Water quality management of the Aras River in Iran using two modeling methods: canonical correlation analysis and principal component analysis

The surface water quality is of vital importance for humans and aquatic organisms (Safizadeh et al. 2021). One of the key issues in the quality management of water resources is the determination of a relationship between various physical and chemical parameters affecting water bodies (Valiallahi 2022). The response to this issue is a fundamental step toward a more detailed exploration of pollution resources and provides appropriate solutions to reduce the adverse effects of pollutants and ultimately to improve the quality of such resources (Sajjadi et al. 2019). Water quality monitoring requires accuracy in the identification of pollution resources and the determination of physicochemical and biological parameters affecting the surrounding environment (Mackialeagha et al. 2022).

The nature of physical pollutants generally depends on the physical attributes and climate conditions of the watershed and the chemical pollutants are often produced by human activities. Hence, the physical and chemical parameters can be considered as dependent and independent variables, respectively. Thus, there is a need for comprehensive studies to determine the correlation between these parameters (Fataei & Shiralipoor 2011). By determining the degree of correlation between these two categories of variables, it can be claimed that the behavior of the dependent variables is a function of the behavior of the independent variables. If an acceptable correlation is found between these two variables, it can be concluded that the origin of both pollutants is anthropogenic activities (Abdolabadi et al. 2014).

To properly manage water quality, it is necessary and inevitable to know the reality of the problem and use models and tools that provide clear, reliable, and practical results as much as possible. Many environmental indicators have been proposed over the past years. One of the simple methods to evaluate the water quality conditions along the river is to use the water quality index (WQI), as one of the widely used indicators to classify surface water quality. In this method, a large amount of data obtained from water quality measurements is converted into a single and dimensionless number, which is defined on a graded scale and has meaning and interpretation. In the WQI method, it is not possible to use all the physicochemical and biological parameters at the same time and it is limited to only nine parameters, including total dissolved solids (TDS), pH, turbidity, phosphate, nitrate, biochemical oxygen demand (BOD), chemical oxygen demand (COD), dissolved oxygen (DO), temperature and coliform (Samadi et al. 2015; Ewaid 2017). Therefore, it is not possible to change variables through this method to determine other parameters affecting water quality (Uddin et al. 2017).

On the one hand, although quality parameters may be accurately measured, index results depend to a large extent on the opinions and judgments of experts. In addition, it has complications and weaknesses in the interpretation of different levels of water quality (very bad to very good) so the analyst has to analyze many water quality variables in some cases. Also, it is difficult to summarize and draw conclusions from a large amount of information in the WQI method, which reduces the value of this method. On the other hand, finding the connection between pollutants using laboratory tests involves huge costs, and it is also not possible to determine the degree of relationship between parameters and clarify their possible common origin using these methods (Sayadi & Gharehmahmoodlu 2019).

Reportedly, the multivariate statistical methods lack these limitations and allow the use of a large and diverse number of parameters (Neissi & Tishehzan 2018). Therefore, the current research decided to use canonical correlation analysis (CCA) and principal component analysis (PCA) in order to reduce the number of water quality parameters in determining criteria and sources of pollution and to reduce data dimensions in water quality management of the Aras River, as one of the important rivers of the Caspian Sea watershed, within the limits of Iran.

Different methods have been reported to determine the correlation between two variables. For example, the standard correlation coefficient or Pearson correlation coefficient (PCC) expresses the direction and degree of correlation between two sets of variables. In addition, non-parametric measurement methods, such as Kendall's Tau and Spearman's rank correlation coefficients, operate on the basis of ranking similarity between two variables. Multivariate regression is a technique capable of estimating the correlation between a dependent variable and a set of independent variables. Therefore, the use of correlation analysis methods, which can explain the correlation between a set of variables, will be effective in analyzing the correlation of multidimensional variables (Eskandare et al. 2015). To this end, many researchers have applied multivariate statistical methods, such as CCA, to engineering problems and subjects (Keskin & Yasar 2007).

In this regard, Noori et al. (2010) used the CCA method to determine the relationship between physical pollutants, including total suspended solids (TSS), turbidity, electrical conductivity (EC) and temperature, and chemical parameters, including BOD, COD, phosphate, nitrate, chloride, and hardness, in the catchment basin of the Karun River in southwestern Iran. Their results indicated a strong correlation between the studied physical and chemical variables. Rasi Nezami et al. (2013) studied the water quality of the Kor River (Iran) using CCA and PCA techniques. The PCA results demonstrated that all the quality parameters selected to check the water quality of the Kor River were of equal importance. According to CCA findings, the correlation coefficient for the first three standard variables was 0.896, 0.848, and 0.593, respectively, indicating that turbidity, DO and TSS among the physical parameters, and BOD, COD, total phosphorus (TP), and total nitrogen (TN) among the chemical parameters had the highest score among the canonical variables. Cheng Chan et al. (2013) analyzed water quality and correlation between physical and chemical variables using PCA and CCA methods in Macau Main Storage Reservoir, China. Among the 28 water quality parameters studied, six variables were less important in explaining the monthly variation of the water quality, and electro-conductivity and chloride compounds were dominant variables in the physical and chemical parameters, respectively. Banerjee et al. (2015) investigated the water quality of the Bakreswar reservoir in the Birbhum district of Bengal using PCA and CCA tests. Their results showed that the main factors affecting water quality were TN, TP, temperature, and relative humidity. The CCA results revealed the correlation of all water quality variables with temporal variation. Neissi & Tishehzan (2018) also applied CCA and PCA to find relationships between physicochemical variables and determine the main variables affecting the Dez River water quality in southern Iran. The results of their study revealed a significant relationship between two categories of physical and chemical variables. Most of the changes were attributed to anthropogenic sources. In addition, the results of the principle components indicated that the main variables were EC, temperature, sodium adsorption ratio (SAR), sulfate, pH, and DO, which often originated from industrial pollutants and other sources of anthropogenic pollution. Soltani et al. (2019) monitored the surface water quality of the Aras watershed (Iran) using statistical multivariate techniques and concluded that the main causes of river water pollution were agricultural activities, domestic sewage, and weathering.

The Aras River is one of the most important and water-rich border rivers in Iran, belonging to the Caspian Sea watershed in the northwest of Iran, which provides the water needed for drinking, agriculture, and industrial purposes in vast areas. The present research aimed to determine the pollution resources and the main parameters affecting the Aras River water quality also to investigate the correlation between the main physical and chemical variables affecting the water quality within the limits of Iran, as well as to present possible control solutions.

Samples were collected from 19 sampling stations during a 2-year period from 20 March 2020 to 19 March 2022 seasonally. Thus, water sampling containers were first washed with detergent and tap water, then with nitric acid, and finally with distilled water. At the time of sampling, they were washed three times with the desired sampling water. Prior to transport to the laboratory, the samples were stabilized by the required chemicals, the air and water temperatures were measured, and the samples were labeled (including sampling station specifications recorded by GPS, sampling time, and weather conditions). The samples were transported to the laboratory under suitable temperature conditions and in the shortest possible time and refrigerated at 4 °C until the test.

Sampling containers for nitrate, sulfate, and heavy metal parameters were made of polyethylene and one liter in volume. Sampling containers for measuring BOD and COD parameters were sterilized with dark glass. The heavy metal samples were acidified (pH up to 2) using concentrated nitric acid to prevent the possible precipitation of cations and the growth of microorganisms, and also to minimize adsorption by the container walls.

In the laboratory, the collected samples were analyzed for 10 parameters based on Standard Methods for the Examination of Water and Wastewater 2015, as described in Table 1. The concentration of heavy metals was measured through inductively coupled plasma mass spectrometry (ICP-MS). Nitrate and sulfate concentrations were measured by a spectrophotometer. DO, pH, EC, and Turb. were recorded by a Hach portable device. The values of BOD and COD were measured using Hach Company's measuring device.

Data were analyzed by SPSS24 and MINITAB15 software using CCA, PCC, and PCA. The normality of the data was checked by the Kolmogorov–Smirnov test. Bartlett's test was used to determine the homogeneity of variances.

This method is used as a technique for cases such as reducing the number of variables, determining the periodic trends of data, water quality, and influential factors, and also extracting important parameters in pollution (Rahnama & Sayari 2019). Data reduction is achieved by substituting the derived factors for the original variables in the data set. Factor analysis pursues three goals, including determining the correlation between variables, recognizing representative variables from a huge set of data, and building a new (smaller) set of variables to replace with principal variables in future analyzes (Mohamed et al. 2015).

The CCA and PCA methods were used to analyze the data obtained from 2 years of sampling at 19 stations of the studied river in order to explore the correlation between physical and chemical parameters and to determine the most important parameters affecting the water quality of this river. Table 2 shows the results obtained from the CCA method, in which we can accurately see that the number of fewer variables, that is, four physical parameters, determined the number of created canonical categories. In this method, after obtaining the U and V sets, their correlations were determined.

According to Table 2, the canonical correlation coefficient for the category of canonical variables 1, 2, and 3 was 0.933, 0.743, and 0.427, respectively, which were relatively high values. The correlation coefficient was 0.240 for the fourth category of canonical variables. According to Tabachnick & Fidell (2000), the correlation between the main canonical variables can be interpreted only with a value greater than 0.3; due to the lower correlation coefficient of the fourth category of canonical variables, it is not used for drawing conclusions. It is also noteworthy that the higher the correlation coefficients of the categories, the more decisively it is possible to judge the existence of a correlation between the categories of the variables under investigation. According to the p-value for each category, it can be seen that only the canonical variables of the first category were valid, because its significance level was less than 0.05, indicating a high correlation between the parameters for the first canonical variables. Therefore, with a confidence interval of 95%, it can be stated that the two categories of physical (dependent) and chemical (independent) variables in the first category, whose correlation coefficient was 0.933, had a high correlation. According to the obtained significance level results, the other three categories of canonical variables did not provide valid results regarding the existence of a correlation between parameters.

More precisely, it could be seen that the correlation between chemical and physical pollution parameters based on the canonical correlation result in Category 4 was very small due to the low correlation coefficient of this category, and it practically indicated the lack of correlation between the two categories of chemical and physical pollution parameters. In general, it was observed that there was an effective correlation between the physical and chemical parameters with a high canonical correlation for Categories 1 and 2, and the canonical variables of Categories 1 and 2 could be the basis of the conclusion that the chemical and physical parameters under study probably had the same pollution sources.

According to Table 2, the principle variables in the first canonical variables (Category 1) included physical variables related to U₁ and chemical variables related to V₁. The CCA results showed that pH and DO variables in column U₁ of physical variables had canonical coefficients of 0.728 and 0.228, respectively, and thus they were more important, indicating the greater influence of the linear combination of U₁ (physical parameters) from these two water quality variables. Among the chemical variables V1, the parameters of NO₃, COD, and SO₄ had the canonical coefficients of 0.760, −0.349, and 0.328, respectively, indicating that the increase in the value of NO₃ parameter had a relatively high correlation with V1 compared to other parameters so that increasing the amount of this parameter enhanced the amount of V1. In the canonical category of physical variables U1, the pH parameter had a relatively high correlation with U1; therefore, as the value of the pH parameter increased, the amount of U1 also elevated.

Based on the CCA results (Table 2), the physical variables of pH and DO with canonical loading of 0.954 and 0.874, respectively, had a significant role in predicting the canonical variable of chemical parameters. According to the squared values of canonical loading (0.910 and 0.764, respectively), in predicting the variance of chemical variables, the share of the pH variable was 91% and the share of the DO variable was 76%. On the other hand, the chemical variables of NO₃, SO₄, and Hg with canonical loadings of 0.932, 0.713, and 0.648, respectively, had a significant role in predicting the canonical variables of physical parameters, with the squared canonical loads of 0.868, 0.510, and 0.420, respectively. Hence, in predicting the variance of chemical variables, the contributions of NO₃, SO₄, and Hg were 87, 51, and 42%, respectively.

According to the results obtained for the correlation of each pair of U_i and V_i, it could be stated that the physical variables of pH and DO and the chemical variables of NO₃, SO₄, and Hg in terms of p-value were the dominant variables among the parameters investigated in this research. Therefore, the results of this research could reveal a strong correlation between two categories of chemical and physical parameters. In fact, it could be claimed that both chemical and physical parameters originated from the same source. Since chemical factors in watersheds are often caused by anthropogenic activities, and physical factors can also be of anthropogenic or natural origin, based on the CCA results, it could be stated that both chemical and physical parameters were of anthropogenic origin in the Aras River watershed in Iran.

In evaluating water quality parameters using CCA and cluster analysis, Abdolabadi et al. (2014) concluded that the physical parameters of TDS and EC as well as the chemical parameters of pH and COD were the dominant variables according to the results obtained for the relationship of each pair of U_i and V_i. Also, Eskandare et al. (2015) presented a new solution for quality management of the Sefidrood River (Iran) based on the CCA method. They announced that EC and T in the linear combination of Ui (chemical parameters category) as well as NO₃ and PO₄ in the linear combination of V_i (physical parameters category) were the most influential parameters, which had the same man-made origin. Sayadi & Gharehmahmoodlu (2019) investigated and predicted water quality parameters of the Gamasyab River using CCA and time series. According to their results, HCO₃ and Mg parameters at Pol-Chehr station with canonical coefficients of 0.938 and 0.933, respectively, were in the first category, and the Na parameter with a canonical coefficient of 0.845 was situated in the second category. At the DoAb station, HCO₃ and Ca parameters with canonical coefficients of 0.945 and 0.0789 were placed in the first category, and also Na and Cl with canonical coefficients of 0.930 and 0.800 were in the second category, respectively. The origin of the first and second categories of conventional variables could be related to the dissolution of limestone and evaporative deposits.

As previously discussed, based on the results of Table 2, there was a strong correlation between pH and DO parameters and NO₃, COD, and SO₄ parameters. The high coefficient of variation of pH and DO parameters was attributed to the discharge of agricultural effluents and sewage from villages and towns located on the banks of the river along the route traveled in Turkey, Armenia, and Azerbaijan. The discharge of these effluents has caused a decrease in the values of both pH and DO parameters at the discharge site.

On the other hand, the increase in NO₃, COD, and SO₄ values could be attributed to some sources of pollution, which are mentioned below:

• – Wastewater returned from Igdir Plain lands and water quality change caused by the volcanic nature of Mount Ararat along the route in Turkey.

• – Wastewater discharge of industrial areas and mines in Armenia.

• – Mineral activities and mineral processing in the East Azarbaijan region of Iran.

• – Effluent of large sturgeon breeding complexes in Armenia and Nakhchivan Autonomous Republic of Azerbaijan.

The discharge of waste mines and active mines in Armenia and East Azarbaijan province in Iran, in which various organic and inorganic chemicals are used to separate metal from the deposit, introduced various chemicals and heavy metals into the Aras River watershed and thus affected its quality.

A scree plot can be used to determine the correlation between the eigenvalues and the number of variables. Components with eigenvalues more than one are selected as influential components. The quality changes between the stations can be interpreted with higher accuracy and confidence through the important parameters of these components (Neissi & Tishehzan 2018). As can be seen in the diagramed scree plot (Figure 3), the first three principal components had an eigenvalue greater than one and so were of the highest significance.

Table 4 presents the PCC matrix for physical and chemical parameters. As it is clear from the correlation between the parameters, among the physical parameters, the EC and pH parameters had a statistically significant correlation with the DO parameter (P < 0.01). In chemical parameters, a significant correlation (P < 0.01) was found between BOD and COD, between Hg and NO₃, and between NO₃ and SO₄, and there was also a significant correlation (P < 0.05) between As and COD and between COD and SO₄. Therefore, comparing the results of canonical correlation (Table 3) confirmed the results of the correlation between variables in the PCC test (Table 4). Similarly, Neissi & Tishehzan (2018) investigated the physicochemical parameters affecting the Dez River water quality in Iran and declared that the physical variables of EC and temperature and the chemical variables of SAR, COD, SO₄^, and NO₃ confirmed each other in affecting the water quality based on canonical and Pearson correlations.

The Aras River is of vital importance in supplying drinking, agricultural, and industrial water in Iran. Meanwhile, many pollutants are discharged into this river through different pollution sources from the countries of Iran, Azerbaijan, Armenia, and Turkey. Therefore, it seemed necessary to evaluate the water quality of this river in a comprehensive and integrated manner in the entire locational range of Iran. The results of using CCA and PCA models in water quality management of this river indicated that the high impact of DO, pH, and EC (physical parameters) as well as NO₃, SO₄, and Hg (chemical parameters) on water quality was attributed to various natural and anthropogenic pollution sources. These sources were residential and industrial sewage, mining effluent, runoff from vast agricultural lands, and erosion and destruction of pastures and forests in the catchment area of this river through neighboring countries.

Therefore, it is necessary to pay special attention to this important river to reduce pollution. The Aras River water quality management programs should focus more on human resources and activities. Based on the research findings, the following suggestions were made for controlling the water quality of the river under study:

• 1. Prevention of the introduction of effluent and drainage caused by mineral activity and processing of active minerals and abandoned mines to the river.

• 2. Construction and improvement of residential and industrial wastewater treatment systems.

• 3. Proper management in controlling the excessive use of agricultural fertilizers and pesticides and establishing a pretreatment system before releasing it into the river.

• 4. Serious effort in controlling floods and soil erosion and implementing soil protection and watershed management plans.

• 5. Compilation of legally binding treaties and conventions in the field of quality protection of river water, emphasizing the principle of ‘prohibition of causing damage to the territory of another state’.

• 6. The necessity of creating an ‘international binding legal mechanism for toxic parameters’.

• 7. Establishment of a joint online quality monitoring system and exchange of monitoring information between neighboring countries.

Based on the research results, the CCA and PCA models worked with high accuracy in determining the correlation between parameters and identifying the most effective parameters in water quality, respectively, and provided valuable information. Therefore, these techniques can be used in the selection of influential variables for the production of water quality indicators in the evaluation of the quality of water resources, especially in different geographical locations. The use of these methods makes it possible to easily access the required information about calculation methods in the selection of variables in the comprehensive assessment of the quality of water resources.

This research has been extracted from the Ph.D. dissertation written by M.M.S. in Environmental Engineering and Science at the Islamic Azad University of Ardabil Branch. Hereby, the authors appreciate the esteemed president and educational and research deputies of Ardabil Islamic Azad University for their cooperation in facilitating the implementation of this project.

No funding.

M.M.S. conceptualized the study; investigated the study; visualized the study; and wrote the article; A.A.I. did formal analysis; M.M.S. acquired funds. M.S. performed the methodology. E.F. did project administration. H.B. collected resources. A.A.I. did software analysis. E.F. supervised the study and validated the study.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.