Water Quality Evaluation and Prediction Using Irrigation Indices, Artificial Neural Networks, and Partial Least Square Regression Models for the Nile River, Egypt

Abstract

Water quality is identically important as quantity in terms of meeting basic human needs. Therefore, evaluating the surface-water quality and the associated hydrochemical characteristics is essential for managing water resources in arid and semi-arid environments. Therefore, the present research was conducted to evaluate and predict water quality for agricultural purposes across the Nile River, Egypt. For that, several irrigation water quality indices (IWQIs) were used, along with an artificial neural network (ANN), partial least square regression (PLSR) models, and geographic information system (GIS) tools. The physicochemical parameters, such as T °C, pH, EC, TDS, K⁺, Na⁺, Mg²⁺, Ca²⁺, Cl⁻, SO₄²⁻, HCO₃⁻, CO₃²⁻, and NO₃⁻, were measured at 51 surface-water locations. As a result, the ions contents were the following: Ca²⁺ > Na⁺ > Mg²⁺ > K⁺ and HCO₃⁻ > Cl⁻ > SO₄²⁻ > NO₃⁻ > CO₃²⁻, reflecting Ca-HCO₃ and mixed Ca-Mg-Cl-SO₄ water types. The irrigation water quality index (IWQI), sodium adsorption ratio (SAR), sodium percentage (Na%), soluble sodium percentage (SSP), permeability index (PI), and magnesium hazard (MH) had mean values of 92.30, 1.01, 35.85, 31.75, 72.30, and 43.95, respectively. For instance, the IWQI readings revealed that approximately 98% of the samples were inside the no restriction category, while approximately 2% of the samples fell within the low restriction area for irrigation. The ANN-IWQI-6 model’s six indices, with R² values of 0.999 for calibration (Cal.) and 0.945 for validation (Val.) datasets, are crucial for predicting IWQI. The rest of the models behaved admirably in terms of predicting SAR, Na%, SSP, PI, and MR with R² values for the Cal. and validation Val. of 0.999. The findings revealed that ANN and PLSR models are effective methods for predicting irrigation water quality to assist decision plans. To summarize, integrating physicochemical features, WQIs, ANN, PLSR, models, and GIS tools to evaluate surface-water suitability for irrigation offers a complete image of water quality for sustainable development.

1. Introduction

Agriculture is vital to the economy and the lives of its people. As a result, water resources are critical for socioeconomic growth and the environmental conservation [1,2]. In addition, water safety has become a serious issue, restricting agriculture’s sustainable and healthy expansion [3,4]. The freshwater availability restricts the sustainable growth of every country in the world, especially in semi-arid areas [5]. As a result, surface waters are a critical resource, providing important ecosystem services that support agricultural supplies in many regions [6]. Surface-water resources are suffering significant concerns including quality challenges and freshwater deterioration, which are related to both human and geogenic processes [7,8].

The sustainable agricultural sector has a crucial role to play in water management in Egypt. By encouraging the use of cutting-edge irrigation techniques, environmentally friendly agricultural methods, and the use of unconventional water sources for agriculture, such as treated wastewater and brackish water, it is feasible to minimize water use in the agricultural sector while maintaining or even increasing crop yields, which can help ease pressure on freshwater resources and guarantee that water is used more effectively. This can contribute to Egypt’s long-term economic and environmental sustainability and ensure that the country’s scarce water resources are used more sustainably and effectively [9]. In general, sustainable agricultural practices can aid in lowering water use, enhancing the quality of the water, and fostering a more reliable water supply for agricultural communities. The agricultural sector may contribute to sustainable water management and aid in ensuring a steady supply of water for future generations by implementing these practices [10].

Rivers are significant components for sustainability through terms of quality of life, as well as environmental and socioeconomic progress, since they are formed by water streams that are subject to anthropogenic, microbial, and contaminant influences [11]. For instance, natural process elements, including evaporation and rock–water interaction and seawater intrusion, have a substantial influence on water quality [12,13,14]. Furthermore, anthropogenic elements such as households, municipalities, animal fertilizers, and agricultural and manufacturing wastes have a substantial impact on water quality [15,16,17].

Egypt’s fresh water scarcity has become a serious concern, affecting crop production and other agricultural operations, whereas overpopulation, expansion, and manufacturing create additional strain on Egypt’s water resources represented in the discharge of inadequately treated sewage, which degrades water quality [18]. Since the completion of the High Aswan Dam, the Nile River’s water quality is usually influenced by the ecosystem and water features of the Nile’s higher reaches. Consequently, multiple environmental impacts in the Nile River ecosystem are observed [19].

In the most recent decades, rigorous regulations and limits have quickly increased in the monitoring of surface water due to water quality deterioration. Water quality is greatly influenced by the physicochemical water bodies and also by their connections with one another. Therefore, assessing water quality is crucial for controlling and preserving surface-water resources, which are becoming highly responsive to physiochemical stresses caused by humans [20]. Therefore, knowledge of the physicochemical variables and geochemical processes that characterize the surface-water features can help to visualize hydrochemical ecosystems, which, in turn, can help to enhance sustainability and water quality management [21].

In general, the comprehensive testing of physical and chemical water quality indicators is required for an accurate assessment of the surface-water standard. One of the main significant issues in evaluating water quality is the collection of parameters that can be continually observed associated with acquiring, analyzing, and evaluating. To address these concerns, irrigation water quality indices (IWQIs) were used to manage the effective water quality categorization through numerous characteristics that were extensively documented as helpful to address these concerns. The WQIs are methods for summarizing the data of several water quality criteria into a single number that expresses the water quality state [22]. These approaches are excellent methods for determining the water quality status of a specific region through terms of a single number and a ranked list.

The IWQIs were created by combining the physiochemical features of rivers and streams to offer a more precise evaluation of water quality variations in different points and can be used to describe water quality efficiently [23,24,25,26]. Consequently, the IWQIs are metrics that are used to describe the grade of these variables in order to assess the total water quality. Therefore, IWQI, SAR, Na%, SSP, PI, and MH are the most essential water quality measurements that define water quality status and appropriateness for agricultural activities [27].

A frequent water quality surveillance strategy is necessary not only to safeguard water savings and minimize water contamination but also to simulate contaminant dispersion, source site, and health risks [28]. The water quality tracking system allows for frequent sampling, physicochemical component assessment, and the transmission of the state of the surface water [29]. The monitoring data for water quality are produced by collecting and analyzing a particular number of physicochemical characteristics at different sample locations throughout the stream network with respect to the criteria of observable variables established by government systems [30,31]. The monitoring datasets for water quality have several constraints in terms of analyzing and generating meaningful findings since they often contain huge quantities of data and complicated interactions among the variables [32].

A mathematical model for translating the investigated parameters of water quality into a dimensionless single number provides an indication of degree of water quality for the terminal [33]. Various techniques such as ANN and PLSR models were established to analyze, assess, and enhance water quality for agricultural purposes [34,35]. These techniques were widely employed to forecast surface-water quality for irrigation purposes due to their enhanced efficacy and accessible availability of water quality data from many sources [36]. Hence, water quality modeling is critical in water quality management, assisting governments and decision makers in managing and improving water resources for humans and agriculture.

In order to manage the environment at a safe level, the prediction of WQIs is a crucial task. In this domain, the previous work has used a number of deterministic models [37,38,39,40,41]. However, the statistical effectiveness of these cutting-edge models is often poor since real-world natural ecosystems are frequently too complex for them. For creating models to estimate various IWQIs, ANN and PLSR may provide simple and reliable methods. A single dataset’s non-linear patterns may be generalized by ANN and PLSR, which can also resolve difficult problems [13]. These approaches can be used to resolve very nonlinear problems [42,43] and were successfully applied to assess the precision of the predicted constituents of water quality [44].

There is limited research examining the connection between the efficiency and precision of different multivariate approaches, particularly, ANN and PLSR, in assessing water quality for agriculture along the Nile River in Egypt. This assessment incorporates GIS data and physicochemical indicators such as IWQI, SAR, Na%, SSP, PI, and MR of surface water. Although these techniques are commonly used for evaluating water quality and demonstrate robust predictive abilities, identifying the best-performing algorithm remains essential. Due to the limitations of traditional approaches for managing surface-water quality, it is essential to use dependable, efficient, and quick methods that can be utilized and assist decision-makers in comprehensively assessing important metrics appropriate for water quality. This study aims to provide useful methods for making educated judgments on surface water to guarantee proper leadership management, identify IWQIs, and describe in detail how sampling practices may be adjusted. Therefore, this research was undertaken to (i) determine surface-water facies and geochemical regulating mechanisms via physical and chemical characteristics and illustrative approaches; (ii) appraise surface-water availability for irrigation using several IWQIs; (iii) determine the accuracy of utilizing the ANN and PLSR models to quantify IWQIs of surface water; and (iv) finally, discuss the feasibility of employing the ANN and PLSR models as advantageous methods for making informed judgments on IWQIs.

2. Materials and Methods

2.1. Sites Description

The Nile River is divided into two main streams, Rosetta and Damietta (Figure 1). The first is approximately 225 km long and 180 m wide, with a depth of 2 to 4 m, starts in EL-Kanater El-Khayria, and terminates at the Rosetta Estuary in Rashid City, 30 km upstream of the sea, where excessive water is released to the Mediterranean Sea through the Rosetta Estuary [45]. On the other site, the Damietta Branch is approximately 242 km in length, with an average width of 200 m and depth of 12 m [46]. The Damietta Branch is receiving a growing amount of sewage discharges from many pollution sources, including industrial, household, and farm [47]. The primary causes of pollution in the branch are the effluents of the fertilizer factory and the electric power station in Talkha, Mansoura city, as well as the discharge of several minor farms, household drains, and nearby communities. As previously stated, in the Delta region, which covers approximately 12,357 km² (Figure 1), the Nile River branches are the primary freshwater supply for most economic activity.

2.2. Samples Collection and Analysis

The surface water points (n = 51) were obtained from various sites along the Nile River branches throughout the year 2021 to explore the hydrochemical characteristics and estimate the IWQIs for determining the validity of surface-water resources for agriculture (Figure 1). The obtained samples were kept in 1000 mg plastic vials in a 4 °C refrigerator before being refined using a 0.45 m membrane for laboratory analyses of the main ions. Then, 13 physical and chemical characteristics were measured in 51 surface-water samples from both branches using standard analytical techniques [48].

Temperature (T°), pH, EC, and TDS were measured in situ utilizing a calibrated digital field multi-parameter sensor (Hanna HI 9033). Volumetric titration techniques were used to assess Ca²⁺, Mg²⁺, Cl⁻, HCO₃⁻, and CO₃²⁻ based on the APHA standard method [49]. The Ca²⁺ and Mg²⁺ were tested by the EDTA titrimetric method and Cl⁻ was assessed using the argentometric titrimetric method, while HCO₃⁻ and CO₃²⁻ were determined using phenolphthalein (P). A flame photometer (AFP 100, Hamburg, Germany) was utilized to assess K⁺ and Na⁺, while a spectrophotometer was used to measure SO₄²⁻ and NO₃⁻ (UV 1600 PC).

The charge balance errors (CBE) were calculated after the field measurements. The principle of neutrality states that the summation of cations (Na⁺, K⁺, Ca²⁺, and Mg²⁺) should be equal to the summation of anions (Cl⁻, SO₄²⁻, HCO₃⁻, CO₃²⁻, and NO₃⁻) in meq/L⁻¹. The error in anion–cation balance is evaluated using Equation (1). According to equation 1, quality control was completed having an error rate of less than 0.05 for all triplicate samples [50]. Quality assurance is achieved via the use of laboratory standard processes and quality control systems. The analytical technique was validated using the detection quality control criteria, such as external calibration, precision, accuracy %, linearity, detection, and quantitative limits.

$$\mathrm{CBE}=\frac{{\displaystyle \sum }\mathrm{Cations}-{\displaystyle \sum }\mathrm{Anions}}{{\displaystyle \sum }\mathrm{Cations}+{\displaystyle \sum }\mathrm{Anions}}\times 100$$

(1)

2.3. Indexing Approach

The appropriateness of the surface water in the research region for agricultural purposes was determined using Wilcox and USSL graphs, in addition to other irrigation indicators including IWQI, SAR, Na%, SSP, PI, and MH, as shown in

Table 1. The IWQIs, the equation, and the citations.

IWQIs	Unit	Formula	References
IWQI	mg/L	${\displaystyle \sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{Q}}_{\mathrm{i}}{\mathrm{W}}_{\mathrm{i}}$	[27]
SAR	meq/L	[Na+/√(Ca2+ + Mg2+)/2] × 100	[51]
Na%	meq/L	(Na+ + K+)/(Ca2+ + Mg2+ + Na+ + K+)] × 100	[52]
SSP	meq/L	[Na+/(Ca2++ Mg2++ Na+)] × 100	[52]
PI	meq/L	[(Na++√ HCO3−)/(Ca2+ + Mg2++ Na+)] × 100	[53]
MH	meq/L	[Mg2+/(Ca2++ Mg2+)] × 100	[54]

. The IWQIs are simple non-dimensional indices for measuring water quality depending on a variety of criteria [34]. All the anion concentrations of different physicochemical parameters obtained from surface-water sample examination were utilized to compute various IWQIs through using the referenced literature (Table 1).

Irrigation Water Quality Index (IWQI)

The IWQI is a reliable technique and an essential indicator of agricultural water consumption, reflecting the appropriateness of irrigated water for soil and plant health [55]. The final index was calculated by combining ratings and weights into a single dimensionless number, which ranged from 0 to 100, with high values indicating no restriction water quality. The IWQI was determined according to [27,56], as in the following equations:

$$\mathrm{IWQI}={{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{Q}}_{\mathrm{i}}{\mathrm{W}}_{\mathrm{i}}$$

(2)

where W_i represents the specified weight of each data set, and Q_i represents the value for quality measurement within the permissible range (Table 2).

$${\mathrm{Q}}_{\mathrm{i}}={\mathrm{Q}}_{\max }-\left(\frac{\left[\left({\mathrm{X}}_{\mathrm{ij}}-{\mathrm{X}}_{\inf }\right)\times {\mathrm{Q}}_{\mathrm{imap}}\right]}{{\mathrm{X}}_{\mathrm{amp}}}\right)$$

(3)

where:

• X_ij = the measured value for each variable.
• X_inf = the value indicating the lowest bound for the class.
• Q_imap = the frequency of the class.
• X_amp = the frequency class that the parameter belongs to.

Table 2. Different ranges for several variables in the quality measurement calculation (Q_i).

Qi	EC (µS/cm)	SAR	Na+ (emp)	Cl− (emp)	HCO3− (epm)
85–100	200 ≤ EC < 750	2 ≤ SAR < 3	2 ≤ Na < 3	1 ≤ Cl < 4	1 ≤ HCO3 < 1.5
60–85	750 ≤ EC < 1500	3 ≤ SAR < 6	3 ≤ Na < 6	4 ≤ Cl < 7	1.5 ≤ HCO3 < 4.5
35–60	1500 ≤ EC < 3000	6 ≤ SAR < 12	6 ≤ Na < 9	7 ≤ Cl < 10	4.5 ≤ HCO3 < 8.5
0–35	EC < 200 or EC ≥ 3000	SAR > 2 or SAR ≥ 12	Na < 2 or Na ≥ 9	Cl < 1 or Cl ≥ 10	HCO3 < 1 or HCO3 ≥ 8.5

Table 2. Different ranges for several variables in the quality measurement calculation (Q_i).

Qi	EC (µS/cm)	SAR	Na+ (emp)	Cl− (emp)	HCO3− (epm)
85–100	200 ≤ EC < 750	2 ≤ SAR < 3	2 ≤ Na < 3	1 ≤ Cl < 4	1 ≤ HCO3 < 1.5
60–85	750 ≤ EC < 1500	3 ≤ SAR < 6	3 ≤ Na < 6	4 ≤ Cl < 7	1.5 ≤ HCO3 < 4.5
35–60	1500 ≤ EC < 3000	6 ≤ SAR < 12	6 ≤ Na < 9	7 ≤ Cl < 10	4.5 ≤ HCO3 < 8.5
0–35	EC < 200 or EC ≥ 3000	SAR > 2 or SAR ≥ 12	Na < 2 or Na ≥ 9	Cl < 1 or Cl ≥ 10	HCO3 < 1 or HCO3 ≥ 8.5

Finally, the W_i values were estimated in the following equation:

$${ \mathrm{W}}_{\mathrm{i}}=\frac{{{\displaystyle \sum }}_{\mathrm{j}=1}^{\mathrm{k}}{\mathrm{F}}_{\mathrm{j}}{\mathrm{A}}_{\mathrm{ij}}}{{{\displaystyle \sum }}_{\mathrm{j}=1}^{\mathrm{k}}{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{F}}_{\mathrm{j}}{\mathrm{A}}_{\mathrm{ij}}}$$

(4)

where:

• F = the auto value of component 1.
• A = the factor j significantly limits parameter i.
• i = physicochemical variables count as selected by the model, which might range between 0 and 1 to n.
• j = the number of variables selected by the model on a scale from 1 to k_ij.

2.4. Back-Propagation Neural Network (BPNN)

The BPNN model is a popular ANN that is described as having three levels [57]. The ANN method is an example of ML strategy that makes use of numerous layers to extract high-level qualities from unprocessed input physicochemical variables (Figure 2). All the chosen parameters (Table 1) were incorporated in the models’ ANN as input datasets to predict the six IWQIs. In the process of training the feed-forward neural networks, back-propagation is frequently utilized. Such networks are composed of three unique layers: (1) the entry layer, functioning as the primary data provider for the system; (2) the concealed layer, serving as a bridge connecting the self-governing input layer to the relevant output layer; and (3) the result layer, responsible for producing output values corresponding to the specified inputs. The network is made up of two hidden layers, the number of which is determined by the regression’s accuracy [58]. The hidden layers correlate to the “activation” nodes and are sometimes referred to as “weight.” The intended value of the parameter is displayed in the output layer. The network was trained using at least 2000 iterations. The training dataset was verified using the LOOV technique to determine the quantity of neurons in the hidden layer. The Broyden–Fletcher–Goldfarb–Shanno (lbfgs) weight optimizer was used to assure the algorithm’s efficiency [59]. To enhance the forecast power of the regression model while reducing dimensionality, the most useful feature was determined using the following formula [60].

$$\mathrm{M}=\frac{{{\displaystyle \sum }}_{\mathrm{j}=1}^{{\mathrm{n}}_{\mathrm{H}}}\left[\left(\left|\mathrm{I}\right|{ }_{{\mathrm{P}}_{\mathrm{j}}}/{{\displaystyle \sum }}_{\mathrm{k}=1}^{{\mathrm{n}}_{\mathrm{p}}} \left|\mathrm{I}\right|{ }_{{\mathrm{P}}_{\mathrm{j},\mathrm{k}}} \right){\left|\mathrm{O}\right|}_{\mathrm{j}}\right]}{{{\displaystyle \sum }}_{\mathrm{i}=1}^{{\mathrm{n}}_{\mathrm{p}}}\left({{\displaystyle \sum }}_{\mathrm{j}=1}^{{\mathrm{n}}_{\mathrm{H}}}\left[\left(\left|\mathrm{I}\right|{ }_{{\mathrm{P}}_{\mathrm{i},\mathrm{j}}}/{{\displaystyle \sum }}_{\mathrm{k}=1}^{{\mathrm{n}}_{\mathrm{p}}} \left|\mathrm{I}\right|{ }_{{\mathrm{P}}_{\mathrm{i},\mathrm{j},\mathrm{k}}} \right){\left|\mathrm{O}\right|}_{\mathrm{j}}\right]\right)}$$

(5)

where M is the input variable’s significance level, ${\mathrm{n}}_{\mathrm{p}}$ is the total quantity of input variables, ${\mathrm{n}}_{\mathrm{H}}$ is the number of nodes in the hidden layer, $\left|\mathrm{I}\right|{ }_{{\mathrm{P}}_{\mathrm{j}}}$ is the weight of the concealed layer’s equivalent in line with the pth input parameter and the jth hidden layer, and ${\left|\mathrm{O}\right|}_{\mathrm{j} }$ is the weight of the output layer in absolute terms matching to the jth hidden layer.

One of the contributions of this study was to compare various attributes while training the BPNN model. By selecting the most significant hyperparameters, we optimized the model. These parameters included the number of neurons in two hidden layers (nr1 and nr2) and the activation function (fun). The essential steps of neural network training, proper hyperparameter estimation, and feature organization were accomplished in the research. A parameter tuning process was used to optimize the hyperparameters of a BPNN model. This involved iterating through different network sizes (the ranges of nr1 and nr2 to be set from 1 to 30) and functions (fun = “identity”, “logistic”, “tanh”, and “relu) to find the optimal combination that minimizes the root mean squared error of validation (RMSEV). The RMSEV serves as a metric to evaluate the accuracy of the model. The number of features in each iteration included color indices such as 13, 12, 11, 10, and so on. Generally, the variables were input into the model randomly in the initial iteration, while less important features were eliminated in subsequent iterations. The top-performing features were arranged in ascending order based on their highest contributions. Finally, the ANN outputs were compared to determine superior variables with the lowest RMSE.

2.5. Partial Least-Square Regression (PLSR)

The PLSR models were utilized in this work for forecasting the six IWQIs. All the chosen parameters (Table 1) were incorporated in the PLSR models as input datasets to predict the six IWQIs. For the PLSR models, the leave-one-out cross-validation (LOOCV) method was used to connect the input variables to the output variables. The appropriate number of latent variables (LVs) was determined based on the root mean square error (RMSE) value in order to adequately characterize the calibration data without either overfitting or underfitting. Three metrics such as R² coefficient, RMSE, and equation slope were used to assess how well the ANN and PLSR models predicted the six IWQIs.

2.6. Data Analysis and Processing

The water quality characteristics were statistically examined utilizing the SPSS program, version 26 (SPSS Inc., Chicago, IL, USA). Furthermore, Surfer software (version 15) was used to create USSL salinity and Wilcox charts [61] by linking the SAR and EC and the Na% and EC, respectively, to define and analyze surface-water accessibility for agricultural activities. The spatial distribution of different IWQIs was carried out using ArcGIS 10.5. Figure 2 summarizes the integrating physicochemical parameter IWQIs supported with ANN and PLSR models to appropriate the surface-water accessibility for agricultural purposes. Finally, the ANN was applied utilizing MATLAB 7.0 (The MathWorks, Inc., Natick, MA, USA), and PLSR was performed with Unscrambler X software version 10.2.

3. Results and Discussion

3.1. Water Quality Status

The physicochemical characteristics of surface water are important in establishing its quality and appropriateness for agricultural application, offering a useful method for spotting specific environmental problems, defining patterns, identifying geochemical factors, describing trends, and transferring knowledge on water supplies and water quality [62,63].

Table 3. The different measured physicochemical properties are given as minimum (Min), maximum (Max), mean, and standard deviation (SD).

	T °C	pH	EC	TDS	K+	Na+	Mg2+	Ca2+	Cl−	SO42−	HCO3−	CO32−	NO3−
	Rosetta Branch, Nile River (n = 21)
Min	27.0	7.40	341.00	218.00	3.11	20.70	8.20	16.00	35.50	11.00	88.40	N.D.	3.68
Max	33.7	8.10	544.00	348.00	16.81	38.91	19.00	36.00	88.70	22.00	134.00	N.D.	12.53
Mean	30.0	7.80	404.00	258.52	9.01	28.91	13.17	23.24	54.13	14.90	108.88	N.D.	6.11
SD	2.4	0.18	50.78	32.48	3.67	4.65	2.74	5.08	13.83	2.98	10.22	N.D.	2.41
	Damietta Branch, Nile River (n = 30)
Min	27.1	7.40	328.00	210.00	5.21	16.29	3.40	22.72	23.00	12.00	60.80	5.00	2.25
Max	28.1	8.40	703.00	450.00	14.41	46.05	22.80	44.00	61.00	33.00	208.60	19.00	8.51
Mean	27.4	7.89	385.40	246.63	8.15	22.36	12.15	28.06	40.10	16.80	104.17	8.87	5.48
SD	0.28	0.19	69.14	44.22	2.27	5.29	3.83	4.17	11.29	4.11	26.43	3.84	1.44
	Data across Two Branches (n = 51)
Min	27.0	7.40	328.00	210.00	3.11	16.29	3.40	16.00	23.00	11.00	60.80	N.D.	2.25
Max	33.7	8.40	703.00	450.00	16.81	46.05	22.80	44.00	88.70	33.00	208.60	19.00	12.53
Mean	28.4	7.85	393.06	251.53	8.50	25.06	12.57	26.07	45.88	16.02	106.11	5.22	5.74
SD	1.9	0.19	62.37	39.89	2.93	5.96	3.43	5.11	14.11	3.77	21.27	5.29	1.91

Note: All selected variables are provided in mg/L, except for temperature (T °C), EC (µs/cm), and pH.

provides the description of the statistical results for the physicochemical characteristics of the present study. The mean values for the ionic contents of K⁺, Na⁺, Mg²⁺, Ca²⁺, Cl⁻, SO₄²⁻, HCO₃⁻, CO₃²⁻, and NO₃⁻ were 8.50, 25.06, 12.57, 26.07, 45.88, 16.02, 106.11, 5.22, and 5.74 mg/L, respectively (Table 3). Thus, the values obtained for the ions followed the sequences Ca²⁺ > Na⁺ > Mg²⁺ > K⁺ and HCO₃⁻ > Cl⁻ > SO₄²⁻ > NO₃⁻ > CO₃²⁻, respectively. These findings demonstrated that Ca²⁺ and HCO₃⁻ exhibited the main ionic dominance in the examined surface-water samples.

After reviewing [64], the suggested criteria for measuring irrigation water quality variables were chosen. The acquired physicochemical characteristics for 51 surface-water locations throughout both branches of the Nile River indicated that the temperature measurements varied from 27.0 °C to 33.7 °C, which was within the FAO [64] recommended limits for irrigation (<35 °C). The pH readings varied from 7.40 to 8.40, indicating strong alkalinity (pH > 6.5) and agricultural applicability [64]. According to FAO [64], the EC readings varied between 328 and 703 µS/cm, indicating that all surface-water samples along the Nile River were appropriate for irrigation (EC < 3000 µS/cm).

The Ca²⁺ ion is the most abundant cation in the Nile River branches, with concentrations ranging from 16.0 to 44.0 mg/L. According to [64], these readings revealed that all obtained samples were within the appropriate limits for irrigation (Ca²⁺ = 400 mg/L). In addition, the second dominating cation, Na⁺, ranged from 16.29 to 46.05 mg/L, indicating that every sample was appropriate for irrigation. (Na⁺ = 919 mg/L). Water with an excessive amount of Na⁺ is frequently not suitable for irrigation as it degrades soil qualities by altering the soil’s structural characteristics and making it alkaline [51,65]. According to [65] (Mg²⁺ = 60 mg/L), the Mg²⁺ level in the collected samples ranged from 3.4 to 22.8 mg/L, indicating that all samples were suitable for irrigation. Moreover, the K⁺ content differed from 3.11 to 16.81 mg/L, indicating that every sample was suitable for irrigation [64]. Additionally, the HCO₃⁻ ion is the predominant anion in the Nile River, with concentrations ranging between 60.80 and 208.60 mg/L, which are within accessible limits for irrigation [64] (HCO₃⁻ = 610 mg/L). The second most abundant anion is Cl⁻, with a range of 23.00 to 88.70 mg/L, followed by SO₄²⁻, with a range of 11.0 to 33.0 mg/L. According to [64], the Cl⁻ and SO₄²⁻ readings showed that all of the samples were within the acceptable range for agricultural use and appropriate for irrigation (Cl⁻ < 1036 mg/L and SO₄²⁻ < 960 mg/L). According to [64] (NO₃⁻ = 10 mg/L), the NO₃⁻ value varied from 2.25 to 12.53 mg/L, with a mean of 5.74 mg/L, revealing that most of the samples were appropriate for agriculture. Finally, the CO₃²⁻ readings with a mean value of 5.22 mg/L indicated that the water quality was accessible for irrigation.

3.2. Surface-Water Facies and Source Identification

Hydrochemical assessment is critical for determining surface-water quality and investigating the variables influencing the water’s hydrochemical composition. Natural processes and human actions, in general, affect the chemical composition and features of surface water [65,66]. Therefore, the arrangement and variety of hydrochemical facies can provide insight on variations in surface-water quality, which reflect water type, quality, and features with respect to the physicochemical parameters [67,68]. Hydrochemical data can assist to explain the geochemical factors that influence water quality [69]. The surface-water facies and geochemical factors in the study area were examined and clearly demonstrated a dominance of HCO₃⁻ among the cations and a much more uniform representation of Ca²⁺, Na⁺, and Mg²⁺ in the anions in both average and maximum values [70]. The hydrochemical properties of the obtained surface-water samples reflected Ca-HCO₃ and mixed Ca-Mg-Cl-SO₄ water facies, which suggested meteoric water type and the beginning stage of evolution. The association between TDS and (Na + K)/(Na + K + Ca) and Cl/(Cl + HCO₃) uncovers the geochemical regulatory processes impacting water quality [71,72]. According to the results, the surface-water sites were influenced by evaporation, which was the key process governing the quality of the surface water along the Nile River.

Bivariate graphs and ionic ratios were applied to track the development of geochemical factors in the surface water of the research area. In general, the link between Na⁺ and Cl⁻ is critical for determining the processes and mechanisms affecting water properties in the studied region (Figure 3a). The surface-water samples were positively correlated with the halite dissolving line (R² = 0.866), which indicated that the silicate minerals were weathering or going through an ion exchange process (Figure 3a). The majority of samples (71%) fell under the equiline on the Na⁺ vs. Cl⁻ diagram, indicating that the dissolution of salt-bearing chlorides contributed less to the high Cl⁻ levels in surface water [65]. However, around 29% of the water samples fell above the equiline (1:1), revealing the silicate weathering and ion exchange processes that lead to Na⁺ enrichment in the surface-water [73]. The Ca²⁺ + Mg²⁺ vs. HCO₃⁻+ SO₄²⁻ plot (Figure 3b), which depicts the ion exchange mechanism, verified the significant contribution of carbonate weathering to salinity. The plot of Na⁺ vs. SO₄²⁻ (Figure 3c) revealed a positive correlation (R² = 0.1885), indicating intercalation between the surface water and the soil components. The ion exchange mechanism was reflected in the relationship between Na⁺ vs. (Ca²⁺ + Mg²⁺) in the surface-water samples (Figure 3d). The surface-water samples were dispersed along the mixing line, and the SO₄²⁻ ion concentrations, which increased linearly as a function of the Cl⁻ ion concentrations (Figure 3e), illustrates the impact of the dissolution of the gypsum and anhydrite minerals. The (Ca²⁺ + Mg²⁺) − (HCO₃⁻ + SO₄²⁻) vs. (Na⁺ − Cl⁻) plot with R² = 0.1343 revealed the ion exchange process in determining surface-water chemistry (Figure 3f).

3.3. Water Quality Indices for Agricultural Use

To the best of our knowledge, agricultural practices, soil characteristics, and water quality significantly affect the optimum irrigation techniques [74,75]. Regarding irrigation, numerous indices were applied to measure the appropriateness of the water quality for irrigation, such as IWQI, SAR, Na%, SSP, PI, and MH. These techniques emphasize the potential dangers of soil salinization, as well as irrigation’s harmful impact on plants and soil. The statistical evaluations of IWQI data and water quality appropriateness for irrigation were provided (

Table 4. Statistical analysis of the different IWQIs for the Nile River.

	IWQI	SAR	Na%	SSP	PI	MH
	Rosetta Branch, Nile River (n = 21)
Min	86.48	0.84	33.94	28.04	64.32	35.98
Max	95.79	1.47	45.92	42.98	85.97	59.00
Mean	90.46	1.20	40.07	36.08	76.25	48.30
SD	2.37	0.21	4.04	5.16	6.49	5.73
	Damietta Branch, Nile River (n = 30)
Min	81.56	0.66	26.98	23.55	57.24	14.12
Max	98.57	1.40	39.95	33.57	81.88	53.52
Mean	93.59	0.88	32.90	28.73	69.54	40.91
SD	3.13	0.14	3.10	3.01	6.05	8.62
	Data across Two Branches (n = 51)
Min	81.56	0.66	26.98	23.55	57.24	14.12
Max	98.57	1.47	45.92	42.98	85.97	59.00
Mean	92.30	1.01	35.85	31.75	72.30	43.95
SD	3.22	0.23	4.98	5.41	7.02	8.35

and

Table 5. Categorization of the different IWQIs across the Nile River.

WQIs	Range	Water Category	Number of Samples (%)
			Rosetta Branch	Damietta Branch	Across Two Branches
IWQI	85–100	No restriction	21 (100%)	29 (97%)	50 (98%)
	70–85	Low restriction	0 (0.0%)	1 (3%)	1 (2%)
	55–70	Moderate restriction	0 (0.0%)	0 (0.0%)	0 (0.0%)
	40–55	High restriction	0 (0.0%)	0 (0.0%)	0 (0.0%)
	0–40	Serve restriction	0 (0.0%)	0 (0.0%)	0 (0.0%)
SAR	2–10	Excellent	21 (100%)	30 (100%)	51 (100%)
	10–18	Good	0 (0.0%)	0 (0.0%)	0 (0.0%)
	18–26	Doubtful or fairly poor	0 (0.0%)	0 (0.0%)	0 (0.0%)
	>26	Unsuitable	0 (0.0%)	0 (0.0%)	0 (0.0%)
Na%	0–20	Excellent	0 (0.0%)	0 (0.0%)	0 (0.0%)
	21–40	Good	9 (43%)	30 (100%)	39 (76%)
	41–60	Permissible	12 (57%)	0 (0.0%)	12 (24%)
	60–80	Doubtful	0 (0.0%)	0 (0.0%)	0 (0.0%)
	>80	Unsuitable	0 (0.0%)	0 (0.0%)	0 (0.0%)
SSP	<60	Safe	21 (100%)	30 (100%)	51 (100%)
	>60	Unsafe	0 (0.0%)	0 (0.0%)	0 (0.0%)
PI	>75%	Excellent	11 (52%)	6 (20%)	17 (33%)
	25 to 75%	Good	10 (48%)	24 (80%)	34 (67%)
	<75%	Unsatisfactory	0 (0.0%)	0 (0.0%)	0 (0.0%)
MH	>50%	Unsuitable	8 (38%)	5 (17%)	13 (25%)
	<50%	Suitable	13 (62%)	25 (83%)	38 (75%)

).

Figure 4 depicts the regional distribution of several IWQIs in relation to the physicochemical characteristics, demonstrating that the quality of the surface water for agriculture deteriorated significantly from upstream to downstream along the Nile River.

3.4. Irrigation Water Quality Index (IWQI)

The IWQI is regarded to be the most efficient way to communicate water quality and provides a comprehensive model for water quality [76]. The IWQI is used to present a lot of information about water quality in a single number, and it displays the general quality of water for any initial purpose [77,78,79]. Therefore, the IWQI was recognized among the important tools for urban planners to analyze agricultural water properties and quality [80]. According to the results of IWQI classified (Table 4 and Table 5 and Figure 4a), approximately 98% of the examined surface-water samples fell into no restriction ranges, and approximately 2% of samples fell under the low restriction ranges, demonstrating the importance of avoiding salinity sensitive crops with regard to irrigated soil features, accessibility, and soil sodicity risks.

Sodium Absorption Ratio (SAR)

The SAR measures the percentage of main and earth alkaline cations that are accessible to crops in water. The SAR is considered a widely used index for assessing the sodium danger related to the agricultural water supply and was used to calculate the probability that Na⁺ might build in the soil predominantly, to the detriment of Ca²⁺ and Mg²⁺, due to the continuous utilization of sodic water [81]. Therefore, the SAR is a helpful indicator to assess the appropriateness of agricultural water based on sodium risk, which is highly related to the soil’s exchangeable salt levels [73,82]. The utilization of high-sodium water in agriculture may increase sodium exchange in the soil, lowering water infiltration and soil properties [52]. Soil treatments may be necessary in agriculture when the water has a high SAR value to reduce the long-term soil degradation due to the replacement of Ca²⁺ and Mg²⁺ in the soil by salt in the water, which can lead to reduced soil infiltration and permeability and might be dangerous for crop growth. According to the SAR categorization of water quality for agricultural purposes, the SAR value varied from 0.66 to 1.47 with an average of 1.01 (Table 4). Table 5 indicates that all the surface-water samples fell into the excellent level, making them appropriate for agricultural utilization with no alkali risk to the crops. According to the SAR spatial distribution map, the quality of surface water for irrigation is steadily continuous with an excellent range from upstream to downstream (Figure 4b). In order to assess the viability of using surface water for agricultural use, an irrigation water classification diagram is shown, with respect to the US irrigation water quality classification standards [83]. Therefore, the SAR value was plotted vs. the EC to rank the irrigated waters [84], which demonstrated that all the surface-water points fell into the C2-S1 group with medium salinity (<750 µS/cm), low sodium content, and high appropriateness for irrigation (Figure 5). As a result, for better agricultural irrigation and reduced salinity dangers, an acceptable drainage mode and adequate soil permeability are required [85].

3.5. Sodium Percentage (Na%)

The salinity level was determined using the Na% with respect to the quantity of Na⁺ and K⁺ in comparison to Ca²⁺ and Mg²⁺. The high content of exchangeable cations in the surface water raises the thickness of the diffuse double layer in the soil and alters its structure; therefore, high Na⁺ soil solution concentrations are notorious for forming unattractive soil structures [86]. The outcomes are attributed to clays that have prolonged double layers around them that expand to a considerable degree. A high Na% diminishes the permeability of the soil; changes the properties of the soil, limiting the drainage system; and finally, destroys the crops. A high EC with high Na⁺ content inhibits plant development and modifies soil characteristics [87]. According to the results, the Na% ranged from 26.98 to 45.92 with a mean of 35.85 (Table 4). The surface-water points ranged in quality from good (76%) to permissible (24%) for irrigating crops (Table 5 and Figure 4c). The Wilcox diagram for identifying irrigation water quality provided the link between the EC and the Na% (Figure 6). As a result of the Wilcox classification, the surface-water points were classified as excellent to good and good to permissible for irrigation, implying that the irrigated water would not cause soil alkalization and would be suitable for agricultural purposes (Figure 6).

3.6. Soluble Sodium Percentage (SSP)

The SSP was utilized to estimate salinity by comparing Na⁺ concentrations to Ca²⁺ and Mg²⁺ concentrations, whereas the irrigated water containing a high concentration of Na⁺ ions dispersed the Mg²⁺ and Ca²⁺ ions. The high content of Na⁺ in the irrigated water triggered the exchange mechanism for Ca²⁺ and Mg²⁺ in soil, which led to a decrease in the permeability of the soil and poor internal drainage [88]. The high content of Na⁺ vs. Ca²⁺ and Mg²⁺ in water produces toxicity, resulting in scorched leaves and destroyed tissues [11]. According to the results, the SSP values of the collected surface-water samples varied from 23.55% to 42.98% with a mean value of 31.75% (

Table 6. Findings of ANN to predict six IWQIs.

Variable	Ranking	Parameters (h1, h2, fn)	Calibration		Validation
			R2	RMSE	R2	RMSE
IWQI	EC, Cl−, Na+, Ca2+, HCO3−, Mg2+	(12, 9, tanh)	0.999 ***	0.028	0.945 ***	0.255
SAR	Ca2+, Mg2+, Na+	(12, 18, tanh)	0.999 ***	0.001	0.999 ***	0.001
Na%	K+, Ca2+, Mg2+, Na+	(21, 9, logistic)	0.999 ***	0.004	0.999 ***	0.009
SSP	Ca2+, Mg2+, Na+	(12, 15, tanh)	0.999 ***	0.005	0.999 ***	0.008
PI	Na+, HCO3−, Mg2+, Ca2+	(18, 21, tanh)	0.998 ***	0.302	0.994 ***	0.417
MR	Ca2+, Mg2+	(18, 21, logistic)	0.999 ***	0.005	0.999 ***	0.029

Note: *** p ≤ 0.001 indicates statistical significance. Where h₁ and h₂ are the number of neurons in the two hidden layers, and fn is the activation function.

). Based on the SSP categorization of water validity for irrigation with regard to salt risk, the surface-water samples were safe for agriculture with SSP levels less than 60% (Table 5 and Figure 4d).

3.7. Permeability Index (PI)

The PI is an important measure for detecting the appropriateness of surface water for irrigation, and it was developed to assess the risk of soil permeability [89]. The PI values for the surface-water samples ranged between 57.24 and 85.97 with a mean of 72.30 (Table 4). According to the PI categorization of irrigation water quality (Table 5), the majority of the surface-water points (approximately 67%) were good for long-term irrigation, while 33% were excellent for irrigation, as shown in Figure 4e. Therefore, continuous irrigation with water with high Ca²⁺, Mg²⁺, and CO₃²⁻ contents has a significant impact on soil mobility owing to soil-accumulated water [90].

3.8. Magnesium Hazard (MH)

Since Ca²⁺ and Mg²⁺ are often responsible for maintaining the water’s balance, salinity increases the amount of magnesium in the water, which has a negative impact on agricultural yield [91,92]. A larger proportion of magnesium in the soil decreases the availability of potassium, which has a severe toxic effect on plants, causing damage to the plant and a significant growth of coppery color on the surface of the leaves [88]. Therefore, the surface water with MH values under 50 is regarded as appropriate for agricultural use, whereas surface water with MH values over 50 is unsuitable since it has a detrimental effect on the soil and decreases plant production [91,93]. Ca²⁺ and Mg²⁺ are in equilibrium in nature; therefore, these ions’ concentrations are essential for soil structure and crop yield. Furthermore, the presence of exchangeable Na⁺ in the soil may explain the high quantity of Mg²⁺ in the surface water [94]. The high Mg²⁺ content decreases the soil’s ability to absorb water when the equilibrium shifts, which eventually lowers crop production. Therefore, high MH values indicate soil structural degradation as a result of an increase in soil alkalinity. The MH values of the obtained water points ranged from 14.12 to 59.00 with an average of 43.95 (Table 5). These values indicated that approximately 75% of samples were fit for agricultural purposes, while approximately 25% of samples were unsuitable for agricultural purposes (Figure 4f).

3.9. Performance of ANN and PLSR for Predicting IWQIs

This quality assessment method’s most important downside is that it demands strong comprehension and understanding of weighting variables in order to compute the score for the IWQI, suggesting that the actual result is uncertain. The WQI is calculated by adding together a variety of values from the physicochemical elements into a single number that shows the water quality level’s appropriateness for irrigation. Several academics have looked at ways to decrease the subjectivity of the existing WQI technology, which was demonstrated to be a more accuracy and a crucial tool for accurate measuring systems by providing crucial ion weights depending on entropy [95]. However, water quality research demands a large amount of data gathering, lab analysis, data management, and testing. [96]. Due to the subjectivity of the computation, there are inconsistencies in WQIs interpretation of the results. As the prior research has shown, there is not a single WQI model that is ideal or comprehensive. It is crucial to implement a workable and economical plan for a reliable evaluation of WQI. In this work, IWQIs were predicted using both models depending on several various parameters, as shown in Table 6 and

Table 7. Findings of PLSR to predict six IWQIs.

Variable	Parameters	LVs	Calibration		Validation
			R2	RMSE	R2	RMSE
IWQI	EC, Cl−, Na+, Ca2+, HCO3−, Mg2+	3	0.914 ***	0.932	0.872 ***	1.164
SAR	Ca2+, Mg2+, Na+	2	0.970 ***	0.039	0.971 ***	0.038
Na%	K+, Ca2+, Mg2+, Na+	3	0.986 ***	0.577	0.982 ***	0.678
SSP	Ca2+, Mg2+, Na+	3	0.992 ***	0.484	0.988 ***	0.571
PI	Na+, HCO3−, Mg2+, Ca2+	2	0.982 ***	0.919	0.972 ***	1.126
MR	Ca2+, Mg2+	4	0.985 ***	1.026	0.982 ***	1.175

Note: *** p ≤ 0.001 indicates statistical significance. LVs is the optimal number of latent variables.

, since the construction of IWQI is quick and does not necessitate many steps.

Recently, ANNs have demonstrated exceptional performance as a regression technique, especially when used for the recognition of patterns and determination of function. Compared to conventional methods, ANN is able to produce conclusions, comprehend incomplete information, and is less susceptible to outliers [17,42]. Figure 7 depicts the neural network topologies that were constructed following the collection of the senior water parameters. The architecture of each network includes fundamental information, such as learning synaptic weights and the number of unseen neuron layers, convergence phases, and total mistakes. Several input variables are combined with a number of hidden neuron layers to construct the network design. For instance, the ANN-IWQI-6 has hidden neuron layers (12, 9); the model ANN-SAR-3 was needed (12, 18), and the model ANN-Na%-4 was needed (21,9). In Figure 8, the contemporary models of ANN-IWQI-6, ANN-SAR-3, ANN-Na%-4, ANN-SSP-3, ANN-PI-4 and ANN-MR-4 are shown. For predicting the investigated parameters, the neural network was trained using the properties of the super elements. Based on the R² values and RMSE (Table 6) and slope (Figure 8), both the Cal. and Val. datasets of six IWQIs were evaluated more precisely using the ANN model. The six parameters included in this model (The ANN-IWQI-6) are essential for forecasting IWQI with R² values of 1.00 for (Cal.) and 0.95 for (Val.). The rest of the models behaved admirably in terms of predicting SAR, Na%, SSP, PI, and MR with R² values for the Cal. and Val. of 1.00, as shown in Table 6. Figure 9 depicts the association of six IWQIs using ANN. These figures also demonstrate a respectable slope of the linear association with measured and forecasted readings for each index, with SSP displaying the highest slope (1.0003) and IWQI displaying the lowest slope (0.93).

Six IWQIs were evaluated more accurately using the PLSR model using RMSE and R² values, as indicated in Table 7 and slope (Figure 9). The PLSR models presented robust forecasts for six IWQIs in the Cal. datasets, with R² values located between 0.91 and 0.99. Furthermore, in the Val. datasets, the PLSR model recorded robust forecasts for six IWQIs, with R² values located between 0.87 and 0.99. The RMSE values for six IWQIs, including IWQI, SAR, Na%, SSP, PI, and MR, as illustrated in Table 7, were 1.164, 0.038, 0.678, 0.571, 1.126, and 1.175, respectively. Figure 8 depicts the association of six IWQIs using PLSR. Furthermore, these figures indicate decent slope values of the linear relationship for each index, with SSP demonstrating the greatest slope value (0.990) and IWQI displaying the lowest slope value (0.960). There was neither overfitting nor underfitting in the datasets used to assess Cal. and Val. of the ANN and PLSR models for six IWQIs. Hence, when it comes to forecasting IWQIs, both of the models utilized in this study perform well and have appropriate accuracy. The models of ANN in the Cal. and Val. datasets presented a small improvement in predicting IWQIs compared to the models of PLSR. These findings are in line with the proposed ANN model, which monitors environmental contamination indicators by using all the sensitive characteristics, considerably improving the model’s performance. According to Gaagai et al. [97], the ANN was the most accurate prediction model, since it had the highest correlation between the IWQIs and the exceptional features. The training and validation sets’ respective R² and RMSE values were 0.973, 2.492 and 0.958, 2.175. The ANN model produced the best results when measuring SAR. The R² score was 0.999 (RMSE = 0.003) in the training set and 0.999 (RMSE = 0.006) in the testing phase. Regarding Palani et al. [98], the findings demonstrated how ANN offers a great deal of potential for modeling water quality problems. According to the Nash–Sutcliffe coefficient of efficiency, the simulation precision in the training and test data ranged from 0.8 to 0.9. Therefore, when measured data are lacking yet water quality models are required, a trained ANN model may be able to provide simulated values for the appropriate locations [99,100,101,102,103]. According to Masoud et al. [101], the PLSR generated a more accurate evaluation of six IWQIs based on R² and slope values. The PLSR model produced reliable estimates for six IWQIs with R² ranging from 0.72 to 1.00 in the validation datasets. Each parameter had a reasonable slope value for the linear relationship between measured and predicted values, and RSC had the greatest slope value (1.00). In addition, according to Elsayed et al. [103], both principal component regression and support vector machine regression showed reliable and powerful t models that predicted the IWQIs, with R² values ranging from 0.48 to 0.99. Consequently, the approach that was used to describe the structure in this situation accurately and robustly predicted the water quality using physicochemical characteristics. A complete image of the suitability of the water quality for irrigation and its influencing elements is provided by the effective integration of physicochemical parameters, IWQIs, ANN, PLSR, and GIS techniques. The future research should evaluate the method suggested in this paper, which integrates ANN and PLSR models, to strengthen its stability under various surface-water resource conditions.

4. The Limitations and Future Outlook

The methods suggested in this study should also be further researched to increase their dependability for maintaining water quality under a variety of circumstances and to encourage decision-makers to use multiple technologies for water quality planning and management. By employing ANN and PLSR models under significant salinization, we forecasted the surface-water quality for irrigation purposes in this work to circumvent the constraints of the conventional approaches. The future values were then predicted using these models.

To determine the best strategy for a reliable forecast, the error estimates of forecasts from the two approaches were compared. There is a significant knowledge gap about the performance of the ANN and PLSR models because they were not applied to the same river system. As a result, it is vital to apply many models to the same river system using the same input data and to assess the model’s performance using quantitative comparisons. Such comparisons would improve the model’s strengths and capabilities by highlighting the “optimal uses” for each model, as well as their process and application-related limits.

Additionally, the study concentrated on a particular set of physicochemical factors associated with water quality. It would be beneficial to incorporate additional pertinent characteristics, such as heavy metals, pesticides, and microbiological markers, to gain a more thorough assessment of the water quality and its possible effects on agricultural and human health. It would also improve awareness of the overall water-quality management and decision-making processes to examine the potential integration of socioeconomic aspects, such as water demand, land use, and socio-cultural practices.

5. Conclusions

This study examines whether the surface water from Egypt’s Nile River is suitable for use in agriculture. To identify surface-water facies and evaluate the effectiveness of the water quality for irrigation, physical and chemical properties, irrigation water quality indicators, supported by PLSR, ANN models, and GIS approaches, were used. The measured physicochemical characteristics showed that under the influence of evaporation, carbonate weathering, and reverse ion exchange processes, the water types Ca-HCO₃ and mixed Ca-Mg-Cl-SO₄ had ionic sequences of Ca²⁺ > Na⁺ > Mg²⁺ > K⁺ and HCO₃⁻ > Cl⁻ > SO₄²⁻ > NO₃⁻ > CO₃²⁻, respectively. IWQIs were used to undertake thorough evaluations of the surface-water quality for the irrigation systems. For instance, the IWQI values for the Nile River branches showed that approximately 98% of samples fell into the category of the no restriction range, which prevents the growth of crops that can withstand salinity, and that approximately 2% of samples fell into the category of low restrictions for irrigation usages.

Combining elements with ANN and PLSR can result in efficient methods for precisely estimating six IWQIs in Cal. and Val. datasets concerning surface water. For instance, The PLSR model produced robust estimates for six IWQIs in Cal. datasets, with R² values recorded between 0.91 to 0.99. Furthermore, in Val. datasets, the model recorded good estimates for six IWQIs, with R² values located between 0.87 and 0.99. These results provide a helpful performance evaluation tool for decision-makers, and more study may be conducted by combining the outcomes of the current study to assist decision-plans in arid regions. Finally, the integration of physicochemical properties, IWQIs, PLSR, ANN models, and GIS approaches is effective and gives us a comprehensive picture of the circumstances that govern the suitability of the surface-water quality for irrigation.