Groundwater Quality Assessment for Drinking and Irrigation Purposes at Al-Jouf Area in KSA Using Artificial Neural Network, GIS, and Multivariate Statistical Techniques

Abstract

Groundwater is an essential resource for drinking and agricultural purposes in the Al-Jouf region, Saudi Arabia. The main objective of this study is to assess groundwater quality for drinking and irrigation purposes in the Al-Jouf region. Physicochemical characteristics of groundwater were determined, including total dissolved solids (TDS), pH, electric conductivity (EC), hardness, and various anions and cations. The groundwater quality index (WQI) was calculated to determine the suitability of groundwater for drinking purposes. The EC, sodium percentage (Na⁺ %), magnesium hazard (MH), sodium adsorption ratio (SAR), potential salinity (PS), and Kelley’s ratio (KR) were assessed to evaluate the suitability of groundwater for irrigation. Effective statistical tests and Feed-forward neural network (FFNN) modeling were applied to reveal the correlation between parameters and predict WQI. The results indicated that approximately all samples are appropriate for drinking and irrigation uses except samples of the Al Qaryat region. The ionic abundance ranking was Na⁺ > Ca²⁺ > Mg²⁺ > K⁺ for cations, and Cl⁻ > SO₄²⁻ > NO₃⁻ for anions. Moreover, the groundwater is dominated by alkali metals (K⁺ and Na⁺) and controlled by the rock–water interaction process. The indicators of groundwater quality for irrigation and drinking according to the following criteria (Na⁺ %, SAR, KR, MH, PS, WQI (WHO), and WQI (BIS)) can be predicted by the FFNN with root mean square errors (RMSE) of 0.136, 0.070, 0.022, 0.073, 2.45 × 10⁻³, 1.45 × 10⁻², and 1.18 × 10⁻², respectively, and R² of 0.99, 1.00, 0.99, 0.99, 1.00, 1.00, and 1.00, respectively.

1. Introduction

In the Kingdom of Saudi Arabia (KSA), demand for drinking and irrigation water increases due to population increase and agricultural expansion [1]. The KSA suffers from water scarcity and depends mainly on seawater desalination and/or groundwater for domestic purposes [2]. Desalinated water and groundwater represented 63% or 2.14 billion m³ and 37% or 1.26 billion m³ of the water distributed per year, respectively [1]. Groundwater is generally used for domestic, drinking, industrial, and agricultural purposes. Around 97% of irrigation water comes from groundwater resources [3]. Groundwater is also favored for drinking purposes because of its safety and quality compared to surface water; however, human activation increased affected its quality and added a lot of physical, chemical, and microbial pollutants [4,5,6].

Groundwater pollution is a global problem with significant effects on human health and environmental security [7]. The quality of groundwater can be influenced by different factors, such as geology, weathering system, the amount of recharged water, and the interaction between rock and water [5,8,9]. Sources of aquifer degradation include lithogenic, anthropogenic, and seawater intrusion [1]. Anthropogenic pollution comes from industrial effluents, fertilizers, pesticides, domestic wastewater, and landfills; meanwhile, geogenic pollution derives from geological sources [10,11]. Carbonate sedimentary rocks enriched with gypsum mineral is a geogenic source of sulfate [12]. Magnesium may be derived from geogenic (ion exchange, ferromagnesium minerals, and seawater) or anthropogenic sources such as industrial wastage and mining activities. The presence of CO₂ in the soil zone and ion exchange can be a source of calcium in the groundwater [13]. Sodium may be attributed to persistent evaporation and cation exchange and can also be derived from processes such as silicate weathering and reactions of cation exchange [14]. Chlorine can be derived from some anthropogenic sources or from different sources, such as weathering of minerals and leaching [15]. Exposure to potentially toxic chemicals can cause serious adverse human health effects, including a variety of cancers, intellectual disabilities, and neurological, cardiovascular, kidney, and bone diseases [8].

Thus, there is more attention to irrigation as well as drinking water quality guidelines according to national and international standards such as Saudi Standards, Metrology and Quality Organization (SASO), Bureau of Indian Standard (BIS), and World Health Organization (WHO) [16,17,18]. The water quality index (WQI) is essential to evaluate the water quality that can be compared with national or international standards [19]. Also, the quality of groundwater for irrigation purposes can be assessed using various indicators such as electric conductivity (EC), sodium percentage (Na⁺ %), magnesium hazard (MH), potential salinity (PS), sodium adsorption ratio (SAR), and Kelley’s ratio (KR) [20,21,22].

Many studies investigated groundwater quality for agriculture and drinking purposes in some regions in KSA. Most of the groundwater pollutants in several cities of KSA are of natural origin [2]. More than two-thirds of wells in the Albaha region are inappropriate for irrigation and drinking uses [4]. Most of the groundwater in the Khulais region, Makkah Province, was unsuitable for irrigation purposes [1]. The water quality of groundwater is acceptable and could be used safely for drinking and domestic purposes in the Jazan region [23]. The groundwater is suitable for irrigation; however, it is unsafe water for drinking at the Salilah village, Madinah Munawara [22].

In fact, artificial neural networks (ANNs) are widely used in different environmental applications to examine the match between the measured and predicted concentrations of certain important parameters [24,25,26,27]. The acceptable performance of ANNs in modeling is two-fold; first, it supports the theoretical analysis, and second, it provides a helpful model to predict the level of output parameters for similar input data. In terms of groundwater management, GIS/ANN modeling is an alternative data analysis to obtain fast results using a less complicated approach whose results are satisfactory [28,29].

Al-Jouf region is one of the agricultural areas in KSA [3]. Due to the increase in drinking water requirements and the unexpected acceleration of agricultural activities in the Al-Jouf area, there is an urgent interest in studying groundwater quality for irrigation and drinking purposes. Moreover, the studies related to groundwater quality in the Al-Jouf region are very limited. Therefore, the main objective of this study is to assess groundwater quality for drinking and irrigation purposes in the Al-Jouf region in KSA. This can be achieved by analyzing some important physical and chemical parameters of groundwater and evaluating WQI, then comparing it with national and international standards. Multivariate statistical techniques to reveal correlations between parameters and GIS for the spatial distribution of parameters and groundwater quality will be applied. Moreover, this study will contribute to testing Feed-forward neural network (FFNN) model to forecast and predict groundwater quality indices for irrigation and drinking purposes. The effect of the important design parameters of FFNNs, such as the number of hidden layers, training algorithm, and activation functions, will be considered.

Geological setting: Al-Jouf region is differentiated geologically by several lithological units. The Sirhan Formation, which dates from the Miocene to the Pliocene Epochs, is composed of friable calcareous sandstone, limestone, and shale layers with small chert and clay beds [30]. The Upper Tertiary and Quaternary rocks of Harrat al Harrah are formed of flood basalts that overlie Tertiary sedimentary rocks (Sirhan Formation) [31]. Calcareous and gypsiferous duricrusts, gravel, eolian sand, alluvium, and sabkha deposits are among the Quaternary deposits. These deposits are difficult to trace individually because they overlap and have broad and ambiguous borders [32].

Unconsolidated gravel deposits, eolian sand, and alluvium (silt, sand, and gravel) dominate the region. The calcareous duricrust is large, cemented, hard, and weathering rubbly, resembling massive limestone. Eolian deposits are typically found atop alluvial materials such as sand sheets or dunes. These dunes are a mix of barchanoid ridge-type and linear (sief) dunes that move east to southeast due to northwesterly and southwesterly wind patterns. In the Wadi as Sirhan and the smaller basins, alluvial silt, sand, and gravel create enormous deposits. Interlayered sheets and lenticular masses of poorly sorted gravels and gravelly sands, moderately sorted sands, and clayey silts dominate these deposits. The notable elongated sabkhas are formed of silt and clay layers interbedded with gypsum and calcite and are situated along the northeast trending axis of the Wadi as Sirhan [3]. Figure 1 shows geological provinces, geological units, volcanics, and intrusives at the Al-Jouf region according to Open-File Report 97-470B by USGS [33].

2. Material and Methods

Figure 2 shows the methodology of the current study, which includes groundwater sampling and analysis, comparing the obtained results with national and international standards, statistical analysis for the different parameters, modeling, and optimization of the different indices, and spatial distribution for sample locations, different parameters, and water quality index using GIS.

2.1. Water Sampling and Analysis

The current study was performed in the irrigated areas of the Al-Jouf region in KSA. A total of 47 selected samples of groundwater were gathered from agriculture farms at different locations, see Figure 3a,b. The farms varied from small (5–10 ha) to medium (20–25 ha) and big size (50–100 ha). The wells varied in depth from 100 to 300 m. The sampling was running for at least 2–3 h to get a representative sample from each agriculture farm. A 300 mL capacity sample was collected in a sterile plastic bottle from each well and sealed properly, then stored in an ice box until subsequent laboratory analysis.

All water samples were analyzed to measure the concentration of different cations (Ca²⁺, Na⁺, K⁺, and Mg²⁺) using 5110 ICP-OES (Agilent technologies, Santa Clara, CA, USA) and anions (Cl⁻, SO₄²⁻, and NO₃⁻) using Ion Chromatography (Dionex, ICS 1600) (Thermo Scientific, Santa Clara, CA, USA). In addition, Orion 5-star plus instrument (Thermo Scientific, Santa Clara, CA, USA). has been used for measuring pH, EC, TDS, and hardness. Three measurements per sample were taken, and the average value was calculated.

2.2. Water Quality Index

2.2.1. Drinking Purpose

The groundwater quality index for drinking use was calculated based on drinking water standards of BIS and WHO. Ten important physicochemical parameters were used for calculating WQI (

Table 1. Weight and relative weight of the used parameters and desirable values for BIS and WHO.

Parameters	Weightage (wi) [10]	Relative Weigh (Wi)	Desirable Values
			BIS [16]	WHO [17]
pH	3	0.09	8.50	8.50
TDS (ppm)	4	0.09	500	500
Hardness (ppm)	3	0.09	200	100
Ca2+ (ppm)	3	0.09	75	75
Na+ (ppm)	2	0.06	200	200
K+ (ppm)	2	0.13	12	12
Mg2+ (ppm)	3	0.06	30	50
Cl− (ppm)	4	0.13	250	200
SO42− (ppm)	3	0.09	200	200
NO3− (ppm)	5	0.16	45	50
	∑ wi = 32	∑ Wi = 1

). WQI has been calculated according to a previous study [20].

2.2.2. Irrigation Purpose

To determine the suitability of groundwater for irrigation purposes, important parameters for groundwater quality such as EC, Na⁺ %, SAR, MH, KR, and PS were calculated according to [20,36,37,38,39] using the following equations:

$${\mathrm{Na}}^{+} \%=\frac{{\mathrm{Na}}^{+}+{\mathrm{K}}^{+}}{{\mathrm{Ca}}^{2+}+{\mathrm{Mg}}^{2+}+{ \mathrm{Na}}^{+}+{ \mathrm{K}}^{+}} \times 100$$

(1)

$$\mathrm{MH}=\frac{{\mathrm{Mg}}^{2+}}{\left({\mathrm{Ca}}^{2+}+{\mathrm{Mg}}^{2+}\right)} \times 100$$

(2)

$$\mathrm{SAR}=\frac{{\mathrm{Na}}^{+}}{\sqrt{\left({\mathrm{Ca}}^{2+}+{\mathrm{Mg}}^{2+}\right)/2}}$$

(3)

$$\mathrm{PS}={\mathrm{Cl}}^{-}+\frac{1}{2}{\mathrm{SO}}_4^{2-}$$

(4)

$$\mathrm{KR}=\frac{{\mathrm{Na}}^{+}}{\left({\mathrm{Ca}}^{2+}+{ \mathrm{Mg}}^{2+}\right)}$$

(5)

2.3. Data Analysis Utilizing GIS

The spatial distribution of groundwater experimental data using GIS has been applied for making groundwater assessments. In the current study, location data of groundwater samples was mapped in Figure 3b through a point layer created by GIS software version 10.8. Each sample point received a different code, which was then inserted into the point attribute table. The database file has separate columns for each sampling site’s chemical parameter values and sample codes. The spatial distribution maps for the different parameters, and WQI were created using the geodatabase. The maps were illustrated using the inverse distance weighted (IDW) interpolation technique from the spatial analyst tools in the ArcGIS V10.8 toolbox has been utilized to produce the spatial distribution maps [40]. The inverse distance weighted technique is one of several interpolation techniques to map the parameters of water quality by interpolating data spatially, and it is one of the most widely use techniques in previously similar studies [41].

2.4. Multivariate Statistical Techniques

The statistical analysis was carried out using MS Excel and Statistical Package for Social Sciences (SPSS 22.0). Minimum (Min), maximum (Max), mean, and standard deviation (SD) of each parameter were calculated. Pearson’s correlation analysis was applied to recognize the relationship between the different parameters. Moreover, dimensionality reduction was performed by running principal component analysis (PCA) procedure. Then, factor analysis (FA) was carried out for more interpretation of latent variables in collected records. It was used to extract the principal factors based on eigenvalues.

The term hydrogeochemical facies indicates to chemical character of water solutions that are found in systems. Various techniques, such as statistical and graphical analysis, can be used to interpret hydrogeochemistry of groundwater, but Gibbs and Piper’s diagrams are commonly utilized to create hydrogeochemical facies [20]. So, they were used to study the similarities and differences in the composition of water and to classify them into certain chemical types.

2.5. Artificial Neural Network

Artificial neural networks (ANNs) are statistical modeling used in various engineering applications to model and predict different environmental problems [24,25,26,27]. In this study, three levels are featured in the Feed-forward neural network (FFNN) that consists of three layers, see Figure 4. The input layer is the layer from which the model’s inputs to the model originate. These inputs come from the analysis of TDS, pH, EC, hardness, and various anions and cations in groundwater samples. The WQI, Na⁺ %, SAR, KR, MH, and PS were calculated to determine the groundwater suitability for irrigation and drinking purposes. There are M neurons in the hidden layer with a particular activation function, as described by tansig in Equation (6), are present in the hidden layer. One neuron in the output layer has a purelin-like activation function, which is used to predict the suitability of groundwater.

$$\mathit{tansig}(x)=\frac{2}{\left(1+\exp \left(-2\ast \mathrm{x}\right)\right) - 1}$$

(6)

Finding the best weights and biases for the connections between layers in the network is the goal of network training, where $x$ is the input to this function. The neuron output can often be written as in Equation (7).

$$o_j=f({\sum }_{i=1}^k{W}_{ij}+b_j)$$

(7)

where,$o_j$. denotes the output of the j-th neuron in a layer that is not an input, $W_{ij}$. denotes the weight of the connection between the i-th neuron in the layer before, $b_j$. denotes the bias value, and $f(·)$ denotes the activation function.

The training process of an FFNN-based predictive model is shown in a flowchart in Figure 5. The training (70%), validation (15%), and testing groups are randomly formed from the data samples (15%). A range of 70–30% is a common test criterion for neural network models in diverse applications. It is widely used in similar previous studies [25,26]. Starting from a random set of weights and biases for all compounds, the FFNN is trained. The root mean square error (RMSE), given in Equation (8), and the coefficient of determination R² are used to evaluate the performance of the model, given in Equation (9).

$$\mathrm{RMSE}=\sqrt{\frac{1}{\mathrm{N}}{\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{T}}_{\mathrm{i}} - {\mathrm{O}}_{\mathrm{i}}\right)}^2}$$

(8)

$${\mathrm{R}}^2 =1 - \frac{\sum _{\mathrm{i}=1}^{\mathrm{N}}{\left({\mathrm{T}}_{\mathrm{i}} - {\mathrm{O}}_{\mathrm{i}}\right)}^2}{\sum _{\mathrm{i}=1}^{\mathrm{N}}{({\mathrm{T}}_{\mathrm{i}} - \stackrel{\mathrm{-}}{\mathrm{T}})}^2}$$

(9)

where N is the sample size (training, validation, or test), ${\mathrm{T}}_{\mathrm{i}}$ is the target value (i.e., the outcome of the experiment), $\overline{\mathrm{T}}$ represents the mean of target values, and ${\mathrm{O}}_{\mathrm{i}}$ is the output of the corresponding model. To track model behavior and avoid overfitting, the RMSE is calculated for both the training set and the validation set. After the training phase is completed, the RMSE for the test set is determined. The entire procedure is repeated for a predetermined number of runs to determine the average model behavior.

3. Results and Discussion

3.1. Groundwater Parameter Analysis

To determine groundwater suitability for drinking, groundwater parameters were compared with the BIS, WHO, and SASO standards [16,17,18], as shown in

Table 2. Descriptive statistical for physicochemical parameters and standard limits for drinking use.

Sample	Unit	Min	Max	Mean	SD	BIS [16]		WHO [17]		SASO [18]
						D	P	D	P	D	P
pH	-	6.92	8.46	7.74	±0.34	6.5	8.5	6.5	8.5	6.5	8.5
EC	(µS/cm)	142.00	19,470.00	2194.47	±3483.80	-	-	-	-	-	-
TDS	(ppm)	71.00	9730.00	1090.40	±1723.08	500	2000	500	1500	1000	-
Hardness	(ppm)	26.82	4594.32	506.36	±803.91	200	600	100	500	-	-
Ca2+	(ppm)	5.30	1010.40	103.93	±172.86	75	200	75	200	200	-
Na+	(ppm)	5.30	2815.10	275.74	±519.32	-	200	-	200	200	-
K+	(ppm)	1.60	138.30	23.88	±29.14	-	12	-	12	-	-
Mg2+	(ppm)	3.10	503.00	59.94	±91.71	30	100	50	150	150	-
Cl−	(ppm)	0.17	61.65	4.77	±10.06	250	1000	200	600	250	-
SO42−	(ppm)	0.30	19.96	3.06	±4.55	200	400	200	400	250	-
NO3−	(ppm)	0.12	0.91	0.30	±0.21	45	-	50	-	50	-

Note(s): D: Desirable limits. P: Permissible limits.

. Figure 6 illustrates the characteristics and variations of the different parameters. The Ca²⁺ concentration is useful for the growth of teeth and bones [13]. The Ca²⁺ concentration in the samples ranged from 5.30 to 1010.40 ppm, with an average of 103.93 ppm. Almost all samples have Ca²⁺ concentration within the permissible limit (200 ppm), except samples in the Al Qaryat region (northern west of Al-Jouf region), see Figure 7a. The Na⁺ concentration varied from 5.30 to 2815.10 ppm, with an average of 275.74 ppm. Most of the samples have Na+ concentration below the permissible limit of 200 ppm, while a Na⁺ concentration higher than the permissible limit was detected for samples at the Al Qaryat region, as observed in Figure 7b. K⁺ concentration varied from 1.60 to 138.30 ppm, with an average of 23.88 ppm. More than 50% of the samples were found to have K⁺ concentration above the permissible limits (12 ppm); see the spatial distribution of K⁺ concentrations in Figure 7c. Mg²⁺ concentration ranged between 3.10 and 503.00 ppm, with an average of 59.94 ppm. The samples were found to be below the permissible limit, while samples in the Al Qaryat region showed higher Mg²⁺ concentrations than the permissible value of 150 ppm, see Figure 7d. The abundance order of cations was Na⁺ > Ca²⁺ > Mg²⁺ > K⁺.

High concentrations of SO₄²⁻ indicate that groundwater is unsuitable for use [12]. Its concentration ranged from 0.30 to 19.96 ppm, with an average of 3.06 ppm, indicating very low concentrations below the permissible limits; see the spatial distribution of SO₄²⁻ in Figure 7e. NO₃⁻ concentration is considered a groundwater pollutant due to industrialization, agricultural activities, and population growth increase [14]. The high nitrate concentration in groundwater results in health problems such as cancer and Blue baby disease [42]. Its concentration varied from 0.12 to 0.91 ppm with an average of 0.30 ppm, indicating very low concentrations below the permissible limits; see spatial distribution in Figure 7f. The high concentration of Cl⁻ in groundwater creates a salty taste and affects human health, causing kidney stones [15]. Its concentration ranged from 0.17 to 61.65 ppm, with an average of 4.77 ppm. All the groundwater samples showed values below the permissible limits; see its spatial distribution in Figure 7g. The results indicate the abundance of anions as Cl⁻ > SO₄²⁻ > NO₃⁻ in the study area.

The groundwater water hardness is due to the presence of magnesium and calcium, and other metal ions [20]. Almost all the samples are below permissible limits and suitable for drinking purposes, while the Al Qaryat region showed hardness values higher than the permissible value of 600 ppm; see Figure 7h. The pH of groundwater varied from 6.92 to 8.46, with a mean value of 7.74, indicating a slightly alkaline nature of groundwater. All pH samples are within permissible limits of the standards; see Figure 7i. The values of EC for groundwater samples varied from 142 to 19470 μS/cm and had a mean value of 2194.47 μS/cm. The high value of EC implies high mineral content and salinity of groundwater with high infiltration and low runoff [43]. The groundwater type can be classified as follow: low salt concentrations (<1500 μS/cm), medium salt enrichment (1500–3000 μS/cm), and high salt enrichment (>3000 μS/cm) [44]. Almost all samples of groundwater have low salt enrichment except samples in the Al Qaryat region; see Figure 7j. TDS in groundwater includes all inorganic salts, which indicate the salinity and water suitability for human consumption [17]. TDS varied from 71 to 9730 ppm, with a mean value of 1090.40 ppm. Most samples were below the permissible limits except the samples in the Al Qaryat region that have TDS values higher than 2000 ppm, as shown in Figure 7k.

3.2. Multivariate Statistical Analysis

3.2.1. Correlation Analysis

Pearson’s correlation matrices are indicated in Figure 8. The r value that is from (±0.80 to ±1.00) gives a strong correlation, (±0.50 to ±0.80) indicates a moderate relationship, and (±0.00 to ±0.50) indicates a weak relationship between parameters of water quality occurred. As illustrated in Figure 8, a strong positive correlation exists for the following: EC with Ca²⁺, Na⁺, Mg²⁺, K⁺, SO₄²⁻, Cl⁻, pH, and hardness; TDS with Ca²⁺, Na⁺, Mg²⁺, K⁺, SO₄²⁻, Cl⁻, pH, hardness, and EC; Cl⁻ with Ca²⁺, Na⁺, Mg²⁺, and hardness; hardness with Ca²⁺, Na⁺, Mg²⁺, and SO₄²⁻; SO₄²⁻ with Ca²⁺, Na⁺, Mg²⁺, and K⁺; both Mg²⁺ and K⁺ with Ca²⁺ and Na⁺; and finally, Ca²⁺ with Na⁺.

3.2.2. Principal Component Analysis

Figure 9 illustrates the results of PCA of the measured parameters, and the numerical results of FA are presented in

Table 3. Numerical results of the main three factors out of FA and the corresponding commonalities (Bold numbers are the most significant values).

Parameters	F1	F2	F3	Commonalities
Ca2+	0.972	−0.033	−0.202	0.986
Na+	0.992	0.022	−0.003	0.985
K+	0.824	0.129	0.511	0.956
Mg2+	0.989	-0.024	−0.045	0.981
SO42−	0.900	−0.008	0.287	0.892
NO3−	0.772	−0.365	−0.072	0.734
Hardness	0.986	−0.029	−0.130	0.991
Cl−	0.972	0.015	−0.163	0.971
pH	0.230	0.946	−0.104	0.959
EC	0.997	0.005	−0.034	0.994
TDS	0.996	0.007	−0.035	0.992
Eigenvalues	8.944	1.049	0.449
(%) of Variance	81.312	9.534	4.078
Cumulative (%) of Variance	81.312	90.846	94.924

. The three principal factors explain 94.92% of the total variance, where F1 explains 81.31% of total variance, F2 accounts for 9.53% of total variance, and F3 explains only 4.08% of the total variance. The parameters Ca²⁺, Na⁺, K⁺, Mg₂⁺, SO₄²⁻, NO₃⁻, Hardness, Cl⁻, EC, and TDS are most significant in the first component (F1) and highly positively correlated with each other and negatively correlated with pH, which is the most significant parameter in the second component (F2), while K⁺ has a notable contribution in F3.

3.2.3. Hydrogeochemical Facies

In the current study, the Piper diagram shows that the plotted samples fall in the sodium and potassium types; see Figure 10a. Gibbs plot can also be used to obtain a better understanding of hydrochemical processes such as precipitation, rock–water interaction, and evaporation in groundwater chemistry [45]. In this study, the ratio of major anions of the water samples calculated using Equation (10) was plotted against TDS values to indicate the mechanism controlling groundwater composition; see Figure 10b.

$$\mathrm{Major} \mathrm{anions}=\frac{{\mathrm{Na}}^{+}+{\mathrm{K}}^{+}}{\left({\mathrm{Ca}}^{2+}+{\mathrm{Na}}^{+}+{\mathrm{K}}^{+}\right)}$$

(10)

The results illustrate that almost all the samples point toward the water–rock interaction domain, with few samples trending toward the evaporation domain. Thus, chemical weathering of rock-forming minerals affects groundwater quality through rock dissolution caused by water below the surface circulates. The samples in the evaporation domain may be attributed to semi-arid environmental climate conditions and surface pollution sources, mainly excessive fertilizer usage, irrigation return flow, industrial outflows, and domestic discharges. Similar results have been found in previous studies [20,45,46].

3.3. Assessment of Groundwater Quality for Drinking

The calculated WQI values ranged from 14.80 to 897.61, with an average of 112.78 for BIS, and from 5.64 to 1050.84, with an average of 129.07 for WHO. The WQI was classified into five types: excellent water type (>50(, good water type (50 and 100), poor water type (100 to 200), very poor water type (200 and 300), and water as unsuitable for drinking < (300) [42]. Figure 11 shows radar maps of WQI based on WHO and BIS. According to WHO standards, 15 samples (31.91% of the total) are excellent, and 21 samples (44.68% of the total) are good water for drinking, see Figure 11a. Regarding BIS, 21 samples (44.68% of the total) are excellent, and 16 samples (34.04% of the total) are good water for drinking, as shown in Figure 11b. However, the Al Qaryat region had yielded WQI values of more than 300, so its water is not appropriate for drinking. Figure 12a,b show the spatial distribution of WQI for WHO and BIS, respectively.

3.4. Assessment of Groundwater Quality for Irrigation

Groundwater quality for irrigation use can be assessed by different parameters such as EC, Na⁺ % MH, SAR, PS, and KR. The groundwater classification for irrigation purposes based on these parameters is indicated in

Table 4. The classification of groundwater for irrigation purposes based on these parameters (EC, Na⁺ %, MH, SAR, PS, and KR).

Parameter	Range	Classification	Number of Samples	Percentage of Sample (%)
EC μS/cm	<700	Excellent	17	36.17
	700–3000	Good	22	46.81
	>3000	Fair	8	17.02
Na+ %	<20	Excellent	0	0.00
	20–40	Good	3	6.38
	40–60	Permissible	27	57.45
	60–80	Doubtful	15	31.91
	>80	Unsuitable	2	4.26
MH	<50	Excellent	45	95.74
	>50	Unsuitable	2	4.26
SAR	<10	Excellent	17	36.17
	10–18	Good	1	23.40
	18–26	Doubtful	9	19.15
	>26	Unsuitable	10	21.28
PS	<3	Excellent to good	30	36.17
	3–5	Good to injurious	7	23.40
	>5	Injurious to Unsuitable	10	21.28
KR	<1	Excellent	23	48.94
	>1	Unsuitable	24	51.06

; meanwhile, spatial distribution is illustrated in Figure 13. The EC represents the dissolved salt concentration in the groundwater and indicates salinity hazard for crops [20,47]. Around 83% of samples are excellent to good and suitable for crop growth. On the other hand, the sodium reacts with the soil and decreases its texture and permeability, so the high sodium percentage in water can cause soil degradation [47]. The values of Na⁺ % varied between 31.63% and 83.02% and had a mean value of 55.75%. Based on BIS standards, the water has Na⁺ % less than 60% suitable for irrigation. Most of the samples (64% of the total) yielded Na⁺ % less than 60%, indicating it is suitable for irrigation use.

MH is a function of Ca²⁺ and Mg²⁺, which are important nutrients for crops, and their high values in groundwater increase the soil pH; moreover, the high quantity of Mg²⁺ ions in the water of irrigation has a negative effect on the quality of soil, where it causes soil alkaline and decreases the crop production [48]. Around 95.74% of all water samples had MH values less than 50 and were appropriate for irrigation. SAR results showed that 36.17% of the samples are considered excellent water, and 23.40% of the samples represent good for irrigation use. Meanwhile, 21.28% of the samples have a high sodium hazard, so they are unsuitable for irrigation. Most of the unsuitable samples were collected from the Al Qaryat region. According to PS results, 63.83% of the samples are excellent to good, which indicates their suitability for irrigation use. However, 21.28% of the samples are inappropriate for irrigation, and most of them were collected from the Al Qaryat region. Regarding the KR ratio, 48.94% of the samples are less than 1 and suitable for irrigation.

3.5. Artificial Neural Network

The groundwater quality for drinking and irrigation purposes has been predicted using FFNN. Model performance is best assessed through multiple runs because the model starts with random weights for connections and randomly divides data patterns. After performing the entire training process in Figure 5 for 20 separate computer runs, the best, average, and standard deviation values of RMSE are presented. A computer with a 2.8 GHz core i5 and 8 GB of memory is used to train the analyzed models. ANN Toolbox, a piece of Matlab software R2020a, is used for simulation [49]. The simulations demonstrate how effectively the process can be modeled by the FFNN with 19 neurons in the hidden layer. The training approach is Bayesian regularization backpropagation, whereas the activation function of the hidden layer is the hyperbolic tangent sigmoid (tansig) (trainbr). To avoid overfitting, the model must be trained for an average of 25 epochs. According to Figure 14, the regression coefficient for each of the patterns’ training, validation, test, and total is above 0.999. The FFNN predictions for the entire data set are highlighted in Figure 15 because they agree well with the experimental patterns. Figure 16 shows the experimental results and the FFNN predictions for the test patterns. To choose the best configurations for the network and its training, the FFNN hyperparameters are also investigated. The numerical results of the hidden layer size evaluation for the test set for a range of values (10, 17, 19, 21, and 25) are shown in

Table 5. Effect of the number of neurons used in the hidden layer of each model.

Model	Measure	10	17	19	21	25
Na+ %	RMSE	0.831	0.136	0.336	0.572	0.456
	R2	0.965	0.990	0.991	0.985	0.988
SAR	RMSE	0.314	0.103	0.070	0.203	0.120
	R2	0.995	0.999	1.000	0.991	0.999
KR	RMSE	0.020	0.045	0.022	0.080	0.048
	R2	1.000	1.000	1.000	1.000	1.000
MH	RMSE	0.386	0.238	0.073	0.509	0.132
	R2	0.981	0.990	1.000	0.988	0.998
PS	RMSE	0.003	0.010	0.002	0.006	0.023
	R2	1.000	1.000	1.000	1.000	1.000
WQI (WHO)	RMSE	0.014	0.052	0.014	0.1114	0.073
	R2	1.000	1.000	1.000	0.999	1.000
WQI (BIS)	RMSE	0.327	0.042	0.011	0.246	0.087
	R2	0.987	1.00	1.00	0.990	1.00

. Compared to using only 19 neurons, using 21 neurons and 25 neurons in the hidden layer leads to a slight improvement in model performance, but the cost of training complexity increases. Performance suffers noticeably when there are 17 fewer neurons in the hidden layer (higher RMSE compared to using 19 neurons).

According to

Table 6. Effect of the training algorithm used in each model.

Model	Measure	trainlm	trainscg	trainbr
Na+ %	RMSE	0.136	1.493	1.349
	R2	0.998	0.886	0.984
SAR	RMSE	0.070	0.676	0.796
	R2	1.000	0.998	0.985
KR	RMSE	0.022	0.120	0.281
	R2	0.996	0.961	0.990
MH	RMSE	0.073	1.415	1.209
	R2	0.992	0.970	0.980
PS	RMSE	2.45 × 10−3	0.142	1.63 × 10−2
	R2	1.000	0.999	1.000
WQI (WHO)	RMSE	1.45 × 10−2	1.110	2.12 × 10−2
	R2	1.00	0.998	1.000
WQI (BIS)	RMSE	1.18 × 10−2	1.870	4.80 × 10−2
	R2	1.000	0.999	1.000

, the training algorithm (trainbr) performs better than (trainlm), Scaled Conjugate, and Gradient Backpropagation (trainscg). (tansig) is the best option for the activation function, while radbas or the triangular basis transfer function (tribas) give worse RMSE results, as shown in

Table 7. Effect of the hidden layer’s activation function in each model.

Model	Measure	Radbas	tribas	tansig
Na+ %	RMSE	2.602	1.285	0.136
	R2	0.996	0.970	0.998
SAR	RMSE	0.081	0.304	0.070
	R2	1.000	0.997	1.000
KR	RMSE	0.071	0.045	0.022
	R2	1.000	1.000	0.996
MH	RMSE	0.451	0.223	0.073
	R2	0.997	0.999	0.992
PS	RMSE	3.12 × 10−3	2.55 × 10−3	2.45 × 10−3
	R2	1.000	1.000	1.000
WQI (WHO)	RMSE	1.65 × 10−2	2.82 × 10−3	1.45 × 10−2
	R2	1.000	1.000	1.000
WQI (BIS)	RMSE	3.73 × 10−3	2.48 × 10−3	1.18 × 10−2
	R2	1.000	1.000	1.000

. In summary, the selection of the ideal FFNN training settings was made considering a wide range of options. The suitability of the developed FFNN for the existing experimental data shows that the model can be used to accurately predict the output of inputs under similar experimental conditions. In addition, the model may need to be re-trained and its structure modified if the experimental conditions or inputs change. However, the procedure described above for selecting the optimal regression model will be useful.

4. Conclusions

The results of physical and chemical analysis, WQI, and GIS showed that most of the groundwater samples in the studied area are suitable for drinking and irrigation purposes, except samples in the Al Qaryat region. According to WHO standards, 31.91% of samples were excellent, and 44.68% of samples were good water for drinking. Regarding BIS, 44.68% of samples were excellent, and 34.04% of samples were good water for drinking. Based on spatial distribution, unsuitable samples for drinking were observed in the Al Qaryat region. The ionic abundance ranking was Na⁺ > Ca²⁺ > Mg²⁺ > K⁺ for cations, and Cl⁻ > SO₄²⁻ > NO₃⁻ for anions. A strong positive correlation existed between the different parameters except for pH. The Piper diagram and Gibbs diagram indicate that the groundwater is dominated by alkali metals (K⁺ and Na⁺) and controlled by rock–water interaction process. The EC results showed that most of the samples were in the excellent to good category for irrigation except those of the Al Qaryat region. The Na⁺ % and MH indicated that 64.00% and 95.74% of samples were suitable for irrigation use, respectively. SAR showed that 36.17% of samples were excellent water, and 23.40% of the samples were good for irrigation use. PS and KR results depicted that 63.83% and 48.94% of the samples were suitable for irrigation use, respectively. Moreover, the applied FFNN was successfully established and showed high accuracy in predicting the indices Na⁺ %, SAR, KR, MH, PS, WQI (WHO), and WQI (BIS). The values of RMSE ranged between 0.136 and 2.45 × 10⁻³, whilst R² reached 1.00 for most output parameters.