The middle route of the South-to-North Water Diversion Project is located between 32.68°N–39.99°N and 111.72°E–116.27°E, which spans eight latitudes and three temperature zones (subtropical zone, warm temperate zone, and temperate zone) from south to north (Luo et al., 2018), with a length of 1,432 km; it needs 2 weeks to divert water from the original reservoir to terminal water bodies. The MRP originates from Danjiangkou Reservoir, which crosses two provinces—Henan (HN) and Hebei (HB), and arrives at Beijing (BJ), the capital of China, and Tianjin (TJ). Water flows aided by gravity along the channel from Danjiangkou Reservoir (whose storage has been expanded) through the intake gate of the Taocha headwater. Along the middle route, canals pass the west Tangbai River region, the watershed between the Yangtze River basin and the Huaihe River–Fangcheng Pass, run across the Yellow River Basin, and finally arrive in BJ. This inter-basin project solves the problem of water shortage and provides clean water for production, living, industry, and agriculture for at least 20 large- and medium-sized cities along the route (Wang, et al., 2021). The MRP canal was closed and separated from the land, so the local catchment cannot influence the water quality, but the rainwater and dry/wet settlement from the air to the channel could affect the water quality. By December 2020, the MRP supplied a total of 34.8 billion cubic meters of water transported northward, benefiting 69 million people (Office of the South-to-North Water Diversion Project Construction Committee, State Council, PRC).

There were 11 monitoring sites chosen to represent the MRP, covering four main provinces including the headstream and terminal places (as shown in Figure 1). The 11 sampling sites were set almost at equal distances; the regional aspect was also considered in the sample site design. S1 (Taocha) was located in the Henan Province, which is the source of headwater. S2 (Shahe), S3 (Lushan), S4 (Chuanhuang South), and S5 (Chuanhuang North) were all located in the HN Province. S6 (Zhanghe North), S7 (Guyunhe), and S8 (Xiheishan) were located in the HB Province. Xiheishan was an intersection site where the main channel divides into two branches: one flows to S9 (Waihuanhe) in TJ and the other flows to S10 (Huinanzhuang station) and S11 (Tuanchenghu) in BJ; the detailed sampling information is given in Supplementary Table S1.

The investigation was conducted every quarter. Parameters including WT, DO, and pH were obtained using a water quality analyzer (YSI, proplus, United States). Velocity and discharge were recorded using a flowmeter (Flowatch, Switzerland). The turbidity was determined using a turbidimeter (WGZ-200S, China). A Secchi disc tool measured SD. The water samples (from a surface depth of 0 cm–50 cm) were collected in 550 mL acid-cleaned plastic bottles and rinsed with surface water before sampling (three replications in each monitoring site). The samples needed to be stored in an icebox and were transferred to the lab to be determined as soon as possible. Samples used for the measurement of PO₄-P, NO₃-N, and NH₄-N were filtered using a GF/C membrane (Whatman GF/C), and then the membrane filters were soaked in 10 mL, 90% acetone solution to extract Chl a. Furthermore, TN and TP were digested at 120°C for 30 min and tested using a UV-spectrophotometer (UV-1780), and chemical oxygen demand (COD_Mn) was determined; the aforementioned parameters were determined following the standard methods for the examination of water and wastewater (APHA., 2017).

To understand the trophic condition of MRP, TLI values were used, which were given a weighting sum based on the correlation between TP, TN, COD_Mn, SD, and Chl a (China Environmental Monitoring Station., 2001). These parameters were applied for calculating the TLI, and the qualification formulas of five parameters were established, which was the calculation result of the investigation of 26 major lakes (reservoirs) in China (Jin et al., 1995).

The TLI has a scale from 0 to 100, with a high value indicating a high trophic level, and was used to assess the eutrophication level of monitoring sites of the MRP. The trophic level is classified into five classes according to the score of TLI (∑): the oligotrophic level (TLI (∑) < 30), mesotrophic level (30 ≤ TLI (∑) ≤ 50), light eutrophic level (50 < TLI (∑) ≤ 60), middle eutrophic level (60 < TLI (∑) ≤ 70), and hyper eutrophic level (TLI (∑) > 70) (Wang et al., 2019).

The water quality index (WQI) is a widely used method to evaluate water quality and reflect the comprehensive impact of different environmental parameters, as well as obtain an objective assessment of the comprehensive water quality without neglecting the overall water quality status because of the deviation of individual factors that were put forward by Pesce et al. (2000). The WQI has been widely used as one of the most effective approaches to performing water quality assessments, integrating multiple measured water quality parameters. In this study, 15 physicochemical parameters collected from eleven sampling sites were selected as evaluation objects to form a 15 × 11 matrix. The principal component formula was deduced by calculating the eigenvalues and the coefficient matrix to quantify the factors. Then, stepwise multiple linear regression analyses were carried out to figure out the key factors affecting WQI and then develop the WQI_min model, which helps to select the influencing key factors.

According to the results of factor analysis and correlation analysis, 11 environmental parameters were selected and weighted. Every single parameter was assigned a weight based on its potential effect on the water quality (Wu et al., 2018). T, DO, pH, COD_Mn, Tur, Chl a, TP, PO₄-P, TN, NH₄-N, and NO₃-N were used to compute the WQI. The formula used was as follows: W Q I = ∑ i = 1 n C i × P i / ∑ i = 1 n P i , (8) where n is the amount of the selected parameters, Ci is the normalized value of parameter I, Pi is the weight of parameter I, the minimum value of P_i was one, and the maximum weight allocated to parameters that influenced the water quality most was four (Supplementary Table S2). These values have been demonstrated in published studies (Koçer et al., 2014; Sun et al., 2016; Tian et al., 2019). The WQI has a scale from 0 to 100, with high values predicting good water quality conditions of each monitoring site of the MRP. Based on the grade of the WQI, the water quality was categorized into five grades: excellent (100–90), good (90–70), medium (70–50), bad (50–25), and very bad (25–0) (Tian et al., 2019).

In order to assess the water quality of the MRP in an easy and effective way, a WQI_min model built during a previous research study was used (Pesce et al., 2000). This model usually uses 3–5 important parameters to evaluate the water quality, per the following formula: W Q I min ⁡ = ∑ i = 1 n C i / n . (9) Here, n is the total number of parameters included in the study; Ci is the normalized value of parameter i (Supplementary Table S2). The selected parameters were introduced into WQI_min on the basis of the results of linear regression analysis. All samples collected from the four seasons were applied for calculating the selected environmental parameters and the relation between WQI with TLI and WQI_min.

Although the water velocity could influence the water quality significantly, spatio-temporal differences in the water velocity of the MRP were not significant, so the water velocity was not used to compute the WQI.

All data analyses were conducted with SPSS 22.0 (IBM SPSS statistical), PAST 3.0 (Paleontological Statistics), and R (Corrplot). Considering the heterogeneous distribution of sampling sites, inverse distance weighting (IDW) interpolation was applied to reveal the spatial variation of physicochemical parameters. ArcGIS 10.2 (Esri Inc., US) was used for the spatial data treatment. The Kruskal–Wallis method was used to evaluate significant differences between the mean of parameters at the spatial and temporal scales. The variations of environmental parameters in the seasonal changes were found using the Kendall test. For grouping the sampling sites, cluster analysis was carried out using single-linkage distance and square-root data transformation. A Spearman correlation analysis was conducted to study the relationship and dependence between environmental factors. Principal component analysis (PCA) is a quantitative analysis of multi-dimensional factors in the system, which can compress multiple variables into individual comprehensive variables that reflect the problem (Tripathi et al., 2019). Stepwise multiple linear regression analyses were carried out to find the key factors affecting TLI and WQI and to develop the WQI_min model, which helps to select the influencing key factors. Based on the Pearson correlation analysis between WQI, TLI, and WQI_min, the slope has explanatory significance.

The average water temperature (WT) was 18.79°C ± 8.78°C, the maximum value was 30.36°C in summer, and the lowest was 6.79°C in winter (Figure 2A). The Kruskal–Wallis test showed significant differences between seasons in WT (p < 0.001). The pH ranged from 7.34 to 9.16, and the mean value was 8.56 ± 0.30 (Figure 2B). The average turbidity was 4.32 ± 5.17 NTU, and the maximum was 7.58 NTU in spring (Figure 2C). In addition, SD ranged from 0.40 m to 6.00 m, and the mean value was 3.19 ± 1.68 m (Figure 2D). The Pearson analysis found that the turbidity and SD had a negative relationship (R ² = −0.79) (Supplementary Figure S1).

COD_Mn was in the range of 0.80 mg·L⁻¹–3.34 mg·L⁻¹, the mean value was 2.15 ± 0.80 mg·L⁻¹, and the lowest value was 1.02 mg·L⁻¹ in winter (Figure 2E). The Kruskal–Wallis test showed that COD_Mn in summer had differences in autumn and winter (p < 0.01). There were significant differences between values in spring and winter (p < 0.01). Pearson correlation analysis illustrated that there was a positive correlation between COD_Mn and Chl a (R ² = 0.57) (Supplementary Figure S1). DO ranged from 4.76 mg·L⁻¹ to 18.10 mg·L⁻¹ (Figure 2F). Pearson correlation analysis showed a negative correlation between DO and COD_Mn (R ² = −0.69) and WT (R ² = −0.83) (Supplementary Figure S1). The content of Chl a was in the range of 0.260 μg·L⁻¹–27.681 μg·L⁻¹, with an average of 4.079 ± 4.760 μg·L⁻¹ (Figure 2G); it first increased and then decreased from south to north. The order of Chl a concentration by season was summer > spring > autumn > winter (p < 0.05). The maximum value of Chl a was 27.681 μg·L⁻¹ at Chuanhuang North in summer. It could be seen that Chl a had notable spatial differences in spring, summer, and autumn (p < 0.05).

TP varied from 0.001 mg·L⁻¹ to 0.077 mg·L⁻¹, with a quarterly value of 0.025 ± 0.024 mg·L⁻¹ (Figure 3A). TP increased significantly from south to north in spring, with an average of 0.063 ± 0.006 mg·L⁻¹, while it was stable in the other three seasons, with an average value of 0.012 ± 0.010 mg·L⁻¹. The Kruskal–Wallis test showed significant differences between spring and the other three seasons (p < 0.01). The concentration of PO₄-P was 0.001 mg·L⁻¹–0.143 mg·L⁻¹, with a quarterly mean value of 0.024 ± 0.028 mg·L⁻¹ (Figure 3B), and the maximum was 0.055 ± 0.007 mg·L⁻¹ at Lushan in spring.

The maximum value of TN was 2.216 mg·L⁻¹, measured at Tuanchenghu during spring. The average TN was 1.786 ± 0.124 mg·L⁻¹ in spring, which was higher than that in summer (1.493 ± 0.190 mg·L⁻¹) (Figure 3C). Meanwhile, TN in the channel showed an increasing variation in spring from south to north (Supplementary Table S3). The range of NH₄-N was 0.001 mg·L⁻¹–0.143 mg·L⁻¹, and the average was 0.068 ± 0.042 mg·L⁻¹. Ammonia (NH₄-N) accounted for 5% of TN (Figure 3D), and there was a notable difference between spring and summer (p < 0.01) as well as in the relationship between autumn and winter (p < 0.01). NO₃-N occupied 70% of TN, and other forms of nitrogen had a percentage of 30%. NO₃-N varied from 0.155 mg·L⁻¹ to 1.696 mg·L⁻¹, with a seasonal average of 0.871 ± 0.457 mg·L⁻¹ (Figure 3E). There were differences between summer, autumn, and spring (p < 0.01). In terms of spatial variation, there was a notable increase in trend along the MRP in spring (Supplementary Table S3).

Single leakage distance clustering analysis showed all samples were clustered into four groups, which reflected a clear seasonal change in the water quality. All samples were bunched into two branches in each season, which indicated notable spatial differences (Figure 4). Thus, water quality has a distinct characteristic of spatio-temporal differences.

Water quality describes the state of the aquatic environment and measures the availability and suitability of water (El-Serehy et al., 2018). The status and variations of water quality can be assessed, which is critical for identifying its spatial and temporal pattern (Wu et al., 2018). According to the TLI and WQI assessment, the trophic level index and the water quality status were classified as “mesotrophic” and “good” in the MRP, respectively. However, a study found that the water quality still had a risk of deterioration from upstream to downstream; TLI and WQI in spring were worse than those in other seasons, which indicated the spatio-temporal heterogeneity of the water quality in the MRP.

A previous study also found that spring had the lowest WQI in the MRP (Nong et al., 2019). This was consistent with our results, that the highest TLI and the lowest WQI were in spring (Figure 5), which indicated that the water quality confronted more risks in spring than in other seasons. The water quality was rated as “poor” in the dry season and as “good” in the wet season in the Aksu River (Şener et al., 2017); WQI was higher in summer than in other seasons in the Luanhe River (Tian et al., 2019); TLI in the wet season reached a lightly eutrophic level, which was higher than that in the dry season in Wuli Lake (Wang et al., 2019). The water quality in these waters exhibited a distinct seasonal variation compared to that of the MRP. A concerning result showed the trophic index and the water quality level reflecting temporal heterogeneity in the MRP, especially during spring.

In terms of spatial variation, it was verified that different regions could show distinct water quality characteristics. It was found that high pollution mainly happened at the beginning and the end of rivers, whereas less pollution occurred in the middle parts of rivers (Hajigholizadeh et al., 2017). In the upstream of the Luanhe River, the water quality was rated as “good,” which was better than that in the downstream, evaluated as “bad” (Tian et al., 2019). WQI classified the Dongjiang River as follows: upstream sites as “good” (WQI = 79.1–83.2) and downstream sites as “medium” (WQI = 63.6–64.1) (Sun et al., 2016). This spatial heterogeneity was also observed in our study. The MRP, with a length of 1,432 km, spans different latitudes and tends to form different regions with different aquatic conditions (McLean et al., 2016). TLI increased and WQI decreased from south to north; the end section (Tuanchenghu) had light eutrophication. The water quality might deteriorate from the initial headwater source (upstream) to terminal water bodies (downstream). In addition, the Chuanhuang project was divided into two parts according to the variation of environmental parameters. Previous research studies revealed that the Yellow River crossing projects as a culver tunnel with a length of 4,250 m; an internal diameter of φ7 m (Chuanhuang South to Chuanhuang North) produced a distinguished difference in the water quality before and after. The special hydraulic project might have a certain impact on the aquatic environment (Luo et al., 2019). In summary, both geographical and seasonal effects resulted in the spatio-temporal pattern of the water quality in the MRP.

WQI_min simplified water quality assessment by using key parameters based on WQI and TLI, contributing to a more rapid and efficient evaluation of water quality statuses (Wu et al., 2018) and making it easy to obtain and represent the major environmental parameters (Pesce et al., 2000). The most influential parameters in the arithmetic of WQI_min were N and COD_Mn in Lake Taihu (China) (Wu et al., 2018), which were also emphasized in the Dongjiang River (China) (Sun et al., 2016) and the Aksu River (SW-Turkey) (Şener et al., 2017). COD_Mn and Chl a played important parts and were included in the WQI_min of the Luanhe River (China) (Tian et al., 2019). In Dianchi Lake (China), TN, Chl a, and COD_Mn were selected in the water quality assessment model, which made great contributions (Yang et al., 2018). It concluded that some factors such as TN, Chl a, and COD_Mn had been widely introduced into WQI_min models because of their great impact on water quality assessment.

Comprehensive and full consideration of the weights of selected parameters improved the accuracy of WQI_min, which was verified in previous studies (Koçer et al., 2014; Barakat et al., 2016; Tripathi et al., 2019). Compared to other water bodies, the proposed WQI_min model in this research was composed of three vital parameters, i.e., TN, Chl a, and COD_Mn, and evaluated the water quality of the MRP, constructing stricter standards on the spatial and seasonal analysis of the water quality. WQI_min could provide simpler measurement methods and lower analytical assumptions (Wang et al., 2019), which has a worthy application for saving testing parameter costs and improving evaluation efficiency.