Groundwater Pollution Source Identification Based on a Coupled PCA–PMF–Mantel Framework: A Case Study of the Qujiang River Basin

Abstract

This study develops an integrated framework for groundwater pollution source identification by coupling Principal Component Analysis (PCA), Positive Matrix Factorization (PMF), and the Mantel test, with the Qujiang River Basin as a case study. The framework enables a full-process assessment, encompassing qualitative identification, quantitative apportionment, and spatial validation of pollution drivers. Results indicate that groundwater chemistry is primarily influenced by three categories of sources: natural rock weathering, agricultural and domestic activities, and industrial wastewater discharge. Anthropogenic sources account for 73.7% of the total contribution, with mixed agricultural and domestic inputs dominating (38.5%), followed by industrial effluents (35.2%), while natural weathering contributes 26.3%. Mantel test analysis further shows that agricultural and domestic pollution correlates strongly with intensive farmland distribution in the midstream area, natural sources correspond to carbonate outcrops and higher elevations in the upstream, and industrial contributions cluster in downstream industrial zones. By integrating PCA, PMF, and Mantel analysis, this study offers a robust and transferable framework that improves both the accuracy and spatial interpretability of groundwater pollution source identification. The proposed approach provides scientific support for regionalized groundwater pollution prevention and control under complex hydrogeological settings.

1. Introduction

Groundwater pollution is a pressing global environmental challenge, threatening freshwater security for over two billion people who rely on groundwater as their primary drinking water source [1]. Anthropogenic activities—such as intensive agriculture, rapid urbanization, industrial discharges, and inadequate wastewater management—have led to widespread contamination of aquifers with nitrates, heavy metals, pathogens, and emerging contaminants [2]. In many regions, including the North China Plain, the Indo-Gangetic Basin, and parts of Europe and North America, elevated concentrations of NO₃⁻, NH₄⁺, As, and Mn have exceeded drinking water quality standards, posing serious risks to human health and ecosystem stability [3,4]. Moreover, the coexistence of multiple pollution sources within complex hydrogeological settings often complicates the accurate identification of contamination origins, thereby undermining the effectiveness of pollution control strategies. Despite recent advances in statistical and modeling tools, there remains a critical need for integrated, robust, and transferable frameworks capable of distinguishing natural from anthropogenic influences and spatially attributing pollution sources across diverse geographical contexts.

Within this global context, groundwater plays a vital role in sustaining water security, particularly in densely populated and rapidly developing regions such as China. It serves as a crucial resource for agricultural irrigation, industrial production, domestic supply, and strategic reserves [5,6]. The Qujiang River Basin, located in western Zhejiang Province, is an essential tributary and headwater region of the Qiantang River, functioning as a key water conservation zone. The basin supports the drinking water needs of millions of residents, underpins agricultural productivity, and maintains regional ecological integrity [7]. However, accelerated urbanization, industrial expansion, and population growth have intensified anthropogenic pressures on groundwater systems, resulting in progressive deterioration of groundwater quality characterized by complex pollutant mixtures [8,9,10]. Notably, elevated levels of NH₄⁺ and Mn have been detected in certain areas, exceeding regulatory thresholds for Class III groundwater quality in China [11]. These emerging contamination issues highlight the urgent need for precise identification of pollution sources and their contributing mechanisms to support science-based management and remediation efforts [12,13].

To characterize groundwater chemistry, classical hydrochemical diagrams—such as Piper and Gibbs plots—and ionic ratio analyses have long been employed and remain valuable tools for initial classification of water types and identification of dominant processes, including water–rock interactions and evaporation effects [1,14,15]. In recent years, multivariate statistical techniques such as Principal Component Analysis (PCA), Factor Analysis (FA), and Cluster Analysis (CA) have been increasingly adopted to extract dominant hydrochemical patterns from complex datasets. For example, Du et al. [16] identified rock weathering and evaporation as primary controls in the Hetao Irrigation District, with agricultural irrigation and drought as major drivers. Mohammed et al. [17] applied PCA and hierarchical cluster analysis to identify rock–water interaction, agricultural activities, and septic tank leakage as key factors influencing groundwater quality in a Nubian aquifer system in Sudan. Positive Matrix Factorization (PMF), which does not require prior information and permits rotational optimization, has increasingly been applied for quantitative apportionment of pollution sources [18,19,20]. Nevertheless, these methods often fail to sufficiently resolve the spatial distribution of pollution sources and their underlying driving mechanisms. The Mantel test, as a spatial statistical approach, can effectively evaluate correlations between hydrochemical components and environmental or anthropogenic factors, thereby providing a robust means to validate spatial drivers of pollution sources [21].

In China, numerous studies have focused on groundwater source apportionment in regions such as the North China Plain [22,23], the Huang–Huai–Hai Plain [24], and the Yangtze River Delta [25,26]. However, most of these studies emphasize single-method analyses, lacking an integrated framework that unifies qualitative identification, quantitative apportionment, and spatial verification. This limitation is particularly pronounced in hydrogeologically complex regions where multiple anthropogenic sources coexist, leading to insufficient accuracy and applicability in identifying pollution control factors. In the Qujiang River Basin, recent studies have characterized hydrochemical patterns and identified dominant natural processes [27], while others have examined surface-water/groundwater interactions using isotopic and hydrological tracers [7]. However, these works did not quantitatively apportion pollution sources or link them to spatial drivers such as land use and geology. Therefore, a systematic framework for source identification that integrates qualitative analysis, quantitative apportionment, and spatial validation remains lacking.

While PCA and PMF have been widely applied in groundwater studies to identify and quantify pollution sources, most such applications remain limited to isolated or sequential use, often lacking explicit spatial validation of the inferred sources. For example, in many studies, PMF-derived source contributions are interpreted qualitatively without testing their correlation with land use, geology, or hydrological gradients [28]. This gap increases uncertainty in source attribution, particularly in complex basins where natural and anthropogenic processes overlap. In the Qujiang River Basin, where groundwater systems are influenced by heterogeneous recharge patterns, variable land use, and legacy contamination, such uncertainty poses a major challenge for effective management. To address this, we integrate PMF with the Mantel test—a spatial correlation method not commonly used in this context—to statistically evaluate whether the identified sources are spatially coherent with environmental drivers. This linkage allows us to move beyond source identification toward source validation, reducing subjectivity in interpretation. Unlike previous PCA–PMF studies that focus on source types and contributions, our approach explicitly tests the spatial plausibility of those sources, offering a more defensible basis for groundwater protection in data-limited, hydrogeologically complex regions.

2. Materials and Methods

2.1. Study Area

The Qujiang River Basin is located in Quzhou City, western Zhejiang Province, and constitutes an important tributary and headwater region of the Qiantang River (Figure 1a). Within Zhejiang Province, the basin covers an area of 1.11 × 10⁴ km² and includes major tributaries such as Changshan Port, Majinxi, and Wuxi River. The study area spans Kaihua County, Changshan County, Qujiang District, Kecheng District, and Longyou County. The region has a subtropical monsoon climate, with a mean annual temperature of 17.3 °C, mean annual evaporation of 938.8 mm, and annual precipitation ranging from 1100 to 2000 mm. Rainfall is concentrated during the plum-rain season (April–July) and the typhoon season (July–September) [27,29].

Topographically, the basin is higher in the north, south, and west, and lower in the central and eastern areas. The upstream is dominated by low mountains and hills, while the midstream and downstream lie in the Jinqu Basin, characterized by hilly red-bed terrain and alluvial plains. (Figure 1b). The upstream region, located along the basin margins, is characterized by low mountains, hills, and uplands, whereas the midstream and downstream regions lie within the Jinqu Basin, dominated by red-bed hilly basins, residual hills, and alluvial plains. The overall geomorphic pattern is described as “three sides surrounded by mountains with a river valley in between, and three rivers interlaced within the basin.” Elevation generally ranges from 50 to 300 m.

The stratigraphy of the upstream basin is dominated by Lower Paleozoic Ordovician and Cambrian formations, whereas the midstream and downstream regions are underlain by Upper Cretaceous deposits. Quaternary strata consist mainly of Middle–Upper Pleistocene alluvial and fluvio-pluvial deposits as well as Holocene alluvium. Based on aquifer media and occurrence, groundwater types include pore phreatic water in loose rocks, pore–fissure water in red-bed clastic rocks, and fissure water in bedrock. In many cases, phreatic water in loose sediments overlies pore–fissure water in red-bed clastic rocks [30]. The hydrogeological map of the study area is shown in Figure 1c, with a typical hydrogeological cross-section illustrated in Figure 2.

2.2. Sampling and Analytical Procedures

2.2.1. Sample Collection

A total of 94 groundwater samples were collected across the Qujiang River Basin in September 2019, including 41 from the upstream, 40 from the midstream, and 13 from the downstream regions (Figure 1a). The sampling protocol followed the Technical Specifications for Groundwater Environmental Monitoring (HJ/T 164-2004) [31]. Prior to sampling, each well was pumped for approximately 15 min to stabilize in situ physicochemical parameters. A multiparameter water quality analyzer (HANNA HI9828, USA) was used to measure pH, temperature, electrical conductivity (EC), dissolved oxygen (DO), and redox potential (Eh) on-site. It should be noted that the dataset represents a single sampling event in September 2019, and therefore may not capture seasonal variations in groundwater chemistry.

In addition, a comprehensive suite of water quality parameters was analyzed to assess both natural geochemical processes and potential anthropogenic impacts. This included major cations (Ca²⁺, Mg²⁺, Na⁺, K⁺), major anions (Cl⁻, SO₄²⁻, HCO₃⁻), nutrients (NO₃⁻, NO₂⁻, NH₄⁺), indicators (F⁻, As), and trace metals (Mn, Ba, Sr, Cd, Pb, Cr). Samples were filtered through 0.45 μm cellulose acetate membranes. Cation and trace metal samples were acidified to pH < 2 with ultrapure HNO₃; anion and nutrient samples were not acidified. All samples were stored at 4 °C and analyzed promptly at Zhejiang Zhongyi Testing and Research Institute Co., Ltd.

The concentrations of F⁻, As, and selected trace metals (Ba, Pb, Cd, Cr) were either below the method detection limits or within the permissible levels specified in the Chinese Standards for Drinking Water Quality (GB 5749-2022) [32]. Given their compliance with regulatory standards and low environmental risk, these parameters are not elaborated upon in the main discussion. Detailed analytical methods and detection limits are provided in

Table 1. Analytical methods and detection limits for groundwater quality parameters.

Parameters	Detection Limit	Detection Method	Parameters	Detection Limit	Detection Method
K+	0.10	flame atomic absorption spectrophotometry	Cl-	0.10	silver nitrate volumetric method
Na+	0.01	flame atomic absorption spectrophotometry	SO42-	0.20	barium sulfate turbidimetric method
Ca2+	0.004	disodium edetate titration	NO3-	0.02	ultraviolet spectrophotometry
Mg2+	0.01	disodium edetate titration	F-	0.01	ion-selective electrode method
HCO3-	5.00	acid titration	NH4+	0.04	UV-Vis spectrophotometry
TH	10.00	disodium edetate titration	Mn	0.005	flame atomic absorption spectrophotometry
TDS	4.00	dry residue method	As	0.01	HG-AFS
Ba	0.005	ICP-MS1	Cd	0.001	ICP-MS1
Pb	0.005	ICP-MS1	Cr	0.01	ICP-MS1

Note: The unit of detection limit is mg·L^−1.; All trace metals (Ba, Pb, Cd, Cr) were analyzed by inductively coupled plasma mass spectrometry (ICP-MS) at Zhejiang Zhongyi Testing and Research Institute Co., Ltd.

.

To ensure the reliability of hydrochemical data, quality assurance and quality control (QA/QC) procedures were implemented during sampling and analysis. Field duplicate samples were collected at approximately 10% of the total sites (i.e., 9 out of 94 samples) to assess sampling and analytical precision, yielding relative standard deviations (RSD) of less than 5% for major ions. Procedural blanks were included to monitor potential contamination during transport and laboratory handling, and no significant contamination was detected.

The accuracy of ionic charge balance was evaluated using the percentage charge balance error (EB%), calculated as:

$$\mathrm{E}\mathrm{B}\%={\displaystyle \frac{\sum \mathrm{cations} (\mathrm{m}\mathrm{e}\mathrm{q}/\mathrm{L})-\sum \mathrm{anions} (\mathrm{m}\mathrm{e}\mathrm{q}/\mathrm{L})}{(\sum \mathrm{cations}+\sum \mathrm{anions})/2}}\times 100$$

(1)

All 94 samples exhibited a charge balance error within ±5%, with 85 samples falling within ±3%, well within the acceptable range for groundwater geochemical studies. These results confirm high analytical accuracy and support the reliability of the dataset used in subsequent multivariate analyses.

2.2.2. Data Processing and Statistical Analysis

Geographic coordinates of the sampling sites were processed in ArcGIS 10.8, where ion concentration maps were generated using the inverse distance weighting (IDW) method with a power parameter of 2 to capture spatial variability. Data organization and descriptive statistics were conducted in Microsoft Excel 2019. To explore the dominant hydrochemical factors, Principal Component Analysis (PCA) was applied using Origin 2021. Source apportionment was further carried out with the EPA PMF 5.0 model, in which uncertainties of chemical constituents were estimated from detection limits and standard deviations following established protocols [33,34]. The uncertainty for each measurement was calculated as follows:

① For concentrations above the detection limit (DL):

$$Unc=\sqrt{(error fraction\times x{)}^2+({0.5\ast DL)}^2}$$

(2)

where x is the measured concentration and error fraction was set to 0.05–0.10 based on laboratory precision.

② For concentrations at or below the detection limit (DL):

$$Unc={\displaystyle \frac{5}{6}}\times DL$$

(3)

and the concentration value was replaced with DL/2, a standard approach to minimize bias while retaining statistical robustness.

The model was run 20 times for each factor solution (from 2 to 5 factors) using the displacement method with a maximum of 100 iterations per run. The optimal number of factors (n = 3) was determined based on the minimum Qrobust/Qtrue ratio, visual inspection of residual patterns (all within ±3), and interpretability of factor profiles. All variables were weighted by their uncertainties, and no re-scaling was applied. The final solution showed a maximum change in Q-value of less than 1% between consecutive iterations, indicating convergence.

To examine the spatial associations between source contributions and environmental variables, the Mantel test was performed in R 4.4.1 with the LinkET package, using Pearson correlation coefficients and 999 permutations [35]. The environmental and land use variables considered included: elevation, slope, percentage of agricultural land, percentage of urban area, and distance to industrial zones. These variables were selected based on their potential influence on groundwater hydrology and contamination sources. Elevation and slope are key topographic controls on groundwater recharge and flow direction; the percentage of agricultural land within a 1 km buffer around each sampling site was included to assess the impact of agricultural activities, which are commonly associated with nitrate and pesticide inputs; similarly, the percentage of urban area was used to represent urbanization intensity and its associated domestic pollution sources (e.g., sewage leakage); distance to industrial zones was incorporated to evaluate the potential influence of industrial discharges on groundwater quality [36]. All variables were derived from 30 m resolution digital elevation model (DEM) and land use data, and were standardized prior to analysis.

In addition, Piper diagrams, Gibbs diagrams, and ion ratio plots were prepared in Origin 2021 to assist in hydrochemical classification and interpretation of controlling mechanisms.

3. Results and Discussion

3.1. Hydrochemical Characteristics and Spatial Distribution

The statistical summary of groundwater hydrochemistry in the Qujiang River Basin is provided in

Table 2. Statistical results of hydrochemical indices in the study area.

Parameters		pH	TDS	Ca2+	Mg2+	K+	Na+	Cl−	SO42−	HCO3−	Mn	NH4+
Upstream	Min (mg·L−1)	5.27	30	4.08	0.30	0.5	1.28	0.3	0.8	32	0.02	0.02
	Max (mg·L−1)	8.98	252	63.00	12.80	24.1	93.00	33.0	105.0	300	0.92	1.59
	Mean (mg·L−1)	6.85	125	26.46	4.43	4.2	9.20	4.4	15.7	108	0.32	0.13
	SD (mg·L−1)	0.65	53	15.06	3.00	5.1	14.29	5.8	20.9	50	0.52	0.34
	CV (%)	9.4%	42.8%	56.9%	67.7%	121.8%	155.3%	132%	132.9%	46.2%	162.4%	270.8%
Midstream	Min (mg·L−1)	5.05	43	1.58	0.13	0.7	2.16	1.5	2.0	20	0.01	0.02
	Max (mg·L−1)	6.95	426	70.10	15.30	35.2	49.80	67.1	100.0	243	3.14	2.17
	Mean (mg·L−1)	6.27	162	33.29	4.12	7.2	11.88	14.2	23.6	86	0.14	0.22
	SD (mg·L−1)	0.53	92	15.50	3.58	7.5	9.52	15.2	21.2	47	0.54	0.49
	CV (%)	8.5%	57%	46.6%	87%	103.5%	80.1%	107%	89.6%	54.3%	372.3%	223.3%
Downstream	Min (mg·L−1)	6.17	58	8.25	1.50	1.0	1.860	2.2	1.9	52	1.50	0.02
	Max (mg·L−1)	7.54	212	49.00	8.95	22.8	20.800	27.6	48.6	158	8.95	0.30
	Mean (mg·L−1)	6.79	1560	29.99	4.57	10.2	12.062	12.6	27.5	93	4.57	0.08
	SD (mg·L−1)	0.40	47	13.23	2.09	7.1	5.346	6.9	14.1	30	2.09	0.10
	CV (%)	5.9%	29.7%	44.1%	45.7%	69.7%	44.3%	54.8%	51.3%	31.8%	45.7%	112.4%

. Groundwater pH values varied between 5.05 and 8.98, with some samples exceeding the threshold for Class III groundwater quality. Spatially, higher pH values (weakly alkaline) were generally recorded in the upstream area, whereas lower values (weakly acidic) occurred in the midstream. Total dissolved solids (TDS) remained below 1000 mg·L⁻¹ across all samples, consistent with freshwater conditions. However, the midstream section exhibited both the highest mean TDS concentration and the largest coefficient of variation (57%), pointing to stronger external perturbations to groundwater chemistry in this part of the basin.

Among the major cations, Ca²⁺, Na⁺, and K⁺ were predominant (Figure 3). Ca²⁺ concentrations were highest in the midstream (mean: 33.3 mg·L⁻¹) and lowest in the upstream (26.5 mg·L⁻¹). Na⁺ concentrations peaked in the downstream (12.1 mg·L⁻¹), while K⁺ exhibited the greatest variability in the midstream (CV = 128%), likely reflecting fertilizer application and localized industrial inputs. In contrast, Mg²⁺ concentrations were relatively uniform across the basin (3~4 mg·L⁻¹). All cation concentrations complied with the Class III groundwater standard.

For the major anions, HCO₃⁻ was dominant, with the highest mean concentration in the upstream and the lowest in the midstream (86.3 mg·L⁻¹). Cl⁻ and SO₄²⁻ were markedly elevated in the midstream relative to upstream and downstream areas, consistent with agricultural return flow and domestic sewage contributions.

Regarding toxicological indicators, NH₄⁺ reached the highest mean concentration in the midstream (0.22 mg·L⁻¹), where localized exceedances appearing as point-source anomalies (Figure 4b). Box plots (Figure 3) further show that Mn concentrations downstream had a median of 4.57 mg·L⁻¹, substantially higher than those in upstream and midstream samples. Elevated Mn levels in the downstream red-bed aquifers (Figure 4c) are primarily attributed to natural geological enrichment, but may also be influenced by local industrial activities, which are commonly found in the downstream area [37].

Overall, the spatial variability of hydrochemical parameters reflects the combined influence of natural geologic settings and anthropogenic activities, providing a critical basis for subsequent pollution source apportionment.

3.2. Dominant Hydrogeochemical Processes

3.2.1. Gibbs Diagram

According to Gibbs (1970), three main mechanisms govern global water chemistry: precipitation, evaporation, and rock weathering [38,39]. As shown in Figure 5, most samples fall within the “rock weathering dominance” field, indicating that groundwater chemistry is primarily controlled by water–rock interactions. Moderate TDS levels (30–426 mg·L⁻¹) and relatively low Na⁺/(Na⁺ + Ca²⁺) and Cl⁻/(Cl⁻ + HCO₃⁻) ratios support this conclusion [40,41].

Given the subtropical monsoon climate of the study area and the presence of Cretaceous red beds, alluvial deposits, and igneous interlayers enriched in easily weathered minerals (e.g., feldspar, pyroxene, dolomite), dissolution is strongly promoted under weakly acidic conditions. Moreover, the hydraulic connectivity between pore–fissure water in red-bed clastic rocks and phreatic water in loose sediments enhances reaction surfaces and rates. Together, these factors create favorable geological conditions for rock weathering.

3.2.2. Correlation Analysis

Correlation analysis provides insights into the co-variation patterns among hydrochemical parameters [42]. The Pearson correlation matrix (Figure 6) shows significant positive correlations among Ca²⁺, Mg²⁺, SO₄²⁻, and HCO₃⁻. Specifically, the correlations between Ca²⁺–HCO₃⁻ and Mg²⁺–HCO₃⁻ are consistent with carbonate mineral dissolution (e.g., calcite, dolomite), whereas the strong positive correlations of Ca²⁺–SO₄²⁻ and Mg²⁺–SO₄²⁻ suggest a potential contribution from sulfate minerals (e.g., gypsum, anhydrite) as important sources of Ca²⁺ and Mg²⁺ [29]. These dual dissolution processes align with the coexistence of carbonates and sulfates in Cretaceous red beds and Quaternary alluvium [7].

In addition, Mn showed a strong positive correlation with TDS, which may reflect the coupled release of Mn²⁺ and other ions during the dissolution of Mn-bearing minerals (e.g., rhodochrosite, manganese oxides), thereby contributing to elevated dissolved solids [37]. This process is particularly pronounced in downstream red-bed aquifers under weakly acidic or reducing conditions. By contrast, NH₄⁺ displayed generally weak correlations with other ions, reflecting its relatively independent sources, mainly associated with anthropogenic activities such as agricultural fertilization and leakage of domestic sewage [7].

3.2.3. Ion Ratio Analysis

Ion ratio analysis is a key tool for evaluating the origins and evolution of groundwater chemistry. In this study, relationships among Cl⁻, SO₄²⁻, HCO₃⁻, Na⁺, Ca²⁺, and Mg²⁺ were examined to elucidate controlling processes in the Qujiang River Basin.

The ratio (Na⁺ + K⁺)/Cl⁻ helps determine sources of alkali metals [43,44]. As shown in Figure 7a, most groundwater samples plot above the 1:1 line, indicating that Na⁺ and K⁺ are not solely derived from halite dissolution, but also from silicate weathering, cation exchange, and anthropogenic activities-- particularly agricultural fertilization (e.g., potassium-rich fertilizers) and domestic sewage discharge (a major source of Na⁺ and K⁺ in urban areas) [45].

The ratio (Ca²⁺ + Mg²⁺)/(HCO₃⁻ + SO₄²⁻) can distinguish between carbonate and evaporite dissolution [38,39]. Most samples fall near the 1:1 line (Figure 7b), suggesting good charge balance and confirming that carbonate and sulfate dissolution are the main sources of Ca²⁺ and Mg²⁺.

To further differentiate contributions of carbonate versus evaporite dissolution, the ratio (SO₄²⁻ + Cl⁻)/HCO₃⁻ was analyzed [46]. Samples plotting below the 1:1 line (Figure 7c) indicate that carbonate weathering is the dominant control on groundwater hydrochemistry.

Anthropogenic activities also exert significant influence, especially on Na⁺, Cl⁻, NO₃⁻, and SO₄²⁻ concentrations [47,48,49]. For example, industrial effluents and sewage increase Na⁺ and Cl⁻ levels, while application of nitrogen fertilizers and animal manure elevates NO₃⁻, posing risks of eutrophication and nitrate contamination in groundwater. Combustion emissions, mining, and industrial wastewater contribute to SO₄²⁻ enrichment. The NO₃⁻/Ca²⁺ vs. SO₄²⁻/Ca²⁺ relationship helps distinguish anthropogenic impacts [50,51]. Samples plotting below the 1:1 line generally reflect agricultural and domestic sources, whereas those above the line are indicative of industrial and mining activities. Given the minimal human impact in the upstream, the spatial trend analysis for anthropogenic sources (Figure 7d) focuses on the midstream and downstream regions. As shown in Figure 7d, several midstream and downstream samples (e.g., LYS155, QZS107, QZS116, QZS111) fall above the line, confirming industrial influence as a key driver of hydrochemical anomalies.

End-member analysis (Figure 8) further demonstrates the dominant role of carbonate dissolution, with most samples clustering in the carbonate domain. Some displacement toward silicate fields indicates contributions from silicate weathering, particularly feldspar-rich rocks. A subset of samples also shows influence from evaporite dissolution (gypsum). Overall, carbonate dissolution is confirmed as the primary control on groundwater chemistry, consistent with Gibbs and ion ratio results.

3.3. Hydrochemical Types

Piper diagrams are widely used to classify hydrochemical types [52]. As shown in Figure 9, most groundwater samples plot in the Ca-type cation field (Region B), with fewer in the Na-type (Region C). For anions, samples mainly fall within the HCO₃⁻-type (Region G), with some in the mixed field (Region D). These patterns indicate that Ca²⁺ and HCO₃⁻ dominate the ionic composition, reflecting the fundamental control of carbonate weathering.

In the central diamond field, samples mainly cluster in Regions 1, 5, and 6, indicating that the principal hydrochemical types are HCO₃–Ca·Mg, Cl–Ca·Mg, and HCO₃–Na–Ca. Spatially, upstream areas are dominated by Ca–HCO₃ water, reflecting early-stage carbonate dissolution. In the midstream, Ca–SO₄–HCO₃ types are more prevalent, associated with intensified sulfate dissolution and irrigation return flow. In the downstream, Ca–HCO₃ water remains common, but local occurrences of Ca–SO₄ water highlight enhanced sulfate dissolution under weakly acidic or reducing conditions.

3.4. Pollution Source Identification and Quantitative Analysis

3.4.1. Principal Component Analysis (PCA)

The Kaiser–Meyer–Olkin (KMO) value was 0.608 (>0.5), and Bartlett’s test showed p < 0.001, confirming suitability of PCA for groundwater source identification [53]. Three principal components with eigenvalues > 1 were extracted, explaining 64.32% of the total variance (

Table 3. Principal component loadings after varimax rotation.

Parameters	Rotated Factor Loadings
	PC1	PC2	PC3
K+	0.233	0.135	0.524
Na+	0.187	0.636	0.209
Ca2+	0.450	−0.101	−0.451
Mg2+	0.465	−0.076	−0.079
Cl−	0.398	−0.116	0.289
SO42−	0.472	−0.308	0.025
HCO3−	0.282	0.594	−0.355
Mn2+	0.169	−0.215	0.383
NH4+	0.039	−0.234	−0.339
eigenvalue	2.57	1.31	1.21
explained variance /%	30.37	20.59	13.36
cumulative explained variance /%	30.37	50.69	64.32

).

PC1 (30.37%) had high loadings on Ca²⁺ (0.45), Mg²⁺ (0.465), and SO₄²⁻ (0.472). These ions reflect carbonate and sulfate dissolution, indicating natural rock weathering as a dominant process [54]. Although Cl⁻ also shows moderate loading (0.398), its spatial distribution (Figure 4a) does not fully align with PC1’s upstream dominance, suggesting mixed sources—possibly minor anthropogenic inputs superimposed on a natural background.

PC2 (20.59%) had high loadings on Na⁺ (0.636) and HCO₃⁻ (0.594). Na⁺ enrichment likely results from irrigation return flows and domestic detergents, while HCO₃⁻ partly reflects carbonate buffering under anthropogenic disturbance (e.g., soil acidification from fertilizers) [54]. Thus, PC2 represents mixed agricultural and domestic anthropogenic sources, particularly in the midstream.

PC3 (13.36%) had high loadings on K⁺ (0.524) and Mn²⁺ (0.383). K⁺ indicates fertilizer inputs, while Mn²⁺ reflects both natural red-bed enrichment and industrial effluents [55]. Hence, PC3 suggests combined agricultural–industrial anthropogenic pollution, dominant in downstream industrial zones and midstream agricultural–industrial transition areas, see Figure 10.

In summary, Principal Component Analysis (PCA) preliminarily identified three driving factors of groundwater pollution in the Qujiang River Basin: natural rock weathering (PC1), agricultural and domestic pollution (PC2), and combined agricultural–industrial pollution (PC3). This qualitative identification provided the basis for subsequent quantitative source apportionment using the Positive Matrix Factorization (PMF) model. Although the rotated factor loadings did not exceed 0.7, this reflects the multi-source and multi-process nature of groundwater chemistry in the Qujiang Basin, and the cumulative variance explained (64.32%) together with KMO and Bartlett’s tests confirm the robustness of the PCA results.

3.4.2. PMF Source Apportionment

In this study, ten representative parameters were selected for PMF modeling to quantify the relative impacts of natural and anthropogenic factors on groundwater chemistry in the Qujiang River Basin [19]. The mass concentrations of these indicators and their uncertainties were input into EPA PMF 5.0, with 20 iterations and factor numbers tested from two to five. Model performance indicated that three factors yielded the most reliable solution, with the QRobust/QTrue ratio reaching the lowest convergence value and the maximum Q-value decrease being less than 1% of QRobust (3.251), confirming the robustness of the fit. Factor contribution plots (Figure 11) and species distribution patterns (Figure 12) were highly consistent with PCA results, validating the integration of the two methods.

F1 was dominated by K⁺ (87%) and Na⁺ (78.9%), accompanied by contributions of NH₄⁺ (30.5%). These signatures point to inputs from potassium fertilizers, livestock and poultry manure, irrigation return flows containing sodium salts, and sodium-based detergents in domestic sewage. The enrichment of K⁺ is widely associated with agricultural activities, particularly fertilizer application and animal waste [55], while elevated Na⁺ in groundwater systems often reflects domestic inputs such as detergents and irrigation return flows [54]. This geochemical fingerprint is characteristic of mixed agricultural and domestic sources. Spatially, F1 accounted for the largest share in the midstream region (45.2%), coinciding with intensive agricultural activities and rural settlements. This pattern strongly aligns with PC2 identified in PCA, confirming the dominance of mixed agricultural and domestic sources.

F2 was characterized by high contributions from Ca²⁺ (62.1%), Mg²⁺ (57.0%), and HCO₃⁻ (69.9%), which are typical products of carbonate dissolution (CaCO₃ + H₂O + CO₂ ⇌ Ca²⁺ + 2HCO₃⁻). This ion assemblage is a well-documented signature of natural carbonate rock weathering, commonly observed in geological terrains rich in limestone and dolomite [54]. The high HCO₃⁻ further supports CO₂-driven dissolution processes under near-natural conditions. The highest spatial contribution of F2 occurred upstream (41.7%), consistent with extensive Ordovician and Cambrian carbonate outcrops. This source corresponds well with PC1 in PCA, representing natural geologic processes dominated by carbonate weathering.

F3 was mainly defined by Cl⁻ (79.7%), SO₄²⁻ (77.3%), and Mn (26.3%), reflecting emissions from industrial sectors such as chlor-alkali production, coal-fired desulfurization, metal processing, and electroplating. The co-enrichment of Cl⁻ and SO₄²⁻ is a recognized indicator of industrial discharges, particularly from chemical and metallurgical plants [37]. Mn, while sometimes of natural origin in red-bed regions, shows extreme enrichment here, strongly suggesting industrial release from electroplating or alloy manufacturing [56]. Downstream areas, where F3 contributed up to 58.6%, exhibited elevated Mn concentrations alongside localized Cl⁻ and SO₄²⁻ anomalies. These findings are consistent with PC3 in PCA and confirm industrial effluents as the dominant downstream pollution source.

While the overall interpretations of PMF and PCA are broadly consistent—both identifying agricultural/domestic, natural weathering, and industrial sources as dominant—minor discrepancies exist in the relative loading patterns. For instance, PCA slightly emphasizes NO₃⁻–Cl⁻–SO₄²⁻ correlations in PC3, whereas PMF isolates Cl⁻–SO₄²⁻–Mn as a distinct industrial signal with quantitative contribution. These differences arise from methodological distinctions: PCA is an unsupervised variance-based technique that identifies covariance structures without assuming source specificity, while PMF is a receptor model that incorporates uncertainty and apportions sources based on mass balance. Thus, PMF provides more interpretable source quantification, whereas PCA offers exploratory insight into variable relationships. The convergence of both methods on the same three-source framework strengthens confidence in the final interpretation.

Overall, PMF apportionment indicated that anthropogenic activities (F1 + F3) contributed 73.7% of total groundwater pollution, with agricultural/domestic inputs (38.5%) slightly exceeding industrial sources (35.2%). Natural sources (F2) accounted for 26.3%. These results corroborate PCA interpretations while providing quantitative resolution, thereby establishing a sound basis for subsequent spatial validation using the Mantel test.

3.4.3. Mantel Test Validation

To validate the spatial relevance of the PMF results, Mantel tests were conducted between factor contributions and key environmental/land use variables (Figure 13). The analysis confirmed significant correlations (p < 0.01) across all three sources, highlighting the consistency between geochemical source signals and their underlying environmental drivers, underscoring the robustness of the PCA–PMF–Mantel framework.

Agricultural and domestic sources (F1) showed strong positive correlations with farmland proportion (r = 0.462) and rural settlement density. This spatial pattern reflects the intensive agricultural practices and widespread use of fertilizers and animal manure in the midstream region, where shallow groundwater is particularly vulnerable to leaching from croplands and septic systems. The correlation reinforces the role of rural land use intensity in shaping non-point source pollution in areas with permeable soils and active recharge.

Natural sources (F2) were significantly correlated with elevation and carbonate rock exposure (r = 0.417), a relationship rooted in the upstream geology, where Ordovician and Cambrian limestone formations dominate. Groundwater in this zone evolves through prolonged interaction with carbonate minerals under forested, high-elevation conditions, leading to characteristic HCO₃⁻–Ca²⁺–Mg²⁺ enrichment. This match confirms that natural weathering, not anthropogenic influence, controls baseline chemistry in the headwaters.

Industrial sources (F3) exhibited the strongest correlation with industrial land use and urban density (r = 0.503), particularly in the downstream lowland, where decades of unregulated discharge from chemical and electroplating plants have left a persistent contamination legacy. The co-location of elevated Mn, Cl⁻, and SO₄²⁻ with industrial clusters—combined with thin vadose zones and dense drainage networks—facilitates rapid contaminant infiltration and downstream accumulation. This linkage highlights the long-term impact of industrial development on groundwater quality in urbanized zones.

Compared with multivariate approaches such as Redundancy Analysis (RDA) or Canonical Correspondence Analysis (CCA), the Mantel test offers distinct advantages in this context. It evaluates correlations between distance matrices without requiring assumptions about linear or unimodal responses, making it particularly suitable for heterogeneous basins. Its permutation-based framework also enhances statistical reliability with moderate sample sizes, as in this study.

In summary, the Mantel test effectively validated the spatial consistency of PMF-derived factors, linking source contributions to environmental drivers. Rather than merely reporting statistical associations, the results illustrate how regional hydrogeological settings and human land use patterns collectively shape groundwater contamination. Together with PCA and PMF, it completes a full-process framework for groundwater pollution source identification—qualitative recognition, quantitative apportionment, and spatial verification.

3.5. Comparison with Other Regional Studies

The above analysis clarifies the major sources and their relative contributions to groundwater pollution in the Qujiang River Basin. To place these findings in a broader geographical and environmental context, it is essential to compare them with results from other representative regions undergoing similar anthropogenic pressures.

To contextualize our source apportionment results, we compared them with findings from other major basins in China (

Table 4. Comparison of groundwater pollution source apportionment results from representative studies in different regions of China.

Study Area	Method	Major Sources Identified	Contribution (%)	Key Tracers
Qujiang River Basin (this study)	PCA–PMF–Mantel	Natural weathering	26.3	Ca2+, HCO3−, Mg2+
		Agricultural & domestic	38.5	NO3−, NH4+, Cl−
		Industrial discharge	35.2	SO42−, Mn
Fenhe River Basin [57] (FRB)	PMF–GIS	Rock weathering & evaporation	~40–50	F−, As, Cr, Ca2+
		Agricultural inputs	~30	NO3−, Cl−
		Anthropogenic discharge	~20–30	Na+, SO42−
Guanzhong Basin [60]	Dual isotopes (δ15N–NO3−, δ18O–NO3−) + Bayesian model	Agricultural fertilizers	45–60%	NO3−, δ15N
		Manure & sewage	25–35%	NH4+, Cl−, δ15N > +15‰
		Atmospheric deposition	10–15%	NO3−, low δ15N
Hetao Plain [59]	Hydrochemical + statistical analysis	Irrigation return flow	~40–50	TDS, Cl−, NO3−
		Evaporative concentration	~30–40	Cl−, Na+, TDS
		Natural dissolution	~10–20	Ca2+, HCO3−
Yangtze River Basin [58]	SIAR + δ15N/δ18O–NO3−	Manure and sewage	43–60%	NH4+, Cl−, δ15N > +10‰
		Chemical fertilizers	23–44%	NO3−, low δ15N
		Soil organic N	21–39%	δ15N~+5‰

). The contribution of agricultural and domestic sources (38.5%) in the Qujiang River Basin is comparable to that in the Fenhe River Basin (~30%) [57] and the Yangtze River Basin (43–60% from manure/sewage) [58], reflecting the widespread impact of rural activities. However, the relatively high contribution of industrial discharge (35.2%) exceeds that in most agricultural-dominated regions, highlighting the growing pressure from local industries. In contrast, natural processes and evaporative concentration play a more dominant role in arid regions such as the Hetao Plain (~30–40%) [59]. This comparison underscores the need for region-specific groundwater protection strategies that account for local hydrogeological and anthropogenic conditions.

4. Conclusions and Suggestions

This study developed and applied a coupled PCA–PMF–Mantel framework to identify groundwater pollution sources in the Qujiang River Basin. The main conclusions are as follows:

• Groundwater hydrochemistry in the Qujiang River Basin is dominated by Ca²⁺ and HCO₃⁻, with HCO₃–Ca·Mg as the principal water type. Carbonate dissolution is the primary control, while spatial variations reflect combined influences of natural water–rock interactions and anthropogenic activities. Specifically, upstream groundwater is mainly Ca–HCO₃ type, midstream samples show higher proportions of Ca–SO₄–HCO₃ type due to sulfate dissolution and irrigation return flow, and downstream groundwater locally exhibits Ca–SO₄ type under acidic or reducing conditions.
• PCA and PMF consistently identified three major sources: natural rock weathering (26.3%), agricultural and domestic anthropogenic activities (38.5%), and industrial anthropogenic activities (35.2%). Agricultural and domestic contributions were most significant in the midstream agricultural belt, industrial contributions were concentrated in downstream industrial clusters, while natural sources dominated the upstream with strong geological background control.
• Mantel test results confirmed significant spatial correlations between pollution sources and environmental drivers: agricultural and domestic sources were positively correlated with farmland proportion and rural settlement density; natural sources were associated with carbonate rock outcrops and topographic elevation; industrial sources were strongly linked to industrial land use and urban density. These findings validate the robustness and spatial rationality of the proposed framework.

Despite these insights, this study has several limitations. First, sampling was restricted to a single season, limiting representation of seasonal hydrochemical variations. Second, certain potential pollutants (e.g., organic contaminants, microplastics) were not analyzed, possibly underestimating composite pollution effects. Third, while the Mantel test effectively verified spatial associations, it remains insufficient for characterizing complex groundwater flow paths and aquifer dynamics in the basin.

Future work should incorporate multi-season, long-term monitoring with a wider range of chemical indicators, coupled with isotope tracing and groundwater flow modeling, to further enhance accuracy and resolution of groundwater source apportionment under complex hydrogeological conditions.