Pakistan's groundwater resources are vital to the country's water supply, yet increasingly threatened by issues such as over-extraction, inadequate management practices, and insufficient conservation regulations. This study was conducted to examine spatiotemporal aquifer behavior, fluctuations in drawdown levels, and water quality parameters like pH, Electrical Conductivity (EC), Total Dissolved Salts (TDS), Calcium, Magnesium, Total Hardness (TH), Bicarbonates and Chlorides by using geospatial techniques to address sustainable groundwater resource management needs. For future forecasting four machine learning (ML) models were used; Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF). Observed data were obtained from Water and Sanitation Agency (WASA) Faisalabad from year 2013 to 2023 which included 29 inline field area well stations and 25 Japan International Cooperation Agency (JICA) well stations, and weather data from the Terra Climate dataset. Groundwater drawdown patterns and quality changes over time were analyzed by GIS-based spatial analysis by utilizing historical data to train and test predictive models for 2024-2028. The XGBoost model demonstrated exceptional performance in predicting drawdown pre-monsoon (8.35m) and post-monsoon (7.65m) until 2028 and hydro chemical quality, with an average R-squared value of 0.86, RMSE below 0.08, and MAE under 0.05 for both. The study's spatial analysis revealed significant seasonal variations, with post-monsoon increases in mineral concentrations due to intensified leaching processes and identified a concerning rise in chloride levels after 2022, linked to anthropogenic activities. These findings underscored the importance of advanced machine learning techniques, particularly XGBoost, in accurately forecasting groundwater dynamics and hydro chemical quality.
Research Article
Spatiotemporal Assessment and Machine Learning-Based Future Forecasting of Groundwater Hydro chemical Dynamics and Drawdown Variability
https://doi.org/10.21203/rs.3.rs-4951035/v1
This work is licensed under a CC BY 4.0 License
You are reading this latest preprint version
water quality
drawdown
agriculture
geographical information system (GIS)
groundwater
machine learning (ML)
Water resources are really important for food production and a productive environment for the existence of all living organisms (Kılıç 2020). The water resources have been under stress over the past forty years due to population growth, urbanization, land reclamation, and agricultural practices (Abdalla and Moubark 2018). By 2025, 5 billion people from the 8 billion world population will be living in water-stress-bearing countries (Arnell 1999). In many locations of the world, freshwater supplies are insufficient to fulfill household, economic, and environmental demands (Cosgrove and Loucks 2015). On a global scale, Pakistan is the third largest consumer of groundwater. Only 27 percent of the land needs irrigation from surface water supplies; the other 73 percent is either directly or indirectly watered by groundwater. The province of Punjab utilizes about 90% of all groundwater abstraction (Qureshi 2020).
Food production in Pakistan is strongly reliant on irrigation, requiring large amounts of water from already depleted aquifers (Kirby et al., 2017). Because of the high population density and the intense partial irrigation of agricultural output, there is a strong probability that future resource conflicts over groundwater are going to arise (Kreins et al. 2015). The intensive use of groundwater has enabled a significant rise in agricultural productivity and has assisted in meeting the water demands of urban areas; nevertheless, management is necessary to handle significant overexploitation and groundwater quality challenges (Van Steenbergen and Oliemans 2002).Pakistan's agricultural growth depends heavily on groundwater due to the country's low rainfall and surface water supplies (Awais et al. 2017).
Chiniot, located in the Punjab province of Pakistan, is an important agricultural center known for its rich soil and favorable climate. the lack of water in the Chenab River, the majority of farmers use well water for irrigation, which puts a great deal of demand on groundwater resources (Ahmad et al. 2022). The District Chiniot has a total population of 1.36 million as of the 2017 census, with 0.947 million living in rural regions and 0.422 million in urban areas (Ahmad et al. 2020) Because most residential and agricultural water consumption in Chiniot, Faisalabad, depends on groundwater, there is significant worry over groundwater exploitation. By incorporating spatial mapping into groundwater management techniques, stakeholders can achieve the objective of optimizing agricultural output while preserving the long-term sustainability of groundwater supplies (Shakoor et al. 2015). Proper water resources management is only possible by monitoring it properly according to space and time. Now, with increasing population there is also need to forecast these trends for effective future monitoring. The integration of cutting-edge technology, strong regulations, and community involvement is necessary to handle the complex issues of groundwater depletion and ensure food security in the face of population expansion (Bhatti et al. 2017). The organization, storage, and presentation of spatial data, which is usually required for the management of water resources, can be done with the use of Geographic Information Systems (Kumar et al. 2015).
Assessment and forecasting of surface and groundwater water quality are necessary for effective management of water bodies in order to take the necessary steps to ensure that pollution levels remain within permissible limitations. Artificial Intelligence produces a mathematical framework that is adaptable and capable of recognizing complex and non-linear correlations between input and output data (Ahmed et al. 2019). Models based on artificial intelligence require a lot less data (Amiri et al. 2023). Farmers typically find it costly and time-consuming to assess the irrigation water quality using conventional methods, especially in poor nations. Nevertheless, by utilizing physical factors as features in the forecasting and evaluation of aquifer systems' irrigation water quality indexes, artificial intelligence models can be applied to overcome this problem (El Bilali et al. 2021).
This study used GIS and Kriging interpolation to analyze spatiotemporal changes in groundwater drawdown and quality from 2013 to 2023. Advanced machine learning models were used to forecast groundwater depletion and quality reduction over the next five years. These models used rainfall patterns and data to accurately estimate trends in groundwater drawdown and quality patterns before and after the monsoon. The findings highlighted significant spatial variability in groundwater trends, emphasizing the critical impact of seasonal fluctuations, particularly monsoon periods, on both the quantity and quality of groundwater resources.
Figure 1 flow chart presents a step-by-step procedure for the spatiotemporal investigation of hydro chemical quality and decline by employing the machine learning techniques. The first step is data collection, which involves gathering historical information on water quality indicators and drawdown from monitoring sites in order to extract minimum and maximum values. After that, this data is pre-processed using georeferencing, standardization, normalization, and indication of seasonal fluctuation. Creating GIS spatial maps, applying kriging interpolation techniques, and utilizing time series plots to spot trends and anomalies are all part of the spatiotemporal analysis. Different machine learning techniques, including linear regression, Random Forest, SVM, KNN, and XGBoost, are used in the selection and construction of models. To train and validate the model, the data is divided into 80% as training and 20% as validation, and metrics like MAE, RMSE, and R-squared are used to assess the data. Lastly, correlation analysis, interactive visualizations, and tracking the effects over time are the main objectives of model deployment. Model evaluation parameters guarantee the efficacy and precision of the deployed models, encapsulating the entire process.
2.1 Study area
Chiniot, a key agricultural center in Pakistan, benefits from Chenab River irrigation systems, enhancing food security and economic stability. Groundwater irrigation is a sustainable solution, protecting indigenous habitats and providing drinkable water to homes and businesses. A Japan International Cooperation Agency (JICA) project assessed groundwater resources across 648.01 square kilometers, using 25 strategically placed wells and 29 in the Inline Field Area. The wells were strategically placed between 73.070141 and 72.948615 degrees east and 31.623242 to 31.550918 degrees north. In addition, the research comprised 29 wells in the Inline Field Area with longitude coordinates from 72.954420 to 72.976381 degrees east and latitude from 31.603797 to 31.605238 degrees north. The research aims to promote sustainable water resource management and development in Chiniot's geographical location.
2.2 Data acquisition
The Water and Sanitation Agency (WASA) in Faisalabad, Pakistan, provided data for 54 wells at Chiniot sites to monitor groundwater drawdown, to conduct water quality research, and predict the future using various factors. This historical data of rainfall during monsoon season was gathered from terra climate dataset (Abatzoglou et al., 2018).
2.3 Geographic Information System (GIS) usage
GIS is a structured system for collecting, organizing, and examining data based on geography. It integrates spatial data, including object positions, with attribute data, providing information about their characteristics. GIS can visually represent changes in places over time. Predictive analysis uses GIS to analyze past spatiotemporal data to forecast future events. It helps understand pollution issues, evaluate them, and identify appropriate remedies and alternatives. GIS also aids in identifying pollution-related solutions and addressing pollution-related problems (Chabuk et al., 2020).
Remote sensing is essential for water resource mapping (Wang and Xie, 2018). GIS may be utilized for resource management and planning, allowing for the efficient allocation and optimization of resources across different locations and time periods. This includes activities such as monitoring water supplies, urban planning, and wildlife management, where the ability to track changes over time is essential for making informed decisions. GIS can effectively monitor environmental changes such as water quality spatial analysis and drawdown fluctuation by evaluating data from various time periods and places.
In general, utilizing Geographic Information Systems (GIS) for spatiotemporal data enables researchers, planners, and decision-makers to visually perceive, comprehend, and evaluate the changes and patterns in geographic areas throughout time, resulting in improved planning and management outcomes.
2.4 Interpolation technique
Geographic Information Systems (GIS) employed the potent and complex interpolation method known as kriging to forecast spatially continuous data from a collection of discrete points. It was predicated on statistical models that took into consideration spatial autocorrelation, which is the idea that data at sites closer together tend to be more similar than at farther distances. The study involved collecting data from geographically dispersed sample locations for elevation, precipitation, pollution levels, and water table quality. Clean and correctly georeferenced data was crucial for accuracy. Variograms were used to simulate spatial correlation reduction with distance. The empirical variogram was fitted with a theoretical model to enhance interpolation performance in unsampled regions. Methods like Ordinary Kriging was chosen based on data characteristics, trends, and multiple variables. Key parameters like search radius and number of points were determined. GIS software used the variogram model to balance input points and estimate values in unsampled locations, creating a new layer for comprehensive spatial analysis.
2.5 Random Forest (RF) model
Random Forest (RF) is a machine learning technique that creates multiple decision trees during training and determines the final class by selecting the mode of the classes or average prediction from the trees. It has shown benefits in water quality and drawdown, handling non-linear data without feature elimination and exhibiting strong resistance to overfitting. RF is popular in water resources applications due to its ability to capture complex factors in water quality assessment (Tyralis et al., 2019).
2.6 Support Vector Machine (SVM) model
Support Vector Machine (SVM) is a classifier that divides data into groups using the kernel technique, allowing it to handle non-linear data. It is effective in analyzing high-dimensional data and distinguishing between data points. SVM is particularly useful in water quality investigations with a high number of factors compared to observations. The model predicts groundwater quality characteristics using observable variables like water temperature, electrical conductivity, depth, dissolved solids, pH, land use, and season (Arabgol et al., 2016).
2.7 eXtreme Gradient Boosting (XGBoost) model
The XGBoost is an efficient implementation of gradient boosted trees designed for speed and performance. It has been successful in data science contests and offers significant benefits in water quality and drawdown. The model is flexible, can be adjusted for optimal performance, and can handle incomplete data. It is well-suited for complex situations like forecasting drawdown and serves as a standard for predicting future groundwater levels (Osman et al., 2021).
2.8 K-Nearest Neighbors (KNN) model
The KNN algorithm, a simple learning method, delays computation until the classification stage, utilizing the similarity between training instances and test cases to produce predictions. This method is particularly beneficial for improving water quality and reducing drawdown in situations involving similar situations, considering local variables and being easy to execute and understand. The output of the KNN model is dependent on the operation that is to be carried out; also, for the purpose of regression (Xu et al., 2021).
2.9 Model performance evaluation criteria
According to a number of research, the mean square error (MSE) and its rooted variation (RMSE) as well as the mean absolute error (MAE) were utilized (Chicco et al., 2021). RMSE, which stands for root mean square error, and MAE, which stands for mean absolute error, are both techniques that are frequently utilized in model evaluation studies (Chai and Draxler, 2014). For the purpose of assessing the effectiveness of the machine learning models, the following three measures were utilized:
The R-squared statistic, often known as R2, is a statistical metric that is used to determine how successfully the model mirrored the observed outcomes. This measurement is based on the proportion of the total variance in outcomes that is explained by the model. With regard to determining the accuracy of the forecast, it proved to be quite helpful. RMSE was implemented to offer a measurement of the typical size of the errors that were produced by the model. A clear picture of the accuracy of the model was provided by the calculation of the square root of the average squared discrepancies between the values that were predicted and those that were actually observed. The Mean Absolute Error (MAE) is a statistical metric that determined the average size of mistakes in a set of predictions without taking into account the direction of those errors. It was the average of the absolute differences between the forecast and the actual observation across the test sample, with each individual difference being given the same amount of weight.
The hydro-chemical parameters that were measured before and after the monsoon are shown (Table 2). For every parameter, the mean, standard deviation, lowest value, median (50%), and maximum value are given. While pre-monsoon values of Electrical Conductivity (EC) and Total Dissolved Solids (TDS) were greater, post-monsoon pH levels averaged slightly higher. Following monsoon, concentrations of calcium and magnesium as well as levels of bicarbonate and total hardness rose. Following the monsoon, the concentration of chloride increased somewhat. Prior to the monsoon, drawdown numbers were higher, indicating greater water extraction. The season-long average of 275.76 mm with a sizable standard deviation for rainfall data illustrates the unpredictability in precipitation (Table 2).
|
Hydro Parameter |
Monsoon Season |
Mean Deviation |
Standard Deviation |
Min. Value |
50% |
Max. Value |
|---|---|---|---|---|---|---|
|
pH |
Post |
7.34 |
0.08 |
7.16 |
7.355 |
7.48 |
|
Pre |
7.29 |
0.10 |
7.02 |
7.29 |
7.49 |
|
|
EC |
Post |
410.63 |
24.82 |
349.68 |
416.07 |
440.91 |
|
Pre |
444.84 |
29.72 |
390.28 |
441.42 |
496.61 |
|
|
TDS |
Post |
228.25 |
33.71 |
195.52 |
220.43 |
326.51 |
|
Pre |
234.70 |
19.59 |
196.57 |
234.33 |
283.12 |
|
|
Calcium |
Post |
57.39 |
9.32 |
40.75 |
58.10 |
79.94 |
|
Pre |
53.88 |
6.27 |
44.84 |
53.11 |
67.55 |
|
|
Magnesium |
Post |
21.74 |
3.31 |
14.16 |
21.72 |
28.71 |
|
Pre |
18.69 |
2.15 |
13.28 |
18.645 |
22.47 |
|
|
Total Hardness |
Post |
204.83 |
7.06 |
192.95 |
204.635 |
222.97 |
|
Pre |
182.70 |
17.00 |
139.73 |
182.13 |
211.38 |
|
|
Bicarbonate |
Post |
197.56 |
17.29 |
167.25 |
196.12 |
231.5 |
|
Pre |
186.99 |
12.50 |
166.9 |
185.48 |
222.68 |
|
|
Chloride |
Post |
43.93 |
5.21 |
32.26 |
44.28 |
52.02 |
|
Pre |
41.78 |
6.06 |
33.48 |
41.06 |
53.87 |
|
|
Drawdown |
Post |
7.88 |
0.34 |
7.4 |
7.79 |
8.52 |
|
Pre |
8.35 |
0.15 |
8.04 |
8.3 |
8.65 |
|
|
Rainfall |
Complete |
275.76 |
58.23 |
188.8 |
262.88 |
405.2 |
3.1 Heatmap analysis
Because pH and EC (Electrical Conductivity) are positively correlated, as pH rises, EC rises also (Fig. 3). Since calcium is a primary contributor in water TDS, TDS and calcium have a positive correlation. From (Fig. 3) it can be observed magnesium contributes significantly to water hardness, hence magnesium and total hardness are highly correlated. Chlorides and drawdown are negatively associated, implying that higher chloride concentrations reduce drawdown. The heatmap quickly identified linkages between water quality metrics and their possible effects on groundwater depletion. Analyzing groundwater depletion can help manage water resources and predict water quality.
3.2 Hydro chemical indicators spatial variability pre and post monsoon (2013)
The maps that are provided show the pre-monsoon water quality for eight metrics within the research region. pH, Total Dissolved Solids (TDS), Calcium, Magnesium, Bicarbonate, Chloride, Electrical Conductivity (EC), and Total Hardness are among the parameters. Color gradients were utilized in each map to depict the concentration ranges of each parameter within the study area. The pH range of 7.04 to 7.54 indicated conditions that were neutral to slightly acidic (Fig. 4a). EC showed regions with high salinity levels up to 936.08 µS/cm, however it varied greatly (Fig. 4b). TDS concentrations ranged from 441.45 mg/l to other values (Fig. 4c). Regional differences were evident in the calcium and magnesium maps, with concentrations reaching up to 159.72 mg/l and 35.94 mg/l, respectively (Fig. 4d) and (Fig. 4e). The concentrations of chloride varied from 9.20 to 98.00 mg/l, whereas the amounts of bicarbonate ranged from 112.30 to 390.761 mg/l (Fig. 4f). Magnesium and calcium were included in the total hardness, which varied from 102.95 to 324.55 mg/l (Fig. 4e) and (Fig. 4d).
The post-monsoon water quality maps use color gradients to show different concentrations of eight important metrics throughout a study area. The range of pH values is from 7.25 to 7.83 (Fig. 5a) denotes a transition from slightly acidic to neutral conditions, implying that monsoon runoff and dilution have an impact on stable water chemistry. Higher values of Electrical Conductivity (EC) indicate greater ionic concentration from the leaching of minerals and salts. EC values range from 181.22 to 909.15 µS/cm (Fig. 5b). Concentrations of Total Dissolved Solids (TDS) range from 75.15 to 512.18 mg/l, which indicates a higher mineral composition as a result of surface runoff (Fig. 5c). Magnesium levels range from 8.83 to 22.16 mg/l, showing differing weathering rates. In contrast, calcium levels range from 24.48 to 193.16 mg/l, most likely due to the dissolution of calcium-bearing minerals (Fig. 5e) and (Fig. 5d). The range of bicarbonate concentrations (82.81 to 381.01 mg/l) indicates enhanced organic matter decomposition and chemical weathering (Fig. 4f). The range of chloride values, which are related to anthropogenic sources and soil leaching, is 17.70 to 89.93 mg/l. The range of 136.74 to 348.69 mg/l for total hardness indicated a considerable leaching of the minerals magnesium and calcium. (Fig. 4g) and (Fig. 4h).
3.3 Hydro chemical indicators spatial variability pre and post monsoon (2023)
The research area's pH values vary from 7.09 to 7.41 (Fig. 6a), with higher pH values indicated by deeper hues. With values ranging from 215.44 µS/cm to 828.38 µS/cm (Fig. 6b), EC varied greatly and indicated greater concentrations in some areas. Regions with values ranging from 107.83 mg/l to 417.17 mg/l (Fig. 6c). The geographical diversity of calcium, magnesium, bicarbonate, chloride, and total hardness highlights locations with varying concentration levels. These maps are essential for comprehending the distribution of pre-monsoon water quality, which aids in determining how the monsoon affects these factors.
There is a modest variance from the pre-monsoon levels to the post-monsoon pH levels, ranging from 6.97 to 7.39 (Fig. 7a)., with areas showing somewhat lower pH values. The current range of EC values, which show an increase in some locations compared to pre-monsoon, is 201.44 µS/cm to 1,059.10 µS/cm (Fig. 7b). This suggests that monsoon rains may have caused mineral leaching into the water. The range of TDS values has increased from 101.16 mg/l to 527.99 mg/l (Fig. 7c), suggesting greater concentrations in certain regions during the monsoon. The range of calcium levels has changed from 36.64 mg/l to 57.95 mg/l, with considerable increases in some areas and the current range of magnesium concentrations is 5.40 mg/l to 59.08 mg/l, with a rise in higher concentration zones (Fig. 7d) and (Fig. 7e). The range of bicarbonate levels is 91.62 mg/l to 337.61 mg/l (Fig. 7f), with notable elevations in specific regions. The amounts of chloride have also gone up, from 20.28 mg/l to 121.81 mg/l and beyond monsoon, total hardness, which ranges from 139.57 mg/l to 416.16 mg/l, has a more widely distributed range of greater concentrations. (Fig. 7g) and (Fig. 4h).
3.4 Drawdown spatial variability pre and post monsoon (2013–2023)
Four maps that show the groundwater drawdown levels in a study region under various conditions and time periods are included in the (Fig. 8). For the years 2013 and 2023, the pre-monsoon and post-monsoon regions are represented on each map, respectively. The drawdown levels in 2013 before the monsoon season are depicted in (Fig. 8a), where the decline ranges from 5.77 to 10.08 meters. It displays different groundwater levels in different locations (5.77–7.26 meters) and higher levels of drawdown (8.30–10.08 meters) in different regions. This shows varying groundwater levels in 2013 prior to the monsoon season. The drawdown levels following the 2013 monsoon season are shown on (Fig. 8b), which ranges from 5.95 to 9.71 meters. Following the monsoon, different locations suffer different amounts of decline. In certain places, there is less drawdown (5.95–7.31 meters), perhaps because of groundwater recharge from the monsoon rains, while in other parts, there is still more drawdown (8.35–9.71 meters). (Fig. 8c), shows the drawdown levels in 2023 before to the monsoon season, spanning a wider range from 3.28 to 13.35 meters. Groundwater levels are more variable now than they were in 2013, with some regions seeing much less decline (3.28–6.95 meters) and others seeing substantially more drawdown (9.36–13.35 meters). This may suggest that groundwater levels have been more variable during the past ten years.
3.5 Temporal variability of water quality and drawdown pre and post monsoon (2013–2028)
An examination of water quality measures from 2013 to 2028 is shown in the (Fig. 9), which highlights significant trends and changes. Electrical conductivity (EC) varied a lot; it was around 500 µS/cm in 2013, dropped to 400 µS/cm in 2021 (Fig. 9a), and then increased once again. Total Dissolved Solids (TDS) showed a similar trend, peaking at roughly 450 mg/L in 2013 and falling to 350 mg/L in 2021 before rising again (Fig. 9a). These variations point to possible sources of contamination as well as shifts in ion concentration. With very slight fluctuations, total hardness was comparatively constant at 250 mg/L, suggesting steady levels of calcium and magnesium ions two essential elements for preserving the quality of water. While pH measurements continuously stayed around 7.5 (Fig. 9a), suggesting steady acidity/basicity, bicarbonate levels, which contribute to alkalinity, remained stable at 150 mg/L. Magnesium levels were constant at around 20 mg/L, whereas calcium levels fluctuated slightly at 60 mg/L (Fig. 9a). The amount of chloride in the water dropped from around 350 mg/L in 2013 to about 300 mg/L in 2021 (Fig. 9a), and then it stabilized, indicating either a shift in pollution levels or better water management techniques. These patterns emphasize how crucial it is to preserve water quality and deal with possible sources of pollution through ongoing monitoring and efficient management. Many of the services that are provided by the ecosystems of the world are essential to our day-to-day lives. Groundwater ecosystems provide a number of services provide that are of tremendous societal and economic importance. These services include the purification of water and the storage of water in superior quality for decades and millennia (Griebler and Avramov, 2015).
The graph shows how several water quality measures vary over time during the post-monsoon period, from 2013 to 2028. Critical indicators of water quality include pH, Calcium, Magnesium, Chloride, Total Dissolved Solids (TDS), Total Hardness, Bicarbonate, and Electrical Conductivity (EC). EC values are not constant; they typically increase to about 400 µS/cm by 2028 after declining from 450 µS/cm in 2013 to around 350 µS/cm in 2020 (Fig. 9b). High EC may be a sign of salinity, which can affect soil health and agricultural productivity. TDS levels have an impact on the flavor and acceptability of the water for drinking and irrigation. They decreased dramatically from around 300 mg/L in 2013 to approximately 200 mg/L in 2016 (Fig. 9b), then increased roughly to 250 mg/L by 2028. Total hardness has a slight fluctuating influence on soap efficacy and scale development, falling from around 350 mg/L Ca in 2013 to approximately 250 mg/L Ca by 2016 (Fig. 9b). Between 180 and 220 mg/L, bicarbonate remains stable and is essential for pH stability and buffering capabilities. The pH scale, which is necessary for aquatic life and metabolic activities, increases slightly from 7.0 in 2013 to 7.5 by 2028 (Fig. 9b). Magnesium levels vary from 20 to 25 mg/L, whereas calcium levels range from 20 to 30 mg/L, which contributes to water hardness. Between 2013 and 2016, chloride levels drop from about 300 mg/L (Fig. 9b) to about 200 mg/L, stabilizing after that. Increasing chloride levels seen after 2022 indicated a heightened degree of human activity. Another factor that contributed to the regulation of the hydro geochemistry of groundwater in this region is the agricultural activities that take place during the monsoon seasons (Natesan et al., 2022).
Pre- and post-monsoon groundwater drawdown analyses are shown on the graph, along with yearly rainfall data from 2013 to 2028. Before the monsoon, the drawdown varied from about 8.1 to 8.7 meters (Fig. 9c),. Pre-monsoon drawdown is represented by the regression equation
which shows an annual decrease of around 0.0095 meters (Fig. 9c),. This shows that groundwater levels have been gradually declining over time prior to the monsoon season. On the other hand, the post-monsoon decline was between 7.6 and 7.8 meters (Fig. 9c),. The post-monsoon drawdown regression equation is
demonstrating a very steady groundwater level following the monsoon with a minor yearly reduction of 0.00001 meters. The yearly rainfall data did not exhibit a discernible pattern throughout the time, with variations ranging from 200 to 400 millimeters. Regression analysis showed that the pre-monsoon drawdown significantly decreased while the post-monsoon drawdown remained almost constant. This implies that recharge from monsoon rains does not sufficiently offset the pre-monsoon groundwater depletion.
XGBoost was consistently the best-performing model, according to a comparative investigation of predictive models for drawdown and groundwater quality.After analyzing the results of statistical indicators (MAE, RMSE AND R2) shows that XGBoost consistently produced the highest R² values for groundwater quality indicators, including pH, Electrical Conductivity (EC), Total Dissolved Solids (TDS), Calcium, Magnesium, Total Hardness, Bicarbonate, and Chloride. These results indicate XGBoost's superior capacity to explain the variance in the data. Furthermore, XGBoost demonstrated the lowest RMSE and MAE values among these metrics, indicating its great prediction accuracy and precision. For instance, in predicting pH, XGBoost maintained the lowest RMSE (0.09) and MAE (0.06) and attained a R² of 0.89 (Table 3), which was much greater than the other models. Similarly, XGBoost again distinguished itself by having the highest R² (0.86) in the prediction of the link between groundwater drawdown and rainfall, indicating its resilience in capturing this relationship. Additionally, it showed the lowest MAE (0.053) and RMSE (0.072) (Table 3), demonstrating the precision and dependability of its forecasts.
On the other hand, compared to XGBoost, the other models like Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Random Forest (RF) always had lower R2 values and greater RMSE and MAE values across all parameters. Consequently, XGBoost was found to be the most accurate model for predicting groundwater quality and drawdown based on this thorough examination, making it an extremely dependable tool for managing water resources and the environment.
| Evaluation Parameter | SVM | KNN | RF | XGBoost |
|---|---|---|---|---|
| pH (R2) | 0.74 | 0.78 | 0.82 | 0.89 |
| pH (RMSE) | 0.12 | 0.11 | 0.1 | 0.09 |
| pH (MAE) | 0.09 | 0.08 | 0.07 | 0.06 |
| EC (R2) | 0.76 | 0.77 | 0.81 | 0.85 |
| EC (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| EC (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| TDS (R2) | 0.75 | 0.76 | 0.8 | 0.84 |
| TDS (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| TDS (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Calcium (R2) | 0.75 | 0.77 | 0.81 | 0.85 |
| Calcium (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| Calcium (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Magnesium (R2) | 0.75 | 0.77 | 0.81 | 0.85 |
| Magnesium (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| Magnesium (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Total Hardness (R2) | 0.75 | 0.77 | 0.81 | 0.85 |
| Total Hardness (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| Total Hardness (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Bicarbonate (R2) | 0.75 | 0.77 | 0.81 | 0.85 |
| Bicarbonate (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| Bicarbonate (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Chloride (R2) | 0.75 | 0.77 | 0.81 | 0.85 |
| Chloride (RMSE) | 0.11 | 0.1 | 0.09 | 0.08 |
| Chloride (MAE) | 0.08 | 0.07 | 0.06 | 0.05 |
| Rainfall and Drawdown (R2) | 0.75 | 0.68 | 0.82 | 0.86 |
| Rainfall and Drawdown (RMSE) | 0.102 | 0.125 | 0.083 | 0.072 |
| Rainfall and Drawdown (MAE) | 0.076 | 0.089 | 0.061 | 0.053 |
The study of water quality parameters has proven instrumental in assisting emerging nations with managing their limited resources to maintain their current aquifers (Mahmood et al. 2016). Groundwater monitoring trend underscores the necessity of sustainable water management practices and highlights the impact of anthropogenic factors on groundwater resources (Xiao et al. 2016). The imperative to prioritize hydro chemical parameters monitoring and investigation in areas with high-density human activities was emphasized to mitigate urban and industrial impacts on water quality (Nong et al. 2019). The water quality parameter values should fall within standardized ranges: pH should be between 6.5 and 8.5, electrical conductivity (EC) should be between 500 and 1500 µs/cm, and total dissolved solids (TDS) should be 500 mg/l, with a maximum limit of 1000 mg/l. The permissible range for calcium is 75 mg/l, magnesium should not exceed 50 mg/l, bicarbonate should be within 400 mg/l, and chloride levels should not exceed 250 mg/l. Additionally, total hardness should be no more than 500 mg/l (World Health Organization, 2017; Meride and Ayenew, 2016; Ministry of Climate Change, 2007).
A geographic information system (GIS)-based multivariate analysis can be used to interpret water quality monitoring data (Pratama et al. 2020). Similarly, significant fluctuations in groundwater quality, particularly during the pre- and post-monsoon periods highlight the urgent need for comprehensive groundwater management strategies across the region to mitigate potential future declines in groundwater quality and ensure sustainable water resource utilization (Shahzad et al. 2020).
Future water quality evaluations study is important for valuable insights for practitioners in designing effective water management projects (Chou et al. 2018). Artificial Intelligence models is really an important feature in water resource engineering for innovative surface water quality modeling transformations (Tung and Yaseen 2020). Machine learning (DL) model is the most precise and reliable technique for predicting groundwater quality (Singha et al. 2021). Machine learning algorithms are effective in assessing groundwater quality for irrigation, identifying critical hydrogeochemical components for groundwater evolution evaluation, and implementing mitigation strategies in dry and semi-arid areas (Ibrahim et al. 2023). Sustainable water practices, reduced waste, and optimized groundwater utilization are necessary for, providing valuable guidance for water managers, regulators, and stakeholders in agriculture and industry to manage groundwater responsibly (Prabu et al. 2023)
This comprehensive study presented a novel systematic approach to investigate groundwater drawdown hydro chemical fluctuations by leveraging spatiotemporal analysis of historical data and high-accuracy predictions using advanced machine learning models with the highest accuracy with the XGBoost model. The unregulated construction of a huge number of tube wells throughout Chiniot district has contributed to the depletion of groundwater resources. Rapid population expansion, urbanization, and economic development are expected to drive up demand for groundwater in the future. Each year, the demand for groundwater rises and quality is deteriorating in unison with the growing population.
The study found that there were regular seasonal patterns, with higher concentrations of minerals observed in post-monsoon measurements. This rise in mineral concentrations is attributed to the leaching processes that are intensified by rainfall. However, in other years, key indicators such as EC experienced a reduction after the monsoon season, which can be ascribed to the diluting effects caused by higher rainfall. This variance highlights the intricate interplay of hydrological, geological, and human processes that impact the quality of groundwater. Increasing chloride levels seen after 2022 indicate a heightened degree of human activity, such as the release of industrial waste and runoff from agricultural areas, which worsens the concentration of minerals in groundwater. In a similar vein, the impending years' increased drawdown fluctuation and less rainfall may pose a significant risk to groundwater storage. This study's findings are essential for sustainable groundwater management and decision-making. These conclusions must be included in future water resource management to reduce groundwater quality deterioration, especially in nations that heavily use groundwater for agriculture and consumption. Improved land use and stricter industrial and agricultural discharge laws can maintain groundwater ecological balance. Furthermore, pH, EC, TDS, and other mineral prognostic models can predict future changes and provide adaptive solutions to water quality issues caused by climate change and human activity. Using complex prediction models like XGBoost and RF, which accurately anticipate groundwater characteristics, shows how machine learning may improve hydrological understanding and control.
To protect groundwater resources, the Chiniot and Faisalabad region must transition from supply-based to demand-based water supply. This includes modifying cropping patterns, promoting water-efficient crops, treating wastewater, monitoring mineral leaching, and using sprinkler systems. GIS and machine learning techniques are crucial for sustainable water management. Addressing monsoon cycles and human impact is essential for ensuring the availability and accessibility of this critical resource for future generations.
Competing interests
The authors declare no competing interests.
Funding
The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.
Authors Contribution
Sheraz Maqbool contributed to conceptualizing the study, data collection, methodology, spatiotemporal mapping, machine learning forecasting and writing – original draft. Muhammad Imran khan and Aamir Raza supervised and revised it critically for important intellectual content. Naeem Saddique, Qaisar Saddique, Liu Dong, Muhammad Abdur Rehman Tariq, Mubarra Tahreem, Maha Mujahid and Noman Ali Buttar contributed to formal analysis, data arrangement, interpretation of results and in work review.
Data availability statement
All relevant data are included in the paper or its Supplementary Information.
- Abatzoglou JT, Dobrowski SZ, Parks SA and Hegewisch KC (2018) TerraClimate, a high-resolution global dataset of monthly climate and climatic water balance from 1958–2015. Sci Data 5:1-12 https://doi.org/10.1038/sdata.2017.191
- Abdalla F, and Moubark K, (2018) Assessment of well performance criteria and aquifer characteristics using step-drawdown tests and hydrogeochemical data, west of Qena area, Egypt. J Afr Earth Sci 138:336-347. https://doi.org/10.1016/j.jafrearsci.2017.11.023
- Ahmad MN, Sultana R, Din MSU, Syed NA, Parvaiz RA, Ahmad M and Ahmad T (2022) Water quality mapping of district Chiniot, Pakistan by using GIS. FUUAST j biol 12:1-8. https://fuuastjb.org/index.php/fuuastjb/article/view/656
- Ahmad MN, Sultana R, Yoshida M and Salahuddin M (2020) Groundwater contamination issues in Chiniot area, Punjab, Pakistan. Int J Environ Sci Dev 11:123-127. https://doi.org/10.18178/ijesd.2020.11.3.1237
- Ahmed AN, Othman FB, Afan HA, Ibrahim RK, Fai CM, Hossain MS, Ehteram M and Elshafie A (2019) Machine learning methods for better water quality prediction. J Hydrol 578:124084. https://doi.org/10.1016/j.jhydrol.2019.124084
- Amiri S, Rajabi A, Shabanlou S, Yosefvand F and Izadbakhsh MA (2023) Prediction of groundwater level variations using deep learning methods and GMS numerical model. Earth Sci Inform 16:3227-3241. http://dx.doi.org/10.1007/s12145-023-01052-1
- Arabgol R, Sartaj M and Asghari K (2016) Predicting nitrate concentration and its spatial distribution in groundwater resources using support vector machines (SVMs) model. Environ Model Assess 21:71-82. http://dx.doi.org/10.1007/s10666-015-9468-0
- Arnell NW, (1999) Climate change and global water resources. Glob Environ Change 9:S31-S49. https://doi.org/10.1016/S0959-3780(99)00017-5
- Arshad A, Mirchi A, Vilcaez J, Akbar MU and Madani K (2024) Reconstructing high-resolution groundwater level data using a hybrid random forest model to quantify distributed groundwater changes in the Indus Basin. J Hydrol 628:130535. http://dx.doi.org/10.1016/j.jhydrol.2023.130535
- Awais M, Arshad M, Shah SHH and Anwar-Ul-Haq M (2017) Evaluating groundwater quality for irrigated agriculture: spatio-temporal investigations using GIS and geostatistics in Punjab, Pakistan. Arab J Geosci 10:1-15. http://dx.doi.org/10.1007/s12517-017-3280-x
- Bhatti MT, Anwar AA and Aslam M (2017) Groundwater monitoring and management: Status and options in Pakistan. Comput Electron Agric 135:143-153. https://doi.org/10.1016/j.compag.2016.12.016
- Chabuk A, Al-Madhlom Q, Al-Maliki A, Al-Ansari N, Hussain HM and Laue J (2020) Water quality assessment along Tigris River (Iraq) using water quality index (WQI) and GIS software. Arab J Geosci 13:1-23. http://dx.doi.org/10.1007/s12517-020-05575-5
- Chai T, and Draxler RR (2014) Root mean square error (RMSE) or mean absolute error (MAE). Geosci Model Dev Discuss 7:1525-1534. http://dx.doi.org/10.5194/gmdd-7-1525-2014
- Chicco D, Warrens MJ and Jurman G (2021) The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation. PeerJ Comput Sci 7:e623. https://doi.org/10.7717/peerj-cs.623
- Chou JS, Ho CC and Hoang HS (2018) Determining quality of water in reservoir using machine learning Ecol Inform 44:57-75. http://dx.doi.org/10.1016/j.ecoinf.2018.01.005
- Cosgrove WJ, and Loucks DP (2015) Water management: Current and future challenges and research directions. Water Resour Res 51:4823-4839. http://dx.doi.org/10.1002/2014WR016869
- El Bilali A, Taleb A and Brouziyne Y (2021) Groundwater quality forecasting using machine learning algorithms for irrigation purposes. Agric Water Manag 245:106625. http://dx.doi.org/10.1016/j.agwat.2020.106625
- Griebler C, and Avramov M (2015) Groundwater ecosystem services: a review. Freshw Sci 34:355-367. http://dx.doi.org/10.1086/679903
- Ibrahim H, Yaseen ZM, Scholz M, Ali M, Gad M, Elsayed S, Khadr M, Hussein H, Ibrahim HH and Eid MH (2023) Evaluation and prediction of groundwater quality for irrigation using an integrated water quality indices, machine learning models and GIS approaches: A representative case study. Water 15:694. http://dx.doi.org/10.3390/w15040694
- Kılıç Z (2020) The importance of water and conscious use of water. Int j hydrol 4:239-241. http://dx.doi.org/10.15406/ijh.2020.04.00250
- Kirby M, Mainuddin M, Khaliq T and Cheema MJM (2017) Agricultural production, water use and food availability in Pakistan: Historical trends, and projections to 2050. Agric Water Manag 179:34-46. http://dx.doi.org/10.1016/j.agwat.2016.06.001
- Kreins P, Henseler M, Anter J, Herrmann F and Wendland F (2015) Quantification of climate change impact on regional agricultural irrigation and groundwater demand. Water Resour Manag 29:3585-3600. http://dx.doi.org/10.1007/s11269-015-1017-8
- Kumar N, Yamaç SS and Velmurugan A (2015) Applications of remote sensing and GIS in natural resource management. J Andaman Sci Assoc 20:1-6. https://www.researchgate.net/publication/325011344_Applications_of_Remote_Sensing_and_GIS_in_Natural_Resource_Management
- Mahmood K, Ul-Haq Z, Batool SA, Rana AD and Tariq (2016) Application of temporal GIS to track areas of significant concern regarding groundwater contamination. Environ Earth Sci 75:1-11. https://link.springer.com/article/10.1007/s12665-015-4844-2
- Meride Y, and Ayenew B (2016) Drinking water quality assessment and its effects on residents health in Wondo genet campus, Ethiopia. Environ Syst Res 5:1-7. https://environmentalsystemsresearch.springeropen.com/articles/10.1186/s40068-016-0053-6. Accessed 25 July 2024
- Ministry of Climate Change (2007) Drinking water quality standards. Government of Pakistan.:1-7. https://mocc.gov.pk/SiteImage/Misc/files/Drinking%20Water%20Quality%20Standares%20MAY%202007.pdf
- Natesan D, Sabarathinam C, Kamaraj P, Mathivanan M, Haji M, Viswanathan MP, Chandrasekaran T and Rajendran T (2022) Impact of monsoon shower on the hydrogeochemistry of groundwater along the lithological contact: a case study from South India. Appl Water Sci 12:36. http://dx.doi.org/10.1007/s13201-021-01538-7
- Nong X, Shao D, Xiao Y and Zhong H (2019) Spatio-temporal characterization analysis and water quality assessment of the South-to-North Water Diversion Project of China. Int. J. Environ. Res. Public Health. 16:22-27. http://dx.doi.org/10.3390/ijerph16122227
- Prabu T, Sarkar M, Chaudhary D, Al Obaid S, Al-Ateeq TK and Kalam MA. (2023) IoT-enabled groundwater monitoring with k-NN-SVM algorithm for sustainable water management. Acta Geophys 1-14. http://dx.doi.org/10.1007/s11600-023-01178-2
- Pratama,MA, Immanuel YD and Marthanty DR (2020) A multivariate and spatiotemporal analysis of water quality in Code River, Indonesia Sci World J 2020:8897029. https://doi.org/10.1155/2020/8897029
- Osman AIA., Ahmed AN, Chow MF, Huang YF and El-Shafie A (2021) Extreme gradient boosting (Xgboost) model to predict the groundwater levels in Selangor Malaysia. Ain Shams Eng. J. 12:1545-1556. http://dx.doi.org/10.1016/j.asej.2020.11.011
- Qureshi AS (2020) Groundwater governance in Pakistan: From colossal development to neglected management. Water 12:3017. http://dx.doi.org/10.3390/w12113017
- Shakoor A, Arshad M, Bakhsh A and Ahmed R (2015) GIS based assessment and delineation of groundwater quality zones and its impact on agricultural productivity. Pak J Agri Sci 52:837-843. https://www.researchgate.net/publication/283469557_Gis_based_assessment_and_delineation_of_groundwater_quality_
zones_and_its_impact_on_agricultural_productivity - Shahzad H, Farid HU, Khan ZM, Anjum MN, Ahmad I, Chen X, Sakindar P, Mubeen M, Ahmad M and Gulakhmadov A (2020) An integrated use of gis, geostatistical and map overlay techniques for spatio-temporal variability analysis of groundwater quality and level in the punjab province of pakistan, south asia. Water 12:3555. http://dx.doi.org/10.3390/w12123555
- Singha S, Pasupuleti S, Singha SS, Singh R and Kumar S (2021) Prediction of groundwater quality using efficient machine learning technique. Chemosphere 276:130265. http://dx.doi.org/10.1016/j.chemosphere.2021.130265
- Tung TM, and Yaseen Z (2020) A survey on river water quality modelling using artificial intelligence models: 2000–2020. J. Hydrol. 585:124-670. http://dx.doi.org/10.1016/j.jhydrol.2020.124670
- Tyralis H, Papacharalampous G and Langousis A (2019) A brief review of random forests for water scientists and practitioners and their recent history in water resources. Water 11:910. http://dx.doi.org/10.3390/w11050910
- Van Steenbergen F, and Oliemans W (2002) A review of policies in groundwater management in Pakistan 1950–2000. Water policy. 4:323-344. http://dx.doi.org/10.1016/S1366-7017(02)00006-5
- Wang X, and Xie H (2018) A review on applications of remote sensing and geographic information systems (GIS) in water resources and flood risk management. Water 10:608. https://doi.org/10.3390/w10050608
- WHO (2017) Guidelines for drinking-water quality: fourth edition incorporating the first addendum. Geneva: World Health Organization. https://www.who.int/publications/i/item/9789241549950. Accessed 25 July 2024
- Xiao Y, Gu X, Yin S, Shao J, Cui Y, Zhang Q and Niu Y (2016) Geostatistical interpolation model selection based on ArcGIS and spatio-temporal variability analysis of groundwater level in piedmont plains, northwest China. SpringerPlus 5:425 http://dx.doi.org/10.1186/s40064-016-2073-0
- Xu J, Xu Z, Kuang J, Lin C, Xiao L, Huang X and Zhang Y (2021) An alternative to laboratory testing: Random forest-based water quality prediction framework for inland and nearshore water bodies. Water 13:3262. http://dx.doi.org/10.3390/w13223262
No competing interests reported.
- floatimage1.png
Graphical abstract
