APPLICATION OF MAJOR ION CONCENTRATIONS FOR THE PREDICTION OF STABLE ISOTOPIC COMPOSITION IN SURFACE WATER

The δ2H and δ18O values in water bodies are essential to the management of water resources because of the ability to insight into hydrological processes. In this study, we have measured and analyzed the major ions (Na+, K+, Ca2+, Mg2+, Cl–, 2– 4 SO and 3 HCO ) and stable H-O isotopes (δ 2H and δ18O) for fifteen surface water samples collected from the Xinbian River in Suzhou, northern Anhui Province, China. The results show that all of the water samples are classified to be Na-HCO3 type, and the mean values of δ2H and δ18O are –42.93‰ and –5.36‰, respectively. Gibbs diagram and the relationship between δ2H and δ18O indicate that both water chemistry and stable isotopes in river water are mainly controlled by evaporation. Correlation analysis reveals that a significant correlation between major ions and δ18O. Predictors (K+, 2– 4 SO and 3 HCO ) have been selected by optimal subset regression analysis were used to model the δ 18O values in the river water. Moreover, the residuals of the model were normally distributed and values between –0.2‰ to 0.2‰ for most water samples, suggesting a strong relationship between the observed and predicted δ18O values.


Introduction
Water is the most important resource for human survival and social development. However, in recent years, water pollution in many areas has become more seriously as a result of human activities, such as households, agriculture and industry (Wang & Yang, 2016;Alam et al., 2020;Singh et al., 2020). Under this situation, the stable isotopes have gained more attention from researchers because it can be used to understand the hydrological processes in different watersheds (Kumar et al., 2019), the relationships between different water bodies (Qian et al., 2013;Zhang et al., 2015), water quality assessment (Brooks et al., 2014) and sources of water recharge Prada et al., 2016;Cao et al., 2018;Kuang et al., 2019).
The global meteoric water line (GMWL, δ 2 H = 8 × δ 18 O + 10) has been proposed by Craig and has been recent years, scholars have constructed various predictive models based on these factors (Kumar et al., 2019;Kuang et al., 2019;Delavau et al., 2011;Terzer et al., 2013;Ogrinc et al., 2018). For example, a model was constructed by δ 18 O, latitude and elevation, and the model residuals were used to reveal hydrological processes in different stretches of the Ganga basin (Kumar et al., 2019). The δ 18 O-elevation relationships in the Yellow River source region and surrounding areas have been modeled, and the results have been applied for paleo-elevation reconstructions (Kuang et al., 2019). The precipitation δ 18 O values have been modeled using five parameters (longitude, latitude, altitude, precipitation and temperature) in Canadian, and the results have certain significance for promoting the research of climate variability and surface hydrology in remote areas (Delavau et al., 2011). A multivariate linear regression analysis was performed on the δ 18 O, longitude, latitude and altitude of rainfall over the Sava River Basin, and the predicted δ 18 O values were further compared with the measured δ 18 O values in river water. The research results showed that spatial variability is the major controlling factor for the surface water isotope ratios in this area (Ogrinc et al., 2018).
In summary, most prediction models for isotopes rely on meteorological or geographic parameters currently. However, the relationship between major ions and stable isotopes has not received corresponding attention, which limits the ability of the stable isotopes for understanding hydrological processes. Therefore, fifteen surface water samples have been collected from the Xinbian River, Suzhou, northern Anhui Province, China, and the contents of major ions and stable H-O isotopes have been measured and analyzed in order to obtain the following information: (1) the compositions of water chemistry and stable isotopes and their controlling factors, (2) the relationship between major ions and stable isotopes, and (3) application of major ions to predict δ18O values in river water using multiple linear regression model.

Study area
Suzhou (33°18′-34°38′N and 116°09′-118°10′E) is located in the junction of Anhui, Jiangsu, Henan and Shandong provinces with the total administrative area of 9787 km 2 . The study area belongs to the warm temperate zone and semi-humid monsoon climate and is in a plain landform with elevation between 20 m and 30 m. Temperatures in the region range from -1 °C to 37 °C, with a multiyear average temperature and relative humidity of 14 °C and 70%, respectively.The average annual precipitation is 830 mm, and 60% of the annual precipitation is concentrated in the summer. In addition, October is a typical representative of the dry season in the study area, while surface water is rarely affected by the mixture of other water bodies, such as rainfall and groundwater (Chen et al., 2020). The Xinbian River, completed in 1970, is a tributary of the Huai River basin ( Figure 1) and is also the largest man-made river in Suzhou at present, with a length of 128 km, an average width of 100 m and a watershed area of about 6562 km 2 , it has played an important role in agricultural irrigation and navigation in the area.

Sampling and analyses
In this study, a total of fifteen surface water samples have been collected from the Xinbian River in October 2019, and the locations of the sampling points are shown in Figure 1. All river water samples were collected using precleaned polyethylene bottles at 50 cm below the water surface. After sampling, the samples were filtered through 0.45 µm membranes in the laboratory and then stored at a constant temperature (4 °C) in the refrigerator. The major ions (Na + , K + , Ca 2+ , Mg 2+ , Cland 2-4 SO ) were measured by ion chromatographs (ICS-600-900, Thermo Fischer Scientific, USA), and -3 HCO was measured by titration with hydrochloric acid at a concentration of 0.049 mol/l, 2 H and 18 O were measured by liquid phase isotope analyzer (LGR, LICA United Technology Limited, CAN) and the analytical results were expressed in δ-notation (‰) relative to the Vienna Standard Mean Ocean Water (V-SMOW). Furthermore, the measurement accuracies are ±0.1‰ and ±0.5‰ for δ 18 O and δ 2 H, respectively. Each parameter of each water sample was tested five times, then the maximum and minimum values were removed and the average of the three remaining values was taken as the final test result. The analysis of -HCO has been completed within 24 hours, and the analyses of other parameters have been completed within a week. For the data analysis, the Minitab (version 14), AqQA (version 1.5) and Rstudio (version 3.6.3, based on the ggplot2 package) had been used.

Contents of the major ions and stable isotopes
The analytical results for major ions and stable isotopes of water samples are shown in Table 1 and Figure 2. As shown in Figure 2a, the mean contents of Cl -, Ca 2+ , Mg 2+ and K + are lower than the mean contents of Na + , SO 4 2and -3 HCO .
-3 HCO has the highest average concentration, whereas Na + and The δ 18 O values range from -5.94 to -5.03‰, with mean value of -5.36‰. In addition, the content of TDS ranges from 641.4 to 770.5 mg/L (mean = 693.2 mg/L), belonging to freshwater (TDS < 1000 mg/L) ( Table 1). As shown in Figure 2b, along the flow direction, -3 HCO content has an obvious upward and downward trend, 2-4 SO and Clcontents show a decreasing trend, while the concentrations of Na + , Mg 2+ , Ca 2+ and K + have no significant change characteristics.
Previous studies have shown that the flow of water from the recharge zone to the discharge zone is accompanied by   (Tóth, 1999;Jalali, 2005). Therefore, classification of hydrochemistry types is essential to understand the evolution of water. In this study, Piper diagram have been used by AqQA software (version 1.5) for classifying hydrochemical types and the results are shown in Figure 3 (Piper, 1944). It can be seen that the cations and anions in all water samples show an enrichment trend towards Na + and -3 HCO , respectively ( Figure 3). Therefore, all of the river water samples were classified to be Na--3 HCO type, which means that the water is under the situation of sufficient recharge.

Process controlling water chemistry and stable isotopes
Gibbs diagram has been proposed by Gibbs and can be used to reveal the controlling mechanisms of surface water chemistry (Gibbs, 1970;Prasanna et al., 2011;Tiri et al., 2018;He & Li, 2019). The three regions in Gibbs diagram correspond to three different controlling mechanisms: evaporation dominance, rock dominance and precipitation dominance (see Figure 4). The calculations of the Gibbs values are: Gibbs I = Cl -/ (Cl -+ -3 HCO ) and Gibbs II = Na + / (Na + + Ca 2+ ) (unit in meq/l). In this study, the Gibbs I values range from 0.33 to 0.42 and Gibbs II values range from 0.87 to 0.91. It can be seen from the Figure 4a that all of the samples were located in the evaporation dominance area. In addition, previous studies have shown that evaporation will change the dominate cation in surface water from Ca 2+ to Na + . As can be seen from the Figure 4b, all of the river water samples have high Gibbs II values (mean = 0.89), implying that the water chemistry have been influenced by evaporation (Gibbs, 1970).
Previous studies have shown that evaporation line (EL) is an important reference for understanding the evaporation of regional water bodies (Chen et al., 2008). In recent years, most scholars have applied a combination of GMWL, NMWL and EL to study the relationship between stable isotopes in the water bodies and provided valuable guidance for regional water management (Kumar et al., 2019;Cao et al., 2018). In this study, due to the lack of rainfall-related data of Suzhou City, the GMWL, NMWL of China (δ 2 H = 7.89 × δ 18 O + 8.16) and MEL (evaporation line in mining area near the study area, δ 2 H = 5.89 × δ 18 O -7.68) have been employed for understanding the mechanisms controlling the stable isotopes in river water (Chen et al., 2008;Zheng et al., 1983). As can be seen from the Figure 5, the relationship between δ 2 H and δ 18 O in this study is similar to MEL, but different from the GMWL and NMWL, and therefore, it can be inferred that the river water has been influenced by evaporation significantly.
The fractionation of stable isotopes in regional water bodies can be quantified by comparison with the d-excess values (δ 2 H -8 × δ 18 O) of GWML and NWML (Dansgaard, 1964). Previous studies have shown that water-rock interactions and evaporation are the major factors influencing the enrichment of δ 2 H and δ 18 O values in water bodies (Zhang et al., 2015). The embankments and bottom of this section of the Xinbian River were artificially constructed with little contact with rocks and minerals, moreover, a very weak relationship with groundwater recharge (Chen et al., 2020). Therefore, in this study, the influence of δ 2 H and δ 18 O from water-rock interactions is limited. Because of the atomic mass of oxygen isotopes is much heavier than that of hydrogen isotopes, with the continues of evaporation, the fractionation of oxygen isotopes will be more significant than that of hydrogen isotopes, which will lead to the decreasing of d-excess value of surface water. In this study, the d-excess values of the fifteen water samples range from -4.52 to 5.74‰ (mean = -0.18‰), which were significantly lower than the d-excess value of GMWL (10‰) and NMWL (8.16‰), and suggesting that isotopic compositions of the river water have been strongly influenced by evaporation ( Figure 6).

Relationship between major ions and stable isotopes
Correlation analysis is a multivariate statistical analysis method, which can be used to evaluate the degree of intrinsic association between two or more variables (Wu et al., , 2020. In the study of geochemistry, a good correlation between elements means that they come from the same source or have been influenced by similar factors (Nwankwoala et al., 2014). It can be seen from Tabel 2 that the δ 18 O has a significant positive correlation with Na + (r = 0.722), and has a significant negative correlation with -3 HCO (r = -0.693) and Ca 2+ (r = -0.605), implying the close relationship between them. However, the relationship between hydrogen isotopes and major ions is insignificant, which was similar to the results obtained by the analysis of d-excess, because the mass of hydrogen isotope is lower than the oxygen isotope, and during the high extent of evaporation, the fractionation of hydrogen isotope will be insignificant. Such results provide the possibility of prediction of δ 18 O values by the major ion concentrations, but cannot be applied for the prediction of hydrogen isotopes (see below).

Major ion selection by optimal subset method
Optimal subset regression is an effective tool for selecting linear regression models with as few variables as possible while ensuring a high degree of fit (Gorman, 1976;Lee et al., 2018). Based on the input dependent variable and multiple independent variables, best subset regression can generate several linear models. Then, it is necessary to select the optimal variables by artificially comparing the goodness-of-fit indexes in each model. In most studies, R 2 , 2 adj R , Mallows C p and S are often used as reference indexes for selecting the best variables (Ke et al., 2009;Sun et al., 2010). The details of the four indicators are as follows: R 2 is a goodness-of-fit indicator for the model, with values ranging from 0 to 1. In general, a larger value of R 2 indicating a better fit. However, even though some variables may be redundant, R 2 still increases as the number of variables increases, which may lead to overfitting of the model. Therefore, 2 adj R is proposed to control the number of parameters in order to optimize the model. Mallows C p has been proposed by Mallows and is often used to evaluate the superiority of different linear regression models. The smaller its value, the more accurate the model is. S is the root value of mean squared error for the model, the smaller the value, the smaller the difference between the predicted and observed values. The formulas for the four indicators are as follows: where SSE is the sum of squared errors, SST is the sum of squared total, MSE is the mean square error, n is the total number of samples and k is the number of selected predictors.
Mallow C p = SSE p / MSE m -(n -2p); (3) where SSE p is the sum of squared errors for the model under consideration, p is the number of terms in the model (including constant) and MSE m is the mean square error for the model with all predictors. The results of the optimal subset regression analysis are shown in Table 3 (Note: X1, X2, X3, X4, X5, X6 and X7 represent Na + , K + , Mg 2+ , Ca 2+ , Cl -, 2-4 SO , and -3 HCO , respectively.). As can be seen in the table, from Model 1 to Model 3, R2adj increases gradually as the number of variables increases, while its value begins to decrease when the number of variables increases to four or more, which is a signal that the model is beginning to overfit. At meanwhile, relative to Model 1 and Model 2, Model 3 has the highest R 2 (value = 0.74), 2 adj R (value = 0.669) and lowest S (value = 0.15899). Thus, K + , 2-4 SO and -3 HCO were selected for the final linear fitting.

Model construction & residual analysis
In summary, the concentrations of K + , It can be seen from Figure 7 that the fitting lines (y = 0.74x -1.39, R 2 = 0.74) have a high degree of coincidence with the 1:1 line (y = x), with only a slight skew. In addition, most of the water samples were distributed near the 1:1 line and fitting line with litter deviation. The difference between the observed and predicted values of δ 18 O for each sample can be seen in Figure 8, and the high degree of similarity between the observed and predicted δ 18 O values further indicates the good fit of the models. Moreover, the residuals should be analyzed to check if the model violates any regression assumptions (Sun et al., 2010). The residual value was calculated by subtracting the predicted value from the observed value, and the statistics results can be seen in Figure 9. The Q-Q plot of the residuals was often used to test whether it was normally distributed. In the Q-Q plot, the presence of severely deviated points indicates that the model violates the assumption of normal distribution. Figure 9a shows that all of the points approximately form a straight line. Therefore, it can be concluded that the model was consistent with the least squares hypothesis. Furthermore, as can be seen in Figure 9b, the maximum value of the residual was 0.239‰, and 81% of the samples (except sample 10 and 12) have residuals between -0.2‰ and 0.2‰, an indication that the strong correlations between observed and predicted values.

Conclusions
Based on the analyses of the major ion concentrations and stable H-O isotope contents of fifteen river water samples in Suzhou, northern Anhui Province, China, the following conclusions were obtained: (1) The density distribution plot indicated that Na + and -3 HCO were the dominant cation and anion, with average concentrations of 308.6 mg/L and 156.2 mg/L, respectively, and all of the river water samples were classified to be Na-HCO 3 type. The mean values of δ 2 H and δ 18 O were -42.93‰ and -5.36‰, respectively.
(2) Gibbs diagram, the relationship between δ 2 H and δ 18 O and d-excess show that evaporation is the major process controlling water chemistry and stable isotopes in river water. Moreover, the close relationship between major ions and δ 18 O were determined by correlation analysis.
(3) Q-Q plot indicated that the residuals of the model were normally distributed. The difference between the observed and predicted values for most water samples was between -0.2‰ and 0.2‰, suggesting the strong relationships between them.