Source Apportionment and Analysis of Potentially Toxic Element Sources in Agricultural Soils Based on the Positive Matrix Factorization and Geo-Detector Models

Abstract

The potentially toxic element pollution of agricultural soils has become a significant environmental threat to food safety and human health. Accurately identifying sources of potentially toxic element pollution is key to developing effective pollution prevention and control measures. In this study, regional potentially toxic element pollution of the soils in the Nanliujiang River Basin was analyzed using the positive matrix factorization (PMF) model and the geo-detector model. First, topsoil samples from the study area were collected to analyze eight potentially toxic elements in the soil, including As, Cd, Cr, Cu, Hg, Ni, Pb, and Zn. The PMF model was used to conduct source apportionment of the potentially toxic element data and identify the primary pollution sources and their contribution rates. Then, the geo-detector model was used to analyze the key factors affecting the spatial distribution of the potentially toxic elements and the influence of natural and human factors on the distribution of the potentially toxic elements. There are four potentially toxic element pollution sources of the agricultural soil in the study area: geological background, agricultural activities, industrial discharge, and river irrigation. The geological background contributed the most. The main factors affecting the spatial distribution of potentially toxic elements included agricultural activities, industrial discharge, and river irrigation. This integrated method can analyze the formation of potentially toxic element pollution in depth from the perspectives of source apportionment and spatial differentiation and provide a scientific basis and decision support for preventing and controlling potentially toxic element pollution in agricultural soils. This study provides a new method and scientific basis for identifying and preventing potentially toxic element pollution sources in agricultural soil and can guide the formulation of targeted soil pollution control measures.

1. Introduction

Potentially toxic elements (PTEs) in agricultural soil, particularly cadmium (Cd) and lead (Pb), pose significant threats to human health and the ecological environment due to their resistance to degradation and propensity for bioaccumulation in living organisms [1,2]. Not only do these potentially toxic elements degrade soil quality and reduce crop yield and quality [3], but they also enter the human body through the food chain, causing various diseases, including nervous system damage, kidney diseases, and cardiovascular diseases [1,4,5]. The pollution of soil by potentially toxic elements can lead to secondary contamination of water and the atmosphere, thereby exacerbating the ecological crisis [6,7]. Therefore, in-depth research on the source distribution of potentially toxic elements in soil is needed to ensure food safety and maintain ecological balance.

It is crucial to accurately identify sources of potentially toxic element pollution to develop effective prevention and control measures. Targeted measures can be taken to reduce or eliminate pollution sources once the source of potentially toxic elements has been traced. This way, the spread of potentially toxic element pollution can be effectively curbed. Potentially toxic element traceability in soil is challenging, and a series of different source apportionment methods have been developed. The main methods include the isotope tracer and multivariate statistical methods [8,9,10]. The isotope tracer method can be used to infer the origin and migration path of potentially toxic elements by measuring their composition and isotope proportions [9]. The multivariate statistical analysis method analyzes the correlation between the potentially toxic elements in the soil and their relationship with various environmental factors to identify possible pollution sources [8,10]. Although soil potentially toxic element source analysis methods can reveal the source and distribution of potentially toxic elements, these methods still have some limitations [8,9,10]. Although isotope tracers are very accurate, the analyses are complex and expensive, complicating their general application in large-scale studies. Sample quantity and weak interpretability result in inaccurate multivariate statistical analysis results.

The positive matrix factorization (PMF) and geo-detector models hold significant advantages for potentially toxic element tracing in soil. The PMF model is an effective factor analysis model that uses the content data of potentially toxic elements in the soil to determine the main pollution sources and their contribution rates through weight calculation and the least squares method. This method can manage complex multi-source pollution problems and is highly flexible and adaptable [11,12]. The geo-detector model uses geospatial information to detect and analyze the key factors affecting the spatial distribution of potentially toxic elements and to explore the influence of natural and human factors on the distribution of potentially toxic elements. This method intuitively reveals the spatial differentiation characteristics of potentially toxic element distribution and its driving mechanisms by calculating the contribution of the independent variables to the dependent variables [13,14,15].

The Nanliujiang River, located in the southeastern part of the Guangxi Zhuang Autonomous Region, is the largest river in Guangxi that empties into the sea. The region’s abundant water resources, the flat river basin, and the alluvial plain provide a convenient basis for regional industrial and agricultural development. However, potentially toxic element pollution associated with industrial development occurs in some areas of the Nanliujiang River Basin [16].

Due to the severity of potentially toxic element pollution in the Nanliujiang River Basin, this study aimed to determine the sources of pollution from potentially toxic elements in the soil by using the positive matrix factorization (PMF) and the geo-detector models. This study attempts to conduct an in-depth analysis of potentially toxic element pollution from source apportionment and spatial differentiation perspectives. This will provide a scientific basis and decision support for preventing and controlling potentially toxic element pollution in agricultural soils. This study will also provide a new method and scientific basis for identifying soil potentially toxic element pollution sources and preventing soil contamination and can guide authorities in the formulation of targeted soil pollution control measures.

2. Materials and Methods

2.1. Overview of the Study Area

The Nanliujiang River originates from the southern foot of the Darong Mountain at Beiliu City, Guangxi, and flows from north to south. It has a total length of 287 km and a basin area of 9320 km² (Figure 1). The whole basin is located south of the Tropic of Cancer and is part of the typical subtropical monsoon climate and tropical marine climate, with an average temperature of 21.5–22.4 °C and an average annual rainfall of 1400–1760 mm [17].

The geological strata of the basin are exposed. Apart from the Quaternary system developed along the river system, the Permian and Triassic granites are widely exposed, covering about a third of the region’s total area. Second, the Neogene, Cretaceous, and Silurian clastic rocks are widely distributed from the east side of Yulin City to the middle reaches of Shahe Town and then to the lower reaches of Wujia Town. The area of exposed carbonate rock is small and mainly located in the northwestern part of Xingye County and the southern part of Beiliu City [18,19].

Figure 2 shows the land use type in the Nanliujiang River Basin. The total forest area is 4748 km², accounting for 50.79%; the cultivated land area is 3005 km² (32.15%); the area of the orchards and other gardens is 969 km² (10.36%); the area of the rivers and lakes is 316 km² (3.38%); and the area of the cities, towns, and villages is 297 km² (3.18%). A small proportion of coastal beaches and other utilization types cover the rest of the land uses in the basin. Chemical, metallurgical, and other industrial factories that cause metal pollution and put pressure on the agricultural soil environment occur throughout the region.

2.2. Sampling and Analyses

The Guangxi Geological Survey Institute has carried out multi-objective geochemical surveys over 22,059 km² of the Beibu Gulf region of Guangxi and collected 7327 topsoil samples [18,19]. This paper’s study area covers 2257 topsoil samples (0–20 cm) in the southern part of the Beibu Gulf Region survey area.

A grid system was used to determine the sampling localities and ensure the spatial uniformity of the sampling points. The usual sampling depth of topsoil samples is 0–20 cm, with a basic sampling density of 1 point/km², and that of deep soil samples is usually 150–180 cm, with a basic sampling density of 1 point per 4 km². The surface plant residues and other debris were first removed before collecting the topsoil (0–20 cm). To ensure uniformity, the soil was sampled vertically from the surface to a depth of 20 cm. To avoid surface pollution when collecting deep soils, a 10–50 cm long soil column was used to collect soil samples below the specified starting depth continuously. Each soil sample’s original weight had to be at least 1 kg.

The specific process for sampling and analysis is as follows:

(1) A 5.0 g powder sample was weighed and pressed under 35 tons of pressure. Eleven elements, including Cr, Pb, and Zn, were determined via X-ray fluorescence spectrometry (XRF).

(2) A 0.2000 g sample was weighed and then processed with HC1, HNO₃, HF, and HClO₄. Thereafter, the processed sample was transferred to a 25 mL colorimetric tube for constant volume measurement. An Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES, iCAP 7400 Radial, Thermo, Waltham, MA, USA) was used to determine 3 elements, including Cu and Ni. The solution indicated above was diluted 10 times, and the Cd content was determined using inductively coupled plasma mass spectrometry (ICP-MS, iCAP Q, Thermo, USA).

(3) A 0.5000 g sample was weighed and decomposed with aqua regia. The solution was then reduced using potassium borohydride. The Hg content was determined using the atomic fluorescence method (AFS). The solution was pre-reduced by reducing the masking agent (thiourea-ascorbic acid) and then reducing this with potassium borohydride. The arsenic content was determined using the atomic fluorescence method (AFS).

The samples were then analyzed according to the standard requirements of DZ/T0258-2014 [20], the national first-class soil geochemical reference materials GBW07401-GBW07408 and GBW07423-GBW07430 (GSS1-GSS12), the effective state reference material GBWO7412-GBW07417, and trace gold reference substances GBW07228, GBW07229, and GBW07243-GBW07248 and other series of samples for analysis data quality monitoring. The logarithmic deviation (∆lgC) or the average relative error (RE%) between the measured mean and the standard value of each measured element was used to test the accuracy of the method. To evaluate the precision of the analytical method, the relative standard deviation (RSD%) between the mean value and the standard value was calculated.

2.3. PMF Model Construction

The positive matrix factorization (PMF) model proposed by Paatero and Tapper and endorsed by the EPA is a robust factor analysis tool [11,21]. It decomposes data into source contribution and composition matrices, aiding in the identification of primary pollution sources and their respective contribution rates. The PMF model rationale assumes that a soil sample X can be represented as an n × m matrix, where “n” represents the number of samples and “m” represents the chemical composition [11,21]. The matrix X can be further decomposed into a pollution source contribution matrix and a pollution source composition spectral matrix. The method calculates the error of each chemical component in the pollutant by weight and then uses the least squares method to determine the main pollution source of the pollutant and its contribution rate. The least squares method is a multivariate factor analysis tool.

The basic principle of the PMF model is to factorize the original matrix X (n × m) and decompose it into a factor contribution matrix G (n × p), a factor component matrix F (p × m), and a residual matrix E (n × m) as follows (Equation (1)):

$$E_{nm}=X_{nm}-{\displaystyle \sum }_{k=1}^p{G}_{np}{F}_{pm}$$

(1)

where X_nm represents m chemical components in n samples; p is the number of sources apportioned; G_np is the source contribution matrix; and F_pm is the source contribution matrix. The elements in the matrix G_np and F_pm are all positive values, meaning they are non-negative constraints, and all parameters in the calculation process outlined above are dimensionless. The PMF defines an objective function Q and minimizes the value of this objective function as follows (Equation (2)):

$$Q \left(E\right)={\displaystyle \sum }_{i=1}^m{\displaystyle \sum }_{j=1}^n{\left(E_{ij}/{\sigma }_{ij}\right)}^2$$

(2)

where E_ij represents the residual of the ith chemical component in the jth sample, and σ_ij represents the uncertainty of the ith chemical component in the jth sample. All parameters in the calculation outlined above are dimensionless.

When the concentration of the chemical component is less than or equal to the corresponding method detection limit (MDL), the uncertainty ($U$) is calculated according to Equation (3) as follows:

$$U=5/6\times \mathrm{MDL}$$

(3)

When the concentration of the chemical component is greater than the corresponding MDL, $U$ is calculated according to Equation (4) as follows:

$$U=\sqrt{{\left(s\times c\right)}^2+{\left(0.5\times \mathrm{MDL}\right)}^2}$$

(4)

where $U$ is the uncertainty; s is the percentage of error; c is the measured potentially toxic element content, with a unit of mg·kg⁻¹; and the MDL is the method detection limit.

2.4. Geographical Detector

A geographical detector (geo-detector) is a method developed by Wang et al. to explore the influence of geographical spatial zoning factors on disease risk [22]. The factor detector, a sub-module, was used in this study. The factor detector measures the contribution of independent variables to dependent variables by calculating the ratio of the sum of the variance of each independent variable and that of the dependent variables after classification. It is used to detect the spatial differentiation of dependent variables and the ability of each independent variable to explain the degree of influence of the dependent variables, which is measured using the q value as follows:

$$SSW={{\displaystyle \sum }}_{h=1}^l{N}_h{\mathsf{\sigma }}_h^2 h=1,2,3,\dots ,l$$

(5)

$$SST={\mathsf{\sigma }}_h^2$$

(6)

$$q=1-{\displaystyle \frac{SSW}{SST}}$$

(7)

where h = 1, 2, …, l represents the number of classifications of the independent variable X; N_h represents the number of units in the partition h; and ${\mathsf{\sigma }}_h^2$ represents the variance of the variable Y in partition h. The SSW and SST represent the sum of the variances for all classes of the independent variable X and the total variance within the region, respectively. The range of q is [0,1]. The greater the value of q, the greater the influence of the independent variable X on the dependent variable Y.

2.5. Analytical Statistical Method

Considering the overall situation of the study area and based on the overall analysis of the literature, 4 potential factors that may affect the distribution of soil PTE contents were considered. Of these, the natural factors include the distance from the riverbank and the parent material of the soil, and human factors include the number of factories and land use types. Microsoft Excel 2016 (Microsoft Corporation, Redmond, WA, USA) and SPSS22.0 (International Business Machines Corporation, Amonk, NY, USA) were used to perform basic statistical analyses; ArcGIS10.8 (Environmental Systems Research Institute Inc., RedLands, CA, USA) was used to analyze the spatial distribution characteristics and topographic parameters of the potentially toxic elements. The other graphics were plotted in Origin (2022) (OriginLab, Northampton, MA, USA) and CorelDRAW 2019 (Corel Corporation, Ottawa, ON, Canada).

3. Results and Discussion

3.1. Content Characteristics of Cd and Other Elements in Topsoil

Table 1. Summary of basic statistics of eight selected PTEs in the topsoil of Nanliujiang.

	Min.	Q25	Mean	Median	Q75	Max.	Std.	CV (%)
As	1.17	6.28	11.76	9.17	13.30	246.00	11.96	101.70
Cd	0.01	0.09	0.16	0.13	0.19	6.18	0.19	119.84
Cr	4.50	31.00	45.48	43.00	53.00	319.00	24.38	53.59
Cu	1.87	10.45	16.56	14.40	20.85	99.70	9.30	56.13
Hg	0.02	0.07	0.13	0.10	0.12	8.11	0.30	225.58
Ni	2.04	8.09	12.67	11.20	15.92	81.20	7.04	55.53
Pb	7.40	21.30	32.25	31.20	41.65	171.00	14.33	44.42
Zn	5.50	30.20	47.89	45.50	59.40	275.00	25.91	54.11

shows the descriptive statistics of the eight potentially toxic elements in the topsoil of Nanliujiang. The statistics include the minimum value, maximum value, mean value, median, quantile (25th and 75th), standard deviation, coefficient of variation, and MDL. The median values of the eight elements were 9.17, 0.13, 43.00, 14.40, 0.10, 11.20, 31.20, and 45.50 mg/kg, respectively. The large gap between the minimum, 75th quantile, and maximum indicates the presence of outliers (i.e., maximum values) in the dataset of the study area, resulting in a positively skewed distribution of all elements (Figure 3). The coefficients of variation of As, Cd, and Hg were relatively high (>100%), indicating that human factors interfered with their concentrations. The moderate spatial variability of Cr, Cu, Ni, Pb, and Zn indicates that these elements came from similar sources and are relatively less disturbed by human activities.

The Spearman rank correlation coefficient was used to quantitatively measure the strength of the relationship between variable pairs of eight potentially toxic elements (Figure 4). Histograms and normal curves indicating the normality of each chemical substance showed that not all the variables followed the standard normal distribution (K-S test < 0.05), with a long right-skewed distribution. Therefore, the Spearman rank correlation coefficient based on non-parametric statistics was selected for correlation analysis.

Generally, a significant strong correlation suggests that these variables have similar chemical properties. This means that they may be controlled by a common source. In addition to the strong positive correlation among Cr, Cu, Pb, and Zn, these PTEs belong to transition metals with similar chemical properties, mainly derived from natural sources such as bedrock and parent materials. Cd and As, however, correlated poorly with the other elements. Most studies have recorded significant agricultural disturbance effects on As and Cd [12], and the distinction between natural factors and human factors on these two elements must be further analyzed in this study.

3.2. Spatial Distribution of Components

The spatial distribution of eight potentially toxic elements in the topsoil of Nanliujiang is shown in Figure 5. The distribution of As and Cd differed from the other elements. The As and Cd concentrations increased near Yulin City (Figure 5a,b). The influence of natural factors and agricultural activities was observed in the regional distribution of manganese rocks and dry cultivated land in the Liujiang Formation. The Cr and Hg concentrations were similar in the Quaternary sediments, indicating that human factors had a significant influence on the topsoil in Nanliujiang.

Furthermore, the concentrations of Cr, copper (Cu), nickel (Ni), lead (Pb), and zinc (Zn) are higher in the north-central region and lower in the southeast. This geospatial division aligns with the boundary of the granites, as shown in Figure 5. The natural mineralization of these granites contributes to the enrichment of Ni, Pb, and Zn in the topsoil of Nanliujiang.

3.3. Source Apportionment of Potentially Toxic Elements in the PMF Model

The PMF model was used to further identify and distinguish the potentially toxic element sources in the topsoil of Nanliujiang. Figure 6 provides an overview of the basic factors, including the contribution percentages and species concentrations. Although the PMF model has advantages that other receptor models cannot match, the determination of the optimal number of factors is still challenging. By adjusting the number of input factors from three to six in this study, we found that when the number of factors was four, the Q value was at a minimum and stable, providing the optimal solution (Figure 6).

The contribution rate of As in Factor 1 was far higher than the other potentially toxic elements, reaching as high as 81.59%. This is because As exists in different chemical forms, such as As³⁺ and As⁵⁺, and because its solubility is much higher than other potentially toxic elements. Human factors, such as agricultural chemical inputs, often control the As concentration. Previous studies have indicated that long-term cultivation methods, which include the application of chemical fertilizers and pesticides, can cause the accumulation of potentially toxic elements, such as As, in soil [23,24]. These relatively high As loads and the high concentrations in cultivated land further validate the results of the PMF model [23,24]. The coefficient of variation of As (101.7%) was also higher than that of the other seven potentially toxic elements, indicating that it was more significantly affected by human activities. The high contribution area of Factor 1 was mainly concentrated in the central and eastern parts of the study area (Figure 7a), and the contribution scores gradually decreased along other directions, similar to the distribution of river channel density. Factor 1, therefore, might represent irrigation water sources characterized by river channel distribution or other agricultural inputs.

Factor 2 was dominated by Cr, Cu, and Ni, which contributed 75.56%, 65.10%, and 66.91%, respectively. In Table 1, the coefficient of variation of Cr was the lowest of the eight potentially toxic elements (53.59%), indicating that the soil Cr content in the study area was stable and less affected by human activities. Cr and Ni are mainly derived from soil parent materials [25,26,27], and human activities in the environment usually do not affect their distribution [25,26]. However, the high contribution region of Factor 2 was widely distributed in the northeast and central regions, like industrial factories’ distribution. Electroplating factories were found in these areas during the site survey. It was tentatively inferred that Factor 2, therefore, may represent a mixture of natural and industrial sources.

Pb and Zn dominated Factor 3, with contribution rates of 81.19% and 67.80%, respectively. Motor vehicle emissions are a major source of Pb in agricultural soils. Vehicle exhaust accounts for about two-thirds of global Pb emissions [28,29,30], while high Zn concentrations are associated with metal manufacturing, mineral aggregates, and mining activities [28,29,31]. However, the high contribution score area of Factor 3 was mainly located in the central and northern parts of the study area (Figure 7c). The contribution decreased in the surrounding areas and coincided with the medium-acid magmatic rocks. The coefficients of variation of Pb and Zn were both low in these areas, indicating that human pollution was relatively limited. Therefore, Factor 3 probably represents a mixture of natural sources and exhaust emissions, while the potential mineralization in the granite area further promotes the further enrichment of Pb and Zn.

Factor 4 was dominated by Cd and Hg, with contribution rates of 71.97% and 72.83%, respectively. The coefficients of variation of both Cd and Hg varied widely, indicating that they were influenced by human factors or other natural sources. Agricultural activities, such as the use of phosphate fertilizers, pesticides, organic manure, and sewage irrigation, often introduce large amounts of Cd into the environment [32,33]. Crop production in China is often accompanied by the large-scale use of fertilizers and pesticides, causing the accumulation of potentially toxic elements in the soil and related environmental problems [32,33,34]. Since Cd had the highest coefficient of variation in this farmland area, agricultural activities are probably the primary source of Cd. Factor 4 might, therefore, be closely related to agricultural activities.

3.4. Source Identification of Geographical Detectors

The spatial distribution of potentially toxic elements in soil in nature results from the joint action of multiple factors [34]. The interaction detector model in the geo-detector was used to conduct interaction analysis for the main drivers of potentially toxic elements in soil (α < 0.05) to further explore how these influencing factors work together (Figure 2 and Figure 8).

The interaction between the parent material and land use type contributed significantly to Factor 1, and its effect was more than the sum of the two effects alone. Specifically, the interaction between the parent material and the land use type explains 7.4% of the spatial distribution of potentially toxic element concentration in the study (i.e., q(SP∩LU) = 0.074). This is higher than the sum of the two factors alone (q(SP + LU) = 0.061). The combined effect of each pair of driving factors was greater than that of a single factor, which is a two-factor enhancement effect. Also, the interaction between other driving factors showed nonlinear enhancement. The distance from the riverbank was the strongest driver explaining the spatial distribution of Factor 1, and its contribution to the interaction with other major drivers was significantly enhanced, far exceeding the interaction between any two drivers.

Observations for Factor 2 were similar to those of Factor 1, in which the contribution of the pair interaction was also greater than the sum of the individual contributions of the two influencing factors. The combined effect of each pair of drivers was greater than any single factor. The distance from the factory was the strongest driver explaining the spatial distribution, and its interaction contribution with other major drivers was also significantly enhanced.

The interaction between the parent material and other driving factors for Factor 3 also showed a two-factor enhancement effect. The parent material was the driver, which best explained the spatial distribution of Factor 3.

For Factor 4, the influence of the interaction between land use and the distance from the factory was less than the sum of the two contributions alone. Land use became the most powerful driver influencing the spatial distribution of Factor 4 and showed nonlinear enhancement.

Combined with the results of the geo-detector model, the main factors affecting the spatial distribution of potentially toxic elements in the study area are as follows: parent material (q = 0.327) > land use type (q = 0.010) > distance from the factory (q = 0.008) > distance from the riverbank (q = 0.007). Of these, the parent material explained the spatial distribution of the potentially toxic elements the best and reflected the significant influence of human activities on the accumulation of the potentially toxic elements. These results not only reveal the complex mechanisms affecting the spatial distribution of the potentially toxic elements but also emphasize the key role of multi-factor synthesis in the distribution of potentially toxic elements.

3.5. Summary of Source Apportionment

The spatial analysis data of the auxiliary dataset (categorical and numerical variables) and the comprehensive model proposed in this study (which combines the PMF with the geo-detector model) provide a deep understanding of the multi-source allocation [12,13].

F1, as obtained using the PMF model, showed a wide range of high values in the central and eastern regions (Figure 5). The spatial pattern of F1 showed high spatial consistency with the distribution of the riverbank when combined with the F1 distribution obtained using inverse distance weights. The distance from the riverbank had the highest and most significant spatial correlation with F1 in the geo-detector model, and the interaction between the distance from the riverbank and other influencing factors showed nonlinear enhancement. Therefore, in summary, F1 can be defined as a source of irrigation water. Urban sewage and agricultural activities are the main sources of As pollution in irrigation systems. As-containing substances enter the irrigation system through runoff and infiltration and then enter the farmland via the irrigation water.

The PMF model showed a high value in the northeast suburbs with strong industrial development and urban expansion for F2. From the spatial distribution, the high value of F2 mainly occurs in the suburbs. The spatial correlation between industrial production and F2 was the highest and most significant in the geo-detector model. In terms of interaction, the q value of the combination of the distance from the riverbank and the distance from the factory increased the most. This further confirms that F2 mainly resulted from industrial activities, and the discharge of wastewater and gas caused the increase in the soil’s potentially toxic elements.

The PMF-derived F3 was high in the south-central region. The parent material had the highest and most significant spatial correlation with F3 in the geo-detector model. Regarding interaction, the PD value of the combined effect of the soil parent material and the distance from the factory was the highest. Therefore, the F3 is related to the parent material. In particular, the granitic area controls the occurrence of Pb.

The F4 obtained using the PMF model showed high values in the midwest and the northeast. The spatial distribution shows that the high F4 value was mainly limited to the farming area. The spatial correlation between cultivated land distribution and F4 was the highest and most significant in the geo-detector model. Regarding interaction, the q value of the combined effect of land use type superimposed by distance from the riverbank increased the most. This further confirms that the main source of F4 is agricultural activities. Agricultural inputs, such as phosphate fertilizer and livestock manure, increase the potentially toxic element content of the soil.

In terms of the contribution of various pollution sources to potentially toxic elements, 59.4% of the potentially toxic elements in the soil can be attributed to human sources, and 40.6% can be attributed to natural sources. Of the human sources, agricultural pollution accounted for 22.9% of the total pollution, and industrial pollution accounted for 36.5%. The main source of Pb and Zn is the soil parent material, and the source of Cd and Hg is agricultural activities such as fertilization. Arsenic mainly comes from irrigation water, while Cr, Cu, and Ni are related to electroplating factories.

3.6. Comparison and Verification of the Results of the Two Models

The PMF model and the geo-detector model agreed in several respects. Both models identified agricultural activities and industrial emissions as the main sources of potentially toxic element pollution in the study area. This is consistent with the characteristics of regional industrial structure. The key factors affecting metal pollution, as analyzed using the geo-detector model, corresponded to the types of pollution sources analyzed using the PMF model. This verifies the reliability of the results. The PMF model can also quantify the contribution of each pollution source, while the geo-detector model reveals the spatial heterogeneity and interaction of the influencing factors. The combined use of the two models allows for the comprehensive analysis of the distribution of potentially toxic element pollution from two-dimensional source apportionment and spatial differentiation, further confirming the effectiveness of soil potentially toxic element traceability research methods [14].

4. Conclusions

This paper combined the positive matrix factorization (PMF) model and the geo-detector model to analyze the regional soil potentially toxic element pollution sources in the Nanliujiang River Basin. The main sources of the potentially toxic element pollution of the agricultural soil in the Nanliujiang River Basin mainly included four categories: geological background, agricultural activities, industrial discharge, and river irrigation. The geological background contributed the most, while agricultural activities, industrial discharge, and river irrigation mainly affected the spatial distribution of the potentially toxic elements. The various pollution sources and their contribution rates were successfully identified through the PMF model, and the key factors affecting the spatial distribution of the potentially toxic elements were further analyzed using the geo-detector model.

Regarding the pollution source apportionment, the geological background was the main source of potentially toxic element pollution, which reflects the abundant exposure in the region’s interior and the critical influence of soil parent material on the potentially toxic element content. Human factors, such as agricultural activities, industrial discharge, and river irrigation, have also contributed significantly to potentially toxic element pollution. The parent material, land use type, distance from the factory, and distance from the riverbank jointly affected the spatial distribution of potentially toxic elements, of which the parent material had the largest influence.

By integrating the PMF and geo-detector models, this study provides a comprehensive analysis of soil PTE pollution sources in the Nanliujiang River Basin. The results highlight the significant contribution of geological background and human activities to PTE pollution, offering valuable insights for targeted soil pollution control measures.