Characterizing the Spatiotemporal Distribution of Water Quality and Pollution Sources in Mountainous Watershed

Abstract

The precise identification of pollution sources constitutes a cornerstone for effective water environment management in mountainous watersheds. This study employed principal component analysis–absolute principal component scores–multiple linear regression (PCA-APCS-MLR) receptor modeling to analyze monthly water quality indicators across the Longxi River Basin. Results revealed comparable water quality between the main stream and its tributaries, with no statistically significant differences identified. Water quality exhibited a distinct spatial pattern, with superior conditions in the upstream and downstream segments compared to the middle reaches. Water quality parameters exhibited significant seasonal variations. During the wet period, the degradation of water quality was primarily driven by diffuse agricultural sources, contributing 42.9%, followed by watershed background levels and surface runoff. In the dry season, rural domestic wastewater (39.3%) was the leading pollution source. For Permanganate index (COD_Mn) exceedance, basin background and agricultural non-point sources in the wet season were the main contributors (46.8% and 44.7%, respectively). For ammonium nitrogen (NH₃-N), wet season agricultural non-point sources (44.4%) and dry season rural domestic pollution (71.8%) were key contributors. Agricultural non-point sources were the dominant pollution source for total nitrogen (TN) in the wet season (84.2%). Effective water quality improvement in the Longxi River Basin hinges on targeted strategies—to mitigate diffuse agricultural sources through optimized fertilization, and to enhance the collection and treatment of rural domestic sewage. This study not only enhances the understanding of pollution source distribution and quantification in mountainous watersheds, but also serves as a vital reference for formulating targeted water environment management strategies.

1. Introduction

Rivers, as a fundamental source of Earth’s surface freshwater [1], constitute the ecological and economic cornerstone of human societies. However, this vital role is increasingly compromised by escalating water pollution, a global crisis exacerbated by rapid urbanization and economic growth [2,3,4,5]. River water quality is governed by both intrinsic natural processes and diverse anthropogenic activities. Natural factors include soil erosion and weathering, mineral oxidation of rocks, extreme weather such as droughts and heavy rainfall [1,6,7]. Anthropogenic factors mainly involve rural and urban domestic sewage discharge, pollution from agriculture, livestock and aquaculture, industrial wastewater discharge, etc. [8,9,10,11]. Particularly in developing countries, anthropogenic activities have significantly contributed to the deterioration of surface water quality [5,12,13]. The Bulletin of Ecological and Environmental Conditions in China 2024 indicates improved water quality in the main stem of the Yangtze River basin, though certain tributaries remain slightly polluted. As integral components of the water cycle, pollution in rivers can propagate to and degrade the quality of surrounding water bodies. Water environment problems in small basins remain prominent, necessitating targeted improvement. Accurately delineating spatiotemporal water quality dynamics and pinpointing pollution sources is a prerequisite for robust water environment assessment and protection [14,15].

At present, many scholars have employed statistical methods in analyses of water quality spatiotemporal patterns to identify contributing pollution sources. Pillsbury et al. employed regression analysis to examine how natural and anthropogenic sources influenced the spatiotemporal variation in water quality in the Hillsborough River [16]. Multivariate statistics were employed by Liu et al. to characterize the spatiotemporal water quality patterns (2008–2020) in the Yangtze River Basin and to identify key drivers of COD_Mn, NH₃-N, and total phosphorus (TP) levels [17]. In the Wushui River Basin, Zhu et al. utilized principal component analysis and thermal distribution to examine spatiotemporal variation and identify main pollutant factors [18]. However, these studies lack quantitative source apportionment for river pollution.

On the other hand, mechanism models have also been employed for quantitative analysis. Li (2018) utilized mechanistic modeling to quantify nitrogen outputs from different agricultural land types in a subtropical watershed, thereby elucidating the spatial variability of N sources [19]. Bai et al. proposed a one-dimensional model-based method for analyzing pollution source contributions, which they then apply to calculate the contributions of a typical transboundary river section in the Yangtze River Basin across different periods [20]. Although mechanistic models can derive river input coefficients, they typically require extensive laboratory-derived data.

In recent years, advances in automated water quality monitoring networks and technologies have enabled management departments to rapidly acquire vast amounts of accurate data. This provides a crucial foundation for receptor modeling and source apportionment based on multivariate statistical techniques. Many scholars have applied multivariate statistical techniques to identify key pollution sources and quantify their contributions in diverse river basins worldwide, such as the rivers of South Florida [10], the Tuo River Basin in China [14], the Imjin River in South Korea [21], the Zhangweinan Canal in China [22], the coastal waters of Hong Kong [23], Danjiangkou Reservoir [24], the Laizhou Bay [25], the Songhua River in Harbin [26] the Karasu River in Turkey [27], and the upper Yangtze River [28]. However, existing applications have predominantly focused on coastal, plain, or more industrially developed watersheds.

However, the dynamics and drivers of pollution in agricultural-dominated mountain watersheds during wet and dry seasons have yet been studied. Rivers in mountainous watersheds faced significant variations in water quantity due to factors such as large channel slopes, small-scale reservoirs, and levee construction [29,30]. Water quality fluctuates in mountainous watersheds due to changes in topography, climate and other factors, especially after heavy rainfall, which accelerates water flow and leads to rapid migration of pollutants [31,32,33]. While point-source pollution has been significantly mitigated in China through improved management and treatment, non-point source pollution, especially in mountainous regions during wet periods, has come to the forefront [34,35,36]. Accurate source assessment in these complex environments requires recognition of the natural hydrochemical baseline, which is governed by catchment geology, soils, and vegetation. Against this natural backdrop, the contributions and transport of anthropogenic pollutants can be more clearly evaluated.

Therefore, this study aims to elucidate the spatiotemporal variations in water quality across seasons and to perform a quantitative source apportionment of pollution in a representative mountainous watershed. For this purpose, we focus on the Longxi River Basin, a first-order tributary of the Yangtze River located in the mountainous municipality of Chongqing, China. This basin is characterized by a karstified low mountainous and hilly terrain, numerous rain-fed rivers, and extensive agricultural land use, rendering it an ideal system to investigate the interplay between natural processes and anthropogenic pressures. The findings are expected to provide a scientific basis and practical recommendations for water quality management in similar mountainous regions.

2. Materials and Methods

2.1. Study Area

The Longxi River Basin (LRB) is situated in the hinterland of the Three Gorges Reservoir Area within the upper reaches of the Yangtze River, between longitude 107°12′–107°24′ E and latitude 29°56′–30°12′ N. Its main stream extends 238 km in length. The Longxi River originates from Liangping District, Chongqing Municipality, and flows through Dianjiang County and Changshou District from the north to the south, and injects itself into the Yangtze River at Fengcheng Street of Changshou District, with a watershed area of 3248 km². The water system is well developed and the tributaries are densely distributed, with about 300 tributaries of different sizes forming a dendritic (treelike) water system. The terrain of the watershed slopes from northeast to southwest, with the highest watershed ridge at an elevation of 1209 m and an outlet elevation of 33 m, making it a typical mountainous watershed. The average annual precipitation ranges from 1150 to 1197 mm, with most of the precipitation occurring from May to September. With an average annual flow of 49 m³/s at the estuary, the annual runoff amounts to 1.55 billion m³.

The Longxi River Basin has a long history of hydropower station development. At present, there are 16 hydropower stations distributed in the main stream and 17 in the tributaries, contributing to flow within the runoff. There are 49 urban domestic sewage treatment plants in the basin, with a total design and actual treatment capacity of 211,400 and 198,800 m³/d, respectively. The water-related enterprises in the basin are mainly agricultural and food processing industry, accounting for 40.74% of the total number of water-related enterprises. The existing rural resident population in the basin is about 1.3 million, comprising over half of the total population. Agriculture along the basin is well developed, with a focus on rice cultivation, and the area also contains extensive orchards and fields dedicated to commercial vegetable production.

2.2. Data Sources and Pre-Processing

This study analyzed the water quality characteristics of the Longxi River Basin from 2019 to 2022 using data from six monitoring sites: US, MS, LS, TR1, TR2 and TR3 (Figure 1). The division of the Longxi River into upper, middle, and lower reaches was based on an integrated analysis of catchment units delineated from Digital Elevation Model (DEM) data and county-level administrative boundaries, ensuring each section represents a coherent unit in terms of both hydrology and management. The US, MS, and LS labels denote the upper, middle, and lower Longxi River, respectively. TR1, TR2, and TR3 represent the main tributaries within the basin. The water quality data were obtained from the Chongqing Ecological Environment Data Center. According to the official website of Chongqing Meteorological Bureau, it designates May–September as the wet season and the other months as the dry season. The analysis of the indicators carried out followed the standard of GB 3838-2002 [37]. All the indicator data below or near the detection limit were deleted in this study.

This study investigated key water quality parameters, including water temperature (WT), pH, dissolved oxygen (DO), COD_Mn, chemical oxygen demand (COD), five-day BOD (BOD₅), NH₃-N, TP, TN, electric conductivity (EC), fluoride (F⁻) and arsenic (As). Their spatiotemporal variations were analyzed to quantitatively assess potential pollution sources within the Longxi River Basin.

2.3. Methods of Analysis

Key multivariate statistical techniques, such as principal component analysis (PCA), the absolute principal component scores–multiple linear regression (APCS-MLR) model, and positive matrix factorization (PMF), are widely used as essential tools for atmospheric and soil studies [38,39,40]; these methods have subsequently been applied to groundwater and specific aquatic pollutants [41,42], as well as for characterizing water quality changes and analyzing pollution sources in rivers [10,14,28,43]. For example, in a study of three rivers in South Florida, Gholizadeh, M.H. et al. applied both APCS-MLR and PMF models, finding that APCS-MLR provided a superior fit to the water quality data, particularly during high-flow periods [10].

In summary, this study developed a novel PCA-APCS-MLR framework that integrates principal component analysis and multiple linear regression for source apportionment in the Longxi River Basin. We used Origin 2021 software to graph the trends of LRB water quality indicators on spatial and temporal scales. SPSS 26.0 and R language 4.1.0 were used to realize PCA-APCS-MLR model construction.

2.3.1. Principal Component Analysis

The first step in constructing the PCA-APCS-MLR model is to apply principal component analysis (PCA) to extract the principal components, eigenvectors, and factor score coefficients from the water quality data. PCA is a dimensionality reduction technique that transforms correlated original variables into a smaller set of uncorrelated composite variables while preserving maximum variance from the original dataset [44,45]. These new composite variables are called principal components, and they can represent the main pollution information exhibited in all monitoring data, facilitating the analysis of pollutants sources. The calculation formula is

$$A_{jk}={\sum }_{i=1}{w}_{ij}{z}_{ik}$$

(1)

$$z_{ik}={\displaystyle \frac{C_{ik}\overline{C_i}}{{\sigma }_i}}$$

(2)

$A_{jk}$: Principal component score j for sample k;

$w_{ij}$: Factor loading (coefficient) of pollutant i on principal component j;

$z_{ik}$: Standardized value of the concentration of pollutant i for k samples;

$C_{ik}$: Measured concentration of pollutant i in sample k(mg/L);

$\overline{C_i}$: Arithmetic mean concentration of pollutant i (mg/L);

${\sigma }_i$: Standard deviation of pollutant i.

The main computational procedure is as follows: (1) Data Standardization: Normalize the original dataset using the z-score method. (2) Data Suitability Test: Verify the suitability for factor analysis using the KMO (Kaiser–Meyer–Olkin) and Bartlett’s tests. (3) PCA Execution: From the standardized data, extract the eigenvalues and eigenvectors of the correlation matrix to determine the number of principal components (based on variance explained). (4) Score Calculation: Compute the composite scores for the retained principal components.

2.3.2. APCS-MLR Model

The APCS-MLR receptor model for water quality is based on the principle of mass conservation and linear source additivity [10]. Its core procedure consists of two steps: (1) PCA and APCS Transformation: Perform principal component analysis (PCA) on standardized data and convert the resulting factor scores into absolute principal component scores (APCSs) [46,47]. (2) MLR Source Apportionment: Conduct multiple linear regression (MLR) with APCS as independent variables and measured pollutant concentrations as the dependent variable to quantify the contributions of potential sources. The APCS values are calculated using Equations (3)–(5):

$$S_{jk}=A_{jk}-A_{0j}$$

(3)

$$A_{0j}={\sum }_{i=1}{w}_{ij}{z}_{0i}$$

(4)

$$z_{0i}={\displaystyle \frac{0-\overline{C_i}}{{\sigma }_i}}$$

(5)

where $S_{jk}$ is the j absolute principal component score for the k samples; $A_{0j}$ is the j principal component score at a value of 0; and $z_{0i}$ is the normalized value when the concentration of the i pollutant factor is set to zero.

In the third step, a multiple linear regression is performed using APCS values (independent) and measured concentrations (dependent). The resulting regression coefficients thereby quantify the contribution of pollution sources corresponding to each principal factor.

$$C_{ik}={\sum }_{j=1}{a}_{ji}·S_{jik}+b_j$$

(6)

$C_{jk}$: Measured concentration of pollutant i in sample k;

$a_{ji}$: Regression coefficient representing the contribution of source j to pollutant i.

$S_{jik}$: Absolute principal component score (APCS) for source j, pollutant i, and sample k.

$b_j$: Constant term for pollutant i, often interpreted as the contribution from unidentified sources

The proportional contributions from both identified and unidentified sources are calculated based on the absolute values of the regression terms. Pollutant i:

Contribution of Identified Source j:

$$P{C}_{ji}={\displaystyle \frac{\left|a_{ji}·\overline{S_{ji}}\right|}{\left|b_i\right|+\sum _j{|a}_{ji}·\overline{S_{ji}}|}}$$

(7)

Contribution of Unidentified Sources (lumped into the constant term b_i):

$$P{C}_{ji}={\displaystyle \frac{\left|b_i\right|}{{|b}_i|+\sum _j|a_{ji}·\overline{S_{ji}|}}}$$

(8)

where $\overline{S_{ji}}$ is the mean of the j absolute principal component factor scores for pollutant i.

In this study, principal component analysis was performed for 10 water quality indicators at 6 sites from 2019 to 2022. The raw data were grouped into wet and dry seasons, and principal components were extracted for each season separately. In this study, we used the diagnostic covariance matrix and the resulting eigenvalues as criteria, and considered eigenvalues greater than 1 as the main contaminating factors identified, which explain the main information of the raw data. Principal component analysis can identify the main pollutants, but in terms of the actual sources and processes controlled, such an interpretation is subjective and has some generalization limitations [48].

Prior to PCA, the data underwent sequential pre-processing: calculation of descriptive statistics, z-score standardization, and evaluation of suitability via the KMO and Bartlett’s tests for both seasons. The results indicate that Bartlett’s test of sphericity yielded statistics of 367.635 (wet season) and 422.144 (dry season), both with a significance level (p-value) of <0.001, confirming that the correlation matrix is suitable for factor analysis. Additionally, the KMO test yielded values of 0.673 (wet period) and 0.714 (dry period). These KMO values confirmed the suitability of the original variables for factor analysis. Following PCA, an orthogonal rotation (e.g., Varimax) was applied to the factor loading matrix to facilitate the interpretation of the extracted components. This procedure simplifies the factor structure and yields a new set of variables termed varimax factors (VFs). To clarify the factor structure, VFs were interpreted based on their correlation coefficients (loadings) with the original variables. Following common practice [49,50], absolute loading values were classified as strong (>0.75), moderate (0.50–0.75), or weak (0.30–0.50).

2.3.3. Model Validation

To resolve the composition and spatial distribution of key pollution sources, the APCS-MLR model was employed to link identified sources with the concentrations of major water quality indicators (COD_Mn, COD, BOD_5, NH₃-N, TP, TN). Multiple linear regression was performed using SPSS 26. To assess the model’s reliability, measured values for each water quality indicator were compared with model predictions (Figure 2). According to the results, the R² of the model’s calculated and measured concentrations of most of the water quality indicators in the wet and dry seasons is greater than 0.5, indicating that the predicted and measured concentrations are in good agreement [10,51]. Furthermore, the predicted-to-measured ratio was close to 1, which confirms the model’s reliability for pollution source analysis in the Longxi River Basin.

3. Results and Discussion

3.1. Spatiotemporal Variations in Water Quality

The overall water quality of the six monitoring sites in the LRB in 2019–2022 meets the standards. BOD₅, COD and F⁻ showed declining yearly trends. However, exceedances of COD_Mn, TP, and NH₃-N were observed in specific periods (Figure 3). The national surface water quality standard thresholds for key parameters are COD_Mn (6 mg/L), TP (0.2 mg/L), and NH₃-N (1.0 mg/L). The number of exceedances for COD_Mn was 7% of the total number of monitors, and COD_Mn concentration exceeding 5.5 mg/L up was to 20%. TP exceeded the standard by 8%, with the highest concentration reaching 0.41 mg/L. NH₃-N exceedances accounted for 2% of the total number of monitoring events.

In terms of water period distribution (

Table 1. Mean and maximum water quality concentrations during the wet and dry period.

Parameters	Wet Season		Dry Season
	Mean	Maximum	Mean	Maximum
WT/°C	26.20	32.80	14.65	26.50
pH	7.73	9.01	7.81	8.98
ρ(DO)/(mg/L)	7.10	15.60	8.72	13.00
ρ(CODMn)/(mg/L)	4.78	7.30	4.25	5.28
ρ(COD)/(mg/L)	14.77	33.00	14.37	28.50
ρ(BOD5)/(mg/L)	2.19	4.40	2.01	4.20
ρ(NH3-N)/(mg/L)	0.20	0.98	0.28	1.43
ρ(TP)/(mg/L)	0.15	0.41	0.12	0.38
ρ(TN)/(mg/L)	2.07	6.60	2.29	5.77
ρ(F−)/(mg/L)	0.296	0.58	0.28	0.66
ρ(As)/(mg/L)	0.0012	0.0069	0.0008	0.0025
EC(μS/cm)	408.18	969.00	458.92	875.00

Notes: mg/L (milligram per liter), μS/cm (microsiemens per centimeter).

), WT, COD_Mn, TP, and As exhibited significantly elevated levels during the wet period, whereas DO, NH₃-N, and EC showed higher values in the dry period. BOD₅, COD and F⁻ showed a yearly decreasing trend, but water period change was not obvious. WT showed the largest change, with an average WT of 26.2 °C in wet season, and 14.65 °C in dry season; the wet season was 78.84% higher than the dry season. As, TP, and COD_Mn concentrations were 61.04%, 25%, and 12.5%, respectively. All three indicators exhibited higher concentrations during the wet season, with As at 0.0012 vs. 0.0008 mg/L, TP at 0.15 vs. 0.12 mg/L, and COD_Mn at 4.78 vs. 4.25 mg/L. DO and WT were negatively correlated (r = 0.5118, p < 0.01), and the dry season saw higher concentrations, with an average concentration of 8.72 mg/L and 7.1 mg/L. NH₃-N concentrations were elevated during the dry season, with an average concentration of 0.28 mg/L and 0.2 mg/L. The maximum value of NH₃-N during the dry season amounted to 1.43, and exceedances often occurred during dry season. BOD₅, COD, and F⁻ showed minimal seasonal variation, with concentrations in the wet vs. dry season of 2.19 vs. 2.01 mg/L, 14.77 vs. 14.37 mg/L, and 0.296 vs. 0.28 mg/L, respectively.

Spatially, the middle reaches emerged as the critical zone with comparatively impaired water quality. Regarding different water quality indicators, concentrations varied between the main stream and tributaries. WT was slightly higher in the main channel, the tributaries exhibited elevated pH yet lower levels of DO, COD_Mn, COD, NH₃-N, and TP, whereas the main stream had lower concentrations of BOD₅, TN, F⁻, As, and EC. The LS site outperformed other tributaries in water quality, attributable to the dilution and mixing effects from the Yangtze River’s backwater. The tributary site LR1 is situated in the Huilong River. Data from the Second National Pollution Source Census and a comprehensive assessment of the Longxi River’s ecological health indicate that the surrounding area in Liangping District has fewer sewage-discharging enterprises, along with lower loads from rural domestic and plantation runoff compared to other regions. Consequently, this site exhibits better water quality than other tributary sections. In contrast, sites MS and LR2 are located in the middle reaches of the Longxi River Basin in Dianjiang County, with poor water quality, and exceedances of COD_Mn, COD, and TP occurred during the monitoring period, with BOD₅ exceeding the standard in March 2019 for section MS, and NH₃-N exceeding the standard in March 2020 for section LR.

3.2. Comparative Analysis of Pollution Sources in Wet vs. Dry Seasons

The principal component loading matrices and total variance for both wet and dry water periods are shown in

Table 2. Matrix of rotated factor loadings in spatial analysis during wet and dry seasons.

Parameters	Wet Season			Dry Season
	VF1	VF2	VF3	VF1	VF2	VF3
WT/°C	−0.258	0.838	0.142	−0.035	−0.762	0.243
pH	0.086	0.882	0.155	0.164	0.708	0.209
ρ(DO)/(mg·L−1)	0.277	0.720	0.004	0.055	0.860	−0.072
ρ(CODMn)/(mg·L−1)	0.778	0.352	0.104	0.338	−0.137	0.651
ρ(COD)/(mg·L−1)	0.667	−0.106	0.115	0.423	0.172	0.641
ρ(BOD5)/(mg·L−1)	0.504	0.179	0.572	0.667	0.222	0.005
ρ(NH3-N)/(mg·L−1)	0.519	−0.152	0.499	0.850	−0.019	−0.030
ρ(TP)/(mg·L−1)	0.162	0.162	0.739	0.772	−0.123	0.119
ρ(TN)/(mg·L−1)	0.750	0.022	0.088	0.539	0.256	0.215
EC(μS·cm−1)	−0.146	0.091	0.784	0.381	0.429	−0.600
Eigenvalue	3.087	1.970	1.311	2.962	2.108	1.000
Total variance/%	30.87	19.70	13.11	29.62	21.08	10.00
Cumulative variance/%	30.87	50.57	63.68	29.62	50.70	60.70

and Figure 4.

During the wet season, the first principal component (VF1) accounted for 30.87% of the total variance, capturing the largest share of information in the dataset. The main loading variables strongly associated with VF1 were COD_Mn and TN (Table 2 and Figure 4). COD_Mn and TN can originate from both point sources (e.g., domestic wastewater) and non-point sources—primarily nitrogenous nutrients in runoff from excessive fertilizer application within the watershed [14,21,52]. For VF1, COD_Mn and TN changed from strong loading in wet season to weak loading in dry season. Therefore, the first principal component in the wet season can be categorized as agricultural non-point sources. The prominence of this agricultural signature in the wet season is also modulated by the watershed’s natural setting—its steep topography accelerates runoff, while the soils and underlying geology influence the leaching and delivery efficiency of nutrients to the river network. The second principal component (VF2) contributed 19.70%, with pH and TW as the main loadings. In this mountainous watershed, pH is primarily governed by the natural buffering capacity of the catchment, which is influenced by the weathering of carbonate and silicate minerals in the local geology [53,54]. Water temperature is primarily a natural factor controlled by seasonal atmospheric conditions, with minor influence from geothermal inputs. Therefore, VF2 is interpreted as representing the natural background signature of the watershed. VF3 explained 13.11% of the water quality variation, with the main loading variable being EC. EC is affected by the concentration of metal ions and the corresponding anions in the water. High salt content increases the conductivity, indicating a great influence by surface runoff. In this study, the VF3 of EC changed from a positive and strong correlation of 0.784 during wet season to a negative and moderate correlation of −0.6 during dry season, suggesting a significant influence by wet deposition of precipitation. So VF3 is categorized as surface runoff and wet deposition. The elevated EC associated with wet season runoff is fundamentally linked to the weathering of soluble minerals (e.g., carbonates, evaporites) in the catchment, which supplies the major ions contributing to conductivity.

During dry season, VF1 explained 29.62% of the water quality variation, with strong loadings for NH₃-N and TP factors. NH₃-N and TP were high in farmland runoff, livestock and poultry wastewater and rural domestic sewage [17,55,56]. In general, agricultural runoff and livestock wastewater flow into the river with surface runoff in the wet season. However, in this study, NH₃-N and TP showed a strong positive correlation in the dry season but only a moderate-to-weak correlation in the wet season. The study area, being a commercial grain base area of China, has a large number of rural residential population, with relatively backward domestic sewage treatment facilities, resulting in untreated sewage being directly discharged into nearby rivers, increasing nitrogen and phosphorus loads. Hence, VF1 in the dry season is categorized as rural domestic sewage. The second principal component (VF2) contributes 21.08% of the water quality variation, and with strong loadings for DO and WT. Given that all aquatic organisms rely on dissolved oxygen to metabolize food for energy, it serves as the best indicator of environmental quality. As WT increase, aquatic organisms become more biologically active, consuming more DO. Therefore, VF2 is interpreted as reflecting the combined effects of physicochemical processes and biological activity [10].

The third principal component of VF3 explained 10.00% of the variation in water quality, and the main loading variables were COD_Mn and COD factors. Both of COD_Mn and COD are pollution indicators of organic matter. Therefore, it suggests that the third main component is related to industrial and domestic wastewater in the basin [27,57]. Along the watershed, there are many agricultural and sideline product processing enterprises with most enterprises treating sewage through centralized industrial sewage treatment plants or urban sewage treatment plants. However, some industrial enterprises with incomplete wastewater treatment facilities directly discharge into natural water bodies. Therefore, the VF3 is categorized as an industrial source.

This comparative seasonal analysis highlights that the relative contributions of pollution sources vary markedly. Crucially, our apportionment identifies a distinct natural background component (VF2) and reveals how other seasonal sources (e.g., agricultural runoff, surface runoff) are expressed through and interact with the physical and geochemical template of the mountainous watershed. In the wet season, due to the mountainous characteristics of the region, the pollutants mainly originate from agricultural non-point sources. In the dry season, pollutants mainly originate from rural domestic sources. Other pollutants were mainly represented by the river bottom sand mineral and other physical and chemical characteristics of the basin both in wet and dry season. According to the official statistical bulletins (Chongqing Municipality and three relevant districts/counties), the rural population in the Longxi River Basin in 2022 was 41.93%, significantly higher than the overall rural population in Chongqing Municipality, which was 29.04%. The area of Longxi River Basin accounted for 3.94% of Chongqing Municipality, but its annual sown area of grain accounted for 9.49% of Chongqing Municipality. This indicates that the mountainous watershed is significantly influenced by its agricultural, rural, topographic, and geomorphic background characteristics.

3.3. Quantitative Assessment of Pollution Source Contributions Across Seasons

The results from these analyses are presented in

Table 3. Mean source contributions to water quality variables across wet and dry seasons.

Variable	Source Contribution in Wet Season				Source Contribution in Dry Season
	S1	S2	S3	UIS	S1	S2	S3	UIS
CODMn	44.71%	46.75%	3.19%	5.35%	19.52%	18.31%	20.18%	41.99%
COD	32.85%	12.14%	3.02%	51.98%	30.75%	28.88%	25.01%	15.37%
BOD5	33.97%	27.90%	20.63%	17.50%	40.34%	31.03%	0.15%	28.49%
NH3-N	44.43%	30.09%	22.92%	2.56%	71.81%	3.42%	1.43%	23.34%
TP	17.28%	40.02%	42.63%	0.07%	45.57%	17.11%	3.96%	33.36%
TN	84.16%	5.75%	5.26%	4.83%	27.57%	30.37%	5.93%	36.13%
Mean	42.90%	27.11%	16.28%	13.71%	39.26%	21.52%	9.44%	29.78%

Notes: In wet season, S1 represents agricultural non-point sources, S2 represents watershed background, S3 represents surface runoff and wet deposition; in dry season, S1 represents rural domestic sewage, S2 represents physicochemical and biological non-point source pollution, S3 represents industrial source; UIS is unidentified influence source.

and Figure 5.

During the wet season, analysis of source contributions indicates that agricultural sources were the predominant contributor affecting water quality in the study area (42.9%), followed by the basin background (27.11%), a signature primarily governed by geological weathering and atmospheric inputs. For the main exceeding factors, COD_Mn, the main sources are natural factors and agricultural sources, with contribution rates of 46.75% and 44.71%, respectively. COD mainly comes from uncertain factors and agricultural sources. Agricultural sources are also the main sources of BOD5, NH₃-N and TN. The contribution rate of agricultural sources to TN reaches 84.16%. TP was greatly affected by the watershed background and surface runoff; the contribution rate is 40.02%.

During the dry season, the study area is most influenced by rural domestic sources (39.26%), followed by unidentified sources, and physicochemical and biological non-point sources, with contributions of 29.78% and 21.52%, respectively. Industrial sources accounted for a smaller share of 9.44%. Rural domestic sewage is the main source of NH₃-N, TP, BOD₅ and COD, with contributions of 71.81%, 45.57%, 40.34% and 30.75%. For COD_Mn, contributions from rural domestic sewage, physicochemical and biological non-point sources, and industrial sources all account for approximately 20% each. For TN, the unidentified sources contribute 34.73%, followed by physicochemical and biological non-point sources and rural domestic sources, with contributions of 31.07% and 26.20%.

The main sources of NH₃-N pollution in both wet and dry season are agricultural non-point sources and rural domestic sewage, with contribution rates significantly higher than that of other sources. Agricultural sources contributed up to 84.16% of the TN load during the wet season. This highlights that nitrogen pollution in the Longxi River Basin is primarily attributed to agricultural and rural activities, particularly nitrogen fertilizer application.

3.4. Recommendation

The preceding analysis has revealed significant spatiotemporal heterogeneity in the water quality of the Longxi River Basin. The identification of key pollution sources in the LRB offers critical support for evidence-based water environment management by local authorities. However, unidentified pollution sources were present in all extracted principal factors, especially in the dry season, when they accounted for a higher proportion of the sources for COD and TN than any other identified category. The results are subject to environmental pressure changes, indicating significant variability. It is necessary to use multivariate statistical methods for qualitative identification of pollution sources alongside regular source surveys [58]. The source of river pollution can be analyzed by multi-model coupling methods, such as forward accounting of pollution sources, to corroborate statistical and mechanistic models, thereby improving the credibility of the findings.

4. Conclusions

This study employed multivariate statistics and a PCA-APCS-MLR model to quantify the spatiotemporal variations and pollution source contributions in the surface water of the Longxi River Basin—a mountainous first-order tributary of the Yangtze River. The analysis was based on 12 key parameters measured at six strategic sampling points across its main stream and tributaries during both wet and dry periods. An evaluation of spatiotemporal water quality patterns showed that the overall water quality of the basin meets the standard, and there are exceedances of COD_Mn, TP, NH₃-N and other indexes in individual time periods. Water quality parameters exhibited distinct seasonal patterns. WT, COD_Mn, TP, and As were significantly elevated during the wet season, whereas DO, NH₃-N, and EC peaked in the dry season. In contrast, BOD₅, COD, and F⁻ demonstrated a consistent annual decreasing trend without pronounced seasonal fluctuations. Spatially, there was no consistent pattern of water quality in the main stream and tributaries. Water quality was superior in the upstream and downstream reaches relative to the middle section.

Three principal components extracted per season via PCA were used to identify potential pollution sources by analyzing their major loading variables. The PCA-APCS-MLR model quantitatively revealed the differences in the contribution of each pollution source to the six water quality parameters of major concern. During the wet season, agricultural non-point pollution was the most dominant factor influencing water quality in the Longxi River Basin (42.9% contribution), followed by the basin background (27.11%), a quantifiable signature reflecting the catchment’s inherent geochemical baseline. In contrast, in the dry season, rural domestic sewage was the most dominant influencing factor in the watershed (39.26% contribution), followed by physicochemical biological non-point sources and industrial sources (21.52% and 9.44%), respectively.

The results demonstrate that the integrated PCA-APCS-MLR approach is a powerful method for analyzing spatiotemporal water quality patterns and identifying pollution sources in complex mountainous watersheds, thereby providing a scientific basis for management. The quality of water environment in the mountainous watershed is greatly influenced by human activities and surface runoff, with the direct discharge of agricultural non-point sources and rural domestic sewage being the main causes for exceeding the water quality standard of the river. To achieve overall improvement in the water quality, multiple measures should be taken for comprehensive management of the water environment in the watershed.

The study helps managers and policy makers gain insight into the major sources of pollution during different water periods and identify priorities for improving water quality. Therefore, for the Longxi River Basin, the priority pollution control measures focus on the following: (1) Optimizing the fertilization strategy, improving the efficiency of nitrogen and phosphorus fertilizer use, reducing pesticide application, and improving pesticide utilization efficiency. (2) Strengthening the control of surface runoff and soil erosion, especially in the middle reaches of the Longxi River (in the territory of Dianjiang County). (3) Improving the capacity of domestic wastewater collection and treatment, and developing advanced technologies to reduce pollutants from decentralized septic tanks and domestic wastewater in rural areas.

Despite these insights, further research is required to evaluate changes in unidentified pollution sources and unmonitored parameters. Nonetheless, this study contributes insight from valuable baseline data syntheses to support water quality management in mountain watersheds.