Spatiotemporal Analysis of Water Quality in the Upper Watershed of Guanting Reservoir Based on Multivariate Statistical Analysis

Abstract

Exploring the spatiotemporal pattern of water quality and identifying pollution sources is crucial for achieving precise management of reservoir watersheds. This study is based on monthly water quality data from 9 monitoring stations in the upstream watershed of Guanting Reservoir in 2024, combined with an improved water quality index method (WQI) and multivariate statistical analysis (clustering, discrimination, principal component and factor analysis), to reveal the spatiotemporal variation characteristics of water quality and pollution sources. The results show (1) significant spatiotemporal heterogeneity. In terms of time, the water quality is worst during the summer rainy season (June August), indicating that the pollution load input from surface runoff exceeds the dilution effect of rainfall. In terms of space, the water quality deteriorates significantly downstream along the river network, with the most prominent pollution occurring in the entrance area. (2) The results also show clear identification of key indicators and dominant pollution sources. Discriminant analysis shows that BOD₅ and DO are key indicators for distinguishing rainy and dry seasons, while TN, TP, COD, COD_Mn, and F can effectively distinguish spatial clusters. Factor analysis further revealed that organic pollution (originating from domestic and industrial wastewater) and nutrient pollution (mainly from agricultural non-point sources) are the dominant factors. This study confirms that pollution input during rainfall is the primary driving factor for water quality deterioration, and human activities have led to the cumulative effect of pollutants along the river network. Based on this suggestion, differentiated and precise governance strategies should be implemented according to the spatiotemporal differentiation characteristics to improve the water environment quality of the upstream watershed of Guanting Reservoir.

1. Introduction

Urban reservoirs play an important role in rainwater collection, groundwater recharge, flood control and drainage, agricultural irrigation, local climate regulation, and the maintenance of ecological balance. Urban reservoirs have become a critical link in ensuring public well-being, supporting sustainable urban development, and promoting harmonious coexistence between humans and nature [1]. The upstream watershed of the reservoir serves as a critical water source for the reservoir area, playing an essential role in supplying water for production, domestic use, and ecological needs in the surrounding regions [2]. Its water quality is directly related to the water quality and ecological stability of the downstream reservoir. Once the upstream water environment is polluted, it can not only compromise the water supply security of the downstream reservoir but also cause long-term impacts on the regional ecosystem. Therefore, ensuring the cleanliness and safety of water resources in the upstream watershed is a key measure for safeguarding the ecological security and sustainable use of urban reservoirs. However, with the continuous population growth, accelerated industrialization and urbanization, and the ongoing expansion of agricultural activities, the water quality in the upstream watershed of the reservoir is facing unprecedented challenges [3]. Industrial point-source wastewater, domestic sewage generated by urban residents, and agricultural non-point-source pollution resulting from the use of fertilizers and pesticides during crop cultivation continuously enter the water bodies. These pollutants continuously introduce nitrogen, phosphorus, organic matter, and heavy metals into the upstream watershed, significantly increasing the concentration of contaminants in the water and posing a serious threat to water quality and ecological stability [4]. Therefore, studying the water quality health status and the spatiotemporal distribution characteristics of water quality in the upstream basin of the reservoir is of great theoretical value and practical significance for achieving refined management of the reservoir, identifying the main pollution sources in the reservoir area, and formulating scientifically effective control strategies.

Long-term water quality monitoring has accumulated a large amount of complex data with multiple sampling points, various indicators, and high-frequency characteristics, covering a wide range of water quality parameters including physical, chemical, and biological factors [5]. How to effectively analyze and interpret the potential water quality characteristics still poses certain challenges. Commonly used water quality assessment methods mainly include the single-factor evaluation method, gray system evaluation method, fuzzy mathematics evaluation method, artificial neural network evaluation method, comprehensive water quality index (WQI) method, and multivariate statistical analysis [6,7,8]. Among them, the Comprehensive Water Quality Index (WQI) method transforms multidimensional water quality parameters into a dimensionless single evaluation value, enabling a systematic representation of the overall condition of the aquatic environment. Compared with traditional single-factor evaluation methods, it offers greater comprehensiveness and intuitiveness. A study using the WQI method to assess river water quality in the Taihu Basin of China found that WQI can effectively evaluate water quality and its spatial variations [9]. Another study applying the WQI method to evaluate water quality in the Middle Route of the South-to-North Water Diversion Project in China found that WQI can accurately reflect the seasonal and spatial variations in water quality [10]. Multivariate statistical analysis methods, which are used to simplify data structures and extract latent information, have become important tools for analyzing water quality variation patterns and identifying potential pollution sources by uncovering the spatiotemporal relationships among water quality parameters. For example, a study used correlation analysis, principal component analysis, and cluster analysis to assess water quality in the Mahanadi River Basin in Odisha, India, revealing the main causes of water quality changes and providing data support for water quality protection and management in the basin [11]. Another study, using the Keban Reservoir in Turkey as a case study, applied multivariate statistical methods such as discriminant analysis, principal component analysis, factor analysis, and cluster analysis to assess the seasonal and spatial variations in surface water quality of the reservoir. The study also identified the total phosphorus content in sediments, water types, and the trophic status of the reservoir, providing scientific evidence for water quality management of large reservoirs [12]. However, the condition of the aquatic environment is influenced by both natural processes and human activities, making its assessment complex and variable. The application of multivariate statistical methods to multivariable water quality assessment has certain limitations when used in isolation. Therefore, it is often necessary to combine multiple multivariate statistical methods to minimize their individual limitations and comprehensively assess the spatial and temporal variations in water quality.

Guanting Reservoir is the first large-scale reservoir built after the founding of the People’s Republic of China and is an important and representative reservoir in North China. It controls a drainage area of 43,402 km², accounting for 92.8% of the total Yongding River basin area (46,768 km²). As a multipurpose project, it serves functions such as flood control, water supply, power generation, and irrigation, and was once one of the main water sources for the capital, Beijing [13]. Guanting Reservoir is part of the Yongding River system within the Hai River Basin. The basin spans from 112°8.3′ E to 116°20.6′ E and from 41°14.2′ N to 38°51′ N. It is located in a mid-latitude region characterized by a temperate semi-arid climate and a continental monsoon climate pattern. In the past decade or so, Guanting Reservoir has strengthened ecological environment management and protection across the entire basin, resulting in significant temporal and spatial changes in water quality within the reservoir area. However, most existing studies have focused on analyzing the temporal and spatial trends of water quality in the reservoir area, with relatively little attention given to the factors influencing water quality changes. In particular, there is a lack of systematic evaluation of water quality in the upstream watershed of the Guanting Reservoir, as well as in-depth analysis of the factors and pollution sources affecting upstream water quality. Therefore, it is urgent to analyze the spatial and temporal variations in water quality and the status of pollution sources in the upstream watershed, in order to provide a scientific basis for improving the water environmental quality and ecological service functions of the reservoir area. To gain a deeper understanding of the water quality health status, spatial and temporal distribution characteristics, and pollution sources in the upstream watershed of the Guanting Reservoir, this study analyzes monthly water quality monitoring data from nine monitoring stations located in the Hebei section of the upstream watershed in 2024. Eight key water quality indicators were selected, and an improved Comprehensive Water Quality Index (WQI) method was applied. This was combined with multivariate statistical analysis methods, including Cluster Analysis (CA), Discriminant Analysis (DA), Principal Component Analysis (PCA), and Factor Analysis (FA), to investigate the spatial and temporal characteristics of water quality in the upstream watershed of the Guanting Reservoir. This study aims to identify the dominant factors and potential pollution sources influencing the spatial and temporal variations in water quality, providing a scientific basis and technical support for the spatial-temporal management of water quality and environmental protection in the upstream watershed of the Guanting Reservoir and other similar watersheds.

2. Materials and Methods

2.1. Study Area and Data

The upstream basin of the Guanting Reservoir includes two major tributaries, the Yang River and the Sanggan River, which are part of the Haihe River system [14]. The Yang River and the Sanggan River converge at Zhuguantun in Huailai County, Zhangjiakou City, and are thereafter known as the Yongding River, which flows into the Guanting Reservoir and continues downstream into Beijing [15]. The Yang River and the Sanggan River originate from Xinghe County in the Inner Mongolia Autonomous Region and Ningwu County in Shanxi Province, respectively [16]. The upstream region of the Guanting Reservoir has a typical temperate monsoon climate, with annual maximum temperatures reaching up to 40.9 °C, minimum temperatures dropping to −26.2 °C, and an average annual temperature of 7.8 °C. This paper focuses on the study of water quality in the upstream tributaries of the Guanting Reservoir located in Zhangjiakou City, Hebei Province, specifically the Dongyang River, Yang River, Sanggan River, and Huliu River. According to published studies within the same basin 16, about 80% of the annual precipitation in the Guanting Reservoir basin is concentrated in the summer months of June to August and early autumn months of September, which are clearly defined as the rainy season. Therefore, this study follows the climate characteristics of the region and defines June, July, and August (i.e., the core period of the rainy season) as the rainy season (WP) for subsequent discussions on seasonal differences and time clustering analysis. The remaining months are defined as the dry season (DP). A geographical overview of the area is shown in Figure 1.

This study selected monthly water quality data from the year 2024 for nine key monitoring stations in the upstream basin of the Guanting Reservoir: Dongyang River (DYH), Zuo Wei (ZW), Xiangshuibao (XSB), Jimingyi (JMY), Bahao Bridge (BHQ), Chuaigutong (CGT), Shixiali (SXL), Wenquantun (WQT), and Hulu River (HLH). Among them, DYH is located in Hebei Province near Zhangjiakou City and is used to monitor upstream water quality. ZW, XSB, and JMY are situated along the Yang River, distributed sequentially from upstream to downstream, to monitor changes in the Yang River’s water quality. BHQ is located near the Guanting Reservoir and serves as a key station for monitoring reservoir water quality. SXL and CGT are positioned along the Sanggan River, also distributed from upstream to downstream, for monitoring water quality variations in the Sanggan River. WQT is located in the upper reaches of the Yongding River and is used to assess the water quality in that area. HLH, located on a tributary of the Sanggan River, is also an important station for monitoring the water quality of the reservoir. The eight water quality parameters selected for this study are: Total Nitrogen (TN), Total Phosphorus (TP), Ammonia Nitrogen (NH₃-N), 5-Day Biochemical Oxygen Demand (BOD₅), Dissolved Oxygen (DO), Chemical Oxygen Demand (COD), Permanganate Index (COD_Mn), and Fluoride (F). The data used in this study were provided by the Guanting Reservoir Administration of Beijing.

During the water quality monitoring process, the layout of monitoring sites, sample collection, preservation and transportation, laboratory analysis, data compilation and processing, as well as quality assurance and control of the monitoring activities, were all conducted in strict accordance with the relevant requirements of the Technical Specifications for Monitoring of Surface Water and Wastewater (HJ/T 91-2002) [17], issued by the Ministry of Ecology and Environment of the People’s Republic of China. In addition, to improve the accuracy of the research results, the collected data were preprocessed prior to water quality assessment, including the handling of missing values and outliers. For certain water quality parameters that were missing at some stations but did not affect the overall data analysis, the missing values were omitted. A small number of outliers were handled in accordance with the Statistical Treatment and Interpretation of Data—Determination and Treatment of Outliers in Normal Samples (GB/T 4883-2008) [18], issued by the Standardization Administration of China. The water quality data were analyzed in accordance with the Technical Specifications Requirements for Monitoring of Surface Water and Wastewater (Standard No. HJ/T 91-2002), issued by the Ministry of Ecology and Environment of the People’s Republic of China.

To understand the overall water quality status of the upstream region of the Guanting Reservoir within Hebei Province, this study employed an improved Water Quality Index (WQI) method to analyze the water quality data from various monitoring stations. The CA method was used to classify the water quality data in terms of spatial and temporal dimensions, in order to explore the characteristics of its spatiotemporal classification. Based on the results of CA, both the standard model and stepwise model methods in DA were employed to assess the accuracy of the spatiotemporal classification of the water quality data. The stepwise model was further used to identify effective water quality indicator variables that can distinguish between different spatiotemporal groups. PCA and FA methods were used to investigate the dominant factors and potential pollution sources contributing to the spatiotemporal variations in water quality in the upstream watershed of the Guanting Reservoir.

2.2. Improved Water Quality Index Method

The Water Quality Index (WQI) is a comprehensive method for assessing water quality. It reflects the overall condition of a water body by converting multiple water quality parameters—such as total nitrogen, total phosphorus, ammonia nitrogen, dissolved oxygen, biochemical oxygen demand, and potassium permanganate, etc.—into a single index value [19]. This method simplifies the analysis and interpretation of water quality data, fully utilizes the information contained in water quality parameters, and provides a comprehensive reflection of the water quality condition. As a result, it allows non-experts to easily understand the water quality status and holds significant value in water resource management. Since the 1960s, the Water Quality Index (WQI) method has been widely applied in the assessment of both surface water and groundwater quality [20].

When applying the WQI method to evaluate water quality data, the Analytic Hierarchy Process (AHP) is used to determine the weights. The Analytic Hierarchy Process (AHP) is a hierarchical weighting decision analysis method proposed by Professor Thomas L. Saaty, an operations researcher at the University of Pittsburgh. It commonly uses the traditional nine-point scale, which employs nine values (and their reciprocals) ranging from 1 to 9 to indicate the relative importance between evaluation elements. This process results in the formation of a judgment matrix and carries a degree of subjectivity [21]. The $WQI$ calculation formula for each water quality data point is as follows:

$$WQI=\sum _{i=1}^n{W}_i\times C_i$$

(1)

In the formula, n is the total number of water quality parameters; $W_i$ is the weight of the i-th water quality parameter calculated using AHP, with the sum of all $W_i$ equal to 1; $C_i$ is the normalized value of the i-th water quality parameter.

In response to the water quality management needs of rivers in the upstream watershed of the Guanting Reservoir in Beijing—taking into account both the protection of drinking water sources and the characteristics of agricultural non-point-source pollution—a normalization process was applied to each water quality parameter. This process was based on the national Environmental Quality Standards for Surface Water (GB3838-2002) and the water pollution prevention requirements of the Beijing-Tianjin-Hebei region, to ensure scientific and rational data processing. This national standard is the basic benchmark for water quality classification in China and serves as the basis for interpreting measurement data in this study. The weights and normalization details of each water quality parameter are presented in

Table 1. Weights and normalization of each water quality parameter.

Parameter	${\mathit{W}}_{\mathit{i}}$	${\mathit{C}}_{\mathit{i}}$
		100	90	80	70	60	50	40	30	20	10	0
TN (mg/L)	0.293	0–0.1	0.1–0.2	0.2–0.35	0.35–0.5	0.5–0.75	0.75–1.0	1.0–1.25	1.25–1.5	1.5–1.75	1.75–2.0	>2.0
TP (mg/L)	0.214	0–0.01	0.01–0.02	0.02–0.06	0.06–0.1	0.1–0.15	0.15–0.2	0.2–0.25	0.25–0.3	0.3–0.35	0.35–0.4	>0.4
NH3-N (mg/L)	0.157	0–0.075	0.075–0.15	0.15–0.325	0.325–0.5	0.5–0.75	0.75–1.0	1.0–1.25	1.25–1.5	1.5–1.75	1.75–2.0	>2.0
BOD5 (mg/L)	0.061	0–0.75	0.75–1.5	1.5–2.25	2.25–3	3–3.5	3.5–4	4.0–5.0	5.0–6.0	6.0–8.0	8.0–10.0	>10.0
DO (mg/L)	0.114	>9	7.5–9	6.75–7.5	6.0–6.75	5.5–6.0	5.0–5.5	4.0–5.0	3.0–4.0	2.5–3.0	2.0–2.5	0–2.0
COD (mg/L)	0.045	0–3.75	3.75–7.5	7.5–11.25	11.25–15	15–17.5	17.5–20	20–25	25–30	30–35	35–40	>40
CODMn (mg/L)	0.033	0–1.0	1.0–2.0	2.0–3.0	3.0–4.0	4.0–5.0	5.0–6.0	6.0–8.0	8.0–10.0	10.0–12.5	12.5–15	>15
F (mg/L)	0.083	0–0.167	0.167–0.333	0.333–0.5	0.5–0.667	0.667–0.833	0.833–1.0	1.0–1.125	1.125–1.25	1.25–1.375	1.375–1.5	>1.5

.

2.3. Cluster Analysis

Cluster Analysis (CA) is an unsupervised learning method, meaning it does not require pre-labeled data for model training. It is used to group objects in a dataset such that objects within the same group (referred to as a “cluster”) have high similarity, while objects in different groups have low similarity. CA is a data classification method based on object similarity [22]. In this study, Ward’s method combined with squared Euclidean distance was used for cluster analysis. A dendrogram was employed to classify and group water quality data with similar characteristics [23]. In this study, temporal clustering was applied to divide the water quality data from the 12 months of the year into different monthly groups. Spatial clustering was applied to group the nine monitoring stations into different categories based on the similarity of the water quality data collected at each site. Spatiotemporal cluster analysis was conducted to explore the spatiotemporal characteristics of the water quality data. Based on the spatiotemporal clustering results, discriminant analysis, principal component analysis, and factor analysis were performed on the water quality data.

2.4. Discriminant Analysis

Discriminant Analysis (DA) is a supervised statistical method based on the results of cluster analysis. It assumes that the study objects have already been classified into several categories in some way, with each category characterized by a set of factors. The method aims to determine how a set of quantitative variables can be used to distinguish between these known categories [24]. In this study, based on the results of spatiotemporal cluster analysis, both standard discriminant analysis and stepwise discriminant analysis were employed to perform the discriminant analysis [25]. The standard discriminant analysis method uses all water quality parameters as independent variables without any selection. The stepwise discriminant analysis method, based on standard discriminant analysis, selects the most statistically significant independent variables with strong discriminative power from all water quality parameters for analysis. Discriminant analysis was used to examine the accuracy of the spatiotemporal classification results of the water quality data and to explore the characteristics of its spatiotemporal variations.

2.5. Principal Component Analysis and Factor Analysis

Principal Component Analysis (PCA) is a commonly used linear dimensionality reduction technique. Its purpose is to project data from a high-dimensional space to a lower-dimensional space through linear transformation while preserving as much of the original data’s variation as possible [26]. Factor Analysis (FA) is a statistical technique that is typically conducted after Principal Component Analysis. It reduces the dimensionality of data by appropriately categorizing a large number of variables. In this process, it extracts factors that can represent the common variation in multiple original variables. The core concept of this method is to reveal and reflect the information contained in the original variables by identifying and utilizing these representative factors [27]. In this study, based on the results of spatiotemporal cluster analysis, Principal Component Analysis (PCA) and Factor Analysis (FA) were performed on the temporal and spatial groups identified through cluster analysis. For each group, components with eigenvalues greater than 1 were retained to explore the dominant factors and potential pollution sources responsible for the spatiotemporal variations in the water quality data.

3. Results

3.1. Statistical Analysis of Water Quality Parameters

The descriptive statistics for measuring water quality parameters are detailed in

Table 2. Statistical Summary of Measured Values for Eight Water Quality Indicators in 2024.

Parameter		Min	Max	Mean	Standard Deviation	Coefficient of Variation
Eutrophication	TN	0.69	13.1	4.81	2.93	0.61
	TP	0.01	1.18	0.13	0.21	1.64
	NH3-N	0.02	1.97	0.30	0.36	1.19
Organic pollution	BOD5	0.25	7.10	2.48	1.44	0.58
	COD	5.10	13.60	9.50	2.01	0.21
	CODMn	2.30	50.40	17.26	9.42	0.55
Self purification of water bodies	DO	0.70	12.50	3.99	1.98	0.50
Inorganic pollution	F	0.52	2.18	0.94	0.41	0.43

. In order to preliminarily evaluate the water quality status, the following discussion will link these measured values with the “Environmental Quality Standards for Surface Water of the People’s Republic of China (GB 3838-2002)”.

Referring to the Environmental Quality Standards for Surface Water of the People’s Republic of China (GB 3838-2002), the selected eight water quality indicators (Table 2) can be categorized into four groups: eutrophication indicators, organic pollution indicators, water self-purification indicators, and inorganic pollution indicators. Among the eutrophication indicators, the average value of TN was 4.81 mg/L, which is far above the Class V water quality standard. The coefficient of variation was relatively small, indicating low data variability. The average value of TP was 0.13 mg/L, which falls within the Class III water quality standard range, but the coefficient of variation was large, indicating high data variability. The average value of NH₃-N was 0.30 mg/L, within the Class II water quality standard range. The coefficient of variation was relatively high, indicating that this indicator’s data is unstable and exhibits large fluctuations. Among the organic pollution indicators, the average value of BOD₅ was 2.48 mg/L, which meets the Class I water quality standard range. The coefficient of variation was relatively small, indicating that the data for this indicator fluctuates little and is relatively stable. The average value of COD was 9.50 mg/L, also within the Class I water quality standard range. The coefficient of variation was small, indicating that the data for this indicator is relatively stable. The average value of COD_Mn was 17.26 mg/L, exceeding the Class V water quality standard range. The coefficient of variation was slightly low, indicating relatively small data fluctuations. The average value of the water self-purification indicator DO was 3.99 mg/L, meeting the Class IV water quality standard. The coefficient of variation was relatively small, indicating low data dispersion and relatively stable data. The average value of the inorganic pollution indicator F was 0.94 mg/L, within the Class I water quality standard range. The coefficient of variation was relatively small, indicating low data variability.

3.2. Spatiotemporal Analysis of WQI

To determine the pollution status of the upstream watershed of the Guanting Reservoir in 2024, the Water Quality Index (WQI) was used to analyze the overall water quality of nine monitoring stations upstream of the Guanting Reservoir. The WQI values for different months at these nine monitoring stations were calculated. The range of the WQI is 0 to 100, and it is a dimensionless number. A higher WQI value indicates better water quality. The WQI values for different months in 2024 at the nine monitoring stations in the upstream watershed of the Guanting Reservoir are shown in Figure 2. According to the WQI ratings, water quality is classified into five levels: excellent (90–100], good (70–90], moderate (50–70], poor (25–50], and very poor (0–25] 28.

The WQI values of the 9 monitoring stations showed a certain trend of change in different months (Figure 2), among which the WQI values of most stations fluctuated greatly in April and August. In August, the WQI values at BHQ, ZW, HLH, and XSB were the lowest of the year, indicating the poorest water quality. At CGT, the WQI in June was the lowest of the year, indicating the poorest water quality. At JMY, the WQI values were relatively low in January and June, reflecting poorer water quality during these months. At WQT, the WQI value was the lowest in April. The WQI at the DYH station shows a month-by-month decreasing trend. The WQI values at different stations fluctuate significantly across months, showing temporal heterogeneity. However, after September, there is an upward trend in WQI values, which gradually tend to stabilize. Overall, most stations recorded lower or the lowest monthly WQI values during the rainy summer period of the year. This indicates that rainfall has a significant impact on water quality in the upstream watershed and can lead to a certain degree of water quality degradation. This may be because the pollutants carried by surface runoff during the rainy season have a greater impact on water deterioration than the dilution effect of rainfall on various water quality parameters.

The boxplots of the monthly WQI values for the nine monitoring stations are shown in Figure 3. This visually illustrates the differences in the annual average WQI among the different monitoring stations. The ranking of the annual average WQI for each monitoring station is as follows: DYH (moderate) > CGT (moderate) > WQT (moderate) > XSB (moderate) > SXL (moderate) > HLH (poor) > BHQ (poor) > JMY (poor) > ZW (poor). DYH is located in the upper reaches of the watershed and is less affected by human activities. Pollutants in the water are washed downstream from the upper reaches, so the water quality at DYH is the best. This discovery is consistent with the research results of similar watersheds in North China. For example, research on Baiyangdian Lake shows that human activities have a greater impact on water quality than natural factors, and different regions of the lake are dominated by different types of human activities [20]. In contrast, ZW stations are likely to be significantly affected by the combined pollution of domestic sewage, industrial wastewater, and agricultural runoff, similar to the severe impact of rivers carrying upstream urban pollutants in the western region of Baiyangdian [20]. BHQ, which is closest to the reservoir inlet, also has poor water quality and can directly affect the water quality of the reservoir area. HLH is located on a tributary of the Sanggan River, and its water quality is influenced by the water quality conditions of the Sanggan River watershed. It can be seen that the water quality status in the upstream watershed of the Guanting Reservoir exhibits spatial heterogeneity.

3.3. Spatiotemporal Grouping of Water Quality Data

To clarify the seasonal characteristics of water quality in the upstream watershed of the Guanting Reservoir, temporal cluster analysis was conducted on water quality data from different months. The dendrogram generated during the cluster analysis process is shown in Figure 4a. Based on the changes in each indicator for each month, the data for the 12 months of the year were divided into two groups at the point where (Dlink/Dmax) × 100 < 10. Cluster 1 includes the three summer months of June, July, and August, which correspond to the rainy period (WP). Cluster 2 includes nine months: March, April, and May in spring; September, October, and November in autumn; and December, January, and February in winter. This period corresponds to the dry period (DP). This seasonal classification of water quality data aligns with the climatic characteristics of Zhangjiakou City in Hebei Province [20].

To clarify the spatial characteristics of water quality in the upstream watershed of the Guanting Reservoir, spatial cluster analysis was conducted on the water quality data from nine monitoring stations. The dendrogram generated during the cluster analysis process is shown in Figure 4b. Based on the changes in each indicator at each monitoring station, the nine monitoring stations were divided into four clusters at the point where (Dlink/Dmax) × 100 < 5. Cluster I includes one station, DYH, located along the upper reaches of the Dongyang River; Cluster II includes one station, ZW, located in the upper reaches of the Yang River; Cluster III includes one station, HLH, located on a tributary of the Sanggan River; Cluster IV includes six stations—XSB, JMY, and BHQ located along the Yang River from upstream to downstream, and CGT, SXL, and WQT located on the Sanggan River. The result of dividing the space into four clusters can be explained by the pollution status of the water quality at the nine monitoring stations. Among these nine monitoring stations, DYH has the highest annual average WQI and the best water quality, forming a group by itself. ZW has the lowest annual average WQI and the poorest water quality, also forming a separate group. HLH is located on a tributary of the Sanggan River, and its water quality is influenced by the Sanggan River watershed, with significant monthly fluctuations in WQI; thus, it forms its own group.

3.4. Spatiotemporal Variations in Water Quality Data

Discriminant analysis was performed based on the two temporal clusters (WP and DP) obtained from the temporal cluster analysis of monthly water quality data to evaluate the temporal variation characteristics of river water quality in the upstream watershed of the Guanting Reservoir.

Table 3. Coefficients of Classification Functions for Discriminant Analysis of the Two Temporal Clusters WP and DP. (Cross-validation accuracy: Standard Model = 81.5%; Stepwise Model = 85.2%).

Parameter	Standard Model		Stepwise Model
	WP Coefficient	DP Coefficient	WP Coefficient	DP Coefficient
TN	0.317	0.331
TP	0.736	−2.979
NH3-N	−4.588	−3.349
BOD5	4.330	3.406	4.824	4.181
DO	4.610	5.451	4.321	5.123
COD	0.016	0.038
CODMn	0.445	0.672
F	5.518	6.163
Constant	−30.692	−36.151	−26.541	−30.656

presents the discriminant functions derived from the standard model and the stepwise model, respectively. The standard model used all water quality indicators as variables and achieved a cross-validation accuracy of 81.5%, indicating that most water quality data were correctly classified into WP and DP. The stepwise model is based on the Lambda method combined with forward selection and backward elimination processes for stepwise discriminant analysis to determine the simplest model [20]. This method iteratively selects the variable that contributes the most to the classification group at each step, while considering the existing variables in the model. Finally, the model selects two effective water quality indicators, BOD₅ and DO, with a cross validation accuracy of 85.2%. By using two water quality indicators, BOD₅ and DO, a step-by-step model can identify water quality data with high accuracy. This indicates that the combination of these two parameters can more robustly and effectively distinguish seasons compared to using all parameters, which may be achieved by reducing noise and multicollinearity [28].

Discriminant analysis was conducted based on the results of spatial cluster analysis of water quality data from each monitoring station (four site clusters) to evaluate the spatial variation characteristics of water quality in the upstream watershed of the Guanting Reservoir.

Table 4. Coefficients of Classification Functions for Discriminant Analysis of the Four Spatial Clusters. (Cross-validation accuracy: Standard Model = 94.4%; Stepwise Model = 95.4%).

Parameter	Standard Model				Stepwise Model
	Class I	Class II	Class III	Class IV	Class I	Class II	Class III	Clas IV
TN	−0.453	1.304	0.113	0.590	0.576	2.308	1.206	1.480
TP	−6.882	−7.207	−35.381	−12.659	−11.160	−9.967	−38.802	−16.592
NH3-N	−2.789	0.435	0.262	−3.018
BOD5	3.791	4.059	2.084	3.498
DO	4.579	4.055	3.844	4.115
COD	0.140	0.718	1.125	0.493	0.247	0.789	1.226	0.589
CODMn	−0.428	1.469	0.245	0.752	0.182	2.470	0.394	1.315
F	31.470	16.592	93.755	35.091	31.932	17.346	92.661	35.668
Constant	−38.453	−48.689	−133.774	−44.980	−14.655	−29.173	−116.528	−25.699

presents the discriminant functions derived from the standard model and the stepwise model, respectively. During the discriminant analysis, the cross-validation accuracy of the standard model was 94.4%, while that of the stepwise model was 95.4%. Furthermore, the stepwise model selected five effective water quality indicator variables: TN, TP, COD, COD_Mn, and F.

3.5. The Dominant Factors and Possible Pollution Sources of Seasonal Differences in Water Quality Data

To investigate the dominant factors and possible pollution sources causing seasonal differences in water quality in the upstream watershed of the Guanting Reservoir, principal component analysis and factor analysis were conducted based on the water quality data from the two temporal clusters (WP and DP). Two component factors were extracted during the rainy season, and the factor loading values are shown in

Table 5. Load values and variance contribution rates of rainy season factors.

	VF1	VF2
TN	−0.493	−0.264
TP	0.259	0.84
NH3-N	0.035	0.801
BOD5	0.909	0.316
DO	−0.273	−0.66
COD	0.82	0.461
CODMn	0.849	0.441
F	0.868	−0.221
Variance Contribution Rate	54.55%	17.52%
Cumulative contribution rate	54.55%	72.07%

; three component factors were extracted during the dry season, and the factor loading values are shown in

Table 6. Load value and variance contribution rate of dry season factors.

	VF1	VF2	VF3
TN	−0.321	0.786	−0.166
TP	0.17	0.691	−0.147
NH3-N	0.024	0.829	0.2
BOD5	0.117	0.137	0.902
DO	−0.235	0.335	−0.729
COD	0.939	0.017	0.195
CODMn	0.848	0.093	0.306
F	0.863	−0.12	−0.02
Variance Contribution Rate	37.41%	23.59%	14.63%
Cumulative contribution rate	37.41%	61%	75.63%

. The cumulative variance contribution rates were 72.07% and 75.63%, respectively. According to previous studies, factor loadings are classified based on the absolute value as strong loadings (>0.700) and moderate loadings (0.500–0.700).

3.5.1. Dominant Factors and Potential Sources of Pollution During the Rainy Season (WP)

During the rainy season, two component factors were extracted, and their factor loading matrices are shown in Figure 5. VF1 explained 54.55% of the total variance and showed strong positive loadings on BOD₅, COD, COD_Mn, and F, reflecting the significant impact of organic pollutants. Based on the actual situation of the watershed, these pollutants are mainly related to the discharge of domestic sewage and industrial wastewater [29,30]. VF2 explained 17.52% of the total variance, with strong positive loadings on TP and NH₃-N, and moderate loadings on DO, indicating that nutrient pollution is also an important source of pollution during the rainy season, mainly related to agricultural non-point sources such as fertilizer application and livestock breeding 16. The decrease in dissolved oxygen (DO) may be related to the large amount of surface runoff carrying organic matter into water bodies during the rainy season, which exacerbates microbial oxygen consumption.

3.5.2. Dominant Factors and Potential Sources of Pollution During the Dry Season (DP)

Three component factors were extracted during the dry season, and their factor loadings are shown in Figure 6. VF1 explained 37.41% of the total variance, with higher loadings on COD, COD_Mn, and F, while TN and DO showed moderate loadings, indicating that point-source pollution (such as industrial and domestic discharge) still dominates during the dry season, and denitrification may be enhanced at low temperatures [31]. VF2 explained 23.59% of the total variance, with significant loadings on TP and NH₃-N, and moderate positive loadings on TN, further confirming the persistence of nutrient inputs from agriculture and daily sources [16]. VF3 explained 14.63% of the total variance, in which BOD₅ and DO showed an antagonistic relationship, reflecting the seasonal balance between water self purification capacity and oxygen consumption process, which may be related to the weakened photosynthesis of algae and reduced external input at low temperatures.

3.6. Dominant Factors and Potential Sources of Pollution for Spatial Differences in Water Quality Data

In order to explore the dominant factors causing the spatial differences in water quality in the upstream basin of Guanting Reservoir and the possible sources of pollution, principal component analysis and factor analysis are conducted based on the results of cluster analysis of four types of monitoring stations (i.e., Class I, Class II, Class III, and Class IV). Among them, two principal components were extracted for Class I, Class II, and Class IV stations based on the criterion of eigenvalues greater than 1. Three principal components with eigenvalues greater than 1 were extracted for Class III stations. The total variance contribution rates for Class I and Class II are 63.465% and 85.085%, respectively. The total variance contribution rate for Class III is 79.175%, and for Class IV it is 59.586%.

3.6.1. Dominant Factors and Potential Sources of Pollution for Class I Sites

Principal component analysis was conducted on the water quality parameters of each space station, and the results showed significant differences between different clusters. In the Class I cluster, VF1 explained 43.343% of the total variance, with strong positive loadings on TP, NH₃-N, COD_Mn, and F, moderate positive loadings on COD, and strong loadings on TN. This indicates that the DYH site is not only contaminated with nutrients such as nitrogen and phosphorus, but also affected by organic pollution from sources such as domestic sewage and industrial wastewater. At the same time, natural processes also have a significant impact on water quality [32]. VF2 explained 20.122% of the variance, showing a moderate positive load on DO and COD, and a strong load on BOD₅, reflecting the influence of organic pollutant inputs on watershed water quality, such as domestic sewage, agricultural non-point sources, or industrial emissions. The load of BOD₅ may be related to the presence of recalcitrant organic matter in the watershed, which contributes significantly to COD but has a relatively small impact on BOD₅ [33]. The load values of each factor are shown in

Table 7. Load values and variance contribution rates of Type I site factors.

	VF1	VF2
TN	−0.71	0.019
TP	0.833	0.189
NH3-N	0.739	−0.199
BOD5	−0.105	−0.728
DO	−0.415	0.661
COD	0.544	0.566
CODMn	0.847	0.309
F	0.723	−0.394
Variance Contribution Rate	43.343%	20.122%
Cumulative contribution rate	43.343%	63.465%

, and the factor load matrix is shown in Figure 7.

3.6.2. Dominant Factors and Potential Sources of Pollution for Class II Sites

In the Class II cluster, VF1 explained 68.887% of the variance and showed strong positive loadings for TP, BOD₅, COD, COD_Mn, and F, as well as strong loadings for TN and DO, indicating the presence of significant organic matter and nutrient compound pollution in the region, which may mainly come from domestic sewage and agricultural non-point-source pollution [30]. VF2 explained 16.198% of the variance and showed a strong positive load on NH₃-N, indicating that the ZW site is more significantly affected by nitrogen eutrophication [14]. The factor loading values are shown in

Table 8. Load values and variance contribution rates of Class II site factors.

	VF1	VF2
TN	−0.558	0.763
TP	0.909	−0.201
NH3-N	0.02	0.968
BOD5	0.929	−0.081
DO	−0.641	0.581
COD	0.931	−0.252
CODMn	0.919	−0.206
F	0.798	−0.19
Variance Contribution Rate	68.887%	16.198%
Cumulative contribution rate	68.887%	85.085%

, and the factor loading matrix is shown in Figure 8.

3.6.3. Dominant Factors and Potential Sources of Pollution for Class III Sites

In the Class III cluster, VF1 explained 41.184% of the variance, showing strong positive loadings for TP, COD_Mn, and F, and moderate positive loadings for NH₃-N. This suggests that HLH sites have high levels of phosphorus and organic pollution, which may come from agricultural activities (such as fertilizers and livestock manure), domestic sewage, or industrial emissions [34]. VF2 explained 24.296% of the variance, showing strong positive loadings for DO and COD, moderate positive loadings for NH₃-N, and moderate loadings for BOD₅, indicating that the water body is affected by both organic and nitrogen pollution, but dissolved oxygen did not significantly decrease, which may be related to the self purification capacity or other reoxygenation mechanisms of the water body [35]. VF3 explained 13.695% of the variance and showed a moderate positive load on TN, indicating a relatively weak impact of nitrogen pollution. The factor loading values are shown in

Table 9. Load values and variance contribution rates of Class III site factors.

	VF1	VF2	VF3
TN	−0.021	0.87	−0.004
TP	0.767	0.401	−0.287
NH3-N	0.463	0.689	0.303
BOD5	0.223	0.302	−0.748
DO	−0.49	0.275	0.677
COD	0.265	0.191	0.793
CODMn	0.709	0.578	−0.201
F	0.929	−0.045	0.106
Variance Contribution Rate	41.184%	24.296%	13.695%
Cumulative contribution rate	41.184%	65.48%	79.175%

, and the factor loading matrix is shown in Figure 9.

3.6.4. Dominant Factors and Potential Sources of Pollution for Class IV Sites

In the IV cluster, VF1 explained 37.818% of the variance, showing strong positive loadings for BOD₅, COD, and COD_Mn, strong loadings for TN, and moderate loadings for DO. This indicates that the water quality in the upstream watershed of Guanting Reservoir is significantly affected by organic pollution, and there may be more organic nitrogen present, while the total nitrogen content is relatively low. Microbial decomposition of organic matter consumes dissolved oxygen, leading to a decrease in DO [36]. VF2 explained 21.768% of the variance, showing a strong positive load on TP and NH₃-N, and a moderate positive load on F, indicating that the region may be greatly affected by point-source pollution such as industrial wastewater and domestic sewage [37]. The factor loading values are shown in

Table 10. Load values and variance contribution rates of Type IV site factors.

	VF1	VF2
TN	−0.785	0.374
TP	0.058	0.802
NH3-N	−0.036	0.71
BOD5	0.76	0.142
DO	−0.686	0.129
COD	0.804	0.231
CODMn	0.826	0.218
F	0.079	0.58
Variance Contribution Rate	37.818%	21.768%
Cumulative contribution rate	37.818%	59.586%

, and the factor loading matrix is shown in Figure 10.

4. Discussion

4.1. Differences in Pollution Between Rainy and Dry Seasons

Previous studies have shown significant or insignificant differences in surface water quality between rainy and dry seasons [38]. This study conducted an in-depth analysis of water quality data in the upstream watershed of Guanting Reservoir during the rainy and dry seasons, and identified some key pollution characteristics that are crucial for understanding the dominant factors of water quality differences in the upstream watershed of the reservoir area and developing effective management strategies. The analysis results of monthly WQI time at various sites in this study show that BHQ, ZW, HLH, XSB, SXL, and JMY sites have significant fluctuations in WQI values during the rainy season, with lower or even the lowest WQI values appearing (Figure 2). The monthly WQI of various stations during the dry season fluctuated significantly before September, but showed a stable trend after September. Compared with the dry season, the water quality at each station did not show a significant improvement trend during the rainy season. This discovery is different from some traditional beliefs, as previous studies have suggested that water quality is usually better during the rainy season due to increased runoff and dilution effects [39,40]. However, in semi-arid agricultural watersheds such as the upstream of Guanting Reservoir, surface runoff formed by strong seasonal rainfall is a key driving force for transporting accumulated pollutants within the watershed [41]. Research in this watershed has found a significant increase in total nitrogen and COD concentrations during the early rainy season, which is mainly attributed to the input of surface runoff from agriculture and urban-rural fringe areas [13]. Similar phenomena have also been observed in this study. As shown in Figure 2, multiple stations (such as BHQ and ZW) have the lowest WQI in August, which is highly consistent with the phenomenon of “sharp deterioration of water quality at the beginning of rainfall” observed in the Fenhe River basin under similar climatic conditions [42]. Meanwhile, factor analysis in this study showed that the first principal component (VF1) during the rainy season has a high load on organic indicators such as BOD₅, COD, and COD_Mn (Figure 5), further confirming the prominent contribution of organic pollution during the rainy season. This indicates that in the study area, the pollutants carried by surface runoff erosion at multiple stations have a higher degree of deterioration on the water body than the dilution of various water quality indicators by rainfall, resulting in poor water quality during the rainy season. This is a supplement to previous research, indicating that the impact of rainfall on water bodies is uncertain, and appropriate protection measures should be developed for water bodies based on the combined effects of local natural conditions and human factors.

The discriminant analysis of the two time clusters obtained from time clustering shows that the organic pollution index BOD₅ and the water self purification index DO have temporal heterogeneity and are effective indicator variables for distinguishing rainy and dry seasons (Table 3). The results of principal component analysis and factor analysis indicate that the first component factor in the rainy and dry seasons has a high degree of similarity. The water quality in both seasons is mainly affected by organic pollutants generated from domestic sewage and industrial wastewater discharge, as well as nitrogen and phosphorus pollution caused by non-point-source pollution such as agricultural fertilizer use, livestock and aquaculture. Similarly [30], studies on Guanting Reservoir and Baiyangdian have shown that watersheds with intensive agricultural activities experience a significant increase in nitrogen and phosphorus loss during the early rainy season due to surface runoff. This is consistent with the high load results of TP and NH₃-N on VF2 in the factor analysis of this study. This may be because during the dry season, due to reduced water flow, the concentration of organic pollutants and nutrient pollutants in the water is higher. Although rainy season rainfall helps dilute the concentration of these pollutants in water bodies and improve their self purification capacity, it also flushes a large amount of pollutants from domestic sewage and industrial wastewater into water bodies. Therefore, targeted water quality management and prevention measures should be implemented to analyze the differences in water quality and pollution sources in different seasons in the upstream watershed of Guanting Reservoir. The focus should be on strengthening the organic pollution control of domestic sewage and industrial wastewater discharge around the upstream watershed of the reservoir area, while also paying attention to nutrient pollution caused by nitrogen and phosphorus.

4.2. Differences in Spatial Pollution

The spatial analysis results of the annual average WQI at various stations in this study show significant spatial heterogeneity in the water quality status of the upstream watershed of Guanting Reservoir. The DYH station located at the upstream is least affected by human activities, with the highest annual average WQI and the best water quality. This is consistent with the general rule observed in other watershed studies that ‘upstream stations are usually used as background values, and water quality is better than that in the middle and lower reaches’ [43]. In sharp contrast, ZW station has the lowest annual average WQI value and the worst water quality. The site is located in the upper reaches of the Yanghe River, and its severe organic and nitrogen pollution characteristics are closely related to the industrial and agricultural layout in the basin. Similarly, in the Huailai Basin upstream of the Yongding River, research has also confirmed that intensive agricultural irrigation and rural domestic sewage are the key reasons for the deterioration of river water quality [44]. The HLH station is located on the Huliu River, a tributary of the Sanggan River. Its water quality is directly affected by the incoming water quality of the Sanggan River basin, highlighting the important contribution of tributary inputs to the water quality of the main stream and the final inflow flux. This main tributary pollution correlation model is very common in complex river network systems [45].

Of particular note is that the BHQ station closest to the entrance of Guanting Reservoir also exhibits poor water quality. This indicates that pollutants in the upstream water migrate with the water flow and are relatively concentrated at the inlet, leading to an increase in pollutant concentration in the area and directly threatening the quality and ecological safety of the reservoir water. The “pollution accumulation effect” formed in the reservoir inflow area is not a special case. In the study of another important reservoir in northern China, Yuqiao Reservoir, it was also clearly found that pollutants form high concentration areas in the inflow and bay areas [46]. Therefore, future governance work should not only continue to strengthen water quality control throughout the entire basin, but also attach great importance to water purification and ecological interception at the critical node of the reservoir inlet, take effective measures to reduce pollution load, and ensure the ecological balance and water supply safety of the reservoir. Based on spatial clustering, four site clusters were obtained, and discriminant analysis results showed that nutrient pollution indicators (TN, TP), organic pollution indicators (COD, COD_Mn), and inorganic pollution indicators (F) were key variables for effectively distinguishing different clusters. The results of principal component analysis and factor analysis further revealed the spatial differentiation of pollution characteristics in different regions: all four clusters were generally affected by organic pollutants. In addition, clusters I, II, and III are also affected by both phosphorus pollution and fluoride (the first principal component is positively loaded on TP and F), while clusters I, II, and IV are generally affected by nitrogen pollution (the first principal component is loaded on TN). The above results indicate that the water quality pollution in different areas of the upstream watershed of Guanting Reservoir has both commonalities (generally affected by organic pollution) and unique characteristics (dominated by different pollutants such as nitrogen, phosphorus, and fluorine). This spatial pattern of “coexistence of common pollution and local characteristic pollution” is highly compatible with the watershed zoning control concept based on source analysis. A study on multiple drinking water reservoirs in eastern China concluded that identifying and controlling these dominant pollution sources with spatial differentiation characteristics is a prerequisite for implementing precise and efficient governance strategies [47]. The different pollution levels in this study area are likely related to factors such as geographical environment, land use type (such as agricultural fertilizer application intensity), and differences in the discharge of domestic and industrial wastewater in each region.

Therefore, in order to more effectively manage and improve the water quality of the upstream watershed of Guanting Reservoir, it is necessary to abandon the “one size fits all” governance model and instead adopt differentiated control strategies based on spatial differentiation characteristics. For example, in Cluster I and Cluster III areas where agricultural non-point-source pollution is prominent, scientific agricultural management measures should be promoted to reduce the use of fertilizers and pesticides. In Cluster II, where domestic sewage has a significant impact, efforts should be made to strengthen the coverage and operational efficiency of township sewage treatment facilities. In areas where organic pollution and specific inorganic pollution (such as fluoride) coexist, it is necessary to strengthen the supervision of water related industries and the deep treatment of wastewater. This management framework of “zoning, classification, and grading” has been proven effective in the management practices of similar water bodies in northern China such as Baiyangdian [48]. At the same time, continuously strengthening the water quality monitoring network and evaluation system and quickly capturing dynamic changes in water quality will provide an indispensable scientific basis for the formulation and adjustment of the above-mentioned refined governance strategies.

5. Conclusions

This study comprehensively applies the improved water quality index method and multivariate statistical analysis method to systematically reveal the spatiotemporal evolution law, dominant driving factors, and potential pollution sources of water quality in the upstream watershed of Guanting Reservoir in 2024. The main comprehensive conclusions are as follows:

(1)

The water quality of the watershed is sensitive to rainfall response, and the pollution input flux is the key to dominant temporal differentiation. Although rainfall has a certain dilution effect, this study found that the summer rainy season (June August) is the worst period for water quality throughout the year. This profoundly reveals that in semi-arid watersheds such as the upstream of Guanting Reservoir, which are strongly influenced by human activities, the pollution migration flux driven by rainfall runoff increases, and the negative effects caused by it systematically exceed the physical dilution effect of rainfall itself. This discovery revises the traditional belief that ‘water quality will inevitably improve during the rainy season’ in such areas and emphasizes the extreme importance of non-point-source pollution control in rainy season management. After the rainy season (starting from September), the water quality tends to stabilize and improve, which confirms that the self-purification ability of water bodies can be effectively exerted after the reduction in pollution input, further proving that controlling external input is the fundamental way to improve water quality.

(2)

The clear spatial differentiation pattern of water quality confirms the basin pollution pattern of “human activity pressure accumulating downstream along the river network and estuary”. Spatial analysis clearly depicts a continuum from a relatively clean state upstream to severe pollution downstream. The water quality at the upstream DYH station is the best, reflecting the baseline state of the background environment. The significant deterioration of water quality at middle and downstream stations (such as ZW and BHQ) directly reflects the migration, mixing, and aggregation of pollution loads generated by human activities such as industry, agriculture, and residential life in the river network. In particular, the BHQ station, which serves as the direct “gateway” to the reservoir, has poor water quality, highlighting the vulnerability of the estuary area as a pollutant “sink” and posing the most direct and serious threat to the overall water quality of the reservoir.

(3)

The analysis of pollution sources reveals the composite pollution characteristics of “point surface combination”, indicating the target of differentiated control. The comprehensive analysis of time and space shows that organic pollution and nutrient pollution are the two core issues that run through the entire basin and time period. However, there are temporal and spatial differences in the priority and sources of their contributions: In terms of time, the pollution during the rainy season is more reflected in the composite pollution carried by agricultural non-point sources and urban runoff; the dry season, on the other hand, is more prominently manifested as the sustained pressure of point-source emissions from daily life and industry. In terms of space, there are significant differences in the dominant pollution sources in different subregions, such as the mixed pollution of daily life and agriculture at the ZW station and the agricultural non-point-source pollution characteristics at the HLH station. This accurately depicts the complexity and diversity of pollution sources within the watershed.

Comprehensive management insights: The conclusion of this study indicates that a refined strategy of “spatiotemporal dual control and source sink linkage” must be adopted for water quality management in the upstream watershed of Guanting Reservoir. In terms of time, an emergency plan for pollution prevention and control should be developed for the rainy season, with a focus on strengthening the interception and buffering of source pollution. In terms of space, differentiated control measures based on sub-basin units must be implemented to reduce emissions at the source in key polluted areas such as ZW in the upper and middle reaches, and ecological restoration and enhanced purification must be carried out at the reservoir inlet (BHQ) to cut off the final channel for pollutants to enter the reservoir. Only by accurately directing limited governance resources towards key temporal and spatial nodes and dominant pollution sources can the water environment quality and ecological security level of the upstream watershed of Guanting Reservoir be fundamentally improved.