Integrated Groundwater Quality Assessment for Irrigation in the Ras El-Aioun District: Combining IWQI, GIS, and Machine Learning Approaches

Abstract

This study focuses on assessing the hydrogeochemical characteristics and irrigation suitability of groundwater in the Ras El Aioun and Merouana districts, using an integrated approach that combines physicochemical analysis, machine learning (ML), and Geographic Information Systems (GISs). Thirty groundwater samples were collected in June 2023 and subjected to extensive analyses, including major ions (Ca²⁺, Mg²⁺, Na⁺, K⁺, HCO₃⁻, Cl⁻, SO₄²⁻), pH, TDS, alkalinity, and hardness. Hydrochemical facies analysis revealed that the Ca-HCO₃ type was dominant (93.33%), with some samples exceeding FAO limits, particularly for Na⁺, K⁺, SO₄²⁻, Cl⁻, Mg²⁺, and HCO₃⁻. Assessment of groundwater irrigation suitability revealed generally favorable conditions based on three key parameters: all samples (100%) were classified as excellent based on the Sodium Adsorption Ratio (SAR < 10), 70% showed good-to-permissible status by Sodium Percentage (Na% < 60), and 83.3% were within safe limits for Residual Sodium Carbonate (RSC < 1.25 meq/L). However, the Permeability Index (PI > 75%) categorized 96.7% of samples as unsuitable for long-term irrigation due to potential soil permeability reduction. Additionally, Total Hardness (TH < 75 mg/L) indicated predominantly soft water characteristics (90% of samples), particularly in the central study area, suggesting possible limitations for certain agricultural applications that require mineral-rich water. GIS-based spatial analysis showed that irrigation suitability was higher in the eastern and western regions than in the central zone. Advanced machine learning algorithms provide superior predictive capability for water quality parameters by effectively modeling complex, non-linear feature interactions that conventional statistical approaches frequently fail to capture. Three ML models—Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were used to predict the Irrigation Water Quality Index (IWQI). XGBoost outperformed the others (RMSE = 2.83, R² = 0.957), followed by RF (RMSE = 3.12, R² = 0.93) and SVR (RMSE = 3.45, R² = 0.92). Integrating ML and GIS improved groundwater quality assessment and provided a robust framework for sustainable irrigation management. These findings provide critical insights for optimizing agricultural water use in water-scarce regions.

1. Introduction

Groundwater is a vital resource for agricultural irrigation, especially in arid and semi-arid regions where surface water is scarce. Nevertheless, the increasing intensification of agriculture, the excessive use of fertilizers, agrochemicals, and pesticides, and the high consumption of groundwater have led to significant degradation of water quality, posing a considerable risk to soil health, crop productivity, and water security in the medium and long term [1,2,3,4,5,6]. Groundwater quality is assessed based on hydrochemical indices and various ions such as cations (Ca²⁺, Mg²⁺, Na⁺, K⁺) and anions (HCO₃, Cl⁻, SO₄²⁻, NO₃⁻) [7]. It is important to note that higher concentrations of sodium contribute to water salinity, exert osmotic pressure on aquatic biota, and can induce hypertension in humans [8]. The quantity and composition of salts present determine the suitability of water for irrigation. Poor-quality water can lead to detrimental effects on soil and plants [9].

Traditional methods of assessing water quality for irrigation rely on physicochemical indicators such as the Sodium Adsorption Ratio (SAR), Permeability Index (PI), and Residual Sodium Carbonate (RSC). The Irrigation Water Quality Index (IWQI) addresses this limitation by integrating multiple parameters into a single, interpretable score [10,11]. Recent advances in Geographic Information Systems (GISs) have further enhanced water quality assessments by enabling spatial interpolation, hotspot identification, and dynamic monitoring [12,13,14,15,16].

Machine learning has proven to be a powerful tool for modeling complex environmental systems, providing highly informative predictions by identifying non-linear relationships in hydrochemical data [16,17]. Algorithms such as Support Vector Regression (SVR), Random Forest (RF), and XGBoost are now widely used for their effectiveness in predicting water quality, especially when combined with GIS for spatial analysis [18,19]. However, few studies have applied this integrated approach to assess the suitability of groundwater for irrigation in regions under heavy agricultural pressure.

Numerous water quality indices have been developed and described in the literature to address this issue and evaluate groundwater quality [20]. In a recent study by [21], the researchers focused on assessing irrigation water quality in the Abu Dhabi Emirate. They established an Irrigation Water Quality Index (IWQI) and utilized GIS-zoning maps to achieve this.

The study aimed to examine various key factors that contribute to the overall quality of irrigation water, such as the Sodium Adsorption Ratio (SAR), electrical conductivity (EC), sodium ratio (%Na), chloride concentration, and Permeability Index (PI). By analyzing these parameters, the researchers identified potential soil-related issues arising from current irrigation practices. Additionally, the IWQI visualization is a valuable tool for decision-makers to identify “Red Zones” characterized by excessive groundwater extraction.

This study investigates the use of machine learning in conjunction with Geographic Information Systems (GISs) to predict and monitor the quality of groundwater in Ras El Aioun, a region heavily reliant on agriculture. GIS enables the spatial analysis and visualization of water quality data, the identification of pollution sources, and the tracking of temporal variations [22,23,24]. Combining the Water Quality Index (WQI) with GIS-based techniques enhances the assessment of groundwater quality through advanced analytics. Machine learning models further improve the accuracy of predictions by uncovering complex hydrochemical patterns and offering a robust framework for the sustainable management of water resources. The proposed approach can be replicated in other groundwater-vulnerable regions.

2. Materials and Methods

2.1. Study Area

The Ras El-Aioun Plain is located in the Hodna Basin, in the eastern part of Algeria, approximately 100 km northwest of Batna. It covers an area of 265 km² and is geographically bounded by several prominent mountain ranges. To the north lie Kef El Bioud and Dj. Tazila, to the northwest are Dj. Guetiane and Dj. Azekkar, and to the southwest Dj. El Gues forms a natural boundary. The region has a semi-arid climate, with an average annual precipitation of 417.7 mm, significantly lower than the evaporation rate of 688.17 mm. This creates an agricultural rainfall deficit of 1044.25 mm, making groundwater irrigation essential for agricultural activities.

Agriculture in the region relies heavily on irrigation, with crops such as cereals, legumes, and vegetables predominant. However, water scarcity and the increasing use of groundwater for irrigation have put considerable pressure on the aquifers in the area, leading to concerns over the long-term sustainability of groundwater resources.

2.2. Geology and Hydrogeology

The Hodna Mountains form the northern outcrops surrounding the study area, with anticlines sheltering Jurassic limestone and dolomite formations. These geological structures serve as a natural barrier, separating the Setifian region to the north and east, where the Hodna River originates from the Belezma Mountains. The Hodna massifs, including Boutaleb, Guetiane, Hadjar Labiod, and Foural-Talkhempt, extend from west to east [25]. To the southeast of the study area lies the northern part of the Belezma Mountains, particularly the Bou Ari area. This area is notable for its hydrogeological significance and consists of sediments from the lower Cretaceous period. The boundary of Djebel Bou (Figure 1) Ari is characterized by a hydraulic gradient of approximately 3%, which influences the region’s groundwater reserves. The Ras El-Aioun Plain, or Synclinal of Oud Chair, is oriented in a NE–SW direction, marked by the upper Pliocene detrital formations that form the relief of Djebel Gues, which is located south of Djebel Guetiane. Jurassic Mestaoua-Messaouda formations have eroded the structure to the east, while Miocene deposits are found along the northern flank near Djebel Guetiane [26]. Above these, Albian limestone marls form an impermeable layer, while marly sandstone and Aptian limestone outcrop along the anticline from Belezma to Dj. Fourhal and Dj. Tazila. These fissured limestone sandstones are confined aquifers, suggesting significant groundwater potential around 140 m northwest of the Tazila anticline foothills.

Additionally, the Mio-Plio-Quaternary aquifer interacts with these calcareous layers. Lithostratigraphic analysis in the southern study area reveals two distinct aquifer systems: a 700 m thick fractured carbonate sandstone formation of Barremian age and a 250 m thick Mio-Plio-Quaternary alluvial unit characterized by interstitial porosity [27,28]. Groundwater chemistry in these aquifers is governed by competing hydrogeochemical processes, including carbonate dissolution, evaporite precipitation, and cation exchange reactions, which collectively control water quality evolution in production wells. The region’s agricultural sector, dominated by water-intensive cereal, legume, and vegetable crops, accounts for over 80% of groundwater abstraction. This demand, compounded by recurrent droughts and inadequate recharge, has led to unsustainable extraction, with observed piezometric declines of 1.5–2 m per year. Such over-exploitation threatens both aquifer sustainability and water quality, particularly through emerging saltwater intrusion in coastal areas, requiring urgent water management interventions.

2.3. Groundwater Sampling and Analytical Procedures

To assess water quality in the Ras El Aïoun plain, 30 water samples were collected from different sites distributed across the districts of Ras El Aïoun and Merouana, during the period of May–June 2023. Samples were rinsed and preserved in plastic bottles, and domestic and agricultural boreholes were pumped for 15 min before sampling. The samples were analyzed in the Chemistry Laboratory of the University of Batna 2, in accordance with the standardized procedures described in Rodier [29]. The physical and chemical parameters examined included pH, Total Hardness (TH), electrical conductivity (EC), and the concentrations of Mg²⁺, Ca²⁺, K⁺, Na⁺, HCO₃⁻, Cl⁻, and SO₄²⁻. The concentrations of Ca and Mg were determined using the complexometric EDTA titration method. Chloride (Cl⁻) was measured using the Mohr method, while sulfate (SO₄²⁻) was analyzed through conductimetric titration. Potassium (K⁺) and sodium (Na⁺) concentrations were measured using a flame photometer, and Bicarbonate (HCO₃) was evaluated using the method of titration. Data collected in (mg/L) were converted to (meq/L).

To calculate the analytical error for each water sample, we divided the total cation and anion concentrations by the ratio of the difference between their sums. The inaccuracy in the ionic charge balance should be limited to a maximum of 5%. If the magnitude exceeds 5%, the analysis should be conducted again. The analytical error of the measured ion concentration was verified using the electroneutrality (charge-balance error) method [30] based on the following formula:

$$\mathrm{C}\mathrm{B}\mathrm{E}(\%)={\displaystyle \frac{\mathsf{\Sigma }\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}-\mathsf{\Sigma }\mathrm{a}\mathrm{n}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}}{\mathsf{\Sigma }\mathrm{c}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}+\mathsf{\Sigma }\mathrm{a}\mathrm{n}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{s}}}\times 100$$

Across all samples, the maximum deviation observed was ±6%.

In the present study, the charge balance was calculated for each of the 30 water samples to validate the consistency and accuracy of the laboratory measurements. The maximum deviation observed was ±6%, slightly above the standard threshold in a few cases. However, most samples were within acceptable limits and the variation remained relatively small. As such, the dataset was considered generally reliable, and the small exceedances were attributed to natural variability and trace ion concentrations not always accounted for in the baseline analyses. Including this verification step reinforces the analytical integrity and quality assurance of the water chemistry dataset used in this research.

These procedures ensured that all collected data were accurate, reproducible, and suitable for assessing the irrigation suitability and geochemical status of groundwater in the Ras El-Aioun plain.

2.4. Multivariate Statistical Analysis

Multivariate statistical analysis is a quantitative and independent technique for classifying groundwater samples, facilitating sample grouping and the identification of correlations between metals and groundwater samples [31].

To analyze and interpret the hydrochemical parameters related to the chemical composition of each groundwater sample, we applied Principal Component Analysis (PCA) and Hierarchical Cluster Analysis (HCA) techniques using the R programming language [32]. We categorized the water samples into distinct hydrochemical facies and identified patterns within the data using HCA, which also helped trace the sources of groundwater contamination. Additionally, we employed Microsoft Excel to perform statistical analysis on the physicochemical properties, calculating the mean, standard deviation (variance), maximum, and minimum values. The diagram in Figure 2 illustrates the steps involved.

2.5. Water Quality Index for Agricultural Purposes

According to [33], the identification of optimal irrigation methods is influenced by various factors, such as water quality, soil types, and agricultural activities. The irrigation water quality assessment requires the determination of SO₄²⁻ concentrations and parameters related to hardness and principal ions, including CO₃²⁻, HCO₃⁻, Ca²⁺, and Mg²⁺ [34]. In general, the assessment of the suitability of groundwater for irrigation primarily relies on analyzing the concentration of sodium ions (Na⁺) in relation to the overall cation content in the system [35]. Due to the significant impact of Na⁺ ions on soil permeability and water infiltration, they directly influence the quality of groundwater [36]. The assessment of water quality for agricultural purposes involved the utilization of different indices, including Kelly’s ratio (KR), Total Hardness (TH), the sodium percentage (Na%), the Sodium Adsorption Ratio (SAR), and Residual Sodium Carbonate (RSC) (

Table 1. Irrigation groundwater quality.

Parameters	Equations	Reference
Sodium Adsorption Ratio (meq/L)	$\mathrm{S}\mathrm{A}\mathrm{R}={\displaystyle \frac{{\mathrm{N}\mathrm{a}}^{+}}{\sqrt{\frac{{\mathrm{C}\mathrm{a}}^{2+}+{\mathrm{M}\mathrm{g}}^{2+}}{2}}}}$	[37]
Sodium Hazard	$\mathrm{N}\mathrm{a}\%={\displaystyle \frac{{\mathrm{N}\mathrm{a}}^{+}+{\mathrm{K}}^{+}}{{\mathrm{C}\mathrm{a}}^{+}+{\mathrm{M}\mathrm{g}}^{+}+{\mathrm{N}\mathrm{a}}^{+}+{\mathrm{K}}^{+}}}\times 100$	[38]
Residual Sodium Carbonate (meq/L)	RSC= [HCO3− +CO32−] − [Ca2+ + Mg+]	[39]
Permeability Index	$\mathrm{P}\mathrm{I}={\displaystyle \frac{{\mathrm{N}\mathrm{a}}^{+}+\sqrt{{\mathrm{H}\mathrm{C}\mathrm{O}}_3^{-}}}{{\mathrm{C}\mathrm{a}}^{2+}+{\mathrm{M}\mathrm{g}}^{2+}+{\mathrm{N}\mathrm{a}}^{+}}}\times 100$	[40]
Kelly’s Ratio	$\mathrm{K}\mathrm{R}={\displaystyle \frac{{\mathrm{N}\mathrm{a}}^{2+}}{{\mathrm{C}\mathrm{a}}^{2+}+{\mathrm{M}\mathrm{g}}^{2+}}}$	[41]
Total Hardness	TH = 2.497 Ca2+ + 4.11 Mg2+	[42]

).

2.6. GIS Analysis Mapping of Water Quality

Geographic Information Systems (GISs) are computer-based technological solutions specifically designed for processing digital representations of geographical data. Their principal functions include the acquisition, storage, manipulation, analysis, and presentation of diverse collections of spatial or geolocated data [43,44]. Ordinary kriging (OK) is a method that offers insights into spatial dependence and autocorrelation among sampling sites; this information is essential for accurate spatial estimation [45]. In our case, we used the default linear Kriging approach without explicitly applying a geostatistical model or conducting variogram modeling. This choice was made for simplicity and to allow for a preliminary spatial estimation without the need for model calibration. We generated kriging maps to achieve a comprehensive understanding of the spatial distribution of underground water quality indexes in the Ras El-Aioun plain, and we used ArcGIS 10.2. Groundwater quality parameter distribution maps were created based on thirty samples collected from the area under assessment.

2.7. Model Performance and Evaluation

Predicting the Irrigation Water Quality Index (IWQI) is crucial for assessing groundwater suitability for irrigation, allowing for informed water management decisions. Traditional methods are often limited in areas with restricted data and complex hydrogeochemical interactions, which makes machine learning (ML) an effective tool for water quality evaluation [46].

Five ML models were selected based on their regression capabilities: Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), XGBoost, and K-Nearest Neighbors (KNNs).

Support Vector Regression (SVR) extends Support Vector Machines (SVMs) to regression by mapping input data into a high-dimensional space and fitting an optimal hyperplane. It was chosen for its effectiveness in handling complex relationships, resilience to outliers, and strong performance with limited data [47].

The SVR optimization problem is formulated as the optimization problem in SVR is formulated as follows:

$${\displaystyle \frac{1}{2}}{‖\mathrm{Ϣ}‖}^2+\mathrm{C}{\sum }_{\mathrm{k}=1}^{\mathrm{n}}\left(\mathrm{p}\mathrm{i}+{\mathrm{p}\mathrm{i}}^{\mathrm{*}}\right),\mathrm{i}=1,2,\dots \mathrm{n}$$

$$\mathrm{y}\mathrm{i}-\mathrm{Ϣ}.\mathrm{ϴ}\left(\mathrm{x}\right)-\mathrm{b}\le \mathrm{\varepsilon }+{\mathrm{p}\mathrm{i}}^{\mathrm{*}}$$

$$\mathrm{Ϣ}.\mathrm{ϴ}\left(\mathrm{x}\right)+\mathrm{b}-\mathrm{y}\mathrm{i}\le \mathrm{\varepsilon }+\mathrm{p}\mathrm{i}\hspace{1em}.\mathrm{p}\mathrm{i}{\mathrm{p}\mathrm{i}}^{\mathrm{*}}\ge 0$$

In this formulation, w represents the weight vector, ‖$\mathrm{Ϣ}$‖ is the complexity referred term and C is the regularization parameter, θ(x) is the mapping function, and ε is the insensitive loss function.

Random Forest (RF) is an ensemble learning algorithm that constructs multiple decision trees and averages their outputs to enhance accuracy and mitigate overfitting. It was selected for its robustness to noise, ability to model intricate feature interactions, and effectiveness in capturing non-linear dependencies [48].

The prediction for Random Forest is given by

• Prediction (Regression, mean of T trees):

  $$Ŷ(x)={\displaystyle \frac{1}{T}}\sum _{t=1}^T{h}_t(x)$$
• Split Criterion (Classification, Gini impurity):

  $$G=\sum _{k=1}^K{p}_k(1-p_k)$$

  where pk is class proportion in the node.

Gradient Boosting (GB) sequentially builds decision trees, with each tree correcting the errors of its predecessors, thereby improving predictive accuracy. It was chosen for its ability to capture complex patterns, reduce bias, and refine IWQI predictions through iterative optimization [49].

• Loss Function (Regression, MSE):

  $$\mathrm{L}\left(\mathrm{y},\mathrm{F}\left(\mathrm{x}\right)\right)={\displaystyle \frac{1}{N}}\sum _{i=1}^N{(y_i-F(x_i)}^2)$$
• Update Rule (Step t):

  F_t(x) = F_t−1(x) + η⋅ht,   where h_t(x) ≈ −∇FL(y,F_t−1(x))
• Optimization: gradient descent on residuals

  r_it = y_i − F_t−1(x_i).

XGBoost (Extreme Gradient Boosting) is an advanced implementation of Gradient Boosting that incorporates regularization, parallel processing, and efficient handling of missing data. It was selected for its superior accuracy, computational efficiency, and ability to model intricate hydrogeochemical relationships [50].

A key characteristic of objective functions is that they consist of two parts: training loss and a regularization term, as follows:

obj(θ) = L(θ) + Ω(θ)

where L denotes the empirical loss function quantifying model fidelity to the training data, while Ω represents the regularization term controlling model complexity. The loss function L evaluates the predictive performance on the training set, with Mean Squared Error (MSE) commonly as

$$L(\theta )=\sum (Yi-Ŷi{)}^2$$

Another frequently used loss function is logistic loss, applied for logistic regression as follows:

$$\mathrm{L}(\theta )={\sum }_i[\mathrm{y}\mathrm{i} \mathrm{l}\mathrm{n}(1+{\mathrm{e}}^{-\mathrm{Ŷ}\mathrm{i}})+(1-y_i\left)\mathrm{l}\mathrm{n}\right(1+{\mathrm{e}}^{\mathrm{Ŷ}\mathrm{i}}\left)\right]$$

Many practitioners overlook the critical importance of the regularization term in machine learning models. This term governs model complexity and is essential for preventing overfitting.

K-Nearest Neighbors (KNNs) is a distance-based algorithm that predicts values by averaging the closest data points in feature space. It was chosen for its simplicity, ability to detect local patterns, and non-parametric nature. KNNs served as a baseline model for comparison in IWQI prediction [51]. To identify the nearest neighbors, the algorithm uses a distance metric to calculate the distance between a new data point x and each data point x_i in the training dataset [46].

• Distance Metric:

  $$\mathrm{d}(\mathrm{x},{\mathrm{x}}_{\mathrm{i}})=\sqrt{{\sum }_{j=1}^p(x^j-x_i^j{)}^2}$$

  where x_i represents a training instance from our dataset, x denotes the novel query point requiring prediction, and p corresponds to the dimensionality of the feature space.
• Regression (Mean of K neighbors):

  $$Ŷ(\mathrm{x})={\displaystyle \frac{1}{K}}{\sum }_{iϵN_{K\left(x\right)}}{y}_i$$

  where N_K(x) = set of K-nearest points to x.

These models were trained on hydrogeochemical data from the Ras El-Aioun plain to determine the most accurate approach for IWQI estimation. Model performance was evaluated using Root Mean Square Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared. R² quantifies the proportion of variance explained, with values closer to 1 indicating better predictive capability. MSE and RMSE measure error magnitude, where lower values denote higher accuracy. MAE provides an intuitive measure of the average deviation between predicted and actual values. The mathematical formulations for these metrics are

$$RMSE=\sqrt{{\displaystyle \frac{{\sum }_1^n{(Xi-\widehat{X}i)}^2}{n}}}$$

$$R^2=1-{\displaystyle \frac{{\sum }_{i=1}^n{(\widehat{X}i-Xi)}^2}{{\sum }_{i=1}^n{(\overline{Xi}-Xi)}^2}}$$

$$MAE={\displaystyle \frac{{\sum }_1^n\left|(Xi-\widehat{X}i)\right|}{n}}$$

$$MSE={\displaystyle \frac{1}{n}}\sum _{i=0}^n\left(Xi-\widehat{X}i\right)$$

where Xi represents the measured value, $\widehat{X}i$ represents the predicted value, and n is the number of samples.

3. Results

3.1. Characterization of Groundwater

The physico-chemical parameters of the groundwater sampled at 30 sites are summarized in

Table 2. Parameter analysis of groundwaters.

Parameters	Unit	Min	Median	Max	Drinking Water Standards	Irrigation Water Standards
pH		6.7	7.19	7.47	6.5–8.5	6.0–8.5
EC	µs/cm	1200	800	1700	500–1500	3000
Ca2+	mg/L	49.32	78.09	406.89	75–200	400
Mg2+	mg/L	29.16	48.60	240.57	30–150	60
Cl−	mg/L	20.59	20.59	102.24	250	1100
HCO3−	mg/L	195.20	317.20	402.60	300–500	600
Na+	mg/L	44.46	123.03	181.60	200	900
K+	mg/L	0.43	1.23	3.64	12	2
SO42−	mg/L	57.60	77.57	119.04	250	1000
TDS	mg/L	457	744	1219	500–1000	2000

and compared with Algerian regulatory standards and FAO irrigation guidelines [52]. pH values ranged from 6.7 to 7.47 (mean: 7.19), indicating neutral to slightly alkaline conditions which were all within the FAO recommended irrigation range (6.0–8.5). Electrical conductivity (EC) varied from 800 to 1700 μS/cm, which is consistent with acceptable irrigation limits (750–2000 μS/cm) [42]. Elevated EC values indicate higher dissolved mineral content, with spatial and temporal variations in total dissolved solids (TDSs) primarily attributed to salinization. While porosity, permeability, and weathering dominate TDS variability in natural aquifers, anthropogenic contributions appear to be secondary. TDS concentrations averaged 744 mg/L (range: 457–1219 mg/L), with 90% of samples classified as freshwater (TDS < 1000 mg/L) [53]. However, three samples (E2, E9, E16) exhibited brackish conditions (1000–10,000 mg/L), likely influenced by proximity to the Salt Lake (Sabkha Essoukhna). Calcium (Ca²⁺) concentrations ranged from 78 to 406.8 mg/L, with 13.3% of samples exceeding the WHO recommended limit. Magnesium (Mg²⁺) concentrations ranged between 20.16 and 240.5 mg/L, with 56.6% of samples exceeding the FAO irrigation threshold (60 mg/L) [52]. This is consistent with the typical geochemical behavior of natural waters, where calcium levels generally dominate magnesium.

The elevated concentrations of these divalent cations are primarily due to the dissolution of carbonate minerals, particularly in the western and southern regions of the study area where Albian limestone and dolomite formations predominate [54]. The dominant reaction for calcium release (CaCO₃ + H₂O + CO₂ → Ca²⁺ + 2HCO₃⁻) also explains the observed bicarbonate (HCO₃⁻) enrichment (195.2–402.6 mg/L), reflecting slightly alkaline conditions [55]. Prolonged water–rock interactions in wells E2, E11, and E16 further enhanced this dissolution process due to extended groundwater residence times.

Alkali metal concentrations showed Na⁺ (44.4–181.6 mg/L) within acceptable irrigation limits, while K⁺ (0.43–3.64 mg/L) exceeded the FAO guideline (2 mg/L) at 11 sampling sites (E1, E5–E8, E12, E13, E16, E17, E21, E30), thereby reducing irrigation suitability [52]. Sodium enrichment likely occurs through base exchange reactions in clay-rich zones, where Na⁺ replaces Ca²⁺ in the aquifer matrix. Chloride (Cl⁻: 20.59–102.24 mg/L) and sulfate (SO₄²⁻: 57.6–119.04 mg/L) had mean concentrations of 55.23 mg/L and 97.83 mg/L, respectively. Notably, site E16 showed a marked enrichment in Na⁺, K⁺, and Cl⁻, suggesting saline intrusion or evaporite dissolution from the El-Hamiet region. These patterns highlight how lithological composition and hydrogeological conditions control groundwater chemistry and irrigation potential across the Ras El-Aioun plain.

3.2. Hydrochemical Facies Type of Groundwater

The Piper diagram is a widely used graphical tool for visualizing the relative concentrations of major ions in water and assessing their geochemical relationships [56]. Its classification system effectively highlights the influence of lithological facies on groundwater composition. As shown in the Piper trilinear diagram (Figure 3) [57], the groundwater samples were classified into four hydrochemical types: (1) Ca-Mg > Na-K, (2) Na-K > Ca-Mg, (3) HCO₃⁻ > Cl⁻ SO₄²⁻, and (4) Cl⁻ SO₄²⁻ > HCO₃⁻. Geochemical analysis revealed a general cation dominance sequence of Ca²⁺ > Na⁺ > Mg²⁺ > K⁺, although in some cases, sodium was dominant (Na⁺ > Ca²⁺ > Mg²⁺ > K⁺). For anions, the dominant patterns were HCO₃⁻ > Cl⁻ > SO₄²⁻ and HCO₃⁻ > SO₄²⁻ > Cl⁻. The groundwater samples were predominantly (93.3%) Ca-Mg-HCO₃ facies, characteristic of calcium and magnesium bicarbonate waters. The remaining samples were classified as Na-K-HCO₃ (3.33%) and Na-K-SO₄ (3.33%) types, reflecting local variations in hydrochemical processes.

3.3. Hydrogeochemical Process

The Gibbs diagram [58] is instrumental in understanding the factors influencing water composition in different environments and visually represents these distinct hydrochemical processes. These processes control groundwater chemistry by determining the relationship between an aquifer’s lithological characteristics and the water’s composition [59,60]. Water composition is primarily influenced by evaporation, rock–water interactions, and atmospheric precipitation. When analyzing the dissolved chemical constituents, it is common to plot the cation ratio (I) and the anion ratio (II) against the TDS value [61,62]. The following equations for anions and cations were used to plot the major ion data from groundwater samples and construct the Gibbs diagram, as follows:

$$\mathrm{Gibbs} \mathrm{ratio} \mathrm{I} (\mathrm{meq}/\mathrm{L})=({\mathrm{N}\mathrm{a}}^{+}+{\mathrm{K}}^{+}/{\mathrm{K}}^{+}+{\mathrm{N}\mathrm{a}}^{+}+{\mathrm{C}\mathrm{a}}^{2+})$$

$$\mathrm{Gibbs} \mathrm{ratio} \mathrm{II} (\mathrm{meq}/\mathrm{L})=({\mathrm{C}\mathrm{l}}^{-}/{\mathrm{C}\mathrm{l}}^{-}+{\mathrm{H}\mathrm{C}\mathrm{O}}_3^{-})$$

Figure 4 shows the groundwater samples on the Gibbs diagram. The ratio I ranges between 0.11 and 0.75, while the ratio II falls within the range of 0.05 to 0.3.

Aquifer and soil parameters are used to locate groundwater on the left or right side of the diagram. If carbonate minerals predominate, lower Na/(Na + Ca) ratios are expected to affect groundwater chemistry [63].

The water samples are mostly rock-dominated. These findings imply that sedimentary geological contexts are the primary factor influencing hydrogeochemistry. Several samples fall within the evaporation process.

3.4. Hierarchical Cluster Analysis

In hydrogeochemical studies, the hierarchical clustering approach has been successfully used to classify sampling sites and geochemical elements with similar properties influenced by comparable processes and sources [64,65,66]. In this study, the approach was implemented using two methods: R-mode HCA was used to analyze 10 hydrochemical variables (EC, TDS, Ca²⁺, Mg²⁺, K⁺, Na⁺, HCO₃⁻, Cl⁻, SO₄²⁻, NO₃⁻) and Q-mode HCA was used to group 30 groundwater variables. The result of the hierarchical clustering is presented in a dendrogram (Figure 5), which shows three main clusters.

The first group, G1, comprises HCO₃⁻, Ca²⁺, and Na⁺, indicative of fossil groundwater. This group accounted for 86.6% of the sampled boreholes and was predominantly found in the northeast zone. These findings demonstrate that the Cretaceous aquifer can be considered a recharge zone. On the other hand, E2 and E9, situated in the east and west Ticherirt Flanks, respectively, exhibit higher TDS (average 1135.5 mg/L). The dissolved salts in these locations primarily comprise Ca²⁺, Mg²⁺, HCO₃⁻, and SO₄²⁻, suggesting that carbonate weathering is the prevailing factor.

The second subgroup (G2) includes HCO₃⁻, K⁺, and pH. This group represents 6.6% of water samples. The good statistical coherence between HCO₃⁻ and K⁺ can be explained regarding groundwater circulation and the water–rock interaction. In wells E20, E16, E27, and E28, there was an exchange of Na⁺ and K⁺ for Mg²⁺ or Ca²⁺ in the aquifer.

The effects of human activity, increasing rainfall, and increased regular groundwater and surface water exchange throughout the sampling period indirectly confirm that the origins of these ions are linked or comparable, such as mineral dissolution and silicate weathering dissolution [67].

By employing chloralk indices (CAI-I and CAI-II), one can analyze the cation exchange between the surrounding environment and groundwater [68]. These are computed using of the following formulas:

$$\mathrm{C}\mathrm{A}\mathrm{I}-\mathrm{I}={\displaystyle \frac{{\mathrm{C}\mathrm{l}}^{-}-({\mathrm{N}\mathrm{a}}^{+}+{\mathrm{K}}^{+})}{{\mathrm{C}\mathrm{l}}^{-}}}$$

$$\mathrm{C}\mathrm{A}\mathrm{I}-\mathrm{I}\mathrm{I}={\displaystyle \frac{{\mathrm{C}\mathrm{l}}^{-}-({\mathrm{N}\mathrm{a}}^{+}+{\mathrm{K}}^{+})}{{{\mathrm{S}\mathrm{O}}_4}^{2-}+{{\mathrm{H}\mathrm{C}\mathrm{O}}_3}^{-}+{{\mathrm{N}\mathrm{O}}_3}^{-}}}$$

Positive cationic exchange values (CAI-I) indicate that the waters exchange sodium (Na) and potassium (K), release Ca, and capture K from the clay minerals (Figure 6).

The subgroup G3, representing the Cl-SO₄-HCO₃ water type characterized by E30, E21, and E20 samples, demonstrates a predominance of alkaline earth metals. This group reflects anthropogenic influences, such as the discharge of sewage effluents and unregulated waste management. The extensive use of chemical fertilizers is also prevalent.

3.5. Irrigation Water Quality Assessment

This study evaluates the suitability of groundwater for agricultural irrigation in the Ras El Aioun region by analyzing several key water quality parameters, including the Sodium Adsorption Ratio (SAR), Kelly’s Ratio (KR), sodium percentage (Na%), Permeability Index (PI), Residual Sodium Carbonate (RSC), and Total Hardness (TH). The results of the Irrigation Water Quality Indices (IWQIs) are summarized in

Table 3. Irrigation water quality of the plain Ras El Aioun.

Wells	SAR (meq/L)	RSC (meq/L)	Na%	PI	TH	KR
E1	1.77	−10.57	20.27	20.49	78	8.89
E2	0.41	−32.25	24.68	24.79	99	21.04
E3	0.82	−19.50	20.31	20.49	64	13.96
E4	1.35	−0.73	10.03	10.69	18	4.64
E5	2.80	−1.60	19.08	19.60	22	5.45
E6	3.22	−0.68	17.62	18.27	16	4.24
E7	1.72	−10.22	20.22	20.46	40	9.1
E8	1.32	−18.12	23.53	23.67	60	13.15
E9	0.79	−23.32	22.59	22.74	74	15.98
E10	0.52	−28.17	22.38	22.50	84	18.01
E11	1.43	1.85	9.62	10.47	16	4.24
E12	2.19	−5.02	19.90	20.26	31	7.27
E13	2.47	−0.07	14.83	15.51	17	4.44
E14	2.97	−2.26	20.47	21.01	23	5.65
E15	1.02	−4.15	11.36	11.79	27	6.46
E16	1.63	−21.54	29.08	29.19	69	14.97
E17	2.048	−4.50	16.90	17.29	26	6.26
E18	1.24	−0.02	8.42	9.19	15	4.03
E19	0.69	−14.14	15.31	15.52	49	10.92
E20	2.93	−0.52	15.63	16.36	15	4.03
E21	2.07	1.65	11.49	12.36	14	3.83
E22	2.31	−0.88	12.44	13.18	14	3.83
E23	2.30	−1.75	14.50	15.10	18	4.64
E24	0.96	−0.73	6.52	7.29	13	3.63
E25	1.26	0.95	7.31	8.27	12	3.43
E26	1.15	1.30	7.27	8.20	13	3.63
E27	2.52	1.25	13.90	14.74	15	4.03
E28	1.97	−3.06	16.68	17.29	20	7.07
E29	2.46	−0.73	15.25	15.90	18	4.64
E30	3.75	−1.64	18.50	19.08	14	3.83

. A comprehensive assessment was also conducted using the USA salinity diagram and spatial distribution maps (Figure 7).

3.5.1. Sodium Absorption Ratio (SAR)

The evaluation of sodium concentration is crucial for determining the suitability of water for irrigation. Calcium and magnesium levels often reduce the toxicity of sodium, with moderate amounts of calcium shown to alleviate damage, while higher quantities provide a preventive effect. Due to the influence of salt and calcium on the SAR value, it is possible to assess the toxicity potential of irrigation water [69,70]. According to [64,71], SAR values are classified into four groups: low (<10), medium (10–18), high (18–26), and extremely high (>26).

The representation of the SAR index values on the Richards diagram [37] is based on the conductivity (Figure 8). In the Ras El Aioun region, SAR values ranged from 0.41 to 3.75, with all samples falling within the excellent quality range for irrigation. The spatial distribution of SAR, shown in the IWQI map, indicated the highest values in the central and western regions, whereas lower values were observed in the western plain. The dissolution of carbonates likely influenced the higher SAR values.

3.5.2. Sodium Percentage (%)

Irrigation water management relies heavily on accurate sodium percentage (%) assessments. Excessive sodium concentrations in water and soil can inhibit plant growth by decreasing soil permeability [72]. According to [38] and [37], the classification of Na% in irrigation water ranges from excellent (<20), good (20–40), acceptable (40–60), doubtful (60–80), to unsuitable (>80). In the study area, 70% of groundwater samples were classified as excellent, while 30% were classified as good, highlighting the region’s low salinity and minimal salt risks. The presence of calcium and magnesium carbonate in such waters can lead to the precipitation of these compounds. However, over time, the precipitation of calcium and magnesium carbonates may increase sodium concentrations, potentially degrading soil structure and negatively affecting plant growth.

3.5.3. Residual Sodium Carbonate (RSC)

RSC is a key parameter for assessing the impact of groundwater on soil structure during irrigation. RSC values are affected by evaporation and evapotranspiration processes, which concentrate on the soil solution and can lead to precipitation of sodium bicarbonate (NaHCO₃), potentially altering soil permeability [73,74]. A negative RSC indicates an excess of sodium ions, which results in the displacement of calcium and magnesium. This causes the precipitation of calcium as CO₂ and leaves sodium as the dominant cation. Conversely, a positive RSC value indicates a higher level of calcium and magnesium due to the formation of calcium bicarbonate and magnesium bicarbonate through the reaction with HCO₃ [75].

In this study, RSC values ranged from −0.02 to 1.85, with 83.33% of samples classified as “good” and 16.6% as “doubtful.” The spatial map of RSC values (Figure 8) demonstrated regional variability, with RSC decreasing from the southeast to the center of the study area and increasing from the northeast. This variation suggests that increased salt absorption may occur in certain regions.

3.5.4. Permeability Index (PI)

Singh [76] studied the effects of mineral-rich water containing ions such as Ca²⁺, Mg²⁺, Na⁺, and HCO₃⁻ on soil permeability. They observed that utilizing this water for irrigation had a significant impact on the permeability of the soil. As a result, the Permeability Index is used to evaluate the water quality and its influence on agricultural soil degradation [77]. Based on criteria developed by [39,40], groundwater was classified into three categories: Class I (suitable), Class II (doubtful), and Class III (unsuitable). The results of PI (Table 3) showed that the Permeability Index (PI) varied between 8.20 and 29.2 for the groundwater of Ras El Aioun, with 96.6% of the samples classified as unsuitable and 3.3% as doubtful. Higher PI values were observed in the central and southwestern parts of the study area, indicating the potential degradation of soil permeability in these regions due to irrigation.

3.5.5. Kelly’s Index (KI)

Kelly’s Index (KI) is another important indicator for determining the suitability of groundwater for irrigation. A KI value below one indicates low salinity, while waters considered safe for irrigation have a KI below one. In contrast, waters with a KI value greater than one are classified as unsuitable [78,79]. Based on the KI results, all groundwater samples in this study were classified as unsuitable for irrigation, with values ranging from 3.43 to 21.04 (Table 3), indicating high salinity levels. The study by Goodarzi et al. [80]. emphasizes the importance of implementing Best Management Practices (BMPs) to enhance groundwater quality. These include reducing the use of agricultural fertilizers and pesticides, as well as ensuring effective waste management. Moreover, regulating anthropogenic activities such as mining, overexploitation of water resources, and landfilling is crucial in mitigating groundwater contamination and promoting long-term water quality sustainability.

3.5.6. Total Hardness (TH)

Total Hardness (TH) is determined by the concentrations of calcium and magnesium in groundwater. Water hardness is categorized into four classes: soft (<75 mg/L), moderately hard (75–150 mg/L), hard (150–300 mg/L), and very hard (>300 mg/L) [81]. In the Ras El Aioun region, TH ranges from 12 to 99 mg/L. Table 3 shows the classification of groundwater quality in the research area, determined by the hardness percentage. Notably, 90% of groundwater samples in the central region were categorized as soft (TH < 75 mg/L), while the remaining 10% fell into the moderately hard category (TH 75–150 mg/L).

The elevated TH (

Table 4. Irrigation water classification according to SAR, KR, Na%, and TH values [37,41,81].

	Value	Class
SAR	<10	Excellent
	10–18	Good
	18–26	Fair
Total Hardness (TH) mg/L	<75	Soft
	75–150
	150–300	Hard
KR	KR > 1	Unsuitable
	KR < 1	Safe
PI	>75	Good
	25–75%	Doubeful
	<25	Unsuitable
RSC	<1.25	Good
	1.25–2.5	Doubeful
	>25	Unsuitable
Na%	<20	Excellent
	20–40	Good
	40–60	Permissible
	60–80	Doubtful
	>80	Unsuitable

) values are attributed to the dissolution of bicarbonate salts of calcium and magnesium and the weathering of limestone in the region. These findings provide valuable insight into groundwater quality for irrigation in the Ras El Aioun region and underscore the importance of continuous monitoring and effective management practices to ensure sustainable agriculture.

3.6. Machine Learning Analysis and Modeling

This study evaluates the suitability of groundwater for irrigation in the Ras El-Aioun plain, northwestern Algeria, using four machine learning models: Support Vector Regression (SVR), Random Forest, Gradient Boosting, and KNNs. The primary goal is to predict the Irrigation Water Quality Index (IWQI) based on key groundwater quality parameters. The models were trained on 30 groundwater samples, and their predictive performances were assessed using key evaluation metrics, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²).

Table 5. Training model performance results.

	Parameters	SVR				Random Forest
		RMSE	R2	MSE	MAE	RMSE	R2	MSE	MAE
Train	KR	2.8576	0.534	8.1658	1.1164	0.5244	0.9843	0.275	0.2422
	SAR	0.1651	0.9617	0.0273	0.1273	0.1757	0.9566	0.0309	0.1337
	RSC	6.7181	0.3359	45.1329	3.398	1.1067	0.982	1.2247	0.6338
	TH	22.4933	0.0321	505.9487	12.2171	4.564	0.9602	20.8298	2.1571
	Na%	2.0685	0.8071	4.2786	1.3398	0.3572	0.9942	0.1276	0.2762
	PI	1.9326	0.8158	3.7348	1.2425	0.381	0.9928	0.1452	0.2892
Test	KR	5.4267	0.1467	29.449	4.2197	1.1899	0.959	1.4159	0.8923
	SAR	0.2129	0.8143	0.0453	0.1736	0.4006	0.3424	0.1605	0.3636
	RSC	12.9647	−0.2189	168.0836	9.2305	2.8167	0.9425	7.9339	2.125
	TH	36.8698	−0.4876	1359.3847	27.5768	4.452	0.9783	19.8202	3.7867
	Na%	5.3252	0.5606	28.3582	4.3102	2.6902	0.8879	7.2372	1.7011
	PI	5.0522	0.5756	25.5249	4.0903	2.7844	0.8711	7.7529	1.9272
		Gradient Boosting				KNNs
		RMSE	R2	MSE	MAE	RMSE	R2	MSE	MAE
Train	KR	0.0005	1	0	0.0003	1.8661	0.8013	3.4822	1.0581
	SAR	0.0048	1	0	0.004	0.3791	0.7979	0.1437	0.2999
	RSC	0.0206	1	0.0004	0.017	3.4421	0.8257	11.848	2.0071
	TH	0.0048	1	0	0.0034	10.5775	0.786	111.883	5.8167
	Na%	0.0005	1	0	0.0003	1.5969	0.885	2.5501	1.1928
	PI	0.0009	1	0	0.0006	1.5212	0.8859	2.3141	1.1102
Test	KR	1.2763	0.9528	1.6291	0.9413	2.3682	0.8375	5.6086	1.971
	SAR	0.4322	0.2347	0.1868	0.3539	0.4786	0.0618	0.229	0.4547
	RSC	1.9734	0.9718	3.8942	1.5466	4.5627	0.849	20.8187	3.2015
	TH	6.1662	0.9584	38.0217	4.7814	9.3748	0.9038	87.8867	6.1
	Na%	2.3419	0.915	5.4846	1.5938	3.313	0.8299	10.9762	2.3841
	PI	2.4017	0.9041	5.7681	1.6547	3.2857	0.8205	10.796	2.3724

presents the model training results for predicting the SAR, RSC, KR, Na%, PI, and TH as main parameters for IWQI assessment. For model development, our dataset (n = 30 samples) was partitioned into training (80%, n = 24) and testing (20%, n = 6) sets using a stratified random split to maintain parameter distributions, with the training set used for model fitting and hyperparameter optimization via five-fold cross-validation to prevent overfitting, while the independent test set was used for the final evaluation of model generalizability on unseen data, ensuring reliable performance assessment despite the moderate sample size.

The predictive performance of machine learning models for estimating water quality parameters—SAR, Na%, PI, RSC, KR, and TH—was evaluated through observed vs. predicted scatter plots (Figure 9). The results suggest that machine learning models can effectively predict certain water quality parameters, but their performance varies depending on the complexity of the variable. The strong correlation for SAR, Na%, and TH suggests that these parameters follow well-defined patterns that the models can learn effectively. However, the higher errors observed in PI, KR, and especially RSC indicate that these parameters may be influenced by additional, unaccounted-for factors or exhibit more complex, nonlinear relationships that the models struggle to capture.

In this study, the results indicate that SAR consistently achieved the lowest error metrics and the highest correlation values (r~0.96–1.00) in both the training and testing phases, making it the most reliable model. In contrast, the TH and RSC models exhibited significant overfitting, with high RMSE and MSE values and a drastic drop in correlation coefficients during testing (TH: r = −0.4876, RSC: r = −0.2189), indicating poor generalization. KR and Na% performed moderately well, maintaining relatively stable error metrics but showing slight declines in correlation.

In the study region, the Jurassic limestone and dolomite formations, particularly in the Hodna and Belezma Mountains, contribute significantly to groundwater hardness through carbonate weathering, as follows:

CaCO₃ + H₂CO₃ → Ca²⁺ + 2HCO₃⁻

MgCO₃ + H₂CO₃ → Mg²⁺ + 2HCO₃⁻

The aquifers in the Tazila anticline and the Foural-Talkhempt massif, composed of fractured limestone and sandstone, exhibit high TH due to enhanced rock–water interactions and longer residence times. Evaporite deposits (e.g., gypsum and anhydrite) can further increase hardness. However, impermeable layers such as Albian marls can restrict groundwater flow and limit mineral dissolution, contributing to spatial variability in TH. Machine learning models in the study showed overfitting in TH predictions, as indicated by a substantial drop in test phase correlation (r = −0.4876). This suggests unaccounted geological heterogeneity, such as localized dissolution processes or differential recharge from carbonate-rich formations. To improve TH prediction accuracy, it is essential to integrate additional hydrogeological parameters, including aquifer permeability, groundwater flow velocity, and residence time.

RSC is governed by bicarbonate-rich lithologies and is calculated as follows:

RSC = (CO₃²⁻ + HCO₃⁻) − (Ca²⁺ + Mg²⁺)

The groundwater in Ras El-Aioun has distinct hydrochemical characteristics influenced by local geology and hydrological processes. Elevated Residual Sodium Carbonate (RSC) values occur predominantly in areas underlain by Cretaceous limestone and marl formations, where the dissolution of carbonate minerals leads to bicarbonate accumulation. However, machine learning models significantly overfitted RSC predictions (r = −0.2189), suggesting complex, non-linear hydrogeochemical interactions. These include cation exchange processes (Na⁺ replacing Ca²⁺ and Mg²⁺ in clay-rich aquifer sediments), evapotranspiration concentrating bicarbonate in shallow groundwater, and variable recharge dynamics periodically diluting carbonate concentrations.

Notably, high sodium percentage (Na%) values in certain wells indicate multiple contributing factors: dissolution of evaporite deposits, silicate weathering reactions, and enhanced cation exchange capacity in fine-grained Mio-Plio-Quaternary alluvial sediments. The abundance of clay minerals in these formations promotes sodium retention through exchange reactions, which can be represented by the following equation:

2Na²⁺ + Ca-X₂ → 2Na-X + Ca⁺

where X represents clay exchange sites

Direct dissolution of halite and gypsum deposits releases sodium ions into groundwater, increasing Na% concentrations. Although the machine learning models predicted Na% with relative stability, slight declines in correlation coefficients suggest minor, unaccounted-for influences, possibly related to hydraulic gradients or anthropogenic sources such as irrigation return flows and fertilizer application.

In contrast, the SAR displayed exceptionally strong correlations with the observed data (r ≈ 0.96–1.00), indicating that dominant geochemical processes were effectively captured. These processes include carbonate weathering (particularly the dissolution of dolomite and limestone), which supplies Ca²⁺ and Mg²⁺ to counteract sodium effects, ion exchange reactions where Na⁺ replaces divalent cations in clay-rich sediments, and saline intrusion or evaporation, which can increase sodium dominance. The highest SAR values were observed in aquifers characterized by limited calcium input and elevated sodium content, such as in tertiary and quaternary alluvial deposits, with additional contributions from evaporite dissolution. The high predictive accuracy of the model for SAR demonstrates its effectiveness in capturing the principal hydrogeochemical processes that control sodium behavior in the study area.

The results of this research were compared with findings from previous studies to better understand groundwater quality trends. For example, a study conducted by Gaagai et al. [82] found that groundwater in similar arid and semi-arid regions showed moderate to high limitations for irrigation use. This aligns closely with our findings, where several water quality indicators also pointed to varying degrees of irrigation constraints. The comparison supports the idea that groundwater quality challenges in these regions are influenced by similar environmental and human factors.

4. Conclusions

This study presents a comprehensive assessment of groundwater suitability for irrigation in a semi-arid region by integrating geochemical characterization, water quality index (WQI) analysis, and machine learning (ML) modeling. Thirty groundwater samples were collected and analyzed to evaluate major ion chemistry, hydrochemical facies, and irrigation water quality parameters.

The dominant hydrochemical facies was identified as the SO₄–Ca type, accounting for 93.3% of the samples. The average concentration order of ions was found to be HCO₃⁻ > Cl⁻ > SO₄²⁻ for anions and Ca²⁺ > Na⁺ > Mg²⁺ > K⁺ for cations. Gibbs diagram analysis revealed that the rock–water interaction governed the geochemistry of 63.33% of the samples, whereas evaporation influenced the remaining 36.66%, indicating processes such as carbonate dissolution and salt accumulation due to high evaporation rates.

Most chemical parameters were within acceptable limits according to FAO [52] and WHO [83] guidelines for irrigation use. However, elevated concentrations of potassium (K⁺) and magnesium (Mg²⁺)—likely resulting from anthropogenic sources such as agricultural runoff—pose potential long-term risks of soil salinization and reduced permeability. Water quality indices, including SAR, RSC, TH, and KR, indicated spatial variability in irrigation suitability, ranging from excellent to marginal across the study area.

Among the applied machine learning models, XGBoost demonstrated the highest predictive accuracy in classifying irrigation suitability, followed by RF KNNs and SVR. These results underscore the value of integrating traditional hydrochemical methods with data-driven approaches to improve decision-making in groundwater resource management.

This study offers a transferable framework for sustainable groundwater management in agricultural regions facing similar water quality challenges. Future research should aim to enhance model generalizability through ensemble techniques, incorporating additional environmental variables such as climate data and land-use patterns, and validate predictions using field-based crop response studies.