Contamination Assessment and Source Apportionment of Heavy Metals in Agricultural Soil Through the Synthesis of PMF and Geogdetector Models

Science of the Total Environment 747 (2020) 141293

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

Contamination assessment and source apportionment of heavy metals in agricultural soil through the synthesis of PMF and GeogDetector models

Xufeng Fei a,b,⁎, Zhaohan Lou a, Rui Xiao c, Zhouqiao Ren a,b, Xiaonan Lv a,b a Zhejiang Academy of Agricultural Sciences, Hangzhou, China b Key Laboratory of Information Traceability of Agriculture Products, Ministry of Agriculture and Rural Affairs, China c School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, China

HIGHLIGHTS GRAPHICAL ABSTRACT

• A novel framework based on spatial analysis for source apportionment is proposed. • Combined with auxiliary data, the new model provides foundations for source analysis. • Cr (80.72%) was derived mainly from natural sources while As and Pb had mix sources. • Cd (73.68%) was closely associated with agricultural activities. • Hg (92.38%) was mainly attributed to industrial activities.

article info abstract

Article history: Heavy metal pollution in soils has attracted great attention worldwide in recent decades. Selecting Hangzhou as a Received 14 May 2020 case study location, this research proposed the synthesis application of positive matrix factorization (PMF) and Received in revised form 24 July 2020 GeogDetector models for quantitative analysis of pollution sources, which is the basis for subsequent soil pollu- Accepted 25 July 2020 tion prevention and remediation. In total, 2150 surface soil samples were collected across the study area. Al- Available online 28 July 2020 though the mean concentrations of As, Cd, Cr, Hg, and Pb in the soils were lower than the National Editor: Filip M.G. Tack Environmental Quality Standards for Soils in China, the mean contents of As and Cd were higher than their corresponding local background values by approximately 1.31 and 1.59 times, respectively, indicating that heavy Keywords: metals have been enriched in topsoil. Agricultural activities, industrial activities, and soil parent materials were Heavy metal the main sources of heavy metal pollution in the soils, accounting for 63.4%, 19.8%, and 16.8% of the total heavy Positive matrix factorization metal accumulation, respectively. Cr was derived mainly from soil parent materials (80.72%). Cd was closely as- GeogDetector models sociated with agricultural activities (73.68%), such as sewage irrigation and application of fertilizer. Mercury was Spatial analysis mainly attributed to industrial activities (92.38%), such as coal mining and smelting. As was related to agricultural Source apportionment (57.83%) and natural (35.56%) sources, and Pb was associated with industrial (42.42%) and natural (41.83%) sources. The new synthesis models are useful for estimating the source apportionment of heavy metals in soils. © 2020 Elsevier B.V. All rights reserved.

1. Introduction

Soil is one of the most important ecosystems for human survival and development, and soil safety is the fundamental guarantee for national ⁎ Corresponding author at: No. 198 Shiqiao Road, Hangzhou, Zhejiang 310021, China. food security and human health (Fei et al., 2019; He et al., 2019; Wang E-mail address: [email protected] (X. Fei). et al., 2019). With the rapid urbanization and industrialization of

https://doi.org/10.1016/j.scitotenv.2020.141293 0048-9697/© 2020 Elsevier B.V. All rights reserved. 2 X. Fei et al. / Science of the Total Environment 747 (2020) 141293 society, soil pollution, especially soil heavy metal contamination and ac- PMF model, and (3) develop a new combined spatial method for quan- cumulation, has become a serious problem, attracting great public at- titative source apportionment and definition through various forms of tention worldwide (Hu et al., 2017a; Huang et al., 2018; Yang et al., auxiliary data. The results of this study provide useful information for 2018). As China is a developing nation that has undergone rapid urban- the prevention and control of heavy metal pollution in soil. ization in recent decades, soil heavy metal pollution has become a main environmental problem (Niuetal.,2013; Hu et al., 2017b; Liuetal., 2. Materials and methods 2019). According to the Chinese National Soil Contamination Survey Re- port released by the Ministry of Land and Resources and the Ministry of 2.1. Study area Environmental Protection of China, approximately 16.1% of soil samples analyzed were contaminated by heavy metals to various degrees (MEP, Hangzhou is located in the eastern coastal area of China (29°11′– 2014). Because of the heavy metal characteristics of bioavailability, per- 30°33′ N, 118°21′–120°30′ E). It is the capital city of Zhejiang Province sistence, and toxicity, heavy metal accumulation in soil can lead to re- and one of the most important cities in the Yangtze River Delta urban ductions in soil fertility and crop production. Heavy metals can also be agglomeration (Fig. 1). The city has a moderate subtropical climate transferred easily and accumulate through biomagnification in the with a hot and humid summer and cold and dry winter. The annual food chain, which poses significant risks to food safety and human average values of temperature, precipitation, and sunshine hours are health, as heavy metals are absorbed by the human body through inha- 17.8 °C, 1454 mm, and 1765 h, respectively. In total, 13 districts/ lation, ingestion, and dermal absorption (Xu et al., 2017; Zang et al., counties are included in the study area, which has a total area of approx- 2017; Zhao et al., 2014a). Therefore, it is necessary to quantitatively imately 16,854 km2. According to the population census in 2017, the evaluate the characteristics, contamination, and sources of heavy metals population of the area is approximately 9.47 million. The northeastern in soil. area consists of plains, but the southwestern area is mountainous. The Generally, heavy metals in soil come from two major sources: natu- main soil types are red soil and paddy soil derived from eluvium and al- ral sources, driven by weathering and pedogenesis processes, and luvial soil parent materials, respectively. In the past few decades, this anthropogenic activities, such as industrial manufacturing, sewage irri- area has experienced rapid industrialization and urbanization, and the gation, vehicle exhaust, agricultural fertilizer, mining, and smelting (Sun consequent increasing accumulation of heavy metals has caused great et al., 2014, 2019; Yang et al., 2013). Therefore, to effectively reduce the concern (Chen et al., 2009; Fei et al., 2018; Zhang and Ke, 2004). cost and workload of soil remediation, it is important to quantitatively clarify the sources of soil heavy metal pollution (Fei et al., 2018; 2.2. Soil sampling and chemical analysis Huang et al., 2018). Many previous studies have used multivariate qual- itative/quantitative statistical methods, such as spatial deviation, corre- According to the agricultural soil vector map, there are 211,931 ha of lation analysis, cluster analysis, geographic information systems, and agricultural soil in the study area. The 1 km*1 km grid layer was over- principal component analysis (PCA), often in combination with multiple lapped with the distribution of agricultural soil to extract the expected linear regression, to determine the variability in and possible sources of sampling point locations in the study region (Ren et al., 2019). In total, heavy metals in soils (Davis et al., 2009; Dong et al., 2019; Gu et al., 2150 surface soil samples were collected across the study area (Fig. 1). 2012; Lv, 2019; Nanos and Martín, 2012). Among these approaches, Five subsamples were collected and mixed thoroughly to obtain a repre- the positive matrix factorization (PMF) model recommended by the sentative sample, and the sampling locations were recorded by a global US EPA, an ideal method for ensuring nonnegative source contribution positioning system (GPS). All soil samples were air-dried at room tem- values and factor profiles through the use of correlation and covariance perature and then ground to 100-mesh size for chemical analysis. For matrices to simplify high-dimensional variables, has been applied the determination of Cd, Cr, and Pb contents, the soil samples were widely and successfully to identify and quantify pollution sources of digested with HNO3–HClO4–HF, and the metals were analyzed by in- heavy metals in soils (Dong et al., 2019; Huang et al., 2018; Wang ductively coupled plasma mass spectrometry (ICP-MS). For the deter- et al., 2019). mination of As and Hg contents, soil samples were digested with

However, most of these methods cannot analyze the effect of cate- HNO3–HCl, and atomic fluorescence spectrometry (AFS) was used for gorical variables such as soil type and soil parent materials, which the analyses. Blind duplicates and standard reference materials (GSS- have an important influence on the accumulation of heavy metals. 3, China National Center for Standard Materials) were used for quality Few studies have taken the spatial information of the sampling points assurance and control. Standard sample recovery ranged between 90 into consideration, and the definition of each principal component ob- and 110%, and the relative standard deviations of duplicate samples tained from a single PCA/PMF model is based mainly on previous re- were between 3 and 8%. searches and experience, thus precluding the ability to quantitatively define the detailed effects and spatial characteristics of every source 2.3. Analysis methods on different heavy metals. Moreover, previous methods are unable to identify the importance of specific sources to the principal components In this work, the analysis consisted of both a statistical and a spatial based on the spatial analytical data, which is vital for controlling local analysis part, as follows. HM pollution and remediation in soils. To address these gaps, this Statistical analysis: (i) Statistical indicators (mean, median, coefficient study proposes the synthesis of PMF and GeogDetector models for soil of variation, etc.) and Spearman correlation analysis were employed to heavy metal source apportionment and evaluates the approach in a describe the basic characteristics of heavy metals in agricultural soil and case study using the city of Hangzhou, China. Compared to previous re- their internal correlation, and (ii) a PMF model was used for pollution ceptor models, by including spatial information and auxiliary data (cat- source apportionment of heavy metals, as well as for calculation of the egorical and numerical variables), such as soil parent material, road pollution component scores of each site. networks, industrial production, etc., this approach could provide an Spatial analysis: (i)Getis–Ord Gi* analysis (Getis and Ord, 1992)was in-depth understanding of multiple source contributions and a better implemented to explore the spatial distribution pattern of the compo- definition of principal factors. To our knowledge, this is the first study nent scores derived from PMF analysis. (ii) Ordinary kriging was used using GeogDetector to interpret results derived from a PMF receptor to draw the spatial distribution map of each component score (grid for- model. mat). (iii) The results of ordinary kriging and the detailed pollution The main objectives of this study were to (1) analyze the contents source dataset (soil parent materials, distance to the main road, etc.) and basic characteristics of heavy metals in soil in the city of Hangzhou, were combined, and the GeogDetector model was employed to deter- China, (2) identify the potential sources of heavy metals through the mine the spatial correlation between each component score and specific X. Fei et al. / Science of the Total Environment 747 (2020) 141293 3

Fig. 1. Distribution of sampling points and soil parent materials.

pollution sources, which could provide additional valuable information metal concentration is greater than the MDL, the uncertainty can be cal- for quantitative source apportionment. culated as follows: The overall flow chart of this study is shown in Fig. 2, and the specific qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 description of each method is as follows: uij ¼ ðÞerrorfraction concentrations þ ðÞMDL : ð3Þ

2.4. PMF model Otherwise, the uncertainty is estimated through the following equation: The PMF model, an efficient e receptor model for pollution source apportionment, was used in this study to analyze the sources of heavy 5 uij ¼ MDL: ð4Þ metals in the soil (Huang et al., 2018; Wang et al., 2019; Zheng et al., 6 2018). In brief, this model decomposes the matrix of the original dataset All calculations follow the US EPA PMF 5.0 User Guide (U.S. Xij into two factor matrices (the source contribution matrix gik and the Environmental Protection Agency, 2014). source profile matrix fjk) and a residual matrix eij. The basic equation is as follows: 2.5. Spatial analysis model Xp ¼ þ ; ð Þ Xij gik f kj eij 1 After the factor values of each component in the samples were esti- k¼1 mated through PMF, Getis–Ord Gi* analysis (Getis and Ord, 1992)was employed to explore the spatial distribution pattern of the component where Xij is the concentration of the jth heavy metal at the ith sam- scores. The spatial clustering of the samples' component scores was de- pling location, gik is the contribution of the kth source to the ith sam- termined through the Z values, which were calculated as follows: ple, fkj is the concentration of the element j from the kth source, and e is the residual error matrix, which can be calculated through the P P ij n w x −x n w minimum value of the objective function Q.ThevalueofQ is calcu- ¼ j¼v1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiij j j¼1 ij ; ð Þ Z j vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u 2 !3 5 u u 2 lated as follows: u Xn u Xn n t1 2− 2 t 1 4 2−∑ 5 x j x n wij wij Xn Xm 2 n n−1 ¼ e j¼1 j¼1 j 1 Q ¼ ij ; ð2Þ u i¼1 j¼1 ij where Zj is the Getis–Ord Gi* Z value of sample j, xj is the component where uij is the uncertainty of the jth heavy metal in the ith sample, score of sample j, n is the total number of samples, wij is the weight of which is calculated from the species-specific method detection limit the neighbor sample i to j, and x is the mean value of the component (MDL), the concentration, and the provided error fraction. If the heavy score in all samples. Z values higher than 1.96 indicate significantly high 4 X. Fei et al. / Science of the Total Environment 747 (2020) 141293

Fig. 2. The overall flow chart of this study. clustering, and values lower than −1.96 mean significantly low cluster- models. Moreover, semivariogram information could provide funda- ing at the 0.05 level (Getis and Ord, 1992; Fei et al., 2016). mental reference for source analysis (Wang et al., 2020; Wu et al., In addition, ordinary kriging interpolation (Fei et al., 2019; Gribov and 2020). Krivoruchko, 2011) was executed to construct spatial distribution maps of each component score to validate the spatial distribution pattern of the 2.6. GeogDetector model sampling points and provide a foundation for the GeogDetector model. As one of the most commonly used geostatistical methods, ordinary The GeogDetector model (Wang and Hu, 2012) was then used to de- kriging has been successfully applied in various disciplines, especially in termine the spatial correlation between each component score and the the assessment of soil heavy metal pollution in recent years (Li et al., detailed pollution source dataset (soil parent material data obtained 2020; Wang et al., 2020; Wu et al., 2020). This tool realizes the optimal from Wu et al. (2013) as a proxy for natural sources; the distance linear unbiased estimation of spatial data based on sampled data and is from the sampling points to the main roads calculated in ArcGIS 10.2 useful for exploring the spatial distribution pattern and variation char- as a proxy for traffic, or vehicle emission sources; and industrial and ag- acteristics of variables (Christakos, 1998). The semivariogram in the or- ricultural production values from the Hangzhou Statistics Yearbook dinary kriging process can be calculated as follows: (http://tjj.hangzhou.gov.cn) as proxies for industrial and agricultural _ 1 ðÞ sources), which provided additional information for source identifica- γ ðÞ¼h ∑nh½XsðÞþ h −XsðÞ2 ð6Þ X 2nhðÞ i¼1 i i tion of each component obtained from the PMF model. The detailed theory of this model is available in previous works (Fei γ⌢ ðÞ where X h is the semivariogram value at the distance of h, X(si + h) et al., 2016; Fei et al., 2018; Li et al., 2013; Wang et al., 2010; Wang and and X(si) are the values of studied variables at the locations of si + h Hu, 2012). Briefly, there are four detectors in this model, i.e., a risk de- and si,respectively,andn(h) is the number of pairs of sample points tector, factor detector, ecological detector and interaction detector at a distance of h (Olea, 2006). Spatial distribution maps of compo- (Wang et al., 2010; Jiang et al., 2020; Zhang et al., 2020). In this study, nent scores derived from ordinary kriging can be used for overlay the factor detector and interaction detector were employed to quantita- analyses with pollution source variables through GeogDetector tively detect the degree of influence of the pollution sources on the X. Fei et al. / Science of the Total Environment 747 (2020) 141293 5 component scores derived from the PMF model. The basic assumption The mean concentrations of these heavy metals, in decreasing order, of the factor detector model here is that if one component extracted were Cr (52.90 mg/kg) N Pb (31.66 mg/kg) N As (8.99 mg/kg) N Cd from the PMF can be defined as a specific source, then the spatial distri- (0.27 mg/kg) N Hg (0.13 mg/kg). Compared with the risk screening bution of this component should be similar to that of its source. For ex- values defined by the National Environmental Quality Standards for ample, if component 1 (F1) has a spatial distribution pattern similar to Soils in China (GB15618-2018), the mean concentrations of these that of the soil parent material, then F1 has a great probability of heavy metals were all lower than the strictest risk screening values, in- representing a natural source. The strength of this spatial similarity dicating that the soil is relatively safe for agricultural production and can be calculated quantitatively through the power determination human health. However, in comparison with the average values in (PD) value, which can be estimated as follows: China (Teng et al., 2014), the contents of Cd, Hg, and Pb were higher than their national values by approximately 2.78, 2.00, and 1.17 times, 1 Xn respectively, whereas the contents of As and Cr were lower than their PD ¼ 1− A V ; ð7Þ AV i i corresponding national averages. Moreover, the concentration of As i¼1 and Cd were higher than their corresponding local background values (Xu et al., 2012) by approximately 1.31 and 1.59 times, respectively, in- where A is the total area of Hangzhou, V is the component score variance dicating that As and Cd were enriched in the topsoil through human ac- of all samples, Ai is the area of subregion i (for the soil parent material, tivities. The mean contents of Cr, Hg, and Pb were lower than their the subregions were defined as every category of parent material; for corresponding local background values. The proportions of each ele- numerical variables such as distance and production, the subregions ment above its background value were in decreasing order of Cd were classified through maximizing/minimizing the dispersion vari- (46.09%) N As (38.47%) N Hg (20.42%) N Pb (17.16%) N Cr (6.33%). ance between/within subregions, respectively), and Vi is the component The highest value of the coefficient of variation (CV) occurred for Cd score variance of samples in subregion i. In this case, the PD value, rang- (0.81), reflecting a wider extent of variability in relation to the mean. ing from 0 to 1, represents the strength of the spatial correlation, from Arsenic (0.69) and Hg (0.62) also had high CV values. The high CV values weakest to strongest. and strongly positive skewness of Cd, As, and Hg indicated that these Then the interaction detector was implemented to assess the com- metals were strongly affected by human activities (Liu et al., 2018; Lv, bined influences of pairs of pollution sources on the component scores 2019). Cr (0.38) and Pb (0.38) showed moderate variation. According (Ren et al., 2019; Jiang et al., 2020). The PD values of each pair of pollu- to the skewness values, kurtosis values, and results of the K-S normality tion sources were labeled PD(S1)andPD(S2), and their interaction influ- test, these heavy metal contents were not all normally distributed. Thus, ence was calculated by the interaction detector as PD(S1∩S2). Finally, Spearman's rho correlation, a nonparametric method, was adopted to the interaction of the two sources was assessed by comparing the explore the relationships between selected heavy metals (Dong et al., value of PD(S1∩S2) to the sum of PD(S1)andPD(S2), which resulted in 2019; Fei et al., 2019). As there was a large number of samples, the cor- five interaction phenomena (Table S1). relation coefficients were small, but all were positive and significant Descriptive statistics and Spearman correlation analysis were con- (Table 2), indicating that these heavy metals were significantly corre- ducted using SPSS 19.0 software (IBM Inc., Armonk, NY, USA). The lated with one another. Additionally, the pairs of Cd-Pb (0.466) and PMF model was applied using US EPA PMF 5.0 (U.S. Environmental Pb-Hg (0.475) had relatively high correlation coefficients, indicating a Protection Agency, 2014), and Getis–Ord Gi* analysis and ordinary high possibility that they had similar anthropogenic sources, such as ve- kriging interpolation were conducted using ArcGIS 10.2 (Esri Inc., hicle emissions, coal-fired power plants, agricultural fertilizer, and sew- USA). GeogDetector software (Wang and Hu, 2012) was employed for age irrigation (Hu and Cheng, 2013; Yoon et al., 2006). The detailed the GeogDetector model. P values at the 0.05 level were considered sig- source analysis is provided in the following parts of this paper. nificant, and all P values and 95% CIs were two-sided.

3. Results and discussion 3.2. Source apportionment of heavy metals in PMF

3.1. Descriptive statistics of heavy metals in soil The PMF model was used further to identify the sources and quantify the contributions of heavy metals in the soils. The number of factors for The summary statistics of the heavy metal concentrations in the soil the model was initially set at 2, 3, and 4, the start seed number was ran- samples are shown in Table 1. The contents of As, Cd, Cr, Hg, and Pb domly obtained, and the number of runs was 20. The most suitable ranged from 2.63 to 43.20 mg/kg, 0.05 to 1.42 mg/kg, 10.50 to number of factors was determined by assessing the smallest and most 104.00 mg/kg, 0.03 to 0.50 mg/kg, and 17.00 to 62.60 mg/kg, with me- stable Q value. Finally, it was determined that three factors resulted in dian values of 7.07, 0.20, 54.40, 0.11, and 30.90 mg/kg, respectively. good model fitting, with prediction residuals normally distributed within −3.0 to 3.0 and a prediction R2 greater than 0.68. Two error estimation models, classical bootstrap (BS) and displacement of factor el- Table 1 ements (DISP), were implemented to assess the bias and uncertainty of Summary statistics of heavy metal concentrations (mg/kg) in Hangzhou soil samples. the PMF results (Hu et al., 2020). The results showed that approximately As Cd Cr Hg Pb 91% of the base factors were reproduced in the BS model, and no factor Min 2.63 0.05 10.5 0.03 17 swaps were observed in the DISP model, indicating that the three-factor Max 43.2 1.42 104 0.5 62.6 PMF solution was stable. Median 7.07 0.2 54.4 0.11 30.9 Mean 8.99 0.27 52.9 0.13 31.66 CV 0.69 0.81 0.38 0.62 0.38 Skewness 2.309 2.431 −0.024 1.602 0.72 Table 2 fi Kurtosis 6.222 6.846 −0.572 3.073 0.454 Spearman correlation coef cients of heavy metal concentrations in the soils. Background of Zhejianga 6.88 0.17 55.99 0.17 35.7 As Cd Cr Hg Average of Chinab 11 0.097 61 0.065 27 c ⁎⁎ Risk screening values 20 0.3 150 0.5 70 Cd 0.361 ⁎⁎ ⁎⁎ Cr 0.249 0.156 a Data from Xu et al. (2012). ⁎⁎ ⁎⁎ ⁎⁎ Hg 0.124 0.305 0.131 b Data from Teng et al. (2014). ⁎⁎ ⁎⁎ ⁎⁎ ⁎⁎ Pb 0.309 0.466 0.266 0.475 c Based on the lowest soil risk screening values in the National Environmental Quality Standards for soil in China (GB15618-2018). ⁎⁎ Significant at the 0.01 level. 6 X. Fei et al. / Science of the Total Environment 747 (2020) 141293

an indicator of industrial emission (Hu et al., 2018; Rachwa et al., 2015). On the other hand, vehicle emissions have been regarded as the main source of Pb in agricultural soil over the past few decades. It has been estimated that automobile exhaust accounts for roughly two-thirds of global Pb emissions (Guo et al., 2014; Cui et al., 2018). Ad- ditionally, coal mining and combustion emissions could be important contributors of Pb in atmospheric deposition (Liang et al., 2017). There- fore, F2 may be considered to represent coal-related industrial activities and transportation. Factor 3 (F3) accounted for 63.4% of the total contribution and was dominated by Cd (73.68%) and As (57.83%). Agricultural activities, such as the use of phosphate fertilizer, pesticides, organic manures, and sewage irrigation, generally provide considerable amounts of Cd (Shao et al., 2016; Zhao et al., 2014b). In China, crop production is usu- ally accompanied by the large-scale use of fertilizers and pesticides, which result in heavy metal accumulation in soils and create environmental problems (Sun et al., 2019). Considering that Cd had the highest CV in this farmland area, it can be inferred that agricultural activities are the main source of Cd. Previous studies also concluded that long- standing farming practices, such as the application of fertilizers and pesticides, can lead to the accumulation of heavy metals such as As in soils (Jiang et al., 2017; Liu et al., 2017; Qiao et al., 2011). Thus, F3 may be associated with agricultural activities.

3.3. Spatial analysis of components

To understand the spatial distributions of the component factors further and provide information for source identification, the spatial pattern of sampling point component scores was detected by Getis–Ord Gi* analysis and the distribution of component factors across the study area was mapped by ordinary kriging (Fig. 4). Samples with low F1 values were clustered in the northern and central-eastern areas, fi Fig. 3. Factor pro les and source contributions of heavy metals from the PMF model. where eluvium and washland alluvium are the main soil parent materials; whereas high F1 values were distributed in a wide region includ- The results of the PMF analysis are shown in Fig. 3. Factor 1 (F1) con- ing the northern, central, and southwestern areas, where various tributed to 16.8% of the heavy metals in the soils and included a heavy eluvium and alluvium soil parent materials are distributed. The kriging loading for Cr (80.72%) and moderately loadings for Pb (41.83%) and interpolation map of F1 was consistent with the Getis–Ord Gi* analysis As (35.56%). The mean concentration of Cr was lower than the local of the samples. It is worth noting that the variogram model of F1 had rel- background values, and only 6.33% of the samples showed a concentra- atively strong spatial autocorrelation, as indicated by the low ratio tion of Cr exceeding the background value, indicating that there were (0.31) of nugget/(nugget+sill) and the long model range (34,527 m). few external pollution sources of Cr. In addition, Cr showed lower CV Furthermore, there were no obvious point sources of F1 (Lv, 2019; values than the other heavy metals. The above statistics imply that Cr Sun et al., 2019). As discussed above, F1 had a high loading on Cr, mainly comes from natural sources. Previous studies have also reported which had a concentration lower than the mean of local background that Cr in soils originates from parent materials (Fei et al., 2019; Sun value and relatively small CV. Consequently, F1, with strong spatial au- et al., 2019; Zhou et al., 2016). Hence, F1 may represent a natural source, tocorrelation and a long dependence range in the soils, was probably such as soil parent materials and pedogenic processes. controlled by parent materials. Factor 2 (F2), accounting for 19.8% of the total contribution, was There was only one sample with a low F2 value, which was located weighted primarily on Hg (92.38%) and Pb (42.42%). Mercury had a in the southwestern area, whereas there were abundant samples with high CV value, indicating that anthropogenic sources were the main or- high F2 values that were clustered in the northeastern suburban area, igin of Hg pollution (Cai et al., 2015; Mamut et al., 2017; Baltas et al., where industrial development and urban expansion are intense 2020). In China, smelting, coal mining, and combustion are regarded (Zhang et al., 2013). The kriging interpolation map of F2, with signifi- as the main sources of Hg pollution in soil, and Hg is typically used as cant point sources in the high-cluster area, was in accordance with the

Fig. 4. Distribution maps of the significantly high/low points calculated from Getis–Ord Gi* analysis along with the factor scores from the PMF model. X. Fei et al. / Science of the Total Environment 747 (2020) 141293 7 point results. Moreover, the variogram model for F2 had moderate spa- Table 4 tial autocorrelation, indicated by the high ratio (0.64) of nugget/ The interaction power determination (PD) value of two pollution source proxies to the component factors. (nugget+sill) and the short dependence range (4136 m). As discussed above, F2 was heavily loaded on Hg, which had a relatively high CV. F1 F2 F3 fi Therefore, F2, with low spatial autocorrelation and signi cant point SPM&IP 0.423NLE 0.282NLE 0.121NLE sources in suburban areas, was probably controlled by intense industrial SPM&AP 0.437NLE 0.121NLE 0.240NLE and urbanization activities. SPM&DR 0.376NLE 0.123NLE 0.099NLE NLE NLE NLE There was no sample with a low F3 value across the study area, but IP&AP 0.198 0.326 0.182 IP&DR 0.054NLE 0.212NLE 0.040NLE many samples with high F3 values were clustered in the southern and AP&DR 0.029NLE 0.064NLE 0.176NLE southwestern rural areas. The kriging interpolation map of F3 was also SPM: Soil parent materials. in accordance with the point results. The variogram model for F3 had IP: Industrial production. strong spatial autocorrelation, indicated by the low ratio (0.21) of AP: Agricultural production. nugget/(nugget+sill) and the short dependence range (4437 m). As DR: Distance to the main road. discussed above, F3 was heavily loaded on Cd, which had a relatively NLE: Non-linear enhancement. high CV and is typically used as an indicator for agricultural activities Bold data was used to highlight the biggest values in each column. (Shao et al., 2016; Zhao et al., 2014b). Hence, F3, with strong spatial autocorrelation in rural soils, was probably controlled by agricultural F1 derived from the PMF model showed low clusters in the north- nonpoint source pollution, such as sewage irrigation and fertilizer ern and central-eastern areas and high clusters in a wide region in application. the northern, central, and southwestern areas (Fig. 4). Combined with the distribution map of F1 produced by ordinary kriging, the 3.4. Source identification in GeogDetector spatial pattern of F1 showed high spatial consistency with the distribution of soil parent materials. Moreover, the variogram model of F1 In the final step, the GeogDetector model was used to determine the also revealed strong spatial autocorrelation and a long model range, spatial correlation between each component factor and the detailed pol- indicating that F1 was more natural and continuous in space, with lution sources. The PD values of each possible pollution source proxy for less nonstructural heterogeneity caused by human activities. Finally, the components extracted from the PMF are shown in Table 3. Soil par- the soil parent material type had the highest and most significant ent material type had a significant spatial correlation with F1 (PD = spatial correlation with F1 in the GeogDetector model, and there 0.336, P b 0.01), which indicated that F1 represents natural sources of was a certain nonlinear enhancement of the interaction between soil pollution. However, industrial production and agricultural produc- the soil parent material and other pollution proxies. In conclusion, tion had significant spatial correlations with F2 (PD =0.178,P b 0.01) F1 could be defined as natural sources. and F3 (PD =0.143,P b 0.01), indicating that F2 and F3 represent an- F2obtainedfromthePMFmodelshowedhighclustersinthe thropogenic sources of soil pollution (industrial and agricultural activi- northeastern suburban area, where industrial development and ties, respectively). urban expansion were intense (Fig. 2). It can also be seen from the The interaction influences of pairs of sources on the spatial distribu- spatial distribution map of the kriging data that high values of F2 tion of the component factors derived from the PMF models are listed in were mainly distributed in the suburbs. In addition, the variogram Table 4. The joint effect of soil parent materials and agricultural produc- model for F2 had moderate spatial heterogeneity and a short depen- tion had the strongest influence on F1 (0.437), followed by the interac- dence range, indicating that F2 had low spatial autocorrelation and tion of soil parent materials and industrial production (0.423). For F2, significant point sources in suburban areas. Finally, industrial pro- the joint effect of industrial production and agricultural production duction had the highest and most significant spatial correlation had the highest PD value (0.326), which further confirmed that F2 with F2 in the GeogDetector model. In terms of interaction effects, mainly comes from anthropogenic activities. In addition, the interaction the combined influence of industry and agriculture had the highest effect of soil parent materials and agricultural production had the stron- PD value, which further confirmed that F2 mainly comes from an- gest influence on F3 (0.240). The PD values of the interaction of the two thropogenic activities. Thus, F2 is mainly controlled by intense in- sources were all greater than the sums of the single PD values, which dustrial and urbanization activities. corresponded to previous studies that found that the interaction effects F3 derived from PMF exhibited high clusters in the southern and of multiple pollution sources showed nonlinear enhancements, indicat- southwestern rural areas. The kriging interpolation map of F3 was ing that the sources of heavy metals in soils are complex and diverse also in accordance with the Getis–Ord Gi* analysis results (Fig. 4). (Ren et al., 2019; Jiang et al., 2020). In addition, the variogram model for F3 had strong spatial autocorrelation and a short dependence range, indicating nonpoint source pol- 3.5. Summary of source apportionment lution. Finally, agricultural production had the highest and most significant spatial correlation with F3 in the GeogDetector model. Based on spatial analytical data with auxiliary datasets (both cate- In terms of interactions effects, the combined influence of soil parent gorical and numerical variables), the proposed comprehensive model material and agriculture had the highest PD value. Hence, F3 is asso- (integrated PMF and GeogDetector models, as well as Getis–Ord Gi* ciated with agricultural activities such as sewage irrigation and fer- analysis and ordinary kriging) in this study could provide an in-depth tilizer application. understanding of multiple source apportionment. Regarding the contribution of various pollution sources to heavy metals, 83.2% of the heavy metals in the soil were ascribed to anthropogenic sources, and 16.8% were attributed to natural sources. Among the Table 3 anthropogenic sources, agriculture accounted for 63.4% of the total pol- The power determination (PD) value of the detailed pollution source proxies to the com- lution, and industry was responsible for 19.8%. The source determina- ponent factors. tion of each heavy metal indicated that Cr was mainly from soil parent Soil parent materials Industry Agriculture Distance to road materials, Cd was closely associated with agricultural activities such as ⁎⁎ Factor 1 0.336 0.028 0.004 0.013 sewage irrigation and fertilizer application, Hg was mainly attributable ⁎⁎ Factor 2 0.075 0.178 0.014 0.033 to industrial activities such as coal mining and smelting, As was related ⁎⁎ Factor 3 0.086 0.024 0.143 0.01 to natural and agricultural sources, and Pb was related to natural and in- ⁎⁎ Significant at the 0.01 level. dustrial sources. 8 X. Fei et al. / Science of the Total Environment 747 (2020) 141293

4. Conclusions Cai, L., Xu, Z., Bao, P., He, M., Dou, L., Chen, L., Zhou, Y., Zhu, Y.-G., 2015. Multivariate and geostatistical analyses of the spatial distribution and source of arsenic and heavy metals in the agricultural soils in Shunde, Southeast China. J. Geochem. Explor. 148, This study developed a new combined spatial method for quantita- 189e195. tive source apportionment and definition of heavy metal pollution in Chen, T., Liu, X., Li, X., Zhao, K., Zhang, J., Xu, J., Shi, J., Dahlgren, R.A., 2009. Heavy metal fi fi soils through various auxiliary data and performed a case study in the sources identi cation and sampling uncertainty analysis in a eld-scale vegetable soil of Hangzhou, China. Environ. Pollut. 157 (3), 1003–1010. city of Hangzhou, China. Although the mean concentrations of As, Cd, Christakos, G., 1998. Spatiotemporal information systems in soil and environmental sci- Cr, Hg, and Pb in the soils were lower than the National Environmental ences. Geoderma 85 (2), 141–179. Quality Standards for Soils in China (GB15618-2018), the mean con- Cui, Z., Wang, Y., Zhao, N., Yu, R., Xu, G., Yu, Y., 2018. Spatial distribution and risk assessment of heavy metals in Paddy soils of Yongshuyu irrigation area from Songhua River tents of As and Cd were higher than their corresponding local back- Basin, Northeast China. Chinese Geogr. Sci. 28 (5), 797–809. ground values by approximately 1.31 and 1.59 times, respectively, Davis, H.T., Aelion, C.M., Mcdermott, S., Lawson, A.B., 2009. Identifying natural and anthro- indicating that heavy metals were enriched in the topsoil through pogenic sources of metals in urban and rural soils using GIS-based data, PCA, and spatial interpolation. Environ. Pollut. 157 (8–9), 2378–2385. human activities. F1 derived from the PMF model showed strong spatial Dong, B., Zhang, R., Gan, Y., Cai, L., Freidenreich, A., Wang, K., Guo, T., Wang, H., 2019. Mul- autocorrelation, a long model range and the highest spatial correlation tiple methods for the identification of heavy metal sources in cropland soils from a with soil parent material; thus F1 was defined as natural sources. F2 resource-based region. Sci. Total Environ. 651, 3127–3138. Fei, X., Wu, J., Liu, Q., Ren, Y., Lou, Z., 2016. Spatiotemporal analysis and risk assessment of (high values were clustered in suburban areas where industrial devel- thyroid cancer in Hangzhou, China. Stoch. Environ. Res. Risk Assess. 30 (8), opment and urban expansion were intense) obtained from the PMF 2155–2168 2016. model showed moderate spatial heterogeneity, a short dependence Fei, X., Lou, Z., Christakos, G., Ren, Z., Liu, Q., Lv, X., 2018. The association between heavy range and the highest spatial correlation with industrial production; metal soil pollution and stomach cancer: a case study in Hangzhou City, China. Envi- ron. Geochem.Heal. 40 (6), 2481–2490. hence, F2 was mainly controlled by intense industrial and urbanization Fei, X., Christakos, G., Xiao, R., Ren, Z., Liu, Y., Lv, X., 2019. Improved heavy metal mapping activities. F3 (high values were clustered in the southwestern rural and pollution source apportionment in Shanghai City soils using auxiliary informa- areas) extracted by the PMF model revealed strong spatial autocorrelation. Sci. Total Environ. 661, 168–177. Getis, A., Ord, J.K., 1992. The analysis of spatial association by use of distance statistics. tion, a short dependence range and the highest spatial correlation with Georg. Anal 24 (3), 189–206. agricultural production; thus, F3 could be attributed to agricultural ac- Gribov, A., Krivoruchko, K., 2011. Local polynomials for data detrending and interpolation tivities such as sewage irrigation and fertilizer application. Agricultural in the presence of barriers. Stoch. Environ. Res. Risk Assess. 25, 1057–1063. Gu, Y., Wang, Z., Lu, S., Jiang, S., Mu, D., Shu, Y., 2012. Multivariate statistical and GIS-based activities, industrial activities, and natural sources accounted for 63.4%, approach to identify source of anthropogenic impacts on metallic elements in sedi- 19.8%, and 16.8% of the total heavy metal accumulation in the soils, re- ments from the mid Guangdong coasts, China. Environ. Pollut. 163, 248–255. spectively. Agricultural activities were the main source of Cd (73.68%) Guo, W., Zhang, H., Cui, S., Xu, Q., Tang, Z., Gao, F., 2014. Assessment of the distribution and provided a portion of As (57.83%); industrial activities dominated and risks of organochlorine pesticides in core sediments from areas of different humanactivityonLakeBaiyangdian,China.Stoch.Env.Res.RiskA.28(4), the contribution of Hg (92.38%) and provided a portion of Pb (42.42%). 1035–1044. In addition, Cr (80.72%) and a portion of As (35.56%) and Pb (41.83%) He, J., Yang, Y., Christakos, G., Liu, Y., Yang, X., 2019. Assessment of soil heavy metal pol- – were closely related to the soil parent materials. These results suggest lution using stochastic site indicators. Geoderma 337, 359 367. fi fl Hu, Y., Cheng, H., 2013. Application of stochastic models in identi cation and apportion- that anthropogenic activities have strong in uences (83.2%) on heavy ment of heavy metal pollution sources in the surface soils of a large-scale region. En- metal accumulation and distribution in the soils of the study area, viron. Sci. Technol. 47 (8), 3752–3760. which calls for the prevention and control of heavy metal pollution in Hu, B., Jia, X., Hu, J., Xu, D., Xia, F., Li, Y., 2017a. Assessment of heavy metal pollution and health risks in the soil-plant-human system in the Yangtze river delta, China. Int. the region. J. Env. Res. Pub. He. 14 (9), 1042. Supplementary data to this article can be found online at https://doi. Hu, B., Wang, J., Jin, B., Li, Y., Shi, Z., 2017b. Assessment of the potential health risks of org/10.1016/j.scitotenv.2020.141293. heavy metals in soils in a coastal industrial region of the Yangtze River Delta. Environ. Sci. Pollut. R. 24 (24), 19816–19826. Hu, W., Wang, H., Dong, L., Huang, B., Borggaard, O.K., Hansen, H.C.B., He, Y., Holm, P.E., fi CRediT authorship contribution statement 2018. Source identi cation of heavy metals in peri-urban agricultural soils of southeast China: an integrated approach. Environ. Pollut. 237, 650–661. Hu, Y., He, K., Sun, Z., Chen, G., Cheng, H., 2020. Quantitative source apportionment of Xufeng Fei: Conceptualization, Methodology, Writing - original heavy metal(loid)s in the agricultural soils of an industrializing region and associated draft. Zhaohan Lou: Formal analysis, Resources. Rui Xiao: Investigation, model uncertainty. J. Hazard. Mater. 391, 122244. Huang, J., Guo, S., Zeng, G.M., Li, F., Gu, Y., Shi, Y., Shi, L., Liu, W., Peng, S., 2018. A new ex- Visualization. Zhouqiao Ren: Validation. Xiaonan Lv: Writing - review ploration of health risk assessment quantification from sources of soil heavy metals & editing. under different land use. Environ. Pollut. 243, 49–58. Jiang, X., Xiong, Z., Liu, H., Liu, G., Liu, W., 2017. Distribution, source identification, and ecological risk assessment of heavy metals in wetland soils of a river–reservoir sys- Declaration of competing interest tem. Environ. Sci. Pollut. R. 24 (1), 436–444. Jiang, Y., Sun, Y., Zhang, L., Wang, X., 2020. Influence factor analysis of soil heavy metal Cd – fi based on the GeoDetector. Stoch. Environ. Res. Risk Assess. 34, 921 930. The authors declare that they have no known competing nancial Li, X., Xie, Y., Wang, J., Christakos, G., Si, J., Zhao, H., Ding, Y., Li, J., 2013. Influence of plant- interests or personal relationships that could have appeared to influ- ing patterns on fluoroquinolone residues in the soil of an intensive vegetable cultiva- ence the work reported in this paper. tion area in northern China. Sci. Total Environ. 458, 63–69. Li, Y., Chen, H., Teng, Y., 2020. Source apportionment and source-oriented risk assessment of heavy metals in the sediments of an urban river-lake system. Sci. Total Environ. Acknowledgements 737, 140310. Liang, J., Feng, C., Zeng, G., Gao, X., Zhong, M., Li, X., Li, X., He, X., Fang, Y., 2017. Spatial dis- fi This work was partially supported by the National Key Technology tribution and source identi cation of heavy metals in surface soils in a typical coal mine city, Lianyuan, China. Environ. Pollut. 225, 681–690. Research & Development Program of China (2018YFD0200500) and Liu, H., Zhang, Y., Zhou, X., You, X., Shi, Y., Xu, J., 2017. Source identification and spatial distri- the National Natural Science Foundation of China (41801302 and bution of heavy metals in tobacco-growing soils in Shandong province of China with – 41671399). The foundation had no involvement in the design of the multivariate and geostatistical analysis. Environ. Sci. Pollut. R. 24 (6), 5964 5975. Liu, J., Liu, Y.J., Liu, Y., Liu, Z., Zhang, A.N., 2018. Quantitative contributions of the major study, collection of the data, analysis or interpretation of the results, sources of heavy metals in soils to ecosystem and human health risks: a case study writing of the report, or the decision-making process for submission of of Yulin, China. Ecotox. Environ. Safe. 164, 261–269. the article for publication. Liu, S., Pan, G., Zhang, Y., Xu, J., Ma, R., Shen, Z., Dong, S., 2019. Risk assessment of soil heavy metals associated with land use variations in the riparian zones of a typical urban river gradient. Ecotox. Environ. Safe. 181, 435–444. References Lv, J., 2019. Multivariate receptor models and robust geostatistics to estimate source apportionment of heavy metals in soils. Environ. Pollut. 244, 72–83. Baltas, H., Sirin, M., Gokbayrak, E., Ozcelik, A.E., 2020. A case study on pollution and a Mamut, A., Eziz, M., Mohammad, A., Anayit, M., 2017. The spatial distribution, contamina- human health risk assessment of heavy metals in agricultural soils around Sinop tion, and ecological risk assessment of heavy metals of farmland soils in Karashahare- province, Turkey. Chemosphere 241, 125015. Baghrash oasis, northwest China. Hum. Ecol. Risk Assess. Int. J. 23, 1300–1314. X. Fei et al. / Science of the Total Environment 747 (2020) 141293 9

Ministry of Environmental Protection of China (MEP), Ministry of Land and Resources of Wu, J., Hu, Y., Zhi, J., Jing, C., Chen, H., Xu, J., Lin, H., Li, D., Zhang, C., Xiao, R., Huang, H., China (MLR), 2014. National soil pollution investigation bulletin. http://www.mep. 2013. A 1: 50 000 scale soil database of Zhejiang Province, China. Acta Pedol. Sin. gov.cn/gkml/hbb/qt/201404/t20140417_270670.htm. 50 (1), 30–40. Nanos, N., Martín, J.A.R., 2012. Multiscale analysis of heavy metal contents in soils: spatial Wu, Z., Chen, Y., Han, Y., Ke, T., Liu, Y., 2020. Identifying the influencing factors controlling variability in the Duero river basin (Spain). Geoderma 189, 554–562. the spatial variation of heavy metals in suburban soil using spatial regression models. Niu, L., Yang, F., Xu, C., Yang, H., Liu, W., 2013. Status of metal accumulation in farmland Sci. Total Environ. 717, 137212. soils across China: from distribution to risk assessment. Environ. Pollut. 176, 55–62. Xu, Y.Y., Shi, J., Zhou, L.Y., 2012. Characteristics of heavy metals distribution in agricultural Olea, R.A., 2006. A six-step practical approach to semivariogram modeling. Stoch. Env. soils of Hangzhou and its environment significances. Environ. Monit. China 28 (4), Res. Risk A. 20 (5), 307–318. 74–80. Qiao, M., Cai, C., Huang, Y., Liu, Y., Lin, A., Zheng, Y., 2011. Characterization of soil heavy Xu, D.M., Yan, B., Chen, T., Lei, C., Lin, H.Z., Xiao, X.M., 2017. Contaminant characteristics metal contamination and potential health risk in metropolitan region of northern and environmental risk assessment of heavy metals in the paddy soils from lead China. Environ. Monit. Assess. 172 (1–4), 353–365. (Pb)-zinc (Zn) mining areas in Guangdong Province, South China. Environ. Sci. Pollut. – Rachwa, M., Magiera, T., Wawer, M., 2015. Coke industry and steel metallurgy as the R 24 (31), 24387 24399. source of soil contamination by technogenic magnetic particles, heavy metals and Yang, S., Zhou, D., Yu, H., Wei, R., Pan, B., 2013. Distribution and speciation of metals (Cu, polycyclic aromatic hydrocarbons. Chemosphere 138, 863–873. Zn, Cd, and Pb) in agricultural and non-agricultural soils near a stream upriver from – Ren, Z., Xiao, R., Zhang, Z., Lv, X., Fei, X., 2019. Risk assessment and source identification of the Pearl River, China. Environ. Pollut. 177, 64 70. Yang, Q., Li, Z., Lu, X., Duan, Q., Huang, L., Bi, J., 2018. A review of soil heavy metal pollution heavy metals in agricultural soil: a case study in the coastal city of Zhejiang Province, China. Stoch. Environ. Res. Risk Assess. 33, 2109–2118. from industrial and agricultural regions in China: pollution and risk assessment. Sci. Total Environ. 642, 690–700. Shao, D., Zhan, Y., Zhou, W., Zhu, L., 2016. Current status and temporal trend of heavy Yoon, J., Cao, X., Zhou, Q., Ma, L.Q., 2006. Accumulation of Pb, Cu, and Zn in native plants metals in farmland soil of the Yangtze River Delta Region: field survey and meta- growing on a contaminated Florida site. Sci. Total Environ. 368 (2–3), 456–464. analysis. Environ. Pollut. 219, 329–336. Zang, F., Wang, S., Nan, Z., Ma, J., Zhang, Q., Chen, Y., Li, Y., 2017. Accumulation, spatio- Sun, L., Liao, X., Yan, X., Zhu, G., Ma, D., 2014. Evaluation of heavy metal and polycyclic ar- temporal distribution, and risk assessment of heavy metals in the soil-corn system omatic hydrocarbons accumulation in plants from typical industrial sites: potential around a polymetallic mining area from the Loess Plateau, northwest China. candidate in phytoremediation for co-contamination. Environ. Sci. Pollut. Res. 21 Geoderma 305, 188–196. (21), 12494–12504. Zhang, M.K., Ke, Z.X., 2004. Heavy metals, phosphorus and some other elements in urban Sun, L., Guo, D., Liu, K., Meng, H., Zheng, Y., Yuan, F., Zhu, G., 2019. Levels, sources, and spa- soils of Hangzhou City. China. Pedosphere 14 (2), 177–185. tial distribution of heavy metals in soils from a typical coal industrial city of Tangshan, Zhang, Z., Su, S., Xiao, R., Jiang, D., Wu, J., 2013. Identifying determinants of urban growth – China. Catena 175, 101 109. from a multi-scale perspective: a case study of the urban agglomeration around Teng, Y., Wu, J., Lu, S., Wang, Y., Jiao, X., Song, L., 2014. Soil and soil environmental quality Hangzhou Bay, China. Appl. Geogr. 45, 193–202. – monitoring in China: a review. Environ. Int. 69, 177 199. Zhang, R., Chen, T., Zhang, Y., Hou, Y., Chang, Q., 2020. Health risk assessment of heavy U.S. Environmental Protection Agency, 2014. EPA positive matrix factorization (PMF) 5.0 metals in agricultural soils and identification of main influencing factors in a typical fi fundamentals and user guide. https://www.epa.gov/sites/production/ les/2015-02/ industrial park in northwest China. Chemosphere 252, 126591. documents/pmf_5.0_user_guide.pdf. Zhao, L., Xu, Y., Hou, H., Shangguan, Y., Li, F., 2014a. Source identification and health risk Wang, J.F., Hu, Y., 2012. Environmental health risk detection with GeogDetector. Environ. assessment of metals in urban soils around the Tanggu chemical industrial district, Model. Softw. 33, 114–115. Tianjin, China. Sci. Total Environ. 468, 654–662. Wang, J.F., Li, X.H., Christakos, G., Liao, Y.L., Zhang, T., Gu, X., Zheng, X.Y., 2010. Geograph- Zhao, Y., Yan, Z., Qin, J., Xiao, Z., 2014b. Effects of long-term cattle manure application on ical detectors-based health risk assessment and its application in the neural tube de- soil properties and soil heavy metals in corn seed production in Northwest China. En- fects study of the Heshun Region, China. Int. J. Geogr. Inf. Sci. 24 (1), 107–127. viron. Sci. Pollut. Res. 21 (12), 7586–7595. Wang, S., Cai, L.M., Wen, H.H., Luo, J., Wang, Q.S., Liu, X., 2019. Spatial distribution and Zheng, H., Xing, X., Hu, T., Zhang, Y., Zhang, J., Zhu, G., Li, Y., Qi, S., 2018. Biomass burning source apportionment of heavy metals in soil from a typical county-level city of contributed most to the human cancer risk exposed to the soil-bound PAHs from Guangdong Province, China. Sci. Total Environ. 655, 92–101. Chengdu Economic Region, western China. Ecotox. Environ. Safe. 159, 63–70. Wang, Y., Duan, X., Wang, L., 2020. Spatial distribution and source analysis of heavy Zhou, J., Feng, K., Li, Y.J., Zhou, Y., 2016. Factorial Kriging analysis and sources of heavy metals in soils influenced by industrial enterprise distribution: case study in Jiangsu metals in soils of different land-use types in the Yangtze River Delta of Eastern Province. Sci. Total Environ. 710, 134953. China. Environ. Sci. Pollut. Res. 23 (15), 14957–14967.