanxi province, with the correlation coefficients of -0.91 and -0.85, respectively. IC3 indicates a semi-annual groundwater signal related to agricultural irrigation water consumption in southern Hebei province, with a correlation coefficient of -0.85. Besides, IC4 suggests the effect of monsoon precipitation and evaporation in front of Taihang Mountain. Hence, multiple driving sources, including unevenly distributed precipitation, intense seasonal evaporation, and devastating coal mining, coupled with extensive agricultural irrigation, jointly restrict the GWSA rise and fall at different time nodes.

How to cite: Feng, T., Shen, Y., Chen, Q., and Wang, F.: Spatiotemporal Characteristics and Drivers of Groundwater Change in North China Revealed by GRACE Time-Variable Gravity, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-993, https://doi.org/10.5194/egusphere-egu23-993, 2023.

17:05–17:15
|
EGU23-8098
|
ECS
|
On-site presentation
Augustin Thomas, Jérôme Fortin, Benoît Vittecoq, and Sophie Violette

Following the evolution of the hydrodynamic parameters (hydraulic conductivity and storativity) of an aquifer over time is difficult and does not exist for an aquitard. The primary reason is the small number of observations, which are mainly pumping or slug tests. However, solutions exist to recover these parameters using only groundwater level monitoring data at a sampling rate of 1 hour, data which is extensively available. These solutions take advantage of the boreholes’ response to different tidal phenomena, including oceanic, earth and atmospheric tides.

Martinique Island, in the Lesser Antilles, is a very interesting field to study these techniques, since 16 years of piezometric level data have been recorded on this volcanic island in a monitoring network of 29 boreholes. The variety of aquifer geometry and geology enables to study different types of responses, in addition the island is submitted to seismic activity that may impact aquifers and particularly their hydraulic conductivity.

The method consists in computing amplitude and phase response of aquifers to both atmospheric, earth and ocean tides. Then, the response of semi-confined aquifers to different loading sources at the tidal frequencies (between 1 and 2 cycles per day) is modelled. The model is a general and adaptable one, in order to consider the various geometries (including the presence or not of an aquitard) and aquifer boundary conditions (confined, semi-confined, connected to the atmosphere or not). Finally, a careful inversion is done to obtain the characteristics of the aquifer.

Our results show that the Martinique aquifers present responses to different sources of loading. We show that using the dominant source response alone, we are able to recover aquifer or aquitard hydraulic conductivity and its evolution over the last 16 years, which we compare with pumping tests results. Using a combination of sources, we are also able to recover elastic properties like the evolution of the loading efficiency and shear modulus. Finally, comparing different sites responses, we are able to assess which aquifer is more vulnerable to pollution through vertical leakage within their aquitard.

How to cite: Thomas, A., Fortin, J., Vittecoq, B., and Violette, S.: Passive characterization of aquifers hydro-mechanical properties using tidal signals, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8098, https://doi.org/10.5194/egusphere-egu23-8098, 2023.

17:15–17:25
|
EGU23-8608
|
ECS
|
Virtual presentation
|
Hafsa Mahmood, Ty P. A. Ferre, Raphael J. M. Schneider, Simon Stisen, Rasmus R. Frederiksen, and Anders V. Christiansen

Temporal drain flow dynamics and their underlying controlling factors are important for understanding the needs for water resource management in tile drained agricultural areas. The use of physics-based water flow models to understand tile drained systems is quite common. These physics-based models are complex and have high computational demand due to the high spatial and temporal dimensionality of the problem. We examine whether machine learning (ML) models can offer a simpler tool for water management.

 

The main aim of our study is to assess the potential of ML tools for predicting drain flow with varying climate parameters and hydrogeological properties in different catchments in Denmark. We rely on unique data containing time series of daily drain flow in 26 field-scale tile drained catchments in Denmark: climate data (precipitation, potential evapotranspiration, temperature); geological properties (clay fraction, first sand layer thickness, first clay layer thickness); and topographical indexes (curvature, topographical wetness indexes, topographical position index, elevation etc.). The ML algorithm XGBoost is used to predict drain flow in the 26 drain catchments based on both static and dynamic variables. This algorithm also provides an independent measure of the value of information contained in variables related to climate, geology and topography for the prediction of tile drain flows.

 

The ML approach examined could provide a more transferable, faster, and less computationally expensive tool to predict drain flow dynamics. Simultaneously, the results of the study offer insight into the underlying factors that control drain flow, allowing for improved data collection and physics-based model development.

How to cite: Mahmood, H., P. A. Ferre, T., J. M. Schneider, R., Stisen, S., R. Frederiksen, R., and V. Christiansen, A.: Use of machine learning for drain flows predictions in tile-drained agricultural areas, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8608, https://doi.org/10.5194/egusphere-egu23-8608, 2023.

17:25–17:35
|
EGU23-16384
|
ECS
|
Virtual presentation
Aline Moreau, Atefe Choopani, Pierre-Yves Declercq, Philippe Orban, Xavier Devleeschouwer, and Alain Dassargues

Interferometric Synthetic Aperture Radar (InSAR) technology has been used to detect the location and magnitude of ground deformation for the past 30 years, providing cost-effective measurements with a fine resolution and precision within centimeters under ideal conditions1. Persistent Scatterer InSAR Interferometry (PS-InSAR) is an InSAR algorithm that has been developed to overcome decorrelation due to changes in the physical characteristics of the surface over time that limit the InSAR applications 2,3.

PS-InSAR processing has been used to identify multiple localized land subsidence in the Antwerp and Leuven areas in Belgium. In Antwerp, the harbour was gradually developed, leading to dock excavations in a compressible estuary polders environment and PS-InSAR was used to detect, map and study the ground displacements4. In Leuven, significant subsidence was observed through the city and the suburbs, potentially due to delayed consolidation in compressible, low permeability aquitards. One of the possible cause of these subsidence phenomena is related to variation in groundwater levels resulting in consolidation processes. To test this hypothesis, geomechanical calculations coupled to groundwater flow models are carried out to simulate the vertical displacements. The results are compared to PS-InSAR-derived subsidence observations for a better understanding of subsurface consolidation mechanisms.

However, there are several practical and conceptual challenges that must be considered when comparing InSAR measurements to results from hydrogeological and geomechanical models. One issue is the choice of the appropriate modeling scale, as subsidence may occur locally but also regionally as influenced by groundwater pore pressure variations occurring at different scales. Another challenge lies in the selection of the appropriate conceptual assumptions linked to the groundwater flow and geomechanical models. Indeed, in addition, uncertainty in the model parameter values is a typical source of uncertainty in the model results. There may also be errors in the InSAR measurements due to various factors such as atmospheric effects and changes in the surface roughness. All these challenges must be taken into account when comparing InSAR measurements to model results.

In conclusion, comparing InSAR measurements with hydrogeological and geomechanical modeling results can provide valuable insights into the actual mechanisms of subsidence. However, it is important to carefully consider the practical and conceptual challenges and limitations linked to this interesting comparison.

 

References:

1 Peng, M., Lu, Z., Zhao, C., Motagh, M., Bai, L., Conway, B. D., & Chen, H. (2022). Mapping land subsidence and aquifer system properties of the Willcox Basin, Arizona, from InSAR observations and independent component analysis. Remote Sensing of Environment271, 112894.

2 Ferretti, A., Prati, C., & Rocca, F. (2000). Nonlinear subsidence rate estimation using permanent scatterers in differential SAR interferometry. IEEE Transactions on geoscience and remote sensing38(5), 2202-2212.

3 Ferretti, A., Prati, C., & Rocca, F. (2001). Permanent scatterers in SAR interferometry. IEEE Transactions on geoscience and remote sensing39(1), 8-20.

4 Declercq, P. Y., Gérard, P., Pirard, E., Walstra, J., & Devleeschouwer, X. (2021). Long-term subsidence monitoring of the Alluvial plain of the Scheldt river in Antwerp (Belgium) using radar interferometry. Remote Sensing13(6), 1160.

How to cite: Moreau, A., Choopani, A., Declercq, P.-Y., Orban, P., Devleeschouwer, X., and Dassargues, A.: Difficulties arising when PS-InSAR displacement measurements are compared to results from geomechanical and groundwater flow computations., EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16384, https://doi.org/10.5194/egusphere-egu23-16384, 2023.

Posters on site: Wed, 26 Apr, 10:45–12:30 | Hall A

Chairpersons: Mark Bakker, Raoul Collenteur, Tanja Liesch
A.173
|
EGU23-764
Indra Mani Tripathi and Pranab Kumar Mohapatra

Due to rapid industrial and population growth, groundwater levels frequently shift, with depletion being the most prominent effect. Numerical models have the potential to save time and money by supplying pertinent information in areas where data is lacking. We investigate the fluctuations in groundwater levels in the secondary cities of India using the well-known MODFLOW-2005 model. The study covers all the wards of three Indian secondary cities (Bhopal, Bhuj and Kozhikode), and we run a simulation for the year 2012–2020. We use groundwater table data from 2012 to 2020, topographic maps and geological maps of the selected cities. Along with these, we use hydraulic conductivity and specific yield values of the aquifer of the study area. We calibrate the model for both steady-state and transient scenarios to match the field conditions up to acceptable standards. Moreover, we validate the model for any one year to justify the acceptability of all the calibrated values. In addition, we perform sensitivity analysis to illustrate which model parameter has the greatest influence on the model output. Finally, we run the model for 2025 to predict the groundwater level changes in the immediate future. From this, we can identify the regions facing serious groundwater-lowering problems. Therefore, more comprehensive policies, as well as laws on groundwater extraction, should be created in order to safeguard this natural resource. In this context, the present study will provide an overview of this urgent issue in Indian cities.

How to cite: Tripathi, I. M. and Mohapatra, P. K.: Modeling of groundwater level changes in the fast-growing Indian Secondary Cities, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-764, https://doi.org/10.5194/egusphere-egu23-764, 2023.

A.174
|
EGU23-15164
Yun-Chi Chung and Li-Chiu Chang

Surface water and groundwater are both important water sources for people’s livelihood, irrigation and industry. Especially, groundwater can supply water stability in times of drought to mitigate drought disaster. Due to the uneven temporal and spatial distribution of rainfall in Taiwan, there is a large difference in rainfall during the wet and dry seasons, and the rapid changes in the terrain slope cause the river flow rapid, which cannot store and utilize water resources. The rapid economic development makes the water demand increase year by year. Therefore, the surface water cannot meet the demand. Groundwater flows slow and replenishment is difficult. Long-term overuse will cause the gradual depletion of underground water sources, resulting in severe damages such as stratum subsidence and seawater intrusion. Therefore, if we can master the situation of groundwater changes and it is helpful to the management and deployment of surface water and groundwater resources.

The purpose of this study is to explore the interaction between surface water and groundwater during typhoon periods using machine learning methods. The research area is the Choshui river basin. According to the long-term monitoring of hydrological and groundwater data, different temporal and spatial distributions of rainfall events are selected to conduct principal component analysis (PCA) for each aquifer. Through the principal component weights, tempo-spatial distribution of rainfall and streamflow to analyze the trend of groundwater observation wells, and then use the recurrent configuration of nonlinear autoregressive with exogenous inputs (R-NARX) to predict the hourly groundwater level. Principal component analysis was used to understand the impact of storm water on the groundwater recharge, and the relationship between principal component scores and flow and rainfall was used to find out the key factors affecting the groundwater level. The results can provide the sensitive areas of groundwater recharge in Choshui river basin for adopting the optimal water resource allocation strategy.

Keywords:Groundwater; Principal component analysis (PCA); Machine learning

How to cite: Chung, Y.-C. and Chang, L.-C.: Impact of storm water on groundwater recharge and discharge using machine learning, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15164, https://doi.org/10.5194/egusphere-egu23-15164, 2023.

A.175
|
EGU23-15725
|
ECS
Yixuan Xing, Quan Liu, Rui Hu, and Thomas Ptak

Abstract: Oscillatory water level is often observed in groundwater monitoring wells when the aquifer is disturbed by some periodic pressure sources, such as oscillatory hydraulic test, ocean tide, and far-field seismic wave. The amplitude attenuation (A) and phase shift (P) between the source and water level response are utilized to estimate aquifer properties in many related studies. Water level response, in essence, is not only affected by aquifer parameters but also the characteristics of disturbance source and wellbore effects, which are determined by multiple parameters and the highly nonlinear well-aquifer system. To clarify the impacts of A and P on all relevant parameters, a global sensitivity analysis of water level response to harmonic aquifer disturbance is conducted in this study. A general numerical model of water level response that integrates different types of sources and considers wellbore effects is first introduced. Based on the quasi-Monte Carlo method, nine relevant parameters regarding wellbore geometry, aquifer property, and characteristics of the source, are sampled within their normal ranges. A data-driven model of the water level response to the selected parameters is trained and cross-validated using the random forest regression method. The global sensitivity analysis of the A and P is then implemented by using the variance-based method.  The first-order Sobol index shows that for both A and P, the oscillatory period of the disturbance source is the most sensitive factor among others. The second-order Sobol index indicates that the interaction between the period of disturbance source and water column height in the wellbore is the most important to A and P. preliminary results show that wellbore effects have significant impacts on water level responses, especially under a high-frequency disturbance.

How to cite: Xing, Y., Liu, Q., Hu, R., and Ptak, T.: Global sensitivity analysis of water level response to harmonic aquifer disturbances considering wellbore effects, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15725, https://doi.org/10.5194/egusphere-egu23-15725, 2023.

A.176
|
EGU23-1840
|
Patrick Haehnel, Gabriel C. Rau, and Todd C. Rasmussen

Freshwater lenses are an important water resource in coastal areas as well as on oceanic islands, and understanding the dynamic forces acting upon this resource is vital for their sustainable management. A key water-management objective is to understand and manage these freshwater lenses, which requires accurate estimates of drawdown and groundwater recharge. Groundwater levels in such systems, however, are dominated by multiple dynamic factors, such as tidally and meteorologically forced ocean level fluctuations, coastal morphology, aquifer properties, recharge, and groundwater extraction. Unfortunately, tidal influences often dominate groundwater levels in these systems, which confounds the quantification of aquifer recharge and extraction.

This work uses regression deconvolution to quantify oceanic influences on groundwater levels by generating an “Ocean Response Function” (ORF) that is used to reveal groundwater recharge and extraction, once influences have been removed. We use groundwater levels from an unconfined and unconsolidated (mostly fine sand) aquifer on the island of Norderney located in the North Sea in Northwest Germany. Confounding tidal influences are removed from observed groundwater levels to reveal underlying processes. Most prominently, seasonal recharge patterns are now clearly visible, along with responses to daily groundwater extraction from a nearby water-supply well. The obtained ORF also constrains the aquifer hydraulic diffusivity, in that higher diffusivities induce faster responses. Overall, this work demonstrates how regression deconvolution leads to improved insight into groundwater processes and properties when applied to coastal and island groundwater observations.

How to cite: Haehnel, P., Rau, G. C., and Rasmussen, T. C.: Disentangling aquifer dynamics in coastal groundwater systems using high-resolution time series, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1840, https://doi.org/10.5194/egusphere-egu23-1840, 2023.

A.177
|
EGU23-3056
|
ECS
Julian Koch, Jacob Kidmose, Jun Liu, Raphael Schneider, Simon Stisen, and Lars Troldborg

Operational forecasts of groundwater levels provide critical real-time knowledge during extreme events, such as floods and droughts. This study proposes a Long Short-Term Memory based Encoder-Decoder (LSTM-ED) model for multi-step-ahead groundwater level forecasting. The LSTM-ED is a well-suited architecture for sequence-to-sequence modelling tasks but has not yet been applied to forecast groundwater levels. The proposed LSTM-ED model is designed in the context of the Danish online monitoring system grundvandsstanden.dk to serve as operational groundwater level forecasting system. In the encoder LSTM model, sequences of past precipitation, temperature and groundwater levels are processed to initialize the decoder LSTM model which, in addition takes in forecast sequences of precipitation and temperature to output a sequence of groundwater levels. We train LSTM-ED models individually for each well, with all data aggregated to daily timescale. We demonstrate the performance of the LSTM-ED architecture for numerous wells from grundvandsstanden.dk and test varying lead times of up to 30 days. The LSTM-ED model forecasts are contrasted with simple benchmark models as well as with a sequence-to-sequence LSTM model that does not incorporate forecasts of precipitation and temperature for outputting the groundwater sequence. Initial results underpin that integrating forecasts of precipitation and temperature is a crucial component, especially for wells with shallow intakes where surface and sub-surface processes are well connected. The sequence-to-sequence LSTM model yields similar accuracy as the simple benchmark models, whereas accuracy clearly improves for the LSTM-ED model. Overall, this study highlights the potential of LSTM-ED models as an operational tool for multi-step-ahead forecasting of groundwater levels.

How to cite: Koch, J., Kidmose, J., Liu, J., Schneider, R., Stisen, S., and Troldborg, L.: Multi-step-ahead forecasting of groundwater levels using a Long Short-Term Memory based Encoder-Decoder (LSTM-ED) model, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-3056, https://doi.org/10.5194/egusphere-egu23-3056, 2023.

A.178
|
EGU23-16594
Qidong Fang, Francesca Pianosi, and A S M Mostaquimur Rahman

Groundwater is the world's largest accessible source of fresh water and plays an indispensable role in the global water cycle. Groundwater supports irrigation, supplies drinking water, and sustains baseflows to the surface expression of groundwater (e.g. rivers, ponds, wetlands). Simulations using physical numerical models are computationally expensive due to the heterogeneity of the actual groundwater flow and the complex initial and boundary conditions. Several surrogate models for reducing the computational burden have been proposed, however, they usually do not follow physics law. In this study, we intend to combine machine learning methods to analyse the feasibility of physics-informed neural networks (PINN) in groundwater modelling and propose a PINN groundwater model for the simulation of groundwater flow to improve computational efficiency while restricted to the physics law.

How to cite: Fang, Q., Pianosi, F., and Rahman, A. S. M. M.: Groundwater modelling based on physics-informed neural networks, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-16594, https://doi.org/10.5194/egusphere-egu23-16594, 2023.

A.179
|
EGU23-14561
Jere Remes and Jarkko Okkonen

In sub-Arctic environment as in Finland, soil moisture and temperature dynamics affect the development of soil frost that controls snowmelt runoff, infiltration and recharge in winter periods. This study was initiated to investigate the performance of integrated hydrology model Amanzi-ATS, in simulating dynamics of soil moisture and temperature at different depths to assess recharge rate in unconfined aquifer in Central Finland. Our objective is to study intra- and inter-annual recharge rates, and the impacts of soil ice on groundwater recharge rate. Hourly soil water content and temperature was measured at ten different depths. The groundwater depth and temperature were measure daily from the borehole located 2 meters from the soil monitoring station and the climate data was obtained around 5 km from the soil station. 1D model with varying soil texture was developed to predict recharge rates. Measured soil water content, soil temperature and groundwater temperature were used to calibrate the 1D numerical model. The implications of this study will be understanding the freezing and thawing of soil on groundwater recharge rate and how recharge rate may change from one year to the next in the sub-Arctic environment.

How to cite: Remes, J. and Okkonen, J.: Numerical simulations of freeze/thaw cycle and its impact on groundwater recharge in sub-Arctic environment, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-14561, https://doi.org/10.5194/egusphere-egu23-14561, 2023.

A.180
|
EGU23-6907
|
ECS
|
Ronja Iffland and Uwe Haberlandt

Due to climate change, an assessment of future changes in hydrological systems is necessary for appropriate planning, particularly for water management. So far, especially flood and low flow events have been studied. But groundwater levels are also influenced by extended dry periods and seasonal shifts in precipitation, as groundwater recharge is directly related to precipitation and evaporation. In the study area of Lower Saxony (Germany), groundwater is an important resource for drinking water and supplies about 86 % of the demand (MU 2021). Therefore, knowledge about possible changes is of highest importance for a possible need for action.

In this study, climate characterising indices are used. Based on the assumed relationship between groundwater levels and meteorological indices, a simplified statistical approach should be used. Therefore, multiple linear regression models were set up for groundwater level estimation. Local models were set up for 734 groundwater monitoring wells in Lower Saxony, Germany. In order to take the persistence of the meteorological indices into account, moving averages and time lags were also included. Using a split validation procedure, which could be carried out for 114 stations with sufficient time series lengths, shows a good performance of the models. In order to make statements about future changes, the models were applied using climate model data based on the RCP8.5 scenario. Analyses for the reference period show that the groundwater levels can be sufficiently estimated. Slight changes were detected for the near future (2021-2050) and for the far future (2071-2100). For the majority of the measuring stations, decreases in mean annual low groundwater levels and slightly decreases in mean annual high groundwater levels are observed for both future periods. The number of low-level months does not change, while the number of high-level months increases slightly. In addition, a delayed timing of the annual extremes can be detected.

How to cite: Iffland, R. and Haberlandt, U.: Groundwater level prediction from meteorological indices, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-6907, https://doi.org/10.5194/egusphere-egu23-6907, 2023.

A.181
|
EGU23-1859
|
ECS
Fang Li, Wolfgang Kurtz, Ching Pui Hung, Harry Vereecken, and Harrie-Jan Hendricks Franssen

As a vital supply of water resources for human society, groundwater plays a significant part in the water cycle, and is closely linked to precipitation, surface water and soil moisture (SM). Groundwater modelling often suffers from a variety of uncertainties, including uncertain forcing data, parameters and initial conditions. To reduce the uncertainties of model predictions, data assimilation (DA) can be used to correct model predictions with observations to improve the estimation of unknown states and parameters. To investigate the effects of assimilation of groundwater data into the integrated model Terrestrial System Modelling Platform (TSMP) on groundwater table depth (WTD) simulations, groundwater assimilation experiments were conducted for the Rur catchment in Germany. 128 ensemble members were generated by perturbing atmospheric forcing variables and saturated hydraulic conductivity, and then the measured daily groundwater data from 2018 were assimilated into the model TSMP by the Localized Ensemble Kalman Filter (LEnKF). The measured data were screened rigorously before assimilation. The spatial autocorrelation analysis of the measured groundwater data and the open loop (OL) simulations showed consistency in the spatial variability of groundwater levels between measurements and simulations. Based on the results of a spatial autocorrelation analysis, three different local radii (10 km, 5 km and 2.5 km) were selected for the assimilation experiments. Comparing the results of the OL and DA experiments, the simulated WTD bias (simulated - measured) and root mean square error (RMSE) were reduced for all DA runs compared to OL. The 10km localization radius gives the smallest RMSE at assimilation locations, with 81% RMSE reduction compared to the OL. Validation with WTD data from independent verification sites shows that localized assimilation improves groundwater simulations only when the distance to assimilated sites is smaller than 2.5km. Independent WTD validation showed a reduction in RMSE of 30% and the best results were from the DA run with 10 km radius. Also soil moisture measurements from the Cosmic Ray Neutron Sensor (CRNS) were used for validation. The simulated SM reproduced the observed temporal fluctuations, with a high correlation between measured and simulated SM (from 0.70 to 0.89, except for the Wuestebach site). However, there was no RMSE reduction of SM for the DA runs compared to the OL.

How to cite: Li, F., Kurtz, W., Hung, C. P., Vereecken, H., and Hendricks Franssen, H.-J.: Water table depth assimilation in integrated terrestrial system models at the larger catchment scale, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-1859, https://doi.org/10.5194/egusphere-egu23-1859, 2023.

A.182
|
EGU23-15276
|
ECS
Doris Wendt, Gemma Coxon, and Francesca Pianosi

Decision-making in groundwater management could benefit from a robust modelling approach that considers the complexity and uncertainty in water availability, dynamic impact of management and modelling setups available. Modelling groundwater is a complex matter on its own given the heterogeneous aquifers, delayed climate signal in groundwater recharge and dynamic influence of water abstractions. Due to this complexity, decision-making models are often simplified to address only the main impacts on the hydrological cycle. Whilst this simplification is necessary, it is important to examine model process controls and uncertainty in parameters of a simplified model setup. 

In this study, we have converted a lumped conceptual socio-hydrological model to an operational tool for supporting decision-making by (1) evaluating and (2) calibrating the model. First, we applied global sensitivity to examine the “consistency” and “leverage” of the model. A model is considered consistent when modelled process controls match our system understanding. Leverage is observed when modelled strategies have adequate influence on modelling output regardless of parameter uncertainty. Results show that even with large uncertainty in parameter values, consistency is achieved for all hydrological variables. Input parameters defining management strategies are found to have significant leverage, as varying them induce noticeable changes in simulation outputs regardless of the physical conditions or uncertainty in parameters. When looking at hydrological extremes, this impact was amplified. 

Second, the model was calibrated for a range of catchments in the UK. The eleven model parameters were constrained using statistical criteria to identify an optimum parameter range. Model outputs were compared to observations (discharge and groundwater level time series using both similarity and signature-based evaluation criteria. Additionally, we sourced open access UK datasets to validate our data-based parameter ranges with local information. In general, calibrated model outputs represent surface water and groundwater features well and in particular, baseflow generation is well-represented. This encourages model applications for examining regional/national policies aiming to protect groundwater-fed streams. Exploratory model runs can also be used to facilitate discussions on new/altered management strategies and may spark further detailed modelling once a suitable strategy is identified. 

How to cite: Wendt, D., Coxon, G., and Pianosi, F.: Developing an operational model to support groundwater management decision-making in the UK, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-15276, https://doi.org/10.5194/egusphere-egu23-15276, 2023.

A.183
|
EGU23-9341
|
ECS
Raoul Collenteur, Ezra Haaf, Tanja Liesch, Andreas Wunsch, and Mark Bakker

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.

How to cite: Collenteur, R., Haaf, E., Liesch, T., Wunsch, A., and Bakker, M.: Results from the 2022 Groundwater Time Series Modeling Challenge, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-9341, https://doi.org/10.5194/egusphere-egu23-9341, 2023.

A.184
|
EGU23-11427
|
ECS
Ezra Haaf, Markus Giese, Thomas Reimann, and Roland Barthel

A new method is presented to efficiently estimate daily groundwater level time series at unmonitored sites by linking groundwater dynamics to local hydrogeological system controls. The presented approach is based on the concept of comparative regional analysis, an approach widely used in surface water hydrology, but uncommon in hydrogeology. The method uses regression analysis to estimate cumulative frequency distributions of groundwater levels (groundwater head duration curves (HDC)) at unmonitored locations using physiographic and climatic site descriptors. The HDC is then used to construct a groundwater hydrograph using time series from distance-weighted neighboring monitored (donor) locations. For estimating times series at unmonitored sites, in essence, spatio-temporal interpolation, extreme gradient boosting and nearest neighbors are compared. The methods were applied to ten-year daily groundwater level time series at 157 sites in alluvial unconfined aquifers in Southern Germany. The controlling site descriptors were analyzed using shapley values, revealing that models of HDCs were physically plausible. The analysis further shows that physiographic and climatic controls on groundwater level fluctuations are nonlinear and dynamic, varying in significance from “wet” to “dry” aquifer conditions. Extreme gradient boosting yielded a significantly higher predictive skill than nearest neighbor. However, donor site selection is of key importance. The study presents a novel approach for regionalization and infilling of groundwater level time series that also aids conceptual understanding of controls on groundwater dynamics, both central tasks for water resources managers.

How to cite: Haaf, E., Giese, M., Reimann, T., and Barthel, R.: Investigating system controls for prediction of groundwater hydrographs at unmonitored sites transferring head duration curves, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-11427, https://doi.org/10.5194/egusphere-egu23-11427, 2023.

A.185
|
EGU23-13918
|
ECS
Jānis Bikše, Inga Retike, Ezra Haaf, Konrāds Popovs, and Andis Kalvāns

Groundwater level data are often spatially and temporary unevenly distributed, and the heterogeneity of hydrogeological systems limit simple extrapolation to data-scarce locations. Here, similarity of hydrogeological systems can be used to transfer dynamics of groundwater levels from monitored to unmonitored locations. Therefore, the major controls of hydrogeological systems need to be investigated. This study aims to uncover dominant patterns of the groundwater levels and to connect them with possible controls or driving features (such as the catchment and topography characteristics, aquifer properties, climate and land use) of hydrogeological systems in the Baltic states. Part of the spatial features were calculated for three different buffer sizes (300m, 3km and 5km). Gapless daily groundwater hydrographs (prepared by Bikše et al., 2023) were used to create groundwater level patterns by clustering approach. Then, random forest classification was performed using a backward variable elimination approach to find the most significant site-descriptive features for each cluster. The results show that the most significant tend to be geological features in the well cross-section, followed by topographic features within the vicinity of the well. We also found that anthropogenic impacts can play a significant role and should be considered when analyzing groundwater level patterns.

References:

Bikše, J., Retike, I., Haaf, E., Kalvāns, A., 2023. Assessing automated gap imputation of regional scale groundwater level data sets with typical gap patterns. EarthArXiv. https://doi.org/10.31223/x5n94x

 

"The study has been funded by Iceland, Liechtenstein and Norway through the EEA and Norway Grants Fund for Regional Cooperation project No.2018-1-0137 “EU-WATERRES: EU-integrated management system of cross-border groundwater resources and anthropogenic hazards" and by the Latvian Council of Science, project “Spatial and temporal prediction of groundwater drought with mixed models for multilayer sedimentary basin under climate change”, project No. lzp-2019/1-0165.

How to cite: Bikše, J., Retike, I., Haaf, E., Popovs, K., and Kalvāns, A.: Revealing controls on groundwater level patterns by backward variable elimination in the Baltic states, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-13918, https://doi.org/10.5194/egusphere-egu23-13918, 2023.

A.186
|
EGU23-2605
|
Inga Retike, Jānis Bikše, Ezra Haaf, and Andis Kalvāns

Uneven measurement frequencies and continuous gaps in hydrographs are among the major challenges when dealing with regional-scale groundwater level data sets, especially if compiled from different countries. A variety of automated gap imputation methods can be applied to infill a large number of missing values, yet the assessment of modeling performance remains a difficult task often performed by randomly introduced missing values. However, large groundwater level data sets rarely have random gaps and more complex gap patterns can be observed. Here we present a new artificial gap introduction technique (TGP - typical gap patterns) mimicking realistic gap patterns characteristic to regional scale groundwater level data sets thus improving the assessment of gap imputation methods. Imputation performance of machine learning algorithm missForest and imputePCA were compared with routinely used linear interpolation to create gapless groundwater hydrographs for the Baltic states (Estonia, Latvia, Lithuania). Our results showed that infilling performance varies among different gap patterns (TGP). Overall, the missForest algorithm significantly outperformed imputePCA and linear interpolation even when infilling up to 2.5 years long gaps, while linear interpolation produced similarly good results to missForest when infilling relatively short (random-like) gaps. It was observed that imputation performance substantially decreased when infilling previously unseen extremes (such as severe drought episodes in 2018) or groundwater hydrographs likely affected by water abstraction (located near major agglomerations).

The study has been founded by Iceland, Liechtenstein and Norway through the EEA and Norway Grants Fund for Regional Cooperation project No.2018-1-0137 “EU-WATERRES: EU-integrated management system of cross-border groundwater resources and anthropogenic hazards”. The research further contributes to the grant TRV2019/45670 awarded by the Swedish Transport Administration (Trafikverket).

How to cite: Retike, I., Bikše, J., Haaf, E., and Kalvāns, A.: Improved assessment of automated gap imputation in large groundwater level data sets, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2605, https://doi.org/10.5194/egusphere-egu23-2605, 2023.

Posters virtual: Wed, 26 Apr, 10:45–12:30 | vHall HS

Chairperson: Inga Retike
vHS.26
|
EGU23-2072
|
|
Cheng-Shin Jang

Groundwater nitrate-N contamination typically involves several natural and anthropogenic factors, such as hydrology, hydrogeology, topography, and land uses. DRASTIC-LU-based aquifer vulnerability can be adopted to explore pollution potentials of groundwater nitrate-N. Furthermore, groundwater nitrate-N pollution frequently has high levels of spatial variability because of various pollution sources and hydrogeological and hydrochemical conditions. Estimates are typically uncertain owing to limited in-situ data. Geostatistics is a commonly used technique of spatial estimates with limited data. To reduce the underestimation and overestimation of ordinary kriging (OK), this study employed regression kriging (RK) with environmental auxiliary information on DRASTIC-LU-based aquifer vulnerability to characterize groundwater nitrate-N pollution in the Pingtung Plain, Taiwan. First, the relationship between groundwater nitrate-N pollution and aquifer vulnerability assessment was determined using stepwise multivariate linear regression (MLR). Then, simple kriging was adopted to estimate residuals acquired from gaps between nitrate-N observations and MLR predictions. The sum of the estimated residuals and MLR predictions was the RK estimates for groundwater nitrate-N. Finally, groundwater nitrate-N distributions were spatially analyzed using RK, OK, and MLR. To reduce groundwater nitrate-N pollution, feasible environmental management strategies were discussed according to the study results. The study results revealed that the orchard land-use types and medium- and coarse-sand fractions of vadose zone were related to groundwater nitrate-N. Moreover, the fertilizer application in orchards was the major source of groundwater nitrate-N pollution. The RK estimates could characterize the characteristics of the pollution source of the orchard land uses, and exhibited higher spatial variability and accuracy via the correction of the residuals than MLR predictions and OK estimates. In addition, feasible management strategies of orchards at the eastern and western regions with high areal ratios of orchard land uses should be implemented to reduce nitrate-N leaching, such as organic fertilizer uses, ground covers, and irrigation with low intensity and high frequency.

How to cite: Jang, C.-S.: Geostatistical estimates of groundwater nitrate-N with spatial auxiliary information on DRASTIC-LU-based aquifer vulnerability, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-2072, https://doi.org/10.5194/egusphere-egu23-2072, 2023.