Evaluation of Water Quality and Heavy Metals in Wetlands along the Yellow River in Henan Province

: Assessing spatiotemporal variation in water quality and heavy metals concentrations in wetlands and identifying metal contamination source are crucial steps for the protection and sustainable utilization of water resources. Using the water quality identiﬁcation index ( I wq ), heavy metal pollution index (HPI), hierarchical cluster analysis (HCA) and redundancy analysis (RDA), we evaluated spatiotemporal variation in water quality and heavy metals concentrations, and their interrelation in wetlands along the middle and lower Yellow River. The average I wq was highest during ﬂood season but the average HPI was lowest in the same season. Meanwhile, the trend in mean HPI across three hydrological seasons was the opposite to that of mean I wq . There was signiﬁcant variation in wetlands water pollution status across seasons. During the ﬂood season, the wetlands in the a ﬀ ected area with hanging river were seriously polluted. In other seasons, pollution in the artiﬁcial wetlands was even more severe. Moreover, serious pollution of wetlands in belt transect #03 (Yuanyang-Zhongmu) was more frequent. Dissolved oxygen and chemical oxygen demand strongly inﬂuenced heavy metal concentrations, while other water quality parameters had di ﬀ erent inﬂuences on heavy metal concentrations in di ﬀ erent hydrological seasons. The causes of water pollution were divided into natural factors and human disturbance (with potential relationships between them). The polluted wetlands were greatly a ﬀ ected by the Yellow River during the ﬂood season while they were more impacted by agricultural and domestic sewage discharge in other seasons. However, heavy metal deposition and leaching into riparian wetlands were still a ﬀ ected by diverse channel conditions. If this trend is allowed to continue unabated, wetlands along the middle and lower Yellow River are likely to lose their vital ecological and social functions.


Introduction
Wetlands are among the most productive and vulnerable ecosystems in the world, have fundamental ecological functions, and play an irreplaceable role in the maintenance of biodiversity and human development [1][2][3]. Despite the relevant departments' efforts to restore natural wetlands for human well-being [4], global change and intense anthropogenic pressure have destroyed more than half of global wetlands during the last century [5]. Riparian wetlands are buffer zones for the water and nutrient budget of the landscape which play an important ecological role. However, because mountain and plain areas in temperature. The annual average precipitation is about 500-800 mm with approximately half in summer in the form of frequent rainstorms [26,27]. The study area, which is one of the national food production bases, has a long history of agricultural cultivation and the landscape is dominated by agriculture. Most of the study area is located at low elevation and is seriously influenced by the Yellow River. East of Huayuankou in Zhengzhou, the lower Yellow River is characterized by gentle flow, wide and shallow channel controlled by dikes, leading to serious silt deposition and risen river bed, thus forming the "hanging river". Changes in river channel coupling with the irrigation practices have resulted in dramatical dynamics of surrounding landscape, which is especially true for the formation and maintenance of wetlands and the appearance and disappearance of ponds.

Sample Collection and Analysis
The study area was divided into 5 belt transects, according to the distance from Xiaolangdi dam and the height of the riverbed above the ground level (Table A1). Based on remote sensing images and field investigation, 18 typical riparian wetlands were selected as sample sites, including both natural and artificial wetlands at the north and south Banks of the Yellow River and/or inside and outside the dykes. Water samples were collected from the middle as in depth-wise of the surface water, 50 cm below the water surface at each site. Typical total stream depths were 1m, but where stream depth was is <0.5 m, we sampled at half the depth of the stream. We sampled water between and 12 April and 20 April 2018 during the pre-flood season (S B ), from 27 to 31 July 2018 during the flood season (S D ) and from 26 October to 1 November 2018 during the post-flood season (S A ). The sampling timing was planned to avoid significant rain events (10 mm over 48 h).
The water samples were stored in prewashed polyethylene bottles and shipped to the lab in a cooled container following dilution with HNO 3 or H 2 SO 4 , depending on the analysis to be conducted. In situ measurements of dissolved oxygen (DO) were performed on each water sample with an SX736 multi-probe. Duplicate co-located samples were collected at each monitoring site and the percent difference in results between duplicates was always less than 10%.
We analyzed 12 chemical and heavy metal parameters. Ammonia nitrogen (NH 3 -N) was determined by Nessler's reagent method, total phosphorus (TP) by the ammonium molybdate method, total nitrogen (TN) by the alkaline persulfate digestion method [28], and chemical oxygen demand (COD) by the dichromate reflux method [29]. Heavy metals including lead (Pb), zinc (Zn), copper (Cu), cadmium (Cd) and chromium (Cr) were analyzed using inductively coupled plasma mass spectrometry (ICP-MS). The determination of arsenic (As) and mercury (Hg) concentrations in water samples was conducted by atomic fluorescence spectrophotometry (DB51/T836-2008). All water quality parameters were determined within one week of sample collection.

Quantifying Water Quality
The comprehensive water quality identification index (I wq ) and heavy metal pollution index (HPI) provide substantial information for water quality assessment. The I wq is based on single-factor water quality identification index (P i ), so P i is calculated firstly [30]. I wq and HPI both are simple, easy to understand [31,32] and can eliminate the variations between different water quality parameters that are used individually [33].

Single-Factor Water Quality Identification Index (P i )
The P i consists of integer and decimal fractions and P i can be expressed by the formula: where C 1 is the integer and shows the grade of water quality; C 2 is the decimal fraction and shows the degree of monitoring data in interval of C 1 class water quality changing; C 3 is the comparison difference of water quality grade and function goal grade. According to the Surface Water Environment Quality Standards of China, when the water quality grade is between class I and V, for the general indicators (TN, COD, TP and NH 3 -N) and DO, C 1 × C 2 is calculated by the following Equations (2) and (3), respectively: where E i is the monitoring value of ith target; E us is the upper limit of ith target in water quality standard interval of class a; E ls is the lower limit of ith target in water quality standard interval of class a; a = 1, 2, 3, 4, 5, based on monitoring data and national standards.
When the water quality is worse than or equal to class V, for the general indicators (TN, COD, TP and NH 3 -N) and DO, C 1 × C 2 is calculated by the following Equations (4) and (5), respectively: where E vus is the upper limit of ith target in water quality standard interval of class V; E vls is the lower limit of ith target in water quality standard interval of class V; m is the correction coefficient, m = 4 in this study [30].
where f l is the goal grade of water environment functional area. Note: when C 3 > 9, f l = 9.

Comprehensive Water Quality Identification Index (I wq )
I wq is used to evaluate general water quality both qualitatively and quantitatively. Water quality parameters (e.g., TP, TN, COD, DO and NH 3 -N) were selected in this study to conduct a comprehensive evaluation of water quality in water body using I wq , which is calculated by the following formula: where X 1 × X 2 shows comprehensive water quality index; P m is single factor water quality index (that is C 1 × C 2 in single factor water quality identification index), and each indicator is weighted evenly; m is the number of indicators; X 3 is the number of indicators which are worse than water quality standards graded among all indices; X 4 represents the comparison results of water quality categories and function zoning category. Surface water quality grades can be determined based on X 1 × X 2 of I wq (Table 1). According to the Surface Water Environment Quality Standards of China, the water quality in this study area should meet class III. Table 1. Comprehensive water quality grade evaluation standards.

Judging Basis
The Water Quality Grade Inferior, but not black and foul Inferior, black and foul

Quantifying Heavy Metal Pollution
The HPI can objectively reflect the quality of water and its suitability for drinking purposes with respect to metals pollution [34]. Heavy metals (e.g., As, Hg, Cd, Cr, Cu, Pb and Zn) were selected in this study to conduct a comprehensive evaluation of heavy metal pollution in the water body using HPI. The pollution critical index (HPI c ) is 100. HPI > 100 indicates that the level of heavy metal pollution in a water body exceeds its maximum acceptable level [34,35].
HPI is based on a weighted arithmetic quality mean method as follows [31]: where W i (weight unit) is calculated as 1/S i where S i is the recommended standard of the relevant metal. Generally, the proportionality constant (k) determined by the condition is set at 1 for simplicity of calculation [34]. n is the number of estimated metals; Q i is the individual quality rating of ith metal and C i is the measured value of the ith metals in µg/L. The standard allowable value (S i ) for each parameter was taken from the criteria for class III of the water quality standard (GB3838-2002) [36].

Multivariate Analyses
Four multivariate statistical methods including the principal components analysis (PCA), hierarchical cluster analysis (HCA), Pearson's correlation analysis, and redundancy analysis (RDA) were utilized in this study. Prior to the multivariate analyses, the water quality and heavy metal parameters were required to conform to a normal distribution [37,38]. Therefore, the normality of the distribution of each variable was checked by analyzing its kurtosis and skewness [39]. The results showed that all parameters were normally distributed and in line with the standard of the statistical analysis. All the tests were conducted using SPSS ® software for Windows 22.0 and Canoco ® software for Windows 5.0, and the graphics were generated in Origin for Windows 9.1 ® .
Firstly, Kaiser Meyer Olkin (KMO) and spherical Bartlett tests were used to analyze the suitability of the water quality and heavy metal parameters for PCA. The KMO index compared the values of correlations between variables and those of the partial correlations [40]. Generally, this index should be greater than 0.5 for a satisfactory factor analysis. Bartlett Test of Sphericity was used to check the null hypothesis that the intercorrelation matrix comes from a population in which the variables are uncorrelated [41]; the null hypothesis was rejected at the significance level of 0.05. PCA was used to identify the parameters which explained the majority of contamination status. PCA selected a small number of important variables through linear transformation of multiple variables, and several principal components were used to explain water quality and heavy metals concentrations, it becomes more meaningful [39,42].
HCA includes both the variable HCA and the case HCA [28,43]. The variations in water quality and heavy metal parameters were addressed using spatiotemporal matrices. Clustering was based on the similarity between observations and successively by case HCA which is the most widely used clustering method for delineating differences or similarities among sampling sites [44,45].
Finally, Pearson's correlation analysis and RDA were used to identify relationships among individual parameters. RDA is a direct gradient analysis method that statistically evaluates the relationship between one or a set of variables and another set of multivariate data [46]. It has been successfully used in various environmental studies, including for correlations among sampling sites and environmental conditions and outcomes. The advantages and disadvantages of RDA relative to other related multivariate analytical techniques have been discussed extensively [47][48][49]. In this study, the standard deviation (SD) of the average parameter concentrations in this study were 1.2 (S D ), 1.0 (S A ) and 1.2 (S B ), respectively, suitable for a regression analysis by a linear model. Therefore, heavy metal elements and water quality parameters are used as the response variables in RDA and the individual and comprehensive effects of environmental factors on these response variables respectively. RDA yields directional indices of the shared variance between two data sets, which can be regarded as predictive of the other. RDA biplot diagrams explain the relationships between them. In the diagrams, the arrows indicate the correlations among parameters: a longer arrow indicates that the corresponding parameter was more important; a small angle between two arrows indicates that the correlation between the two corresponding parameters was strong [50].

Water Quality Status
The mean I wq value was 5.033, 5.032 and 2.995 the in S D , S A and S B , respectively. A bar chart shows that the I wq value in S B was obviously lower than that in S D and S A (Figure 2a). Based on the I wq classification, the water quality was rated as "class II" in S B , but "class V" during other seasons ( Table 1). There were six sample sites where the I wq value was highest in S D and lowest in S B . 83% of which were natural wetlands within the dike. However, there were 10 sample sites where the I wq maximum value occurred in S A and the minimum value occurred in S B , all of which were artificial wetlands. Only two natural wetlands did not follow this order. The water quality met the standard (I wq value ≤ 4) in S B . In S D and S A , water quality of major wetlands was worse than class III standard. The average I wq value of belt transect #03 and #05 in S A were slightly better than in S D , while the other three sample belts were the opposite. The mean I wq value of two wetland types in the three sampling periods in this study area is shown in Figure 2b. The water quality situation of natural wetlands was slightly better than that of artificial wetlands in S A and S B . The water quality situation of artificial wetlands was obviously better than that of natural wetlands in S D .
Overall, water pollution was most serious in S B , followed by the S A . We expected to observe a higher concentration of water quality parameters during low-flow periods [15][16][17]. Meanwhile, because the selected wetlands were in a major agricultural province, the increased runoff in summer and the fertilization and sewage discharge from farmlands around the wetlands in S B and S A may be the main cause of pollution [51][52][53]. Water quality was excellent in S B , possibly due to the effects of a constructed reservoir on control of river water and sediment [54]. More river runoff would increase the quantity of eroded material and cause more serious water pollution in S D . Because most of the selected natural wetlands were close to a river channel, their water quality was worse than that of artificial wetlands in S D . In S A and S B , artificial wetlands were strongly disturbed by human activities, while natural wetlands were less disturbed. The water body of riparian wetlands had some self-purification ability, and the water quality of natural wetlands was slightly better than that of artificial wetlands.

Heavy Metal Pollution
The mean HPI value was highest in S B ( Table 2). All HPI values were lower than HPI c in S D , indicating that heavy metals were not polluting the wetlands. Other than two sites, heavy metal concentrations were lower than the pollution threshold in S A . However, the HPI value of the polluted sample sites exceeded the HPI c values 2-3 times over, indicating serious heavy metal pollution. The HPI of the sample site (1-S-I-I) with severe human disturbance was close to exceeding the standard, and was obviously higher than that in the adjacent natural wetland (1-S-I). The HPI value of all sample sites exceeded the HPI c value by an order of magnitude or more in S B , primarily because Hg content was very high. Moreover, there were some differences in the temporal pattern of HPI. There were 11 sampling sites whose HPI value was highest in S B and minimum value appeared in S A , among which artificial wetlands accounted for 64% and five of all were within the dike. Meanwhile, there were seven sampling sites whose maximum HPI value appeared in S B and minimum value appeared in S D , among which artificial wetlands account for 57%. Five of these were within the dike. In terms of wetlands types, HPI also showed some seasonal variation. The mean HPI value in natural wetlands was higher than that in artificial wetlands in S D and S A while it was the opposite in S B .
Generally, the spatiotemporal variations in metals concentration in water bodies depend on many factors, including climate, soil type, and pH. Compared with the natural environment, human activities such as urbanization and industrialization aggravate the heavy metal pollution in water [9]. High rainfall and water volume in the Yellow River in S D may accelerate the water flow in wetlands, diluting heavy metal elements. Meanwhile, dense plants in S D also have a certain adsorption effect on heavy metals. Field investigation shows that there were no industrial sources of heavy metals around the selected wetlands. The flood season was in the period of agricultural cultivation in Henan province, and the content of heavy metal elements in agrochemicals and fertilizers is relatively high. Therefore, the excess concentration of heavy metals in the wetlands could be caused by agricultural activities in the surrounding farmlands in S D .
Heavy metal elements were deposited in large volumes (mainly Hg) in S B . The river runoff was relatively small and the wetland plants just germinating, which would cause the heavy metal deposition in the wetlands. The HPI value of artificial wetlands was higher than that of natural wetlands, indicating that the excessive heavy metal concentrations could be caused by human factors, or the vegetation of artificial wetlands would be less than that of natural wetlands. Compared with artificial wetlands, natural wetlands were closer to the river channel and most of them live within the dike of the Yellow River. Heavy metals carried by the Yellow River could be deposited in natural wetlands, resulting in the heavy metal pollution in them being more serious.

Spatiotemporal Analysis
Most water body parameters had significant spatiotemporal variation. Among them, the water body parameters with large spatiotemporal changes include CV of TN, TP, Hg, Cd, Cr, Cu, Pb, and Zn, all of which were over 100%. COD and As had small spatiotemporal changes compared with the other parameters and the standardized coefficient of variation (CV) were 68.85% and 68.79%, respectively (Table 3).

Water Parameter Seasonality
The mean TP concentration was highest and lowest during flood season and pre-flood season, respectively. The maximum TP value appeared in S D , while the minimum value appeared in S A (Table 3). However, the seasonal sequence of average DO concentration was opposite to TP, while the maximum and minimum value were also opposite. The mean NH 3 -N, COD, and TN concentrations were highest and lowest in S A and S B respectively. Among them, the maximum concentration of NH 3 -N appeared in S A and the minimum appeared in S B . The maximum COD concentration appeared in S B , and the minimum appeared in S D . The maximum value of TN concentration appeared in S A and the minimum value appeared in S D . The chemical status of each water body studied in S B was better than those of other seasons, consistently with the I wq results.
Based on the Surface Water Environment Quality Standards of China, the mean DO concentration only reached class III in S B , possibly because temperature, microbial metabolism, and organic matter degradation are higher in this season [55,56]. The average concentrations of TN, NH 3 -N and TP in the majority of sites exceeded the standard threshold in S D . Meanwhile, the concentrations of TN and NH 3 -N in many sites also exceeded the class III in S A , while these three parameters all met the standard in S B . The reason may be the discharge of domestic and industrial water around the wetland (such as the waste water generated by the excrement of the farm and the water discharge from the sewage treatment plant, etc.) and the drainage after the excessive use of chemical fertilizers in the farmland [57][58][59]. N and P concentrations in wetland water were low in S B , which may be associated with artificial adjustments of river water and sediment. After the operation of the Xiaolangdi dam, ammonia nitrogen management obviously improved near the dam, and thus, the downstream wetland water quality of ammonia nitrogen pollution might be reduced [54].
Higher COD aids estimation of organic matter pollution. Our results agree with previous findings that the mean concentration of COD exceeded the class III in three seasons [56]. The excessive concentration of COD could be related to the leaching and transport of natural, domestic sewage, agricultural and industrial pollutant, the high surrounding population density, and the extent of construction and other social factors [39,41,45]. The seasonal sequence of average COD concentration was similar to TN and NH 3 -N sequence. The excess concentration of COD could be the same as the source of N, which was the discharge of agricultural and domestic sewage around wetlands.
Heavy metal concentrations exhibited distinct seasonal variation ( Table 3). The mean Cr concentration was highest and lowest in S A and S D , respectively. However, five other elements' concentrations had different seasonal dynamics, and the temporal difference between the maximum and minimum values was also great. Among these, Hg exceeded the standard seriously in S B , and its difference between seasons was also the greatest. Due to the very high value of a single parameter, the HPI value was affected mainly by this element.
The concentration of As in all selected sample sites was lower than class III allow in all three seasons. The concentration of Hg in all selected sample sites was lower than the detection line in S D ; there were three sample sites that exceeded the standard in S A , which is 1-3 times higher than the standard. In S B , the concentration of Hg exceeded the standard and was many times or even hundreds of times of the threshold value in all selected sites. It is worth noting that Hg concentration was low in almost all wetlands water in S D and S A . In S D , except for one wetland in belt transect #03, the concentration of Cd from other wetland water exceeded the standard threshold, and the highest concentration was 28 times exceeded than the III class standard of water quality. The concentration of Cd was below the standard in the other two hydrological seasons. In S D , Pb concentration exceeded the standard at three sites in belt transect #01, #02 and #03, respectively, while it was normal in two other hydrological seasons. In S D , Cr concentration was lower than the detection line in all sites. However, Cr concentration in two sites exceeded the standard in S B and S A . The Cu and Zn concentrations all were lower than the standard values.
High Cd concentration was almost certainly a result of anthropogenic activities, especially wastewater discharge from industry and the overuse of agrochemicals and fertilizers [60][61][62]. Excessive Pb in wetland water is associated with anthropogenic origin, and travels via the freshwater input from rainfall and freshwater from rivers [63]. If humans ingest the water or fauna from these wetlands, the Pb can be harmful to human health [64]. Regarding seasonal difference, the main factor may be human-caused. For example, all sites with higher Cr concentration were artificial wetlands. High Hg concentration in wetlands could be related to lower water supply and excessive human interference in S B . The excessive Hg may not only be caused by the human disturbance but also by natural factors such as decreased runoff from the Yellow River and the plants in juvenile stage [64][65][66][67][68].

Analysis of Water Parameters' Spatial Characteristics
KMO had a value of 0.673 greater than 0.5, which can take a satisfactory factor analysis. Meanwhile, the null hypothesis of Bartlett Test of Sphericity was rejected at the significance level of 0.05, but in this study, the value is 0, which is small enough to reject the null hypothesis. In fact, the PCA successfully identified underlying interrelationships amongst the parameters.
In this study, four principal components with eigenvalues greater than 1 were extracted; they explained 71.167% of the total variance in the water dataset ( Table 4). The first and second principal components accounted for 51% of the total variance among the original variables, and then the indices with the highest load value > 0.7 were Cu (0.721) and Cd (0.788) [69], indicating that these two water quality parameters can explain more than 50% of the water quality change information of wetlands ( Table 4). The contribution of the third and fourth principal components were 11% and 9%, respectively. The parameters with the highest load value were TN and COD, respectively, and their values were 0.591 and 0.755. PC3 and PC4 including organic and nutrient variables can be attributed to anthropogenic pollution sources, may be associated to influences from municipal and industrial point-source discharges, agricultural nonpoint sources, livestock operations, and/or domestic sources [41,69]. HCA groups sample sites according to their spatial similarities in water quality parameters and heavy metal parameters [70]. By analyzing the principal component load matrix, the change of water quality in wetland landscapes along the Yellow River can be better explained by the COD, TN, Cd and Cu. Then, the wetlands in different hydrological seasons were statistically clustered by four characteristic parameters. We built a dendrogram from the case HCA using Ward's method and grouped the 18 sampling sites into four statistically significant groups at the rescaled squared Euclidean distance (SED) < 5.000 (Figure 4).
Cluster 1 included most sample sites, located in all belt transects in each hydrological season. In S D , Cluster 2 was consistent with four sites, most of which were inside the dike or on the north side of the Yellow River. TN, TP, COD, NH 3 −N and Zn in Cluster 2 sites were obviously higher than in Cluster 1 wetlands, DO was the opposite of the former (Figure 3a). The overall pollution was serious in Cluster 2 sites in S D . By analyzing the records of the ambient environments, Cluster 2 was found to be in the affected area with hanging river. The Yellow River had great water volume and carried a large amount of material in S D , which influences the wetlands along the middle and lower Yellow River through the lateral infiltration mechanism, especially the sites within the great dike of the Yellow River. In the north bank of the belt transect #03, the wetlands were also seriously polluted outside the dike. Lateral infiltration was significant in this area due to topography and river situation; still, it was an artificial wetland and the influence of human interference on it cannot be ignored.
In S A , Cluster 2 contained four sites, two inside and two outside the dike. Three were located at the south side of the Yellow River (Figure 3b). Cr, Hg, Cu, Zn and TP in Cluster 2 were obviously higher than in Cluster 1. The concentration of several heavy metal elements in Cluster 2 was higher than that in Cluster 1. Cluster 2 included only artificial wetlands, except one. There was a high concentration of heavy metals in S A , mainly due to human disturbance. Belt transect #03 (Yuanyang-Zhongmu) in southern region of the Yellow River are located in Yuanfang of Kaifeng. According to field surveys, the intense human development in the region (large areas of cultivated land, amusement parks, plants, etc.), may have caused the high heavy metal concentrations.
In S B , Cluster 2 contained three sites, two sites on the south bank inside the dike and one on the north bank on the outside the dike. Cluster 3 was a fish pond with serious human disturbance (Figure 3c). Cluster 2 was mainly characterized by serious organic pollution. Two of the clustering sites were in the belt transects #03 and were artificial wetlands. Their surrounding areas were residential areas and farmland, so the pollution source might be agricultural pollution and domestic wastewater. Another wetland in Cluster 2 was located near the channel and fish pond on the south bank of the Yellow River, and was also contaminated by human disturbance. Cluster 3 was a typical artificial fish pond. An artificial oxygen pump raised its DO concentration. Due to the human disturbance, the content of heavy metal elements was much higher than that of other wetlands.
Water quality differed among hydrological seasons. The covariability of water quality parameters in three seasons was diverse, which indirectly indicates that water pollutants come from different sources. Overall, the affected area with hanging river was seriously polluted, and the pollutants may be accumulated through natural factors (the Yellow River lateral infiltration) in S D . The artificial wetlands were seriously polluted in S B and S A , mainly by agricultural and domestic wastewater. There was frequent serious pollution of wetlands in belt transect #03.

Pearson Correlation Analysis
COD was significantly positively correlated (p < 0.05) with NH 3 -N and TN, and there was a highly significant positive correlation (p < 0.01) between TN and TP, indicating that there was covariance between organic pollution and inorganic pollution during the flood season (Table 5). Meanwhile, the correlation of nitrogen and phosphorus concentrations in water was extremely significant (p < 0.01). Both these elements may cause serious eutrophication in water. If the discharge of domestic sewage and agricultural waste can be controlled eutrophication in S D may be avoided by reducing pollutant inputs. There was no significant correlation (p > 0.05) between water quality parameters in S A . However, there was a significant positive correlation (p < 0.05) between COD and TP in S B , indicating that COD increases with increasing TP, which indirectly proves that COD is driven by domestic sewage and agricultural sewage. According to the Pearson correlation analysis (Table 6), only As and Zn were significantly correlated with water quality (though As and water quality had no significant correlation in S A ), as was negatively correlated with TN in S D . And it was also significantly negatively correlated (p < 0.05) with DO, while extremely significantly positively correlated (p < 0.01) with COD and TP in S B . Zn was negatively correlated with DO and positively correlated with NH 3 -N in S D . Zn was also positively correlated with TP in S A and TP was positively correlated with NH 3 -N in S B . Heavy metal elements were difficult to degrade in water, but the concentration of some heavy metal elements in water decreased with increasing dissolved oxygen content in some wetlands, which may be partly due by aerobic microorganism activities [71]. Although the correlation between heavy metals concentrations and various water quality indexes was different, all of them were positively correlated, except DO. In other words, these heavy metals and water quality likely covaried with each other [72]. Compared with S B , the TP and As, DO and Zn show significant negative correlation, while the correlation between COD, TP and heavy metals became insignificant in S D (Table 6). River surface runoff, rainfall, and agricultural irrigation increased in S D . Zn and TP in agrochemicals and fertilizers enter into the wetland system through irrigation water or rainfall/agricultural runoff [54,61]. Because of temperature and hydrological conditions in S D , DO concentrations often decreased greatly. As is also not deposited in wetlands water. Thus, DO and Zn show significant negative correlation and water quality results from S B can differ from the "first flush" effects on the metal concentrations.

Redundancy Analysis
The results show that the first two axes explained 82.3% variation of water quality data. Among the five water quality parameters, COD (IF = 1.3) and DO (IF = 0.9) had significant impacts on the concentration of heavy metals in S D . Pb and As were in the first quadrant, which contributed to the potential factors of DO and COD with their response values to the heavy metals greater than 0. Zn was in the third quadrant, which contributed to the potential factors of TP, NH 3 -N, and TN. In contrast, NH 3 -N had the largest influence on Zn, while TP had the smallest. The arrow direction of COD was opposite to that of Cd and Cu and the included angle was obtuse, indicating that there was a negative correlation between them, but it was not significant (Figure 4a). The first two axes explained 96.89% of variation in water quality data. Among the water quality parameters, DO (IF = 3.2), COD (IF = 1.8) and TP (IF = 1.0) had significant influences on the content of heavy metal elements in S A . Hg and Cu were in the first and fourth quadrant, respectively; TP, TN, and COD may have influenced Hg, among which TP had the most significant influence. The Zn was located in the second quadrant and the Cr, Cd, As in the third quadrant, DO and NH 3 -N had certain potential influences on them, among which DO and NH 3 -N had the most significant influences on Cr and As, respectively (Figure 4b).
The results show that the first two axes explain 86.81% of variation in water quality. COD (IF = 1.8), DO (IF = 1.3), TN (IF = 1.3) had significant impacts on the content of heavy metal elements in S B . The Zn and As were in the first quadrant and the Cu and Cr in the fourth quadrant; TP, TN, COD and NH 3 -N may have influenced them. TP had the most significant impact on As and COD had a greater impact on these heavy metal elements. The Hg and Pb were in the fourth quadrant DO influenced them and Hg was relatively affected by DO (Figure 4c).
Part water quality parameters and heavy metals concentrations were correlated with each other, may indicate a common origin of pollutant. Correlations between metals may reflect their common source and similar migration behavior [45]. Previous research results indicate that reduced river flows and increased pollution values and agriculture-driven water depletions have caused a critical situation in water quality of the downstream Yellow River [24,54]. The wetland water in S B is rich in heavy metals and the release of heavy metals can change the physical conditions of freshwater environments, threatening aquatic organisms and reducing the richness and diversity of benthic species [73,74]. However, studies of the relationship between water chemical parameters and heavy metal content are less frequent. We showed that there were diverse potential relationships between multiple factors, that sources of various pollution types may be similar. Among them, DO had a great influence on heavy metal concentrations. DO concentration is an essential parameter that maintains the equilibrium of aquatic ecosystems. It is commonly used to assess water resource quality. Generally, a high DO concentration in the natural state indicated a better water state [55,56]. Prior research had shown DO exhibited weak negative correlations with the heavy metals, proving that it could not be the potential factors of influence on the heavy metals in natural river [45]. In contrast, because artificial oxygen pumps raised wetland water DO concentration in our study area, DO had potential positive influences on heavy metals including Pb, As, Cr, Cd and Hg in different hydrological seasons, indirectly proving that heavy metal pollution is caused by human interference. The effects of environmental parameters on wetland habitats are significant [12]. High concentrations of heavy metals in the surrounding environment can result in reduced relative abundance and diversity of organisms by affecting the balance of the food web, resulting in a considerable potential risk to the wetland ecosystem [75]. The lack of necessary infrastructure and proper management caused environmental pollution at the study region [62] and our study can provide suggestions for water treatment and management.

Conclusions
The mean I wq value was highest and lowest during the flood and pre-flood seasons respectively. However, the mean HPI value was lowest in the flood season and highest in the pre-flood season. The excessive I wq value may be due to the pollution of agricultural and domestic wastewater. The main reason for the excessive HPI value may be the small water volume in the Yellow River and the low precipitation in study area, which lead to slow flow of wetlands water and the deposition of heavy metal elements.
There were significant spatiotemporal differences in water quality parameters and heavy metal concentrations in wetlands along the middle and lower Yellow River. TN, COD, Cu, and Cd explain the spatiotemporal changes of water body status. The spatiotemporal difference of water quality parameters may be mainly caused by the discharge of agricultural and domestic sewage around the sample sites. Excepting Hg, the heavy metals likely originate from human sources (mainly agricultural influence) and the seasonal difference of heavy metal concentrations was caused by different degrees of human activity across the year. Due to human disturbance and a decrease in runoff from the Yellow River, Hg element was deposited in wetland waters during the pre-flood season. During flood season, the affected area with hanging river was seriously polluted and the pollutants may be accumulated mainly through natural factors (the Yellow River lateral infiltration mechanism) [76]. The artificial wetlands were seriously polluted during the post-flood and pre-flood seasons. Sites along belt transect #03 were divided into a cluster with serious pollution, indicating that this belt transect was seriously polluted.
There was a potential correlation between the water quality parameters and heavy metal elements, but their mean concentration was distributed in reverse during different hydrological season. Among them, dissolved oxygen had a great influence on heavy metal concentrations. There were diverse potential relationships between multiple factors, suggesting that sources of various pollution types may be similar.  Acknowledgments: The authors would like to thank Xunling Lu, Guofu Liang, Qiongqiong Kang, Xiaoyu Ji, Yapeng Ding for they help when the sampling. We would like to express our sincere thanks to the anonymous reviewers. Their insightful comments were helpful for improving the manuscript.

Conflicts of Interest:
The authors declare no conflict of interest.

Appendix A
We added the more detailed characterization of environmental settings of the studied wetland in Table A1.