Effects of Climate Variables and Human Activities on Groundwater Level Fluctuations in Unconsolidated Sedimentary Aquifers: A Data-Driven Approach

Abstract

Groundwater level (GWL) in unconfined aquifers is highly susceptible to climate variables and human activities, exhibiting nonlinear fluctuations; these can further contribute to or exacerbate environmental hazards, such as land subsidence. Understanding the relationship between GWL changes and external conditions is essential for effective groundwater resource management and ecological protection. However, this relationship remains unclear and variable. This study systematically analyzes the correlations between climate and human factors and GWLs, using data from monitoring stations in the unconsolidated sedimentary aquifers of Beijing, China. It evaluates the importance of influencing factors on GWL simulation accuracy and tests how different inputs affect simulation performance. The results indicate that human factors are more strongly correlated with GWLs, yet climate factors hold higher importance scores. In GWL simulations, different input variables yield varying accuracy, with the inclusion of precipitation notably decreasing simulation precision because of its lagged or indirect effects on groundwater levels. The variation in accuracy across monitoring stations further suggests that the primary differences may stem from the GWL data itself. These findings underscore the need for high-resolution, localized data and tailored input selection to improve GWL projections and inform adaptive water-resource strategies under changing climatic and anthropogenic pressures.

1. Introduction

Groundwater plays a pivotal role in the global hydrological cycle, serving as a principal source of freshwater for a range of human activities, including drinking, irrigation, and industrial uses [1,2]. A reduction in groundwater level (GWL) can result in soil salinization, particularly in arid or semi-arid regions, which may have an impact on drinking or irrigation water [3,4]. Additionally, coastal areas are susceptible to seawater intrusion due to the lowering of GWLs, which represents a threat to freshwater resources [5,6,7]. Furthermore, the excessive exploitation of groundwater may also lead to ground subsidence, which in turn affects the stability of buildings and infrastructure [8,9]. It is therefore of great practical significance to be able to accurately model and predict changes in GWLs in order to manage regional water resources effectively and to prevent geological disasters.

Groundwater aquifers and hydrogeological conditions determine the characteristics of changes in GWLs [10,11]. Groundwater aquifers dominate the pathways of groundwater recharge and discharge [12,13]. An unconfined aquifer in groundwater exists above the first stable aquifer below the surface and is mainly recharged by precipitation and surface water infiltration and is susceptible to rapid responses to changes in moisture due to precipitation, evapotranspiration, and human activities, which can affect its water level fluctuations and water quality [14,15]. Hydrogeological settings—whether porous, fractured, or karstic—govern the hydraulic conductivity, storativity, and degree of hydraulic connectivity between the saturated zone and the vadose zone, thereby modulating the intensity of groundwater–surface exchange (recharge and discharge) [16]. Unconsolidated sedimentary aquifers, often characterized by high porosity and permeability, can facilitate rapid recharge but also exhibit a high vulnerability to contamination and surface impacts [17,18,19]. Therefore, the study of unconfined aquifers in areas of unconsolidated sedimentary aquifers is crucial.

The water tables of unconfined aquifers exhibit heightened sensitivity to climatic perturbations and anthropogenic pressures [20,21]. From a climate perspective, factors such as precipitation variability, temperature fluctuations, and prolonged droughts can directly influence groundwater recharge and evaporation rates, leading to fluctuations in the water table. Increased temperatures may also intensify evapotranspiration, further reducing GWLs [22]. From the perspective of anthropogenic activities, the extensive extraction of groundwater to meet the demands of agricultural irrigation, industrial operations, and urban expansion has been identified as a significant factor contributing to the rapid depletion of groundwater reserves and the subsequent decline of the water table [23]. Additionally, deforestation and urbanization alter the natural recharge patterns, further exacerbating the pressure on unconfined aquifers [24,25]. These combined effects underscore the need for effective groundwater management to mitigate potential impacts on water availability and ecosystem health.

Because traditional methods struggle to capture the nonlinear variations in groundwater levels, researchers have increasingly turned to machine learning techniques for improved modelling accuracy [26,27,28]. Physical-based models for groundwater simulation have been widely used to understand and predict groundwater dynamics [29,30,31,32]. While these models provide valuable insights, they often require extensive parameters and can be computationally intensive [33]. In recent years, data-driven models have emerged as a promising alternative for GWL simulation [34,35,36]. These models leverage large datasets and machine learning techniques to predict GWLs, offering a potentially more efficient and scalable approach. Among these models, the long short-term memory (LSTM) model stands out due to its ability to effectively handle time series data and capture the complex, nonlinear relationships between groundwater levels and various driving factors. The vanishing gradient problem occurs during the training of neural networks, particularly deep networks, where gradients become exceedingly small, making it difficult for the network to update the weights and learn effectively. This issue is addressed in LSTM models [37]. The LSTM model’s unique “gate” structure allows it to address the vanishing gradient problem that often plagues traditional recurrent neural networks, making it particularly well-suited for simulating the dynamic and nonlinear behavior of groundwater systems.

This study aims to investigate the effects of climate and human activities on GWLs through data-driven modeling approaches. Specifically, we analyze the correlation between key driving factors (climate factors and anthropogenic factors) and GWL changes, assess the relative importance of these factors in influencing groundwater dynamics, and investigate the effects of combining different factors on GWL prediction using LSTM models. We evaluated the effectiveness of these models in simulating GWLs and identified which combinations of input factors lead to the best model fit. By providing insights into the primary drivers of groundwater variability, this research contributes to improved prediction and management strategies, offering a scientific basis for sustainable groundwater resource management in the face of climate change and intensified human activities.

2. Materials and Methods

2.1. Datasets

Located in northern China, Beijing (39.4°~41.6° N, 115.7°~117.4° E) has experienced a rapid population growth that has led to the over-exploitation of groundwater due to limited water resources (Figure 1), results in various geological disasters in the plain region [38,39]. The annual precipitation is about 644 mm, while the precipitation exhibits pronounced orographic gradients: it is markedly higher along the windward piedmont slopes of the northeastern and southwestern ranges, significantly lower in the deep mountainous areas of the north and northwest, and intermediate across the plains. The city governs 16 districts with a total area of 16,410 km², urban and construction land amounts to approximately 3576 km², while agricultural land totals about 11,476 km², yielding a relative coverage ratio of urban to agricultural land of roughly 1:3.2. By the end of 2018, Beijing’s comprehensive forest coverage had attained 43.5%, of which 28.5% was concentrated in the plains. The corresponding forest stock volume was 17.98 million m³. Urban green coverage reached 48.44%, yielding a per capita provision of public green space of 16.3 m². The Beijing Plain comprises quaternary alluvial fans and plains with thicknesses ranging from tens of meters to over five-hundred meters. The aquifer system is complex with variable sediment thickness and lithology. From the alluvial fans to the plains, the sediment thickness increases and grain size decreases, aquifer systems change from a single gravel aquifer to multiple aquifer systems of sand layers separated by silt and clay layers. The water bearing layers within approximately 50 m of the surface constitute the shallow aquifer; those situated below this depth are designated as deep. The large part of the shallow aquifer is unconfined. It receives all the natural groundwater recharge. The majority of agricultural wells are located in the shallow aquifer. Some water supply well fields are installed in the shallow aquifer on the top of alluvial fans. The deep aquifers are confined, recharged chiefly by leakage from the shallow system, and contain the majority of industrial water supply wells along with some drinking water well fields [40]. In this study, five different degrees of aquifer enrichment (the discharge of a single well when the drawdown is 5 m, expressed in m³/d, is classified as follows: less than 500 m³/d is extremely low water richness (EL); 500 ~ 1500 m³/d is low water richness (L); 1500 ~ 3000 m³/d is moderate water richness (M); 3000 ~ 5000 m³/d is high water richness (H); greater than 5000 m³/d is extremely high water richness (EH)) within the unconsolidated sedimentary aquifers formations were randomly selected [41]. The data, which cover the period from 2018 to 2020 with a daily frequency, were sourced from the Groundwater Center of the China Geological Survey Institute.

For the climate variables that were closely linked to GWLs, we selected five driving factors: air pressure, temperature, precipitation, evaporation, and sunlight. Air pressure represents the atmospheric pressure at the surface; temperature refers to the surface temperature; precipitation measures the accumulation of liquid and frozen water that falls on the earth’s surface, including rain and snowmelt; evaporation measures the amount of water that evaporates from the surface and transpires from vegetation, with negative values indicating evaporation and positive values indicating condensation; sunlight measures the total amount of solar irradiance accumulated over a period and is converted into a standard test (irradiance 1000 w/m²). The climate variables data used were sourced from the National Oceanic and Atmospheric Administration (NOAA). The data for these five climatic variables are at a daily frequency. Additionally, five factors related to human activities were selected as input variables: supply water, domestic water consumption, environmental water consumption, industrial water consumption, and agricultural water consumption. Among these, “supply water” denotes the total volume of groundwater that is extracted, treated at water plants, and delivered to the urban distribution network, while “environmental water consumption” refers to groundwater recharged to rivers, lakes, wetlands, or other ecological targets for environmental purposes. The data for these five human activities factors are at a yearly frequency. The human activities data were obtained from the Beijing Water Resources Bulletin.

2.2. Methods

Pearson correlation analysis was used to measure the linear relationship between GWLs and 10 driving factors (including climate factors and anthropogenic factors) among 5 sites. The Pearson correlation coefficient is a correlation coefficient that measures linear correlation between two sets of data [42]. It is the ratio between the covariance of two variables and the product of their standard deviations (Equation (1)). It is essentially a normalized measurement of the covariance, such that the result always has a value between −1 and 1.

$${\rho }_{X,Y}={\displaystyle \frac{cov (X,Y)}{{\sigma }_X{\sigma }_Y}}$$

(1)

To evaluate the impact of climate and human activity on GWL simulation, we utilized Recursive Feature Elimination (RFE) to analyze the importance of the driving factors. RFE works by reducing the size of feature sets based on the weight of each factor [43,44]. For this particular study, we chose to use random forests as our base model due to their robust anti-interference and anti-overfitting capabilities. Based on the GWL datasets from 5 monitoring stations, the important scores of 10 driving factors were calculated 5 times to counteract the impact of the random initial weights of the fitted model on model accuracy.

To accurately assess the impact of various driving factors on GWL simulation, a long short-term memory (LSTM) model was utilized to combine 10 factors (5 climate and 5 anthropogenic variables) with data from 5 groundwater monitoring stations with different levels of aquifer richness. This resulted in 50 distinct datasets for simulation and prediction purposes. The LSTM model is a recurrent neural network that utilizes a “gate” structure to effectively address the challenge of exploding/vanishing gradient problems that can arise during training [45]. Our study employed an LSTM model with a 3 × 100 structure (number of layers × number of neurons), trained over 50 epochs. To mitigate the effects of randomness in neural network simulations, each of the 50 datasets was trained 5 times and averaged for more reliable results, while the dataset was temporally split with 85% used for training and the remaining 15% for validation; a dropout rate of 0.2 was applied after each LSTM and dense layer, and the Adam optimizer (learning rate 0.001, β₁ 0.9, β₂ 0.999) was adopted because its combination of momentum and adaptive learning rates ensures rapid convergence and robust performance in groundwater-level time series forecasting.

In order to better show the effect of model simulation, we used Taylor diagrams [46] with standard deviation (STD) (Equation (2)), correlation coefficient (r) (Equation (3)), and root mean square error (RMSE) (Equation (4)) to represent the performance of the LSTM combined with different driving factors. Standard deviation is an indicator that describes the degree of data dispersion, which represents the magnitude of change in GWL in our study. The correlation coefficient was used to measure the strength of the linear relationship between simulated and actual groundwater level. RMSE was used to measure the difference between simulated and actual groundwater level. These three indexes can measure the effect of LSTM simulation from three aspects (magnitude of change, correlation, and deviation), and shown in a single graph, each simulation result was normalized by dividing the standard difference of the observation data.

$$STD=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^n{(y_i-\overline{y})}^2}{n}}},$$

(2)

$$\mathrm{r}={\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{i}}-\overline{\mathrm{x}})({\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{y}})}{\sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{x}}_{\mathrm{i}}-\overline{\mathrm{x}})}^2{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{y}}_{\mathrm{i}}-\overline{\mathrm{y}})}^2}}}$$

(3)

$$RMSE=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^n{(x_i-y_i)}^2}{n}}}$$

(4)

3. Results

3.1. Correlation Analysis of Driving Factors

The absolute correlation between anthropogenic factors and GWLs was found to be higher than that observed between climate factors and GWLs (Figure 2, The specific values of the Pearson correlation analysis are presented in

Table A1. This is the results of the Pearson correlation analysis.

Driving Factors	Extremely High	High	Moderate	Low	Extremely Low
Air pressure	0.102	0.021	0.346	0.315	0.006
Temperature	−0.254	−0.115	−0.541	−0.264	−0.026
Precipitation	−0.093	−0.029	−0.179	−0.044	0.025
Evaporation	0.243	0.111	0.425	0.187	−0.118
Sunlight	0.068	0.001	−0.115	−0.322	−0.008
Supplying water	0.652	−0.015	0.401	−0.738	0.486
Domestic water	0.482	−0.057	0.324	−0.721	0.605
Environmental water	−0.691	−0.026	−0.401	0.626	−0.297
Industrial water	0.557	−0.043	0.361	−0.745	0.576
Agricultural water	0.688	0.009	0.409	−0.687	0.386

in Appendix A). The former ranged from −0.745 to 0.688, while the latter ranged from −0.541 to 0.425. Furthermore, no significant correlation was identified between the climatic factors (air pressure, temperature, precipitation, and sunlight) and GWLs in Group EL. In Group H, only two factors have been identified as exhibiting a statistically significant correlation with the GWL: temperature and evaporation. The findings suggest that, within the same hydrogeological type, the driving factors and GWLs are not necessarily correlated.

Among the climatic factors, there is a weak positive correlation between air pressure and GWL (in Group L, M and EH), with the value of 0.254 ± 0.102 (mean ± standard deviation). The correlation between temperature and GWL is predominantly negative, with the exception of Group EL, and the value is −0.294 ± 0.178. However, there is no significant correlation between precipitation and GWL which correlation coefficient is −0.136 ± 0.061. The correlation between evaporation and sunlight and the water table is also inconclusive, with both positive and negative correlations observed.

In the correlation calculation of anthropogenic factors and GWL, the results of Group L are diametrically opposed to those of the other groups. In Group L, only the environmental water factor exhibits a significant positive correlation with GWL, while the others demonstrate a significant negative correlation with GWL. In Groups EL, M, and EH, there were significant positive correlations between GWL and the following four water types: supplying water, domestic water, industrial water, and agricultural water. Conversely, GWL is significantly negatively correlated with environmental water.

3.2. Importance of Driving Factors

In general, the driving factors related to climate were considered to be of greater importance than those related to human activity (Figure 3). This discrepancy may be attributed to the quality of the input data, as there was comparatively little change in the variables related to human activity in comparison to those related to climate. With regard to the climate factors, precipitation was assigned the lowest importance score (0.657 ± 0.240), while the remaining variables were ranked within the top four. The importance rating for sunlight and temperature in the GWL simulations was 2.341 ± 0.614 (mean ± standard deviation) and 2.000 ± 0.390, respectively. This is likely due to the fact that they are directly related to evaporation and plant transpiration, which play a key role in the groundwater recharge and depletion process. Furthermore, evaporation and air pressure were also identified as significant factors, with ratings of 1.673 ± 0.292 and 1.259 ± 0.492, respectively. These findings reinforce the crucial role of climatic variables in influencing GWLs.

Despite the relatively low and similar ratings of anthropogenic factors, they do exert some influence across different groups. The ratings of anthropogenic factors in group H are relatively high in comparison to the other groups. This may be attributed to the fact that the monitoring stations were selected in areas with high anthropogenic impacts, where human dependence on groundwater is higher. In group EH, the importance ratings of anthropogenic factors other than environmental water were low, which may indicate that groundwater extraction is limited in this area.

3.3. GWL Simulation Under Different Driving Factors

The evaluation of groundwater simulation results for various input variables was presented using the Taylor diagram (Figure 4). Our models were named after the driving factors, and we grouped GWL simulation assessment based on different levels of water richness. The simulation results within each group were found to be relatively concentrated. Notably, the results from EL and EH groups showed a higher correlation with the observed data, with correlation values close to 1 and low RMSE values. However, the results from L and M groups were comparatively poorer.

The findings indicate high correlations between simulations and observation in GWL, with r values ranging from 0.56 to 0.99. Air pressure was the most fitting model in two groups (EL, EH), with high r values of 0.98 and 0.99. Among the remaining three groups, sunlight, industrial water, and evaporation had the highest fitting coefficients, with 0.82 of group L, 0.80 of group M, and 0.93 of group H, respectively. For all the groups, at least three of the top five relevant models were anthropogenic factors. This suggests that human activities play a beneficial role in the simulation of GWL. However, regardless of the group, the precipitation simulation results had the least correlation, with r values ranging from 0.56 to 0.96.

The RMS measures the disparity between simulated and observed GWL and ranges from 0.07 m to 0.20 m. Among the EL group, domestic water has the lowest RMSE value, while three anthropogenic factors also have lower RMSE values within this group. For the L group, sunlight has the lowest RMSE value. Meanwhile, industrial water, evaporation, and air pressure have the lowest RMSE values in the M, H, and EH groups, respectively. In these three groups, anthropogenic factors have lower RMSE values, ranking as the top six. Precipitation has the highest RMSE value in each group, ranging from 0.11 m to 0.20 m. The exceptional results of the anthropogenic driving factor in both correlation coefficient and RMSE highlight the importance of adding anthropogenic variables in GWL simulation.

The STD reflects the range of variation within GWL data. A higher STD indicates large fluctuations, while a lower STD suggests relatively stable fluctuations. In our comparison, we utilized the STD of observed GWL within each group as the comparison object. If the STD of the simulated value exceeds that of the observed value in the group, the simulation does not accurately reflect the actual GWL mutation. Conversely, if the STD is lower, the simulation may introduce noise, leading to inaccurate results. Our findings reveal that group L and group M have the highest STD of observed values, suggesting the model poorly simulated sudden changes. In group EH, the STD of observed values is the lowest in the group, and the highest value belongs to precipitation (with STD of 0.48 m). In the EL group, precipitation also has the highest STD (0.40 m), while temperature has the lowest (0.36 m).

4. Discussion

4.1. Driving Factors Affecting Groundwater Level

The importance score places considerable weight on climatic factors, reflecting the long-term and far-reaching impact of climate change on groundwater recharge and depletion patterns. To illustrate, air pressure directly affects the pressure and flow state of groundwater system, and air pressure fluctuations can have a significant effect on GWL [47]. Precipitation represents a significant source of groundwater recharge, with the potential to exert an influence on GWLs in the short, medium, and long term [48]. However, the findings of this study indicate that the incorporation of precipitation has an adverse impact on the modelling of GWLs. This is likely due to the presence of impermeable surfaces in the vicinity of the monitoring sites and the correlation between precipitation and GWLs, which subsequently influences the observed response. Evaporation can be defined as the transfer of water from soil, water bodies, and vegetation surfaces to the atmosphere. This process results in the absorption of water from the soil, which in turn recharges soil water. A reduction in groundwater recharge will lead to a corresponding reduction in the water table, a phenomenon that is particularly evident in arid regions [49]. The influence of sunlight and surface temperatures on soil temperatures and evaporation rates is a significant factor in the regulation of GWLs. Therefore, under the combined influence of global warming and intrinsic climate cycles, long-term GWL projections must explicitly incorporate both the periodic and secular components of climatic forcings to accurately resolve the underlying multi-scale variability [50].

Unlike the effects of climate, human activities impart a more spatially heterogeneous influence on GWLs. To illustrate, factors such as industrial water use, agricultural irrigation, and residential water use are subject to rapid change and are susceptible to significant fluctuations in GWLs in the short term [51]. Therefore, although the impact of anthropogenic factors is low in the overall score, they can significantly disturb the dynamic balance of groundwater due to their short-term and controllable nature [52]. In practical applications, the model should be adapted to accommodate the short-term fluctuations in water levels induced by anthropogenic activities, which can be regarded as disturbances, in order to optimize the responsiveness of the prediction model.

The superimposed effects between climate and human activities are manifested in complex, nonlinear relationships in GWL changes. In arid or semi-arid regions, where agricultural water demand is on the rise and groundwater is extensively utilized for agricultural irrigation, this heightened demand serves as an amplifier in changes in GWLs, potentially leading to a scenario where agricultural water usage may outweigh climatic factors [53]. Conversely, in regions characterized by higher precipitation levels and sufficient groundwater recharge, agricultural water demand is diminished, leading to a reduced reliance on groundwater for human activities [54]. Consequently, the impact of climatic factors can be more significant. The complex superposition effect necessitates the incorporation of the interaction between climate and human activities into GWL change models in order to achieve more accurate predictions in practical applications. This interaction analysis can elucidate the sensitivity of human activities to the groundwater system under different climatic conditions, thereby providing a scientific basis for effective groundwater management.

Furthermore, data quality, spatial resolution and temporal resolution all have an impact on the importance score. Climatic factors (temperature, precipitation, evapotranspiration, etc.) can typically be obtained with high temporal resolution through long-term monitoring and satellite remote sensing. Consequently, the model is able to capture seasonal and interannual variations of climatic factors during the simulation process. In contrast, anthropogenic variables (e.g., agricultural water use, industrial water use, and domestic water use) typically lack data records with high temporal resolution and exhibit less variability in the data, which may be considered on an annual or longer time scale. Consequently, models may be unable to accurately represent the immediate effects of human activities on GWLs, resulting in relatively low significance scores. Enhancements in the precision and temporal resolution of anthropogenic data could facilitate a more comprehensive representation of their impact on GWLs.

4.2. Groundwater Level Simulation Based on Data Modelling

Firstly, the performance of single-input models generally outperformed multi-input models, suggesting that data constraints may play a significant role. When numerous variables are introduced, each requires a substantial dataset to provide stable, high-quality input. Insufficient data coverage or quality in certain features may reduce the model’s ability to capture complex relationships, leading to diminished performance when multi-input approaches are employed. Thus, single-input models may yield more accurate predictions because they rely on more robust, focused data.

Secondly, the simulation accuracy varied across monitoring stations despite using a consistent training process. This inconsistency may be attributed to localized hydrological conditions that influence groundwater dynamics, such as unique geological formations or micro-climatic patterns not fully captured by a general model. Additionally, differences in data quality across stations may introduce variability in the accuracy of predictions, indicating a need for location-specific model optimization. An additional potential explanation is the discrepancy between the spatial distribution of the impact factor data and that of the water table data.

Interestingly, the inclusion of precipitation as an input factor sometimes led to poorer simulation outcomes, due to its lagged or indirect effects on GWLs (Figure A1). The process of precipitation input to groundwater is complex and can be influenced by local factors such as soil permeability and vegetation cover, which may not be fully captured by the current dataset. Future work should consider incorporating additional variables and conducting careful calibration to better account for these complex interactions.

5. Conclusions

The present study delved into the impacts of climate variables and anthropogenic activities on groundwater levels and their implications for groundwater modeling using a data-driven approach. The correlation analysis revealed a pronounced connection between human activities and groundwater levels, with Pearson correlation coefficients ranging from −0.541 to 0.425 for anthropogenic factors, compared to −0.745 to 0.688 for climate factors. However, the importance scores indicated that climate variables exerted a more significant influence on groundwater level simulation. The ranking of factor weights is as follows: sunlight > temperature > evaporation > air pressure > environmental water > supplying water > agricultural water > domestic water > precipitation > industrial water.

The simulation of groundwater levels was found to be influenced by a multitude of input variables, each with a distinct effect. Single-input models generally outperformed multi-input models, suggesting that data constraints play a significant role. The simulation accuracy varied across monitoring stations, which may be attributed to localized hydrological conditions and differences in data quality. The inclusion of precipitation as an input factor sometimes led to poorer simulation outcomes, due to its lagged or indirect effects on groundwater levels.

This study underscores the need for tailored data inputs and localized approaches to enhance groundwater level predictions. Future research could further explore the synergistic effects of climate and human activities on groundwater levels. Utilizing higher-resolution meteorological and anthropogenic data may elevate the precision of simulations. Additionally, integrating different machine learning approaches, such as ensemble learning or hybrid models, holds promise for optimizing predictions of groundwater levels. Such an approach would better account for the uncertainties stemming from climate and anthropogenic activities, thereby providing a more robust scientific foundation for groundwater management and improving water protection strategies amidst climate change.