Groundwater assessment is critical for addressing global and regional water security challenges, particularly in transboundary areas and conflict zones such as Ukraine. These regions often experience shifting water balance patterns due to excessive groundwater abstraction, damage to water infrastructure, and large-scale population displacement to safer border areas. Modeling transboundary groundwater storage and flows in such contexts is challenging due to uneven data availability across borders, inconsistent hydrogeological descriptions, restricted fieldwork, and limited local capacity to maintain data collection infrastructure. Effective water management in these areas requires pooling global expertise and resources, increasingly leveraging satellite observations, and fostering close collaboration among partner countries engaged in transboundary aquifer modeling.

The Groundwater Resilience Assessment through Integrated Data Exploration for Ukraine (GRANDE-U) project addresses these challenges through an organizational and technological framework uniting researchers from six countries—the U.S., Ukraine, Poland, Latvia, Lithuania, and Estonia. The project integrates physics-based and machine learning models for transboundary aquifers with downscaled satellite remote sensing data. Building on the foundations of the NSF-funded AccelNet Transboundary Groundwater Resilience project and the European EU-WATERRES project, GRANDE-U employs the following methodology:

-             Developing a spatial database of water-related indicators for transboundary areas, including geology, water resources, land cover, monthly precipitation, evapotranspiration, runoff, soil moisture, and other characteristics at observation points and a 0.1"-0.25" grids covering the aquifer;

-             Creating algorithms to downscale GRACE/GRACE-FO-based terrestrial water storage (TWS) and groundwater storage (GWS) data to resolutions of 0.25" and finer for specific regions with the best available hydrogeologic data sufficient for GRACE/GRACE-FO adjustment; and

-             Developing machine learning models to describe GWS dynamics, utilizing as predictors the monthly averages organized in the spatial database.

Initial results highlight the application of various machine learning models for accurate TWS-GRACE prediction, emphasizing hyperparameter tuning and encoding spatial dependencies. As the database and algorithms evolve, these models aim to improve transboundary groundwater monitoring and management.

An additional novel component of the GRANDE-U collaboration involves analyzing global expertise in transboundary groundwater research. Using a co-authorship network analysis, the project identifies key contributors, emerging topics, knowledge gaps, and collaboration patterns across hydrogeological subdomains and related disciplines. The analysis tracks the formation and evolution of expertise clusters and explores subsets of the network based on environmental, socio-economic, and data-related issues mentioned in publication titles and abstracts. This network analysis is implemented on the SuAVE (Survey Analysis via Visual Exploration, suave.sdsc.edu) visual analytics platform, using OpenAlex, an open-access bibliographic database, to extract and tag relevant publications with keywords and aquifer names. The system provides interactive visualizations of the academic landscape and computes fragmentation and centrality measures for individual researchers and network subsets, offering valuable insights for enhancing international collaboration on transboundary groundwater issues.

GRANDE-U funding under the “International Multilateral Partnerships for Resilient Education and Science System in Ukraine” (IMPRESS-U) initiative, led by the U.S. National Science Foundation, is gratefully acknowledged.

How to cite: Zaslavsky, I., Samalavičius, V., Solovey, T., Brzezińska, A., Janica, R., Śliwińska-Bronowicz, J., Stradczuk, A., Bikše, J., Žaržojus, G., and Kunsakova, A.: Data-driven Modeling of Transboundary Aquifers in Conflict Zones: Challenges and Solutions from the GRANDE-U International Collaboration, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7285, https://doi.org/10.5194/egusphere-egu25-7285, 2025.

Groundwater resources in hard-rock terrains are particularly susceptible to stress due to their intricate geological formations and limited recharge capacity. This study presents a novel methodology for assessing aquifer stress within the Barakar River Basin of the Chotanagpur Plateau, leveraging the integration of the Soil and Water Assessment Tool (SWAT) and advanced deep learning models. A comprehensive evaluation was conducted using 20 hydrogeological and socio-economic parameters, including precipitation, slope, land use, and aquifer lithology. Deep learning techniques, notably Convolutional Neural Networks (CNN), were utilised to classify aquifer stress zones into four categories: Low Stress, Moderate Stress, Semi-Critical, and Critical. The CNN model demonstrated superior performance, achieving an accuracy of 94% and effectively capturing aquifer conditions' spatial and temporal dynamics. Field validation via Electrical Resistivity Tomography (ERT) surveys substantiated the reliability of the model's predictions. Findings indicate that approximately 34% of the basin experiences moderate to critical stress levels, underscoring the urgency for targeted management strategies. This integrated approach offers a scalable and robust framework for sustainable groundwater management in hard-rock terrains, with significant implications for mitigating global water scarcity.

How to cite: Bera, A., Dutta, L., Kumar, R., and Pal, S. K.: Aquifer Stress Assessment in Hardrock Regions of the Chotanagpur Plateau Using Integrated SWAT and Deep Learning Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-803, https://doi.org/10.5194/egusphere-egu25-803, 2025.

Estimating saturated hydraulic conductivity K_f from particle size distributions (PSD) is very common with empirical formulas, while the use of machine learning for that purpose is not yet widely established. We evaluate the predictive power of six machine learning algorithms, including tree-based, regression-based and network-based methods in estimating K_f from the PSD solely. We use a dataset of 4600 samples from the shallow Dutch subsurface for training and testing. The extensive dataset provides not only PSD, but also measured conductivities from permeameter tests. Besides training and testing on the entire data set, we apply the six algorithms to data subsets for the soil types sand, silt and clay. We further test different feature/target-variable combinations such as reducing the input to PSD-derived characteristic grain diameters d₁₀ , d₅₀ and d₆₀ or estimating porosity from PSD. We test feature importance and compare results to K_f estimates from a selection of empirical formulas. We find that all algorithm can estimate K_f from PSD at high accuracy (up to R²/NSE of 0.89 for testing data and 0.98 for the entire data set) and outperform empirical formulas. Particularly, tree-based algorithms are well suited and robust. Reducing information in the feature variables to grain diameters works well for predicting K_f of sandy samples, but is less robust for silt and clay rich samples. d₁₀ also shows to be the most influential feature here. An interesting, but not surprising outcome is that PSD is not a suitable predictor for porosity. Overall, our results confirm that machine learning algorithms are a powerful tool for determining K_f from PSD. This is promising for applications to e.g. deep-drilling data sets or low-effort and robust K_f -estimation of single samples.

How to cite: Zech, A., de Rijk, V., Buma, J., and Veldkamp, H.: Predicting saturated hydraulic conductivity from particle size distributions using machine learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4425, https://doi.org/10.5194/egusphere-egu25-4425, 2025.

Preparing for drought conditions is complicated by the episodic nature of droughts and by our limited understanding of water systems’ response to extreme events. In this, models are useful tools to simulate a range of plausible to low likelihood drought conditions. Water managers may use these simulations to make plans and consider consequences for both normal and extreme drought events. However, critical in this is the representation of water system resilience to drought conditions and simulated management decisions to in/decrease drought resilience. Decision-making in groundwater management could herein benefit from a robust modelling approach that considers the complexity and uncertainty in water availability, dynamic impact of management and modelling setups available.

In this study, we have converted a lumped conceptual socio-hydrological model to an operational tool to support groundwater management in Great Britain by applying a response-based and a data-based model evaluation.  In the response-based evaluation, we first examined the model consistency with our understanding of the system functioning, and the influence of modelled management scenarios on model predictions. In the data-based evaluation, we tested the accuracy of heavily influenced discharge and groundwater level predictions in three catchments representative of typical hydrogeological conditions and water management practices in Great Britain.

Results show consistent simulations across catchments and identified pointers for influential model parameters in drought conditions. Modelled water management interventions have varying influence on simulated model output. Most effective drought management scenarios have (elements of) integrated water storage use, which minimises shortages in water demand. The data-based analysis shows that calibration can be focused on either low flows or groundwater storage, with reasonable results for both model outputs. We provide a source-specific and ‘best overall’ calibration approach that capture groundwater levels and low flows, which also indicates how model parameters (dis)agree with open-source data and our model perception of the modelled water system.

How to cite: Wendt, D. E., Coxon, G., Salwey, S., and Pianosi, F.: SHOWER: A tool for groundwater drought management , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6890, https://doi.org/10.5194/egusphere-egu25-6890, 2025.

The Alcantara River basin, located in Sicily (Italy), encompasses an area of 606 km², including the northern slopes of Mount Etna, Europe's highest active volcano (3357 m a.s.l.). The river stretches for 55 km, originating from the Nebrodi Mountains at 1400 m a.s.l. and discharging into the Ionian Sea approximately 5.5 km south of Taormina (Messina). Groundwater significantly sustains the river’s flow. Irrigation and drinking water wells extract an average of 0.23 m³/s, while a drainage gallery collects water from three springs, supplying the Alcantara aqueduct with an average flow rate of 0.48 m³/s, as measured from January 2009 to December 2022.
To better understand the aquifer-river interactions and assess the impacts of groundwater extractions, the aquifer system was modeled using MODFLOW 6, a finite-difference numerical code developed by the U.S. Geological Survey (USGS). The study covered 14 years (2009–2022), leveraging monthly groundwater withdrawal records provided by the aqueduct operator. Hydraulic conductivity was calibrated using PEST, a software tool for parameter estimation and uncertainty analysis. Two scenarios were considered: (1) the current condition, including all known groundwater extractions, and (2) a hypothetical scenario without extractions. The model was validated by comparing observed and simulated discharge trends from the drainage gallery.
Simulation results revealed that groundwater extractions reduce natural spring discharge to the river by an average of 22%. This reduction shows significant seasonal variability, with the most pronounced impacts in spring and less severe effects in winter. Interestingly, despite this reduction, the midstream section of the river did not experience zero discharge, even during the driest periods (e.g., the summers of 2020 and 2021). This discrepancy suggests the influence of additional unaccounted factors, such as unauthorized or unrecorded water withdrawals, or potential hydrogeological changes induced by seismic activity.
These findings emphasize the need for systematic monitoring of groundwater and surface water resources. Enhanced monitoring would provide a deeper understanding of aquifer-river interactions, identify drivers of hydrological regime alterations, and inform strategies to optimize groundwater use and mitigate its impacts on river systems. Such efforts are essential for protecting the natural environment and ensuring the long-term availability of water resources.

How to cite: Silipigni, M., Di Salvo, C., Preziosi, E., Borzì, I., and Bonaccorso, B.: Hydrological Modeling of a Fractured Volcanic Aquifer to Analyze Interactions Between Anthropogenic Water Resource Usage and the Natural System, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6734, https://doi.org/10.5194/egusphere-egu25-6734, 2025.

This study investigates the effects of groundwater extraction on saltwater movement in Berlin/Brandenburg's lower Spree catchment, a critical freshwater resource increasingly impacted by climate change and water scarcity. A high-resolution groundwater flow model was developed to simulate transient flow and saltwater dynamics. The model incorporates recharge data (1979–2019), pumping records (1994–2021), and a detailed geological framework derived from borehole data and cross-sections. Artificial neural networks (ANNs) were used to capture spatial heterogeneity, with the model discretized into ≈ 3.5 million active flow cells using a finite-difference approach with 100 m horizontal, 5 m vertical, and monthly temporal resolutions. The initial conditions determined through a spin-up period. Model calibration, supported by PEST, ensured robust performance in both steady-state and transient conditions.

Results reveal significant interactions between freshwater and saline zones, with prolonged extraction driving saltwater upconing. Scenario analyses highlight the sensitivity of saltwater movement to climate change, projecting accelerated saltwater intrusion under intensified pumping and reduced recharge conditions.

These findings underscore the need for adaptive groundwater management strategies, such as optimized pumping schedules and integrated management practices, and Managed Aquifer Recharge (MAR)to mitigate saltwater intrusion and ensure sustainable freshwater availability under changing climatic and resource pressures.

How to cite: Abdelrahman, A. A. A., Sauter, M., and Engelhardt, I.: Impact of groundwater extraction on saltwater movement in the lower Spree catchment under climate change and water scarcity, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3175, https://doi.org/10.5194/egusphere-egu25-3175, 2025.

The Tulare Lake Basin (TLB) of California, a vital agricultural hub that covers the southern part of the Sierra Nevada and the Central Valley, is experiencing severe water scarcity due to climate change, rising water demand, and extensive groundwater depletion. This study leverages an integrated hydrological model to quantify Surface Water-Groundwater (SW-GW) interactions in the TLB. We focus on understanding seasonal water balance trends, variability in groundwater recharge, and the impact of snowmelt on groundwater storage. The integrated SWAT+gwflow model was calibrated and validated using observations of streamflow, evapotranspiration, snow water equivalent, and groundwater head. Our results revealed a decreasing trend in groundwater storage (-4.77 mm/year), with a greater deficit during prolonged droughts.  Most of the groundwater fluxes have negative trends, including SW-GW exchange, saturated excess flow, and lateral flow. Boundary inflow exhibits a positive trend due to inflow from adjacent regions driven by hydraulic gradients caused by local groundwater depletion. Snowmelt emerged as a critical driver of groundwater recharge in the TLB, showing a stronger correlation in the spring (R² >0.58) than fall and winter seasons (R² < 0.5). These findings suggest exploring alternative means for groundwater recharge, as mountain snowpack is expected to decline in a warmer climate, such as capturing winter flood flows and the need for adaptive strategies to mitigate long-term water stress in the basin.

Keywords:  Coupled Model; SW-GW Interactions; TLB

How to cite: Sinshaw, B. G., Viers, J. H., and Safeeq, M.: Modeling Surface Water - Groundwater Interactions in the Tulare Lake Basin, California, USA, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7287, https://doi.org/10.5194/egusphere-egu25-7287, 2025.

Single-well push-pull tracer tests are broadly employed to estimate effective parameters for solute and heat transport in aquifers. Tracer recovery curves obtained from these tests serve as inputs for solving an inverse problem to infer effective transport parameters such as porosity, thermal and solute longitudinal dispersivities, and retardation factors. However, the inherent non-uniqueness of the inverse calibration problem and associated uncertainties in field measurements create a bottleneck for multiparametric calibration. To address these challenges, we employed a computationally efficient optimization framework based on surrogate modeling via Gaussian process regression (GPR) to approximate the objective function based on six effective transport parameters to be calibrated simultaneously, which yields plausible parameter combinations. For training and model evaluation, we implemented a 1D finite-difference (FD) representation of the advection-dispersion equation for sorbing tracers featuring an adaptive explicit time stepping scheme adhering to numerical stability criteria while minimizing numerical diffusion, where an analytical radial flow field serves as input based on well hydraulic properties. The FD model includes the measured input time series of temperature and concentration as transient boundary conditions, as well as well-bore storage to accurately model push-pull test conditions. We applied this framework to push-pull tests conducted in a sandy aquifer in Horonobe (Hokkaido, Japan) using heat and three solute tracers: uranine, lithium, and iodide. The confidence intervals for field measurements were included by using repeated under identical conditions tests. The surrogate model facilitates parameter optimization by balancing the exploration of high-uncertainty regions with the exploitation of high-probability regions through a weighted probability function. The posterior parameter distribution reveals reduced uncertainty intervals for porosity and both solute and thermal dispersivities while indicating low sensitivity for the solute retardation factor. The results demonstrate the necessity of high-precision measurements for concentration and highlight the value of utilizing multiple tracers to enhance calibration accuracy under parametric and measurement uncertainty. The developed framework highlights the benefit of using machine learning techniques combined with physics-based models to efficiently address stochastic parameter optimization under parametric uncertainty. The developed framework is a useful tool that enables time-efficient stochastic evaluation of computationally expensive models and optimization of push-pull tests.

How to cite: Petrova, E., Selzer, P., Kranz, S., Zeilfelder, S., Hebig, K. H., Machida, I., Marui, A., Blöcher, G., and Scheytt, T.: Surrogate model supported optimization of a multitracer push-pull test in Horonobe aquifer (Japan) under parametric uncertainty, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5774, https://doi.org/10.5194/egusphere-egu25-5774, 2025.

Groundwater is becoming more important in sustainable water management, particularly in the context of climate change and intensive human interventions. Given that groundwater varies in space and time, it is important to predict both its dynamic processes and static patterns. However, lack of reliable groundwater data restricts the development of large-scale groundwater monitoring systems linking observations with modeling at spatial scales relevant for local decision making. In this study, we leverage existing physically-based modeling data and water table depth observations in the Contiguous United States (CONUS) and develop a machine learning-based downscaling tool to downscale 1-km modeling data to 1arcsec (~ 30 m). The modeling data were generated daily for the water year 2003 using ParFlow, a three-dimensional integrated hydrologic model. In addition, we input a range of meteorological, topographic, geological, and land use data, including daily precipitation and temperature, elevation, hydraulic conductivity, mean soil and clay contents, and land cover types. Based on tree-based machine learning models running on GPUs, the downscaling tool outputs a 1 arcsec water table depth map for the CONUS daily in a relatively short time. The resulting hyper-resolution water table depth map incorporates groundwater pumping and uncertainty, significantly advancing our understanding of groundwater dynamics across various scales, from the continental to small scales relevant to local decision-making. We also obtain the importance of input variables based on the results of the machine learning models, which is helpful for the future development of the groundwater monitoring system over the CONUS. While the downscaling tool is developed for the CONUS, it can be adapted to other regions with similar hydrogeological settings and substantial modeling data.

How to cite: Ma, Y., Tijerina-Kreuzer, D., Defnet, A., Condon, L., and Maxwell, R.: Advancing Large-Scale Hyper-Resolution Groundwater Modeling Using a Machine Learning-Based Downscaling Tool, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13673, https://doi.org/10.5194/egusphere-egu25-13673, 2025.

Rapid population growth and agriculture development have led to unsustainable exploitation of groundwater, a trend likely to be exacerbated by future climate change. Managed Aquifer Recharge (MAR) emerges as a cost-effective strategy for replenishing depleted aquifers, thereby supporting long-term groundwater sustainability. However, the highly heterogeneous nature of groundwater processes necessitates fine spatial and temporal resolution models for designing and evaluating MAR. The high computational burden of physics-based model simulations constrains existing MAR studies to limited scenarios, missing opportunities to evaluate MAR potential across the full range of uncertainties from climate change, policy shifts, and infrastructure development. Recent advances in artificial intelligence offer a promising solution to the trade-off between high spatial-temporal precision and computational efficiency through surrogate models. In this study, we leverage recent advances in attention-based Graph Neural Networks (aGNN) to develop a surrogate model for MAR (GNN-MAR), which allows us to capture multi-scale network structures across river systems, groundwater flow, and MAR infrastructure. Trained on a high-resolution physics-based integrated surface water and groundwater model, GNN-MAR is tailored for two MAR approaches, i.e., in-channel recharge and agriculture MAR (Ag-MAR). We apply GNN-MAR to the Baoding Plain in the North China Plain, one of the world’s most severely groundwater depleted regions. The search for optimal MAR schemes is conducted within large ensembles generated under the XLRM (eXogenous uncertainties, policy Levers, Relationships, Measures) framework, which encompasses climate change scenarios, groundwater pumping policies (X), MAR schemes (L), GNN-MAR (R), and groundwater sustainability targets (M). The framework enables identification of MAR schemes robust to deep uncertainties. Our study provides valuable insights for the development of high-fidelity surrogate models for integrated surface-groundwater systems and demonstrate the potential of AI-based surrogate model for robust decision-making in groundwater recharge management.

How to cite: Guo, W., Li, M., Chen, H., and He, X.: Robust managed aquifer recharge (MAR) design aided by graph neural networks, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7006, https://doi.org/10.5194/egusphere-egu25-7006, 2025.

Finding the initial state groundwater configuration of a catchment is one of the major challenges when simulating the hydrological cycle with an integrated hydrological model. The choice of this initial condition has a large impact on the results of the subsequent simulation, and it is often found by repeatedly running the hydrological model with constant atmospheric settings until the system equilibrates. These spin-up computations are computationally expensive and often require many years of simulated time, especially if the initial groundwater configuration before the spin-up computations is far from this steady state.

We hypothesize that existing large-scale groundwater simulations at steady state can be used to machine learn how steady-state depth-to-water tables (DTWTs) for groundwater depend on readily available data sources like large-scale conductivity and surface slopes. But how well can steady-state DTWTs be estimated by such ideas? How much computing speed can be gained with improved initializations of spin-up simulation? And how well does the estimation of improved initializations generalize across different geological settings and climate?

To answer these questions, we developed the machine learning emulator HydroStartML to accelerate the spin-up computation. HydroStartML is trained on converged steady-state DTWT distribution, and it generates a configuration of the DTWT of the respective watershed. This configuration is used as the starting configuration for spin-up computations. Doing so reduces the overall computational effort compared to the typical approach of initiating spin-up computations with a uniform DTWT across the whole catchment. HydroStartML is trained on the entire contiguous United States on spatially distributed patches with a fixed set of parameters.

Spin-up computations with these DTWT configurations as starting configurations converge faster and with a reduced computational effort compared to spin-up computations with other initial configurations. We found that HydroStartML is indeed able to generate DTWT configurations that are close to the steady state, even on unseen terrain. Although the generation of shallow DTWTs is possible with especially small errors, the strongest reductions in spin-up effort occurs in regions with deep DTWTs. This work opens the door for hybrid approaches that blend machine learning and traditional simulation, enhancing predictive accuracy and efficiency in hydrology for improving water resource management and understanding complex environmental interactions.

How to cite: Pawusch, L., Scheurer, S., Nowak, W., and Maxwell, R.: Development of a Combined Machine Learning and Physics-based Approach to Reduce Hydrologic Model Spin-up Time, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10229, https://doi.org/10.5194/egusphere-egu25-10229, 2025.

Groundwater nitrate (NO₃^-) pollution is a pressing issue linked to agricultural practices, urbanization, and industrial activities. This study focuses on Taiwan’s groundwater nitrate nitrogen (NO₃-N) contamination by integrating satellite remote sensing, groundwater monitoring, and various environmental factors using GIS. Data from 451 monitoring stations, sampled quarterly from 2020 to 2024, reveal that NO₃-N concentrations generally range between 1–10 mg/L, while approximately 2% exceed Taiwan’s Drinking Water Quality Standards of 10 mg/L for NO₃-N (equivalent to 44.3 mg/L NO₃^-). In this study, machine learning models, including Random Forest (RF), Multilayer Perceptron, and Support Vector Classifier, were employed to predict NO₃-N contamination risk at three ranges of concentrations (<1, 1–10, >10 mg/L) using different feature combinations: (1) all features, (2) selective environmental factors, and (3) vegetation indices (VIs) alone. RF demonstrated the highest overall accuracy across all combinations, achieving 87% in Feature Combination I. For Feature Combination III, which only used VIs derived from remote sensing, RF achieved an OA of 68%, highlighting its potential for practical and efficient application without ground-based survey data. Key findings highlight the pivotal role of environmental variables, including VIs derived from Sentinel-2 multispectral imagery, terrain parameters from digital elevation models, and meteorological data in mapping contamination hotspots. Future work should integrate higher-resolution satellite imagery and more advanced parameters to improve model performance and decision-making accuracy.

How to cite: Hsu, Y.-C., Li, K.-Y., Podgorski, J., and Berg, M.: Remote Sensing-Driven Prediction of Groundwater Nitrate Risk: Insights from Machine Learning Applications in Taiwan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6502, https://doi.org/10.5194/egusphere-egu25-6502, 2025.

Groundwater is a critical source of drinking water and as demand increases on a global scale under climate change and population growth, a suitable quantity of clean groundwater must be ensured. Groundwater chemistry depends on environmental factors such as geology, climate, groundwater table and residence time, land use, and recharge source and rate. Geogenic compounds, such as arsenic (As), manganese (Mn), phosphorus (P), ammonium (NH₄⁺), and iron (Fe), often occur in groundwater and are important determinants of groundwater quality. When exceeding recommended concentration limits in groundwater, these compounds can pose risks to human health and the environment, and cause problems in water treatment and distribution. In this study, we applied machine learning, i.e., classification algorithms and feature importance analysis, to investigate the spatial patterns of selected geogenic compounds and their governing factors. We used groundwater chemistry measurements from over 7,000 well intakes with mean depth of 47.8 m distributed across Denmark and 34 covariate maps including soil, geology, and hydrogeology information. Models are developed for As, Mn, total P, NH₄⁺ and Fe, and achieve a balanced accuracy between 76% and 88%. The main results are prediction maps of 100 m resolution showing the probability of the selected geogenic compounds exceeding the concentration limits in groundwater recommended by Danish legislation. Our analysis advocates that the spatial variability of all selected compounds depends mostly on geological factors such as the thickness of Quaternary, accumulated clay deposits above chalk, and the depth to chalk formations. High concentrations of all studied geogenic compounds are predicted in areas with thick Quaternary and clay deposits, while low Mn and P predictions occur in areas where chalk is present at lower depths. Overall, we found that groundwater exceeds recommended concentration limits for As, Mn, P, NH₄⁺ and Fe in 9.6%, 67.5%, 48.5%, 73.5% and 83% of Denmark’s area, respectively. Our results enhance the understanding of the processes driving groundwater quality in Denmark, which may be transferable to other domains with similar hydrogeological settings, e.g. northern America. The generated prediction maps can guide for identifying optimal locations for new wells and water treatment techniques, improving the overall groundwater resource management at national scale. Accordingly, this study shows how public high-quality databases can aid groundwater management.

How to cite: Xenakis, G. I., Jessen, S., Koch, J., and Kazmierczak, J.: Spatial machine learning predictions of geogenic compounds in groundwater, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11772, https://doi.org/10.5194/egusphere-egu25-11772, 2025.

Meat production is a major contributor to global environmental degradation, including groundwater nitrate contamination driven by intensive fertilizer use and manure production. This study explores the environmental implications of substituting conventional meat products (beef, poultry, and pork) with alternative protein sources—plant-based, insect-based, and cultured meat—using the U.S. meat market as a baseline. Employing an eXtreme Gradient Boosting (XGBoost) model, we quantify the risk of groundwater nitrate exceedance and compare resource requirements such as fertilizer, water, and land use for conventional and alternative proteins.

Results indicate that a 10% substitution of meat protein with alternatives reduces fertilizer use by 3.4%, manure production by 10.7%, and water usage by 4.5%, leading to a 20% reduction in groundwater nitrate exceedance risk. Plant-based alternatives show the lowest environmental impact, while insect-based options demonstrate high feedstock efficiency. Cultured meat, despite its potential, currently exhibits higher resource demands due to production constraints. The study further highlights regional variations in substitution effects, driven by agricultural practices and climatic factors.

These findings underscore the environmental benefits of transitioning to sustainable protein sources, providing actionable insights for achieving Sustainable Development Goals (SDGs) related to water quality, food security, and climate resilience. This shift not only reduces environmental risks but also ensures the sustainable management of groundwater resources.

How to cite: Guo, Z. and Zhan, Y.: Shifting Protein Sources to Reduce Groundwater Nitrate Contamination: Insights from the U.S. Meat Market, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2241, https://doi.org/10.5194/egusphere-egu25-2241, 2025.

Identifying contaminant sources is crucial for managing groundwater contamination, particularly in complex fracture networks. Traditional methods for source identification often face limitations such as sensitivity to data perturbations, reliance on simplified hydrological models, and challenges in handling the ill-posed nature of the inverse problem. This study introduces a novel application of Bayesian Evidential Learning (BEL) to quantify contaminant source uncertainty in fracture networks(Hermans et al., 2018; Thibaut et al., 2021).

BEL relies on learning a direct relationship between the target parameters (source location, release time, and concentration) and predictors (breakthrough curves (BTCs) and their statistical features). The learning step relies on the sampling of target parameters for which the release and transport of contaminant is simulated, and the resulting BTCs at the observation point extracted. The complexity of the training process was mitigated by incorporating falsification to classify the prior model(Yin et al., 2020). One-hot encoding then was employed to discretize potential source locations, enhancing the correlation between predictor and target using principal component analysis (PCA) and canonical correlation analysis (CCA) (Figure 1). Experimental data and numerical simulations of solute transport in fracture networks were then employed to validate the BEL framework (Figure 2).

Results demonstrate that BEL not only achieves accurate predictions on the source location, release time and concentration, but also provides robust uncertainty quantification for contaminant sources. These findings highlight BEL's potential as a powerful tool for improving source tracking and remediation strategies in groundwater systems. Future research should consider uncertainty in the fracture network and hydraulic properties of the fractures.

Figure 1. Multivariate analysis of training data. A. The explanatory power of data across different PCs in the PCA space. B-E are the bivariate distributions of predictor and target data in the CCA space.

Figure 2. Posterior distribution predictions for contaminant source information. Red lines correspond to test data

References

Hermans, T., Nguyen, F., Klepikova, M., Dassargues, A., & Caers, J. (2018). Uncertainty Quantification of Medium-Term Heat Storage From Short-Term Geophysical Experiments Using Bayesian Evidential Learning. Water Resources Research, 54(4), 2931–2948. https://doi.org/10.1002/2017WR022135

Thibaut, R., Laloy, E., & Hermans, T. (2021). A new framework for experimental design using Bayesian Evidential Learning: The case of wellhead protection area. Journal of Hydrology, 603, 126903. https://doi.org/10.1016/j.jhydrol.2021.126903

Yin, Z., Strebelle, S., & Caers, J. (2020). Automated Monte Carlo-based quantification and updating of geological uncertainty with borehole data (AutoBEL v1.0). Geoscientific Model Development, 13(2), 651–672. https://doi.org/10.5194/gmd-13-651-2020

How to cite: Miao, K., Huang, Y., Zhang, L., Guo, L., and Hermans, T.: Quantifying Groundwater Contaminant Source Uncertainty in Fracture Networks Combining Falsification and Bayesian Evidential Learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11456, https://doi.org/10.5194/egusphere-egu25-11456, 2025.

Machine learning models, particularly Random Forests (RF), are increasingly used to regionalize environmental pollutants based on point measurements. Spatial variants of RF are emerging to account for geospatial data characteristics, such as spatial autocorrelation and non-stationarity. However, systematic comparisons of these spatial RF variants remain limited.

This study evaluates seven spatial RF variants and compares them to non-spatial RF, universal kriging (UK), a well-established geostatistical method, and multiple linear regression (MLR). Using nitrate concentrations in groundwater from two contrasting hydrogeological macro-regions in Germany, we assess predictive performance (mean absolute error) across varying prediction distances using spatial cross-validation.

The results show minor differences among spatial RF variants, except for the notably lower performance of Random Forest Spatial Interpolation (RFSI) at long prediction distances. Over short distances (within the practical range of spatial autocorrelation), spatial RF variants outperformed non-spatial RF and MLR. The RF-oob-OK method, which applies ordinary kriging on the out-of-bag errors, demonstrated consistently strong performance with acceptable computational efficiency. However, it did not substantially surpass UK in predictive performance.

Computationally manageable spatial RF variants, such as RF-oob-OK, represent viable alternatives to traditional geostatistical methods for spatial prediction of environmental pollutants, effectively exploiting both spatial predictors and autocorrelation.

How to cite: Frank, J., Suesse, T., Jiang, S., and Brenning, A.: Do spatial random forest variants improve the regionalization of environmental pollutants? - The case of groundwater nitrate concentration, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18245, https://doi.org/10.5194/egusphere-egu25-18245, 2025.

Accurate prediction of the contaminant plumes in groundwater systems are critical for effective pollution management and risk assessment. Effective simulations for reliable predictions require two key pieces of information. The first is detailed knowledge of the aquifer system, including subsurface structures and the hydrogeological heterogeneity of hydraulic parameters. The second is data about the contaminant plume, including its source, spatial distribution, and concentration. However, acquiring and analyzing this data is often costly and labor-intensive due to the extensive collection efforts and complex processing techniques required.

To address these challenges, we developed an innovative machine learning prediction approach. The architecture of the model combines fully connected layers followed by convolutional layers. The training dataset for the machine learning model was generated using a numerical simulation model of groundwater flow and contaminant transport processes in a synthetic aquifer. Monitored contaminant concentration data were used as inputs to the machine-learning model, while contaminant plume distributions (e.g., concentration fields spanning from the initial contaminant release to 10 years in the future) served as outputs. The machine learning models are trained and evaluated under two scenarios: (1) assuming aquifer properties are well-known, (2) aquifer properties are unknown. According to the results, in scenario 1, the prediction of the contaminant field at various time is highly accurate: the predictions resemble the reference at high degree. In scenario 2, prediction accuracy decreased but remained effective: the predicted contaminant plume closely matched the overall structure of the reference distribution. The main advantage of this machine-learning approach is its capability to directly analyze monitoring data and predict the transient groundwater contaminant transport processes, the labor-intensive steps of aquifer characterization and initial contaminant field determination are eliminated. Moreover, the results not only forecast future evolution but also allow for historical tracing, all the way back to its initial release point, thus it provides a comprehensive understanding of the contaminant's lifecycle.

How to cite: Wang, C., Dou, Z., Yang, Y., Chen, Z., Hu, R., Zhao, Y., and Wang, J.: A Machine Learning Approach for Predicting Contaminant Plume Evolution in Groundwater systems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20774, https://doi.org/10.5194/egusphere-egu25-20774, 2025.

Fluctuation of shallow groundwater (GW) characteristic in plains and lowlands is regulated by recharge from precipitation and groundwater evapotranspiration under natural conditions. Here a simple 1D (pointwise) model is presented that is capable of reproducing the dynamics of shallow GW levels at monthly, weekly or even daily time scale at the location of groundwater level monitoring wells using precipitation and potential evapotranspiration as input variables. The model utilizes the empirical curve describing the dependence of GW evapotranspiration on the depth of water table, revealed by analyzing historical measured GW level and evapotranspiration time series in the Great Hungarian Plain. Besides GW levels, the model calculates groundwater recharge and evapotranspiration. Soil and hydrological parameters of the model are calibrated based on measured GW levels.

Potential fields of application of the model are shown through Hungarian examples: 1) Using precipitation and potential evapotranspiration data of various climate models, expected alterations of shallow GW levels due to climate change can be predicted. 2) By comparing measured and simulated GW levels, alterations in GW levels and fluctuation due to anthropogenic activity (e.g. GW abstractions) can be revealed. 3) Data gaps in measured GW level time series can be filled by the model. 4) Extending the time series of measured GW levels into the past allows for historical analyses, e.g. the temporal changes of GW supply of dependent ecosystems.

How to cite: Ács, T., Kozma, Z., Decsi, B., and Simonffy, Z.: Modelling shallow groundwater fluctuation based on water table depth – groundwater evapotranspiration relationship, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16851, https://doi.org/10.5194/egusphere-egu25-16851, 2025.

Groundwater recharge is a key input in groundwater resources management. However, it is not directly measurable, and even the parameters for indirect estimation are difficult to obtain and verify. We developed a hydrological model that facilitates reliable calibration of specific yield of an aquifer, thus enabling groundwater recharge estimation by the water table fluctuation method. The novel soil moisture deficit model is based on existing bucket models. It features two parallel soil water reservoirs which capture all incoming precipitation. Groundwater recharge is generated only when the reservoirs are overfilled after satisfying evapotranspiration. The only input data required are easily measurable (and typically available) time series of precipitation and air temperature, and long-term record of water table fluctuations in wells for calibration. The model parameters, most importantly specific yield, are calibrated by comparing the modelled water table to the observed water table. The calibrated specific yield and the observed water table levels then serve as inputs for water table fluctuation method for estimation of groundwater recharge. The model was tested on 9 wells in the lowland along the river Elbe (Czech Republic). The wells are situated in highly permeable alluvial aquifers with unconfined water table. The wells were selected for the study because decades of water table records are available, and their water table exhibits multi-year fluctuations. The values of specific yield obtained by model calibration ranged from 5% to 17%, which is realistic for the studied aquifer type. The results were compared to a previous study conducted in the Czech Republic, in which the long-term mean groundwater recharge was estimated. Our results lie within the range indicated for the sites. Furthermore, the presented method provides the temporal distribution of groundwater recharge, thus broadening the knowledge of groundwater recharge dynamics. Besides this, the method can potentially be used to estimate the groundwater recharge under future climate conditions.

Acknowledgements

The study was supported by project SS02030040 "PERUN - Prediction, Evaluation and Research for Understanding National sensitivity and impacts of drought and climate change for Czechia", co-financed with state support of the Technology Agency Czech Republic as part of the Program Environment for Life. We would like to thank Mgr. Tomáš Ondovčin, Ph.D. for consultation of saturated zone modelling, and Mgr. Martin Lanzendörfer, PhD. and Ing. Jan Černý for aid with mathematical formulation of the SMD model.

How to cite: Šabatová, K. and Bruthans, J.: Calibration of the water table fluctuation method based on groundwater recharge model using easily available data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8737, https://doi.org/10.5194/egusphere-egu25-8737, 2025.

The interaction between surface water and groundwater plays a crucial role in effective water resource management. In Saxony, Eastern Germany, lakes and reservoirs contribute significantly to the public water supply, alongside groundwater to a lesser extent. Despite growing attention to comprehensive studies of the region's water resources, the complex dynamics between streamflow and groundwater levels remain insufficiently explored. This research aimed to address this gap by providing a detailed analysis of these interactions using time series data.

To bridge this gap, the research framework integrated preprocessing techniques, feature-based characterization and clustering of groundwater level time series, and Convergent Cross-Mapping (CCM) applied to coupled groundwater level and discharge datasets. CCM, which uses time series data to identify causal relationships within dynamic systems, their direction and strength, was used to study the interactions between groundwater and streamflow in different regions of Saxony, Germany. Data from 597 groundwater level and 190 discharge time series have been used. The study also employed R, MATLAB, and QGIS for data processing and analysis, based on publicly available GitHub repositories and official documentation from previous studies.

The results revealed significant spatial variability in groundwater-stream interactions, with high levels of interaction identified in catchments such as the Elbe and Schwarze Elster, and lower levels of interaction in urban and agricultural areas. In regions such as Lausitz, geological and soil factors strongly influenced the streamflow-groundwater dynamic, with more complex interactions in areas with loess and highland soils. Factors like land cover and soil type played a significant role, as urbanization and land use changes can reduce groundwater recharge rates and disrupt natural water pathways. These findings underscore the importance of spatially distributed data for understanding the drivers of water system behavior and regional water resource management.

In conclusion, this study demonstrated the value of integrating time series data analysis methods, such as CCM, to enhance the understanding of hydrological dynamics in Saxony. The results provided insights into areas of high groundwater-streamflow interaction, highlighted the role of influencing factors, and emphasized the need for spatially detailed hydrological assessments to inform future water resource management strategies in the region.

How to cite: Vela Castillo, M. A., Hartmann, A., and Liu, Y.: Assessing Surface-Groundwater Interactions Using Time-Series Clustering and Convergent Cross Mapping: A Case Study of Saxony, Germany, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18480, https://doi.org/10.5194/egusphere-egu25-18480, 2025.

The Yellow River basin (YRB) is the second-largest river basin in China , flowing through arid regions. The development and utilization of water resources, including irrigation, urban water supply, and industrial use, face significant challenges (Lin et al., 2019;Qu et al., 2020). Although groundwater resources are abundant, they are constrained by excessive extraction and declining water tables (Lin et al., 2020), posing substantial challenges for water resource management, especially as the global water scarcity issue becomes increasingly prominent. It is challenging to estimate groundwater level at a regional or catchment scale due to its natural heterogeneity.

Here, we use a large sample of groundwater observations, together with datasets from the Global Land Data Assimilation System (GLDAS) and the Gravity Recovery and Climate Experiments (GRACE), to build a machine learning approach — random forest— for predicting regional groundwater levels in the Yellow River Basin of China.

We demonstrated the robustness of this model, with an R² of 0.95 at calibration mode and R² of 0.91±0.009 at a 10-fold cross-validation mode with 100 repetitions. Compared to the spatial predictability, its temporal predictability is less accurate, with R² value of 0.72 for a test period of April-May in 2023. The spatial distribution maps of the groundwater levels in Yellow River Basin showed strong seasonal declines in fall and winter, with severe decreases concentrated in the middle and lower reaches. Overall, this paper shows that it is promising to estimate regional groundwater levels based on machine learning with a large sample of groundwater observations, providing a robust and comprehensive data foundation for groundwater analysis.

References

Lin, M.,  Biswas, A., & Bennett, E. M. (2019), Spatio-temporal dynamics of groundwater storage changes in the yellow river basin. Journal of Environmental Management, 235, 84-95.  https://doi.org/10.1016/j.jenvman.2019.01.016.

Lin, M.,  Biswas, A., & Bennett, E. M. (2020), Socio-ecological determinants on spatio-temporal changes of groundwater in the yellow river basin, china. Science of The Total Environment, 731, 138725. https://doi.org/10.1016/j.scitotenv.2020.138725.

Qu, S.,  Wang, L.,  Lin, A.,  Yu, D.,  Yuan, M., & Li, C. a. (2020), Distinguishing the impacts of climate change and anthropogenic factors on vegetation dynamics in the yangtze river basin, china. Ecological Indicators, 108, 105724. https://doi.org/10.1016/j.ecolind.2019.105724.

Keywords:Random forest model; Groundwater level depth; GLDAS; GRACE

How to cite: Cao, Y. and Zhang, Y.: Estimating regional groundwater level by fusing satellite, model and large-sample observations inputs, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5408, https://doi.org/10.5194/egusphere-egu25-5408, 2025.

The development of purely data-driven approaches for groundwater level prediction is crucial for sustainable groundwater management, offering the ability to predict groundwater levels across numerous monitoring wells and large geographical regions. Especially in arid regions, groundwater resources are under pressure, as seen in areas like Brandenburg, Germany, which is characterized as the driest federal state with the highest number of lakes. Here, data-driven approaches can enable fast and accurate seasonal groundwater level predictions, supporting local authorities in managing sustainable utilization.

Unlike traditional numerical models, which are computationally expensive and require complex parameterization when applied to large geographical areas, data-driven models provide a scalable solution. Recent studies have demonstrated the potential of machine learning (ML) approaches, particularly different deep neural network architectures, to provide accurate groundwater level predictions. In these studies, recurrent neural networks, such as Long Short-Term Memory networks (LSTMs), as well as recently developed architectures like the Temporal Fusion Transformer (TFT), which combines LSTMs with the self-attention mechanisms, and Neural Hierarchical Interpolation for Time Series Forecasting (N-HiTS), which is a time-series decomposition algorithm based on multilayer perceptrons (MLPs), have been used. Another recently developed architecture, the Time-series Dense Encoder (TiDE), which is based on MLPs and residual blocks, has further expanded the toolkit for time-series prediction.

In this study, we evaluate and compare the performance of four deep learning (DL) architectures (LSTM, TFT, N-HiTS, and TiDE) in predicting groundwater levels up to 16 ahead, using a wealth of spatial and temporal information for over 1,000 monitoring wells across Brandenburg. Input features to our models include historical groundwater level measurements, climatic variables, and static physical characteristics, such as groundwater recharge and land cover. Our analysis identifies the environmental conditions under which these models achieve a good predictive performance accuracy and assesses their ability to capture varying groundwater dynamics, thereby testing their alignment with hydrogeological system understanding. Furthermore, we assess whether the static features enhance the models performance and facilitate generalization across monitoring wells with similar static features levels, which we test through ablation studies and spatial out-of-sample cross-validation.

Our findings provide valuable insights into the strengths and limitations of different DL architectures for groundwater level prediction, highlighting their potential to support sustainable groundwater management in regions facing water scarcity.

How to cite: Kunz, S., Wetzel, M., Engel, M., and Broda, S.: Deep Learning Models for Seasonal Groundwater Level Prediction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15537, https://doi.org/10.5194/egusphere-egu25-15537, 2025.

Groundwater which is the most valuable resource available on the Earth`s surface and is used for drinking, irrigation, livestock, etc is depleting day by day. Groundwater level prediction faces complex challenges to sustainably manage this vital resource. Predicting groundwater level is crucial for water resource management. Here this study explores the use of some hybrid machine learning models such as SVM-FFA, SVM-PSO and compared with the stand alone SVM approach. Then introducing an innovative approach for predicting groundwater level with improved accuracy and to enhance the performance of the model and face the challenges developed during the process. This work investigates hybrid machine learning techniques to improve the accuracy of groundwater levels predictions, which are constrained in conventional hydrological models. The study uses long time series of monthly data from 2008-2024 taking precipitation, evaporation, temperature, and relative humidity as input features from Balipatana block of Khordha district of Odisha, India. In this analysis, performance metrices like Root Mean Square Error (RMSE), Coefficient of determination (R²), Mean Absolute Error (MAE) and Willmott Index (WI) were employed. It is found that the value of RMSE, R², MAE, WI are 8.5832, 95.6756, 10.9438, 94.2981; 13.2287, 93.0073, 15.9084, 91.6327; 21.9627, 88.2165, 24.1689, 86.8491 in case of SVM-FFA, SVM-PSO and SVM respectively. The results demonstrates that the SVM-FFA models performs much better than the SVM-PSO and standalone methos in improving their accuracy and their robustness. With high prediction exactness and strategic versatility, the proposed model proved a powerful selection for forecasting groundwater levels.

How to cite: Biswakalyani, C., Samantaray, S., and Satapathy, D. P.: Prediction of Groundwater Level Using Hybrid SVM-FFA Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1501, https://doi.org/10.5194/egusphere-egu25-1501, 2025.