Sources and routes from terrestrial exogenous pollutants affect phytoplankton biomass in reservoir bays Open Access

Hydraulic engineering has greatly prevented and alleviated natural disasters caused by floods and droughts and plays a vital role in water supply, irrigation, power generation, and so on (Chen et al. 2016). Until now, there have been more than 50,000 dams built of height over 15 meters all over the world (Lehner et al. 2011); however, reservoir operation has changed the natural hydraulic mechanism of rivers so that even water connectivity is prevented, which has caused water self-purification capacity to decline and the river ecological environment to change (Gao et al. 2018; Li et al. 2021; Xiang et al. 2021). These changes will further threaten water quality.

Reservoir bays, semi-enclosed water bodies at terrestrial and water boundaries, are affected by their morphological characteristics, and water mobility is turned down in these places. Because they adjoin land, they become the most sensitive places to overland pollutants (Li et al. 2020). There have been quite a lot of reports of water eutrophication events in reservoir bays (Yang et al. 2018; Chuo et al. 2019; Huang et al. 2020), which have aroused extensive concern. What causes the nature of eutrophication in water is phytoplankton blooms; therefore, the key to controlling eutrophication is to inhibit the growth of phytoplankton in water.

Studies on water eutrophication have shown that the input of nitrogen, phosphorus, and other nutrients is the vital inducement for algae outbreaks and water quality deterioration (Paerl et al. 2017; Mamun et al. 2020). According to Liebig's law of the minimum, the required nutrients for plant growth are provided by the external environment; if the amount of a certain nutrient approaches the lower bound, then this nutrient is the limiting factor for plant growth (Chen et al. 2013). The ratio of N:P could serve as an index that represents the nutrient limitation for phytoplankton growth when compared with the average composition of nutrients assimilated in algae (C106:N16:P1) (Fujimoto et al. 1997). Early experimental research established that a high concentration of P and a low N:P supply ratio (<29:1) were favorable for the production of algae blooms (Smith 1983). There have been some successful practices that reduce phosphorus (P) inputs based on the premise that P universally limits primary productivity but not all (Elser et al. 2007; Lewis et al. 2011). The theory this was based on was that nitrogen flow into water could be degraded to gas form through denitrification and other biochemical processes, then circulated at the water-gas interface. However, phosphorus has no gas form in nature, so it is difficult for phosphorus to be degraded in water. As the water flow slows down, phosphorus is deposited in the water and becomes a part of the internal pollution. Therefore, the accumulation rate of phosphorus in water is higher than nitrogen. Many traditional views thought that water eutrophication could only control the input of phosphorus, thus ignoring the limitation of nitrogen (Schindler et al. 2008). Lately, researchers in China, America, Africa, and Europe have discovered that many lakes exhibit varying nutrient limiting and cycling patterns, including periods of P or N limitation, as well as periods where N and P act in concert to facilitate biomass production (Conley et al. 2009; Harpole et al. 2011; Paerl et al. 2011; Chen et al. 2013). Therefore, the management strategy for eutrophic water bodies should overall consider different situations of nutrient limitation, not only one.

Nutrients enriched in lakes or reservoirs are mainly considered to be sourced from overland flows, soil release, and domestic and industrial sewage discharges (Figure 1) (Santos et al. 2017; Zhou et al. 2017). When overland runoff erodes the underlying surface, sediments and contamination will flow into the river then afflux into the lake or reservoir (Chen et al. 2018). The concentration and category of pollutants are influenced by hydrometeorology, land use, and land cover, soil property, and landscape pattern in a watershed (Shi et al. 2013; Yan et al. 2013; Ai et al. 2015; Li et al. 2020). As the extensive management of agricultural, nitrogen and phosphorus fertilizer used in crops always surpasses its use in practice, meanwhile fertilization activity tends to coincide with the rainy season, thus it was inevitable that part of the superfluous fertilizer would erode into surface water, part of which will enrich or infiltrate into the soil (Gao et al. 2004; Menció et al. 2011; Lacher et al. 2019; Xie et al. 2019). Apart from the artificially added nutrients, soil inherently contains nitrogen, phosphorus, organic matter, and nutrients, which will also increase the flow of contamination into the water in dissolved or particulate states (Tong & Chen 2002; Neary et al. 2009; Zhou et al. 2012; Lacher et al. 2019). Besides nonpoint pollution, residential and industrial wastewater discharges construct another pollutant source. Studies in Xiamen Bay – Jiulong River Basin have shown that increasing anthropogenic discharges and agricultural fertilizer loads over the past 30 years were the main causes of high P loadings in water (Chen et al. 2013). Dirk et al. used geospatial statistical methods to discover the response relationships between land use and water quality factors under different spatial scales. The results showed that human activities mainly affected the types and concentrations of water pollutants through large-scale urbanization, and with the expansion of agricultural activities, the natural land use mainly affected the variation of water quality at small scale (Vrebos et al. 2017). Excessive nitrogen inputs from sources such as fertilizer from farmland or domestic wastewater from residential areas have been identified as major contributors to stream nitrogen loading (Berka et al. 2001; Jiang et al. 2014; Zhou et al. 2017; Yi et al. 2020). Previous studies have focused on clarifying the sources and contributions to a certain high loading of water nutrient; however, the category of water pollutants tends to be different in different situations, such as flood or dry seasons, land use and land cover, landscapes, and so on. Therefore, under different nutrient limited types, exploring the sources and routes from terrestrial exogenous pollutants affecting phytoplankton biomass in reservoir bays remains an unsolved problem. This study aims to clarify the dominant nutrient limited types for reservoir bays and the restricting factors for the concentration of Chl a, based on the a priori assumption of: terrestrial exogenous input (agricultural non-point pollution, domestic sewage discharge, soil release) – water nutrient concentrations (nitrogen and phosphorus nutrients) – and water phytoplankton biomass (the concentration of Chl a) to construct the path model, and further identify the sources and transport routes of water pollution from the environmental background.

Humans construct reservoirs because of their indispensable roles in water resource supply, regulation and flood storage, agricultural irrigation, entertainment, and so on; so we should discover the hidden risk of non-point pollution immediately and make scientific guidelines. In this work, we chose reservoir bays as the study area for their special morphological characteristics which made them the most sensitive place for the convergence of non-point pollution. Based on the a priori causal path of terrestrial exogenous input – water nutrients concentration – water phytoplankton biomass, structural equation models were built to identify non-point pollution sources form terrestrial exogenous pollutants to water nutrients in both the reservoir storage period and discharge period. The major objectives of this study were: (i) to identify the temporal and spatial distribution of chlorophyll-a concentration in storage and discharge periods of reservoir bays, (ii) to extract the dominant environmental variables that mainly contribute to water contamination, (iii) to elaborate the influence paths and effects across terrestrial exogenous input- water nutrients – phytoplankton biomass in DanJiangKou reservoir bays. The results will help researchers and local decision-makers to understand the source and dynamics of Chl a; therefore, it is essential to control the input of non-point pollution and eutrophication effectively.

Danjiangkou (DJK) reservoir (32°36′–33°48′N, 110°59′–111°49′E) is located in the upstream of Han River, the largest tributary of the Yangtze River, and is the water source for the Middle Route Project under the South-to-North Water Transfer Scheme, initiated to overcome the spatially unevenly distributed water resources in China (Ai et al. 2015). The area of the DJK reservoir is around 1,023 km² and the watershed of the reservoir is about 17,916 km². The study area contains the DJK reservoir bays and their catchments, and covers an area of 689 km² (Figure 2). The crest elevation was increased from 162 to 176.6 m in 2012, and in 2014 the project was officially opened to supply water to over 20 cities along the line. After the dam was raised, the ecological environment around the reservoir greatly changed, and water quality became an important concern for local and national policymakers. It is because of the critical role of water supply and the high sensitivity for water quality security that we choose the DJK reservoir as a representive place to study the causes and sources of water quality variance.

The elevation of the study area ranges from 0 to 957 m, high and steep in the northwest, low and gentle in the southeast. This area has a typical subtropical monsoon climate, rain and heat during the same period. The average yearly temperature is −16 °C, and the average annual precipitation is between 800 and 1,000 mm, most of which falls during the monsoon season (June to October), thus leads to higher intensity rainfall and overland flow afflux into the reservoir. According to the hydrometeorological rules and flood control requirements, the reservoir usually discharges water from June to October and stores water from May to November of the following year. Based on the regulation of reservoir operation, May was deemed to be the storage period, and September was deemed to be the discharge period. According to the Chinese soil classification system, the major soil types include yellow-brown soil, limestone soils and purple soil (National Soil Survey Office 1998), which correspond respectively to Alfisols, Entisols and Entisols in the USA Soil Taxonomy (Soil Survey Staff 1999). The main land-use type in this watershed is forest, at nearly 70%. Farmland and residential areas are concentrated along the river. The major crops are corn (Zea mays L.) and wheat (Triticum aestivum L.).

Reservoir bays were divided based on the digital elevation model (DEM), with a resolution of 25 m by 25 m. Using Geography Information System (GIS) technology practiced in the hydrological analysis module, according to the threshold of the catchment area, 500 ha to extract bay catchment areas. Follow the rules of homogeneity and universality, we choose 62 reservoir bays, which overall considered altitude, slope and land use types.

We sampled water at reservoir bays from 2015 to 2019 in May (storage period) and September (discharge period). Three sample points were laid at every bay, respectively boundary, center and mouth, the distance between points over 200 m. Water temperature (Temp), pH, dissolved oxygen (DO), turbidity (NTU) and the concentration of Chl a were measured in situ using a YSI EXO2 (YSI Inc., Yellow Springs, Ohio, USA) water quality multiparameter analyzer (Li et al. 2021); other water quality indexes were tested at the laboratory. The concentration of total nitrogen (TN) was determined using the method of alkaline potassium persulfate digestion -ultraviolet spectrophotometry (CSEPB 2002). The concentration of total phosphorus (TP) was tested by ammonium molybdate spectrophotometry (CSEPB 2002). The nitrate nitrogen (NO₃-N) and ammoniacal nitrogen (NH₄-N) concentration were measured by AA3 flow analyzer (FLAstar 5,000 Analyzer). We collected water 10 m away from the shore, depth between 0 and 20 cm, then filled it into a 500 mm polythene plastic bottle, add H₂SO₄ until pH less than 2, cold storage at 2–5 °C, analyze within 24 hours. The total water quality indexes and analysis methods are shown in Table 1.

The primary data used in this study included the Digital Elevation Model (DEM) in the DJK reservoir watershed with a resolution of 25 m by 25 m that was purchased from the National Geomatics Center of China. Climate data were available from nine meteorological stations located within or close to the watersheds. The soil type map (1:100,000 scale) with related soil properties were derived from the Soil Survey Office of Hubei Province. The drone orthophoto map was obtained from the Chang Jiang River Water Resources Commission, with a special resolution of 0.5 m by 0.5 m. The land-use map was based on the orthophoto map via visual interpretation in ArcGIS 10.4.

Topographical variables include elevation, slope, slope length and soil erosion modulus were calculated based on DEM within ArcGIS 10.4. Soil variables contained SOM, STN, STP, SNO₃-N, SNH₄-N. Land-use type comprised water bodies, forestland, shrubland, grassland, garden land, paddy field, dry land, urban land, rural land and unused land, which could be reclassified into water, natural vegetation (forestland, shrubland, grassland), farmland (garden land, paddy field, dry land) and residential area (urban land, rural land) (Figure 3).

The export coefficient (E_i) describes the pollutant load exported from each land use type per unit area per unit time (t/km²·yr) in the catchment, where L is the loss of nutrients (t), A_i is the area of the catchment occupied by land-use type i (km²), or the number of livestock type i, or people, I_i is the input of nutrients to source i (t), and P is the input of nutrients from precipitation (t).

Landscape variables were obtained from the land-use map calculated by the software FRAGSTATS 4.1, which is widely accepted for landscape metrics quantification (McGarigal 2002). The landscape index used in this study included landscape shape index (LSI), contagion (CON), and Shannon's diversity index (SHDI). Abbreviations and descriptions of the selected variables for the environmental characteristics are shown in Table 3.

To analyse the distribution of concentration of Chl a in different stages we used the descriptive statistical analysis method to map the distributions of Chl a concentration by ArcGIS 10.4. To figure out the source and contribution of water contamination from the environmental background, we used the redundancy analysis method, which coupled the principal component analysis (PCA) with Multiple Linear Regression (MLR) has been widely used in water quality research (Santos et al. 2017). The results included the relationship between environmental variables and the ordination axis, what's more, the quantitative decomposition of the independent variables at the ordination axis due to the influence of environmental variables, which showed the relative contribution of explaining variables to predictive variables.

According to the various sources from terrestrial exogenous pollutants – water nutrients – phytoplankton biomass, the PLS-SEM model was built. The first step of PLS-SEM was to propose a conceptual model to reflect the mechanism according to the research problem and the existing understanding of nonpoint pollution. The path routes were built based on the following hypothesis (proven results). (1) Nutrients in the water of reservoir bays were mainly derived from exogenous inputs from the environmental background, including agricultural fertilizer, soil endogenous release and anthropogenic discharge (Onderka et al. 2012; Pratt & Chang 2012; Ai et al. 2015; Pearce et al. 2017). (2) Exploring the influence of water nutrients on Chl a: a multivariate statistical model analysis pollutant concentration can be viewed as the linear sum of the elemental contributions from pollution sources (Helena et al. 2000; Singh et al. 2005). (3) According to the ratio of N:P, reservoir bays could be divided into different nutrient limited classes (N limited, P limited, N + P limited), which results in the major difference for the pollution sources and route effects contribute to phytoplankton (Paerl et al. 2017; Rattan et al. 2017).

The structural equation model (SEM) is a statistical tool based on the established theory of causality hypothesis, which can calibrate and validate the fitness of a model. Wright (1934) first introduced the method in biological population research, and it has been expanded to a wide range of research areas, including social sciences, psychology, chemistry, and biology (Hung et al. 2007; Kashy et al. 2008). SEM combining factor analysis with route analysis contains the relationships of observed variables, potential variables, reveal the direct effect, indirect effect and total effect of independent variables on dependent variables. SEM comprises two basic models, the measurement model and the structural model. The measurement model is composed of potential variables and observed variables, which is the linear function of observed variables, it is usually represented by a rectangular shape. Potential variables are the abstract concept of observed variables, which cannot be measured but can be reflected by observed variables, it is usually represented by an ellipse shape. The structural model declares the causality of variables if variables represent the reason called exogenous latent variables, in this study are the terrestrial exogenous pollutants. If variables represent the results called endogenous latent variables, such as the concentration of nutrients and the biomass of phytoplankton in water.

In the structural model, the direct effects are illustrated by the corresponding path coefficients (β_ji). Indirect effects between potential variables are those pathways that include intermediary variables. The total effects describe the overall relationships from one causative latent variable to another resultative variable, which are the sum of the direct and indirect effects. The model calculation was conducted using R and the R package ‘plspm’ (ver. 0.4.7; Sanchez 2013).

The box chart (Figure 4) draws upon the statistical analysis of the concentration of Chl a from eight times sampling data. The results showed that the range in May (storage period) was always larger than that in September (discharge period), partly because the rainfall was lighter in May; during this period the water mobility was weak in bays, thus making a dispersion distribution of ρ (Chl a). In turn, in September with the increase of precipitation, water mobility increased and caused a more even distribution of ρ (Chl a). Besides, with the increase of exogenous input, the median and average of ρ (Chl a) were generally higher than that in May.

To better understand how the concentration of Chl a was distributed spatially, we mapped the mean concentration of Chl a in two periods within ArcGIS 10.4 (Figure 5). The results showed that in the storage period (May) the high-value zones were scattered. Conversely, this turned to distribution in high-value group intervals with low-value groups in the discharge period (Sep.). This demonstrated that the increase of runoff and the regulation of reservoirs changed the mobility of water and raised the diffusion ability of pollutants. This results indicate that the increase of rainfall-runoff erosion in the rainy season is the driving force of non-point pollution. On the one hand, it increases the total amount of exogenous pollutants input into water; on the other hand, the increase of water fluidity enhances the ability for pollutant migration and diffusion in the water (Ai et al. 2015; Zhou et al. 2017; Xie et al. 2019).

Using redundancy analysis to perform multivariant regression should eliminate the influence of redundancy variables. We use the forward selection method to pick out environmental variables and validated through the Monte Carlo method, the result is presented in Table 4. The slope in topography factors, residential area and natural vegetation in land-use types, STN, SNO₃-N, SNH₄-N in soil variables were the dominant environmental variables that influence water quality in the storage period, while landscape pattern indexes were not so significant in this period. In the discharge period, elevation, slope length, soil erosion modulus among topography factors, residential area and vegetation among land-use types, and LSI, CONTAG, SHDI among landscape pattern variables, could strongly explain the variance of water quality.

In the RDA ordination graph (Figure 6), the proportion of variance explained by the environment variables is shown in Table 4. The results showed that all of the water quality indexes were positively correlated with Chl a; what's more, there were two groups of the cluster in water quality indexes, one being physical water indexes, the other was water nutrient indexes. Among the environmental background variables, the residential area accounted for the largest variation, covering 39%, and this had a higher correlation with TOC, Chl a, COD_Mn and DO. This indicated that residential sewage discharge was the dominant factor that contributed to the increase of water nutrients and phytoplankton biomass, which led to the enrichment of organic pollutants in reservoir bays (Zhou et al. 2012). Soil endogenous nutrients' release was another important source (Pearce et al. 2017); STN, SNO₃-N, SNH₄-N together with slope length contributed 22%. Natural vegetation negatively affected water pollutants, consistent with previous researches (Xiao & Ji 2007), it is proved that vegetation can increase surface roughness and intercept pollutants from overland flow.

As is shown in Figure 6, in the discharge period pH was negatively correlated with the other water quality factors. This indicated that with the increase of external input and the change of water dynamic mechanism, water acidity increased and pH reduced. In this stage, water contamination is mainly attributed to land use, topography and landscape pattern. Residential discharge still accounted for a large part of external input, covering 30%. The contribution of topography factors was quite sophisticated: slope length and soil erosion modulus were positively correlated, with water contamination contributing 19%, while elevation effect was opposite, which coincided with CONTAG in landscape pattern (Pratt & Chang 2012; Vrebos et al. 2017). With the increase of altitude, the interference of human activities reduced and was mainly distributed continuously with vegetation, which could effectively retain non-point pollution (Sliva & Dudley Williams 2001). From the perspective of the landscape, vegetation plays the role of ‘sink’, so the large area of continuously distributed vegetation negatively affects non-point pollution. Landscape indexes LSI and SHDI represent the categories and abundance of patches in a unit area that were positive for water contamination. With the increase of fragmentation and connectivity in the landscape, the fluidity of surface pollutants will increase and the concentration of non-point pollution will aggravate (Uuemaa et al. 2005). Soil endogenous release was not so influential in this period.

The well-established results for nutrient limited study in inland hydrostatic waters indicated that when the ratio of N:P concentration is higher than 23, the limiting nutrient referred to P; when the ratio of N:P concentration is lower than 9 that tended to N limited; otherwise, the nutrients were N + P limited (Paerl et al. 2017). According to this prior classification criteria, the nutrient limited level could be divided into N-limited, P-limited and N + P limited (Figure 7). Different nutrient limited types determine the kinds and amount of nutrients available to phytoplankton, which causes the difference in the biochemical process between phytoplankton and water environmental factors. Therefore, it is necessary to establish a path analysis model based on the nutrient limited classes from the sources of nitrogen and phosphorus.

The results of nutrient limited classification during both periods (Figure 7) showed that the most commonly distributed types were P-limited (35%–47%) and N + P limited (35%–59%), while N-limited situations appeared quite rarely, accounting for less than 20%. In the storage period, P limited (47%) was the dominant distribution type; meanwhile, the N limited situations were the most in this stage (18%). In the discharge period, the pervasively distributed situation was N + P limited (59%); with rainfall increasing, nutrients were transported to reservoir bays with surface runoff becoming more comprehensive, thus quite a lot of reservoir bays converted to N and P joint limited (Paerl et al. 2017; Li et al. 2020).

One source of nutrients concentrated in reservoir bays was from the endogenous release of sediment and another was from the exogenous input of the terrestrial environment. Considering that there existed a water level fluctuation zone between the high and low water level period, part of the external input of the terrestrial environmental background were released from the submerged soil (Shan et al. 2014; Wu et al. 2017). The amount of pollutant confluence in reservoir bays with surface flow was reliant on the types of land cover via along the way. For the application of fertilizer in agricultural practices and the discharge of domestic and industrial wastewater in human activities, the agricultural and residential land uses played the role of ‘source’ for non-point pollution and led to the positive contribution to pollutant concentrations (Zhao et al. 2019). To clarify the sources and effects of input from exogenous pollution, agricultural fertilizer, soil release and anthropogenic discharge were considered as the exogenous latent variables for PLS-SEM. Because of the external input of nitrogen and phosphorus the concentration of nitrogen and phosphorus increased in water, and then influenced the biomass of phytoplankton in reservoir bays. Based on these theories, pollutant transport routes model PLS-SEM was built (Figure 8).

In PLS-SEM, the available nutrients for phytoplankton depended on nitrogen salt and phosphorus salt in water, the exogenous sources of N and P in reservoir bays mainly came from fertilizer, soil release and artificial discharge. N and P fertilizer are mainly from dry land, paddy field and garden land uses. The soil release of N was related to the STN, SNH₄, and SNO₃. While, the soil release of P was considered as related to STP, SAP. The artificial discharge of N and P was concerned with rural and urban land uses (Hua et al. 2019). PLS-SEM were constructed based on N limited, P limited, and N + P limited classes respectively. The total effect, direct effect and indirect effect, of the results of PLS-SEM are shown in Table 6.

The results of PLS-SEM (Figure 9) show that whatever the nutrients limited classes were, the higher effect on phytoplankton biomass was all phosphorus salt. This phenomenon was especially obvious when N and P were joint limited (effect of PS = 0.62). The effect of nitrogen salt was higher than the other nutrient limited types if it was N-limited.

The results of fertilizer, soil release, and anthropogenic discharge effects on nitrogen salt, phosphorus salt and phytoplankton biomass were distinctive among the three nutrients limited types (Figure 9). In N limited reservoir bays, nitrogen salt mainly came from soil release (total effect = 0.6) and nitrogen fertilizer (total effect = 0.43), of which the indirect effect from fertilizer to soil release was 0.3. The sources of phosphorus salt showed a subtle difference that the effect of fertilizer (total effect = 0.22) was a little bit higher than soil release (total effect = 0.17). And the contributions of fertilizer and soil release (total effect = 0.15) were the same as the biomass of phytoplankton. That means when it comes to an N limited situation, fertilizer was the main source that should be paid attention to; what's more, a part of superfluous fertilizer may infiltrate into the soil and provide for phytoplankton growth through overland flow and submerged soil release (Chen et al. 2021; Han et al. 2021).

In P limited reservoir bays, all of the three sources almost act higher on nitrogen salt, phosphorus salt and phytoplankton biomass. Fertilizer to nitrogen salt (total effect = 0.62) and anthropogenic discharge to phosphorus salt (total effect = 0.5) contributed most. In P limited reservoir bays, governance strategies should be overall consideration of the three sources of contamination.

The results of PLS-SEM in N + P limited reservoir bays showed that the effects of anthropogenic discharge were much higher than the other two sources on nitrogen salt (total effect = 0.66), phosphorus salt (total effect = 0.6) and phytoplankton biomass (total effect = 0.43), which were the dominant causes for the deterioration of water quality. Thus the major practices to alleviate water pollution in this type of reservoir bays should focus on the discharge of residential and industrial wastewater (Ai et al. 2015).

In this study, we choose the study area as reservoir bays gave a new suggestion for reservoir water quality research as they combine the influence from both natural and artificial, water and land surfaces. The results of the distribution of Chl a showed the spatial discrepancy during two stages, which represent two different dynamic water mechanisms. The results of water quality factors response to external input indicated that the water pollutants in the storage period mainly originated from domestic sewage and soil release, while in the discharge period, with the increase of precipitation and surface flow, the disturbance from the environment background added topography and landscape pattern. Surface pollutants were easily delivered and transported in complex and broken landscapes but diluted and sedimented by continuous vegetation. To further make out the nutrient sources and contributions in different types of reservoir bays from terrestrial environment background, the PLS-SEM statistical analysis method was used. The main findings were that phosphorus always had a higher effect on phytoplankton biomass, whatever the nutrient limited types were. However, it was also necessary to take measures to control both sources of nutrients. The prevention of eutrophication should emphasize controlling fertilization and anthropogenic discharge first, then the transfer of the nutrients from the overland flow and infiltration to soil and water also cannot be neglected. This study hoped to propose a new solution to the research on the sources and potential risk warning of eutrophication in reservoirs and expected to provide scientific guidance for the control of non-point pollution.

PLS-SEM provides an effective solution to assess the coupled relationships between predictors and water quality characteristics. Nevertheless, for promoting the knowledge of the sources and routes of water nutrients from terrestrial exogenous pollutants that affect water pollution, the PLS-SEM used in this study could be improved by considering more environmental factors and further sources of pollutants such as the pollution loads from livestock production in driving water quality changes at different temporal and spatial scales. We expect that the methods presented in this paper will be useful for hydroecologists wishing to apply PLS-SEM.

Financial support for this research was provided by the National Natural Science Foundation of China (No. 4160010685). We are grateful to the following people for their assistance to the paper. Zhihua Shi: Supervision, review. Lu Li: Resources, Investigation. Lei Ai: Investigation, Funding acquisition, Resources. Jian Wang: Data collection. Wei Yin: Resources, Investigation. Haiyan Jia: Resources. Jianfeng Xu: Investigation, Resources. Xuan Huang: Conceptualization. Yin Zhou: Conceptualization. Nanxin Li: Data collection, Investigation, Field experiments, Review. Xiaolin Zhao: Data collection, Field experiments. Cheng Liu: Data collection, Field experiments. Peipei Han: Data collection, Field experiments. Lingyu Xia: Data collection, Field experiments. Jiacheng Sun: Field experiments. Guoying Zhang: Field experiments. Gen Liu: Field experiments. Qiaozhi Chen: Field experiments. Huaqing Liu: Field experiments. Rui Hao: Field experiments. Zhiming Zhong: Field experiments.

Data cannot be made publicly available; readers should contact the corresponding author for details.