Multimedia Nitrogen and Phosphorus Migration and Source Control Using Multivariate Analysis and XGBoost: The Case Study in a Typical Agricultural Basin, Danjiangkou Reservoir

Abstract

Preventing eutrophication requires a deep understanding of nutrient sources and migration processes. The Guanshan River in the Danjiangkou Reservoir was selected as a typical agricultural basin. River water, sediment, and soil samples were collected to determine nitrogen (N) and phosphorus (P) contents and forms. Multivariate statistical analysis, buffer analysis, and extreme gradient boosting regression (XGBoost) were used to investigate the nutrient sources, the sources’ scale effects, and migration. The results showed that the exogenous sources of river nutrients were fertilizer (provided phosphate (PO₄-P) and ammonium (NH₄-N)), domestic wastewater (provided nitrate (NO₃-N) and organic nitrogen), and natural soil and rock leaching (provided PO₄-P and organic phosphorus). Fertilization within 300–1200 m and domestic wastewater discharge within 150 m of the river affected the contents of the river nutrients significantly (with R ranging between 0.40 and 0.73, p ≤ 0.01). The sediment was the N source and P sink of the overlying water. With NH₄-N/PO₄-P compound fertilization, the NO₃-N and calcium-bound phosphorus (Ca-P) exhibited co-enrichment. Appropriately increasing NH₄-N application could stimulate P biological uptake, thus inhibiting P emigration from agricultural soil under N-limited states. In conclusion, this study effectively recognized river nutrient sources and their scale impacts and also explored more effective fertilization strategies, which are beneficial for the optimized management of agricultural activities.

1. Introduction

Eutrophication is a crucial concern for the water supply, especially in large water sources such as the Danjiangkou Reservoir [1]. As human society develops, river and lake contamination becomes significant. The elevated nutrient load causes severe eutrophication and ecological imbalance [2]. Nitrogen (N) and phosphorus (P) are important nutrients in water that operate as limiting variables for biological growth [3,4]. Understanding N and P sources and migration is crucial for preventing eutrophication; thus, it has attracted broad attention.

Source identification is the cornerstone of the control of eutrophication [5]. Considerable progress has been made in discriminating nutrient sources and their contributions [6,7,8,9,10,11]. Rock and soil weathering, as well as atmospheric deposition, are major natural sources of water nutrients [6,7,8]. Primary anthropogenic sources include excessive fertilizer usage and the discharge of untreated domestic, industrial, and poultry wastewater [12]. Agricultural activities generally provide a larger amount of nutrients to surface water compared with other sources [12]. For example, fertilizer is the dominant source of P in water bodies, and the total phosphorus (TP) loads in lakes from agricultural land (61%) far exceed those from forestland (21%) and urban land (18%) [13]. Many studies have made achievements in identifying anthropogenic sources and their contributions to nutrients [10]. However, only definite spatial influence scopes of pollution sources could provide accurate land use regulation suggestions; these have received little attention so far [14].

Furthermore, during migration at water–sediment or water–soil interfaces, nutrients may undergo morphological transformations driven by physicochemical and biological processes, which may have an impact on nutrient accumulation in water [15,16]. N and P interactions in different forms considerably impact nutrient adsorption and release in environmental media [11,16,17]. Previous research has proved that NO₃-N accumulation in the sediment maintains an aerobic environment, hence inhibiting phosphate (PO₄-P) released from iron/aluminum-bound phosphorus (Fe/Al-P) [18,19,20,21,22]. However, NO₃-N accumulation processes, such as nitrification and mineralization, could also produce proton (H⁺) [23] and organic acid, thus dissolving detrital P in the sediment [24,25], resulting in PO₄-P enrichment in the water [26]. NH₄-N can stimulate biological activity in the soil/sediment, which has diverse impacts on PO₄-P content in the water under different nutrient limitations [27,28]. PO₄-P can also affect microbial activity and redox conditions, further influencing the nitrogen transformation in water and solids [29]. Comprehending the mechanism behind the interactions between N and P could assist with the synergistic management of nutrients [16,30]. Further research is still required due to the complexity of the N and P coupling interaction under different conditions.

Remote sensing (RS) and geographic information systems (GISs) provide trustworthy data and platforms for the spatial characterization of pollution sources [31]. Some research has applied a buffer tool within GIS software (Arcgis 10.2) to analyze RS data and water quality metrics and has discovered that as the distance from the water body increases, the impact of pollution sources on water quality diminishes, indicating that pollution sources have scale effects [14,32,33]. However, due to the improper selection of fundamental buffer units and the overgeneralization of river water quality, only a few research studies have accurately evaluated the spatial scopes of scale effects [33]. Further exploration is needed to improve the precise identification of source scale effects [14,33]. Deep data mining is required for studies on the transport of nutrients among environmental media. Machine learning provides a variety of methods for the mining, but most of them have high requirements in terms of data size, making them unsuitable for environmental sample data analysis [34]. Extreme gradient boosting (XGBoost) is a highly advanced nonlinear machine learning approach that is resistant to overfitting and suitable for small datasets [35]. It is conducive to the solving of inherent nonlinear correlation and the accurate path prediction of environment nutrients [34].

The Danjiangkou Reservoir (DJKR) is the water source for the Middle Route of China’s South to North Water Transfer Project. The water quality of the DJKR is related to water supply safety in North China. Although the water quality of the DJKR was stable and good in 2015–2019 [10,36], the water deterioration (mainly eutrophication) in the major tributaries flowing into the reservoir was evident [7,37,38], which poses a potential threat to the water quality of the DJKR. It is urgent to explore nutrient migration and regulation in the tributaries. The Guanshan River is a typical agricultural polluted tributary located in the southwest of the DJKR [37,39]. Although the basin area of the Guanshan River is small, the nutrient output per unit area was relatively high [39]. Short runoff [40] and intensive agricultural activities in the basin entrained large amounts of N and P into the DJKR [39]. However, existing studies in the Guanshan River mainly focused on monitoring water quality in the estuary [37,39,40], with little attention to N and P regulation at the basin scale.

To enhance the precise management of the nutrient sources and the synergistic regulation of N and P at the basin level, while considering the protection of the DJKR, the Guanshan River Basin was chosen as the study area. Multimedia sampling and interdisciplinary approaches were employed to identify diverse sources of nutrients in different forms and to explore the interactions of N and P across various environmental media at the basin scale. Furthermore, the study tried to optimize buffer analysis combined with GIS technology, improving the precision in the identification of the scale effects of nutrient sources on the river water quality and providing regulation suggestions from multiple perspectives. The findings of this study could provide insights for the guidance of eutrophication management within the DJKR and other comparable agricultural basins.

2. Materials and Methods

2.1. Study Area

The Guanshan River (110°48′00″–111°34′59″ E, 32°13′16″–32°58′20″ N) is a typical agricultural tributary of the DJKR, with a catchment area of 465 km². The Guanshan River Basin falls into a subtropical, semi-humid climate zone with an annual average temperature and precipitation of 15.9 °C and ~880 mm, respectively. The rainy season of the basin extends from July to September. Its elevation ranges from 150 m in the northeast to 1635 m in the southwest, with an average elevation of 645 m.

Quartzite, granite, tuff, and metasedimentary rocks, such as quartz schist, constitute the majority of the basin’s rock, with Quaternary loose sediments around the river. The basin is dominated by forest land, followed by agricultural and residential land, with only a few industrial enterprises and livestock farms. The primary crops grown in agricultural areas are corn and wheat, and in the summer, N-P-K compound fertilizers are heavily applied.

2.2. Sampling and Measurement

A field study was conducted in the Guanshan River Basin in July 2021. Samples of water and sediment were collected from the Guanshan River’s main channel, its tributaries, the Guanshan Reservoir, and the Han Reservoir (HR, the western part of the DJKR). At the reservoir sites, samples were mixed from the shallow, medium, and deep water layers. For a comprehensive understanding of nutrient sources, the soil samples, including undisturbed soil and agricultural soil, were obtained from the upper to lower reaches. In total, 55 water samples (including 50 river samples and 5 reservoir samples), 44 sediment samples, and 26 soil samples (13 samples of undisturbed soil and agricultural soil, respectively) were collected (Figure 1).

For the water samples, dissolved inorganic nitrogen [ammonium (NH₄-N), nitrate (NO₃-N), nitrite (NO₂-N)], total dissolved nitrogen (TDN), total nitrogen (TN), dissolved inorganic phosphorus [phosphate (PO₄-P)], total dissolved phosphorus (TDP), total phosphorus (TP), cations [K⁺, Na⁺, Ca²⁺, Mg²⁺, and dissolved Si (DSi)], and anions (Cl⁻ and SO₄²⁻) were considered. The organic nitrogen (ON) [or organic phosphorus (OP)] contents in the water were calculated by the difference between the TDN (or TDP) and dissolved inorganic N (or P) forms [41]. For the soil and sediment samples, exchangeable N (Ex-N, including NH₄-N, NO₃-N, and NO₂-N), TN, primary inorganic P forms [including iron/aluminum oxidated P (Fe/Al-P), calcium-P (Ca-P)], inorganic P (IP), OP, and TP were considered. The detailed test method is listed in the Supplementary Material.

2.3. Data Analysis

2.3.1. Multivariate Analysis

Multivariate statistical methods have been widely utilized for pollution source identification [7]. In this study, cluster analysis (CA) and redundancy analysis (RDA) were combined for nutrient source identification in the Guanshan River Basin. CA is a multivariate statistical analysis approach based on an unsupervised support vector machine that classifies objects according to the similarities of the samples; it is widely used to identify pollution sources using feature elements [42]. RDA is a sequencing method that combines regression analysis with principal component analysis, which was utilized to determine the nutrient transport and contribution among environmental media [43]. Pearson’s correlation was utilized to investigate the linear relationship of N and P within different media (water, sediment, and soil). Multivariate statistical analysis was performed in Origin Pro 2021 and SPSS 23 software, and the results were plotted using R 4.1.2 software.

2.3.2. Buffer Analysis

Buffer analysis is a spatial layer overlay analysis method that relies on RS, GIS, and global positioning systems technology [29,32,33]. Previous studies mainly used reaches to conduct buffer zones, which generalized the characteristics of rivers to some extent [29,32]. This study created buffer zones based on uniformly distributed river sample points to increase the identification accuracy of the scale effects of nutrient sources. Multiple buffer zones were built around 22 river water sample sites uniformly distributed from the upper to lower reaches using ArcGIS 10.6 software. Precomputation identified 6 radiuses for constructing buffer zones, namely 50 m, 150 m, 300 m, 1000 m, 1200 m, and 1400 m. The agricultural area (residential area) of each buffer zone was extracted in ArcGIS using land use/land cover (LULC) data (Sentinel-2, 10 m resolution) from Esri 2020 Landcover (https://livingatlas.arcgis.com/landcover/ (accessed on 1 May 2020). The LULC data indicated that the agriculture area accounts for 77% of the total area occupied by anthropogenic land types (Figure 1). Following the extraction of agricultural and residential areas, Pearson correlation analysis was conducted to explore the correlations between the agricultural and residential areas and the nutrients in the river (PO₄-P, TDP, TP, NO₃-N, NO₂-N, NH₄-N, TDN, and TN) in each buffer zone, in order to identify the spatial influence scales of nutrient agricultural and domestic sources.

2.3.3. XGBoost Regression

To solve the inherent nonlinear characteristics of nutrient concentrations, XGBoost analysis was utilized. XGBoost is a machine learning algorithm based on a gradient boosting decision tree that is used to solve nonlinear regression problems [35]. It is one of the most advanced machine learning algorithms and is capable of achieving very accurate prediction performance by combining a sequence of models with low performance [44]. Because of the robustness to overfitting, XGBoost largely facilitates the model selection and performs well in the presence of small datasets [34]. Generally, XGBoost aids in the development of conventional regression models based on the decision tree method using following equation (Equation (1)) [44].

$$f\left(L\right)={\sum }_{i=1}^{n}l\left({y^i,y}_L^p\right)+{\sum }_{j=1}^L\phi \left(f_j\right),$$

(1)

where n is the sample numbers; $y_L^p$ is the prediction of $i$ sample at iteration L; $l$ is the loss function; and $\phi$ is the regularization event. XGBoost regression was performed in Python 3.1.2.

3. Results

3.1. Distribution of Nitrogen and Phosphorus in Surface Water

3.1.1. Nitrogen in Surface Water

The summary of the physical and chemical properties of the surface water samples in the Guanshan River Basin is shown in Table S1. The concentrations of NO₃-N, NO₂-N, NH₄-N, TDN, and TN in the Guanshan River ranged between 0.20–1.95 mg/L, 0.01–0.53 mg/L, 0.01–2.58 mg/L, 0.64–5.37 mg/L, and 0.98–6.06 mg/L, respectively. ON occupied 8.11–89.62% of the total dissolved nitrogen. As shown in Figure 2, the lowest concentrations of N in the Guanshan River were found in the upper tributaries. Liuliping Town, which is close to the Han Reservoir entrance, had the highest N concentrations (GS02: NO₃-N: 1.6 mg/L; NH₄-N: 2.58 mg/L; TDN: 5.37 mg/L; TN: 6.06 mg/L).

3.1.2. Phosphorus in Surface Water

The PO₄-P, TDP, and TP concentrations ranged between 0.03–0.32 mg/L, 0.04–0.64 mg/L, and 0.05–1.15 mg/L, respectively. As with the N, Liuliping Town had the highest P concentrations (GS02: PO₄-P: 0.32 mg/L; TDP: 0.64 mg/L; TP: 1.15 mg/L). OP occupied 6.21–79.57% of the total dissolved phosphorus. The PO₄-P concentrations were generally lower than those of TDP and TP, indicating that organic P and particulate P accounted for a large proportion of the water P.

The nutrients in the Guanshan Reservoir and Han Reservoir were notably lower than those in the Guanshan River. The detailed nutrient contents of the reservoirs are listed in the Supplementary Material. Compared to similar reservoirs and lakes in the Yangtze River Basin, the TN and TP contents in both reservoirs were at high and median levels, respectively [1,45,46].

3.2. Distribution of Nitrogen and Phosphorus in Sediment

3.2.1. Nitrogen in Sediment

The N concentration in the sediment was highest around densely populated areas and agricultural zones, such as Liuliping Town and Lvjiahe Village. The concentration of NO₃-N_sediment (0.25–42.95 mg/kg) and NH₄-N_sediment (0.33–70.73 mg/kg) was considerably higher than that of NO₂-N_sediment (0.01–3.02 mg/kg) (Figure S1). Compared with the river sediment, the Han Reservoir and Guanshan Reservoir sediments contained more N (especially NH₄-N).

3.2.2. Phosphorus in Sediment

The concentration of Fe/Al-P_sediment, Ca-P_sediment, IP_sediment, OP_sediment, and TP_sediment ranged between 1.90–314.94 mg/kg, 5.56–32.33 mg/kg, 49.37–380.66 mg/kg, 184.94–1562.31 mg/kg, and 311.98–1772.32 mg/kg, respectively. The spatial distribution of OP_sediment and TP_sediment showed remarkable consistency, and the OP/TP ratio ranged from 0.59 to 0.97, indicating that OP dominated the sediment P. Most of the sites had higher Fe/Al-P concentrations than Ca-P, particularly the reservoirs and densely populated areas in the middle and lower reaches.

As a small part of the Yangtze River Basin, the concentration of TN_sediment in the Guanshan River Basin was relatively low compared with the mean value of the sediment nutrient in the Yangtze River Basin (TN: 1503.14 mg/kg; TP: 890.13 mg/kg) [47]. In contrast, the TP_sediment concentration in 65.90% of the sampling sites was higher than the mean value. Thus, in the Guanshan River Basin, P enrichment in the sediment poses a greater issue than N.

3.3. Distribution of Nitrogen and Phosphorus in Soils

3.3.1. Nitrogen in Soils

As shown in Figure S2, the distribution of N and P varied in the undisturbed and agricultural soils along the river. The mean values of NO₃-N, NO₂-N, NH₄-N, and TN were 4.73 mg/kg, 0.25 mg/kg, 5.22 mg/kg, and 354.59 mg/kg in the undisturbed soil and 5.69 mg/kg, 0.36 mg/kg, 4.33 mg/kg, and 372.02 mg/kg in the agricultural soil, respectively. The majority of the sample sites had greater levels of TN_agricultural (276.89–434.46 mg/kg) than TN_undisturbed (190.05–590.89 mg/kg), most likely as a result of fertilization. However, the opposite phenomenon occurred in several upstream sites, which might be because N is easily dissolved and transported with runoff (such as irrigation water and rainfall) [48], and farming practices might accelerate the loss of N from agricultural soil [49].

3.3.2. Phosphorus in Soils

The mean values of Fe/Al-P, Ca-P, IP, OP, and TP in the undisturbed soils were 46.37 mg/kg, 8.98 mg/kg, 75.12 mg/kg, 929.16 mg/kg, and 1001.97 mg/kg, respectively, whereas those in the agricultural soils were 139.35 mg/kg, 15.13 mg/kg, 167.10 mg/kg, 1020.46 mg/kg, and 1179.60 mg/kg, separately. The mean values of all the P forms in the agricultural soils were higher than those in the undisturbed soils from upstream to downstream, implying that agricultural activity substantially influences P storage in soil. Fe/Al-P in the agricultural soil was higher than that in the undisturbed soil, especially in the middle stream and downstream. This might relate to the adsorption of PO₄-P from fertilizers on iron and aluminum minerals in soil. OP was the main form of P in all the soil samples. Owing to the abundance of organic matter that accumulated in the forests in the summer, the content of OP in the undisturbed soil was often higher than that of the agricultural soil [50,51].

3.4. Contribution of Nutrients from the Guanshan River Basin to the Danjiangkou Reservoir

The water quality in the Guanshan River has been continuously deteriorating since 2014 (2014: TN: 2.5 mg/L, TP: 0.2 mg/L; 2016: TN: 4.67 mg/L, TP: lack; 2021: TN: 2.54 mg/L, TP: 0.34 mg/L) [37,39]. In this study, 9.52% of the water samples had a TP level in class V or worse according to the Environmental Quality Standards for Surface Water in China [52]. And 87.27% of the water samples showed a TN level equal to or over the limited value of Class V. In addition, according to the Guidelines for Sediment Quality and Conservation Management issued by the Ontario Department of Environment and Energy (Table S5), 80% of the river sediment’s TP contents fall within the range of low to severe effect levels. Thus, the Guanshan River Basin poses a potential pollution risk to the DJKR.

In 2014, the Guanshan River contributed 270.41 t/a TN and 4.75 t/a TP to the DJKR [37]. The TN load increased to 503.79 t/a in 2016 [39]. The calculated results in this study showed that the loads of TN and TP increased to 622.3 t/a and 83.3 t/a in 2021, respectively (the calculation method is presented in the Supplementary Material, and the water flux of the Guanshan River was cited in the previous study [39]). Since the input of N and P from the Guanshan River to the DJKR has increased significantly, it is urgent to identify and control nutrient sources and migration pathways in the Guanshan River Basin.

4. Discussion

4.1. Nitrogen and Phosphorus Sources and Their Scale Effects

4.1.1. Sources of Nitrogen and Phosphorus

Earlier research explored the major sources of N and P in the DJKR [7,9,38]. However, there have been few efforts to subdivide the sources of different nutrient forms. Consequently, CA was used to identify the sources of diverse nutrient forms in the Guanshan River based on the NO₃-N, NO₂-N, NH₄-N, TDN, TN, PO₄-P, TDP, TP, Na⁺, K⁺, Mg²⁺, Ca²⁺, DSi, Cl¯, HCO₃¯, and SO₄²¯ data in the water. The dendrogram derived from CA revealed two categories, encompassing four different classes (Figure 3).

The first class comprised Ca²⁺, HCO₃⁻, SO₄²⁻, K⁺, and Mg²⁺ (Figure 3). In the study area, the weathering of various minerals like feldspar, calcite, and dolomite contributes significantly to the presence of these ions. Thus, the first class primarily represented the leaching of rock and soil (natural source). The second class included PO₄-P, TDP, TP, and NH₄-N (Figure 3). Studies have proved that the prevalent use of N and P fertilizers in farming would result in a significant input of NH₄-N and P into surface water [53,54,55,56,57]. Field investigation discovered that NH₄-N and PO₄-P were the primary chemical constituents of fertilizer in the study area. Therefore, the second class primarily represented the input of fertilizer. Subsequently, the initial two classes were amalgamated into a single category, indicating that the leaching of soil and rocks could also contribute slightly to the phosphorus accumulation in the river water. NH₄-N was excluded because of its further placement in the dendrogram.

The third class comprised Na⁺ and Cl⁻ (Figure 3). The presence of Na⁺ and Cl⁻ in the surface water could originate from evaporite dissolving [58], along with sewage and manure in urban areas [59,60,61]. Given the lack of evaporite and livestock farming in the basin, Na⁺ and Cl⁻ in the river water may be mainly from domestic wastewater. The fourth class comprised TDN, TN, and NO₃-N, which then merged with the third class. NO₃-N and ON are the majority of N in surface water [41]. The merging of these two classes showed that domestic wastewater might contribute NO₃-N and ON to river water.

The dendrogram suggested a unique source for DSi and NO₂-N in the river water. The primary source of NO₂-N is attributable to redox reactions during nitrogen transformation [23]. DSi is related to silicate weathering but is also strongly influenced by phytoplankton uptake [62]. The unremarkable relationships between DSi, NO₂-N, and other nutrients implies that the effect of the biological cycle on N and P in river water is limited.

Following an initial investigation into the sources of water nutrients, RDA was employed to delve deeper into how nutrients from the soil contribute to the river (Figure 4). The factor contribution results from the RDA are listed in Tables S2 and S3.

The nutrients in the agricultural soil explained 53.8% of those in the sediment (Figure 4a, Table S2), in which the contribution of Fe/Al-P in the agricultural soil was notable (20%). In the agricultural soil, the direction of the Fe/Al-P arrow aligned with that of Fe/Al-P and NH₄-N in the sediment, showing a strong correlation and suggesting a significant role of agricultural soil in enriching Fe/Al-P and NH₄-N in sediment. The IP (mainly Ca-P) and OP in the undisturbed soil showed certain contributions (10% and 12.8%, respectively) to the sediment nutrients (Figure 4a, Table S2), suggesting an additional natural source of sediment P. Furthermore, there was a significant correlation between OP in the undisturbed soil and TN in the sediment. The evidence shows that ON constitutes over 90% of the nitrogen in the soil or sediment [41] and that both ON and OP have a strong correlation with organic matter [63]. Thus, organic matter in undisturbed soil could also contribute to ON in sediment.

Figure 4b illustrates that NH₄-N in the sediment contributed predominantly (75.5%, Table S3) to the water nutrients. In addition, NO₃-N and TN in the agricultural soil contributed slightly to the nutrients in the water (7.5% and 2.6%, respectively; Table S3). The RDA results indicated that water NO₃-N was affected not only by domestic wastewater but also by NH₄-N transformation from the sediment. Fe/Al-P in the sediment contributed modestly (2.7%) to P in the water, indicating a relatively small release of P from the sediment.

In summary, there were three primary sources of nutrients in the Guanshan River Basin, including fertilizer (which contributed abundant PO₄-P and NH₄-N to the water and Fe/Al-P and NH₄-N to the sediment), natural soil, and rock leaching (which contributed a little PO₄-P and OP to the water and abundant Ca-P, OP, and ON to the sediment), and domestic wastewater (which contributed abundant NO₃-N and ON to the water).

4.1.2. Scale Effects of Nutrient Sources

Buffer analysis is feasible for the identification of the scale effects of river pollution sources [29,32,33]. In this study, the agricultural area (representing the fertilizer source) and residential area (representing the domestic wastewater source) were considered in the buffer analysis (Figure 5). However, the residential areas were small and scattered along the river in the basin (excluding Liuliping Town) (Figure 1), resulting in the low resolution of residential areas in the LULC map and insignificant correlations between the residential areas and river water nutrients. Therefore, the results presented below focused on the relationship between the agricultural areas and the river nutrients in the buffer zones.

As shown in Figure 5, the scale effects of the agricultural areas on the nutrients were insignificant both within the 50 m buffer zone and beyond the 1400 m buffer zone. PO₄-P and NH₄-N showed significant relationships with the agricultural areas in the 300 m buffer zone (R = 0.40, 0.41, respectively, p ≤ 0.05), and the correlations of P (PO₄-P, TDP, and TP) and NH₄-N with the agricultural areas gradually increased from the 500 m to 1200 m buffer zones (R varied between 0.42 and 0.73, p ≤ 0.05). The notable relationship between the agricultural areas with P and NH₄-N in the 300–1200 m buffer zone indicated that the fertilization in the agricultural areas contributed abundant P and NH₄-N to the water, corroborating the findings discussed previously (Section 4.1.1). The insignificant relationship between the river water quality and agriculture areas in the >1400 m buffer zone could be attributed to the retention effect of the riparian zone [64,65,66].

Fascinatingly, the analysis results within the 150 m buffer presented typical features of domestic wastewater contamination (strongly correlated with NO₃-N, TDN, and TN in the river, R = 0.55, 0.50, 0.50, respectively, p ≤ 0.05). This could be attributed to the confusion between the residential and agricultural areas near the river, as mentioned above. Despite the coarse resolution of residential areas in the LULC data, this phenomenon still allowed us to identify the possible control scopes of domestic wastewater discharge. As a result, fertilizer application within the 300–1200 m area and domestic wastewater discharge within the 150 m area surrounding the Guanshan River should be regulated.

4.2. Migration and Transformation of Nitrogen and Phosphorus in the Guanshan River Basin

Pearson’s correlation and XGBoost were employed to investigate N and P migration and transformation among the environmental media (Figure 6, Figure 7, Figures S3 and S4). A conceptual model of the nutrient migration and transformation among the environmental media is summarized in Figure 8.

4.2.1. Transformation of N and P within Soils

Pearson’s correlation analysis revealed that the correlations between the N forms were insignificant in the undisturbed soil, but remarkable for the P forms (Figure 6). There was a significant negative correlation between IP_undisturbed, TP_undisturbed, and OP_undisturbed (R = −0.72~−0.65, p ≤ 0.01); meanwhile, the OP content in the soil of the study area was much higher than the IP content (Figure S2), indicating that IP might originate from the degradation of natural OP in the undisturbed soil [67]. The transformation relationship between Ca-P_undisturbed and OP_undisturbed (R = −0.62, p ≤ 0.01) could also prove this inference. Fe/Al-P_undisturbed, Ca-P_undisturbed, and IP_undisturbed exhibited significant positive interrelations (R = 0.82~0.87, p ≤ 0.01), possibly because PO₄-P tends to be adsorbed on Fe, Al, and Ca minerals, leading to the co-enrichment of IP forms [68].

In the agricultural soil, positive correlations among the IP forms were observed that were similar to those of the undisturbed soil, but the transformation relationship between IP and OP was not significant (Figure S3), because IP in the agricultural area mainly comes from fertilizer input rather than OP degradation. Moreover, the correlation between the N forms was stronger (R between NO₃-N_agricultural and NO₂-N_agricultural was 0.78, p ≤ 0.01), indicating that agricultural activities (plowing and fertilization) disturbed N transformation. The synchronous accumulation of NO₃-N, NO₂-N, and Ca-P in the agricultural soil was identified (R = 0.73–0.83, p ≤ 0.01), which was consistent with earlier studies [16,69], because NH₄-N and PO₄-P were the primary chemical constituents of the fertilizer in the study area. Ammonium phosphate (NH₄H₂PO₄) could lead to the creation of CaHPO₄ precipitation (Equation (2)) [69], thus causing the synchronous accumulation.

$${\left({\mathrm{N}\mathrm{H}}_4\right)}_2{\mathrm{H}\mathrm{P}\mathrm{O}}_4+{\mathrm{C}\mathrm{a}}^{2+}=2{\mathrm{N}\mathrm{H}}_4^{+}+{\mathrm{C}\mathrm{a}\mathrm{H}\mathrm{P}\mathrm{O}}_4$$

(2)

Meanwhile, exogenous NH₄-N may be rapidly nitrified to NO₃-N and NO₂-N under aerobic conditions after fertilization [23,70]; the H⁺ released from nitrification could cause the acidic dissolution of detrital apatite and the re-adsorption of PO₄-P onto CaCO₃ (Equations (3)–(5)), causing a synchronous enrichment of Ca-P, NO₃-N, and NO₂-N [24,25,68]. Ca-P is a more labile inorganic P form than detrital apatite; thus, fertilizer should be used sparingly since it promotes the formation of surplus transferable P.

$$2{\mathrm{N}\mathrm{H}}_4+3{\mathrm{O}}_2\to 2{\mathrm{N}\mathrm{O}}_2^{-}+4{\mathrm{H}}^{+}+2{\mathrm{H}}_2\mathrm{O}$$

(3)

$$2{\mathrm{N}\mathrm{O}}_2^{-}+{\mathrm{O}}_2\to 2{\mathrm{N}\mathrm{O}}_3^{-}$$

(4)

$${\mathrm{C}\mathrm{a}}_5{\left({\mathrm{P}\mathrm{O}}_4\right)}_3\mathrm{O}\mathrm{H}+10{\mathrm{H}}^{+}={\mathrm{H}}_2\mathrm{O}+3{\mathrm{H}}_3{\mathrm{P}\mathrm{O}}_4+5{\mathrm{C}\mathrm{a}}^{2+}$$

(5)

4.2.2. Migration of N and P from Soils to River

The undisturbed soil contributed obviously to the N and P of the river sediment, but only slightly to the river water (Figure 6 and Figure 7). Positive correlations among IP_undisturbed, OP_undisturbed, TP_undisturbed, IP_sediment, and OP_sediment (R = 0.59–0.66, p ≤ 0.01) were revealed by Pearson’s correlation analysis. XGBoost showed similar results, demonstrating that IP_undisturbed and OP_undisturbed had higher feature importance for Ca-P_sediment and OP_sediment, respectively (Figure 7b). The undisturbed land in the study area is predominantly forested and rich in organic matter, potentially leading to high OP leaching during the rainy season [50,51,71].

In the agricultural soil, IP_agricultural and Fe/Al-P_agricultural showed notable positive correlations with most of the nutrients in the river water (except for NO₂-N) and with NH₄-N and Fe/Al-P in the sediment (R = 0.73–0.92, p ≤ 0.01, Figure S3). XGBoost analysis also revealed a notable influence of Fe/Al-P_agricultural on nutrients in the river water, as well as IP_agricultural and OP_agricultural on TP and OP in the sediment (Figure 7b). The SHAP values indicated that Fe/Al-P_agricultural was positively correlated with the nutrients in the river water (Figure 7c). Fertilization simultaneously introduced PO₄-P and NH₄-N in the study area; however, Fe/Al-P was more like a tracer for nutrient migration than NH₄-N. Based on the N and P contents, the agricultural soil in this study area was in N-limited states (TN/TP < 9) [72], leading to rapid biological uptake and the nitrification of NH₄-N [73], while the excess PO₄-P would migrate or be adsorbed onto Fe/Al minerals and have a slower biological turnover rate than NH₄-N [68], increasing the suitability of Fe/Al-P to indicate nutrient migration. Interestingly, though IP and NH₄-N were both from fertilizer, the significant negative relationship between NH₄-N_agricultural and IP_sediment was revealed by Pearson’s correlation analysis (R = −0.83, p ≤ 0.01) and the SHAP value (Figure S5). NH₄-N addition could stimulate biological activity and promote P absorption under N-limited conditions [27,74], thereby inhibiting P emigration from agricultural soil, causing a negative correlation. Therefore, when fertilization is necessary, properly increasing the NH₄-N/PO₄-P ratio of the fertilizer could avoid nutrient surplus and loss in the study area.

4.2.3. Migration of N and P between River Water and Sediment

The nutrients in the river water showed positive correlations with Fe/Al-P_sediment and NH₄-N_sediment (with R ranging between 0.80 and 0.97, p ≤ 0.01, Figure S5); meanwhile, XGBoost analysis revealed that the nutrients in the river water and sediment positively correlated with each other (Figure 7c,d), suggesting the co-enrichment of nutrients in the river water and sediment. Furthermore, XGBoost revealed that sediment N (especially NH₄-N) had a higher feature importance for the water nutrients than sediment P, while river water P (especially PO₄-P) had a greater feature importance for the sediment nutrients than water N. This demonstrated that sediment could act as a N source and P sink for the river water. During eutrophication control, the endogenous release of sediment N into surface water should be noted.

Although the research findings do point to the existence of nutrient movement processes among various environmental media, comprehensive recommendations for managing eutrophication would be improved by more dense soil sample analysis and longer-term monitoring that takes into account different hydrographic times of the year instead of just the wet period.

5. Conclusions

This study improved the resolution of the scale effects of river nutrient sources and accurately identified various sources for different nutrient forms in the river. The key processes of N and P migration and transformation across environmental media were explored.

Nutrient accumulation in the Guanshan River water progressed from the upper to lower reaches, with notable enrichment at the DJKR entrance (TN: 6.06 mg/L, TP: 1.15 mg/L). The Guanshan River contributed approximately 622.3 t/a TN and 83.3 t/a TP to the DJKR in 2021, posing a risk of eutrophication to the reservoir. The primary sources of river nutrients in the study basin were fertilizer, natural soil and rock leaching, and domestic wastewater. Optimized buffer analysis revealed that fertilization within 300–1200 m and domestic wastewater discharge within 150 m of the river affects nutrients in river water significantly. Sediment was the N source and P sink of the overlying river water. The anthropogenic activities have disturbed the N and P transformation in the agricultural soil. The application of NH₄-N/PO₄-P compound fertilizer could indirectly promote NO₃-N and Ca-P enrichment. Appropriately increasing NH₄-N applicability could inhibit the emigration of P from agricultural soil in the study area.

The findings are meaningful for nutrient synergistic control at the basin scale. Future research could conduct indoor experiments to verify N and P interactions and support the field findings.