You are on page 1of 14

Journal of Hazardous Materials 438 (2022) 129468

Contents lists available at ScienceDirect

Journal of Hazardous Materials


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

Research Paper

Apportionment and location of heavy metal(loid)s pollution sources for soil


and dust using the combination of principal component analysis,
Geodetector, and multiple linear regression of distance
Weibin Zeng a, b, Xiaoming Wan a, b, *, Lingqing Wang a, b, Mei Lei a, b, Tongbin Chen a, b,
Gaoquan Gu a, b
a
Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
b
University of Chinese Academy of Sciences, Beijing 100049, China

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• PCA- Geodetector-MLRD was used to


determine the location of the sources,
not just the type of source.
• The river, main road, mining area, metal
plant, and community can significantly
explain the PCA factors.
• A distance-based prediction model for
heavy metal concentration was estab­
lished and its effectiveness was verified.

A R T I C L E I N F O A B S T R A C T

Editor: Jörg Rinklebe The accurate identification of sources for soil heavy metal(loid) is difficult, especially for multi-functional parks,
which include multiple pollution sources. Aiming to identify the apportionment and location of heavy metal
Keywords: (loid)s pollution sources, this study established a method combining principal component analysis (PCA),
Dust Geodetector, and multiple linear regression of distance (MLRD) in soil and dust, taking a multi-functional in­
Ecological risk
dustrial park in Anhui Province, China, as an example. PCA and Geodetector were used to determine the type and
Heavy metal(loid)
possible location of the source. Source apportionment of individual elements is achieved by MLRD. The detection
Soil
Source identification results quantified the spatial explanatory power (0.21 ≤ q ≤ 0.51) of the potential source targets (e.g., river and
mining area) for the PCA factors. A comparative analysis of the regression equation (Model 1 and Model 3)
indicated that the river (0.50 ≤ R2 ≤0.78), main road (0.47 ≤ R2 ≤ 0.81), and mine (0.14 ≤ R2 ≤ 0.92) (p < 0.01)
were the main sources. Different from the traditional source apportionment methods, the current method could
obtain the exact contributing sources, not just the type of source (e.g., industrial activities), which could be useful
for pollution control in areas with multiple sources.

* Corresponding author at: Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China.
E-mail address: wanxm.06s@igsnrr.ac.cn (X. Wan).

https://doi.org/10.1016/j.jhazmat.2022.129468
Received 5 May 2022; Received in revised form 10 June 2022; Accepted 24 June 2022
Available online 27 June 2022
0304-3894/© 2022 Elsevier B.V. All rights reserved.
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

1. Introduction Therefore, this research has the following objectives: 1) to establish a


source analysis process from the screening of potential source ranges to
Heavy metal(loid)s are among the main pollutants of sites (Safari and the identification of specific source factors, 2) to identify potential
Delavar, 2019; Zhao et al., 2015; Wu et al., 2021). The land used by sources using quantitative methods, and 3) to directly identify the
industrial enterprises presents environmental problems, such as soil and location of pollution sources via the use of regression equations.
waste pollution (Pyatt et al., 2002; Z. Zhang et al., 2012a; Zhang et al.,
2012b; Jia et al., 2019; Dhaliwal et al., 2020). Thus, risk control stra­ 2. Method and material
tegies should be applied to these areas (Liu et al., 2020). The identifi­
cation of potential pollution sources is the basis of risk control, and the 2.1. Overview of the research area
source analysis of heavy metal(loid)s in soil has been one of the most
important topics in recent years (Yang et al., 2017, 2020; Hu et al., The study area was located in a prefecture-level city in the middle of
2020). Anhui Province on the lower reaches of Yangtze River (Fig. 1). This
Principal component analysis (PCA), geostatistics, absolute principal location has a subtropical humid monsoon climate, with a multiyear
component score-multiple linear regression (APCS-MLR) analysis, the average temperature and precipitation of 16.4 ◦ C and of 1400 mm,
positive matrix factorization (PMF) model, and the isotope method are respectively (Li et al., 2012; Wang et al., 2019). The area was rich in
commonly used in source analysis (Wang et al., 2021a; Yang et al., mineral resources and has a long mining history of pyrite, copper, and
2021a; Parra et al., 2014; Lv, 2019; Deng et al., 2020), including the other minerals. This industrial park was established in the beginning of
source apportionment of heavy metals or organic matter in sediments the 21st century. After years of construction and development, a leading
(Pekey and Doğan, 2013), agricultural soils (Guan et al., 2018), and industry focusing on copper processing and utilization and the me­
urban topsoils (Chen and Lu, 2018; Qiao et al., 2021; Paatero and chanical manufacturing of electronic components was formed. In the
Tapper, 1994). In recent years, APCS-MLR has been carried out often west, the park was bounded by rivers, which flow from south to
with the PMF model to obtain sources of contamination (Lv, 2019; Wu northwest. A mining area was situated on the upper reaches of this river.
et al., 2020; Zhang et al., 2020; Li et al., 2021). One of the advantages of The river water was used for irrigation of agricultural land in the area in
the PMF and APCS/MLR methods is that they do not rely on the chemical the early years. Subsequently, the farmland was transformed into con­
composition spectrum analysis of the source, which is highly convenient struction land and became part of the park. The industrial park has
and efficient (Lv, 2019). But most of the results obtained by the PMF or various functional zoning. In addition to industrial land, the area has
APCS/MLR methods pertain to the source type, and few pertain to the residential and school lands.
source location (Guan et al., 2018; Chen and Lu, 2018; Qiao et al., 2021;
Zhang et al., 2020; Li et al., 2021). For the risk control of a region, the 2.2. Sampling and analysis
type and location of the source are very important, especially when
multiple sources belong to the same type (Yang et al., 2021a). Therefore, Soil and dust are interconnected and may be sources of heavy metals
how to achieve a more operational and intuitive source apportionment to each other (Chang et al., 2009; Gabarrón et al., 2017a; 2017b; Han
should be emphasized. et al., 2021). Comparing the two with each other can be used to syn­
Considering these, a modified model (the MLR model of distance thesize the source results. In this study, 68 sampling points were set up.
from sampling sites to sources [PCA-MLRD]) was proposed (Huang Soil and dust were simultaneously collected from most of the points,
et al., 2018). PCA was a widely used method to determine source type. while only soil or dust was collected in the other points, that was, soil
The regression equation obtained by the combination of PCA and MLRD from 51 points and dust from 48 points. The sampling points basically
directly reflects the explanatory power of a certain source (Huang et al., covered the study area. In some areas, because the land has not been
2018). This differed from the PMF approach where the degree of match developed, the number of sample points is relatively small. For sensitive
between qualitative sources and factors was still need to be assessed receptors and enterprises with pollution potential, more sampling points
quantitatively. To obtain a better fitting model, Wang et al. (2020a) were set. The expected setting of dust and soil has a one-to-one corre­
conducted data conversion using the work of Huang et al. (2018) as a spondence. In the actual sampling process, the ground was covered by
reference and obtained a regression model with a higher degree of fit cements for a few points, and the soil near the dust points was selected.
than APCS-MLR and PMF. It can be found that the PCA-MLRD method Soil samples were collected from 0 to 30 cm layers with a stainless-steel
provides a new attempt to locate the source, but this method currently shovel. Dust samples were collected from roads or other ground covered
has few applications. And it was unclear which type of regression model by cements with clean plastic brooms and dustpans (Gabarrón et al.,
should be used. These existing studies mainly rely on the comparison of 2017a, b).
different models to verify results, while other verification and quanti­ Approximately 0.100 g of each homogenized sample was weighed
fication methods are rarely used. and then digested in a mixture of HF-HNO3-HClO4 on a hot plate. The
Thus, a quantitative approach, that is, the use of Geodetector, can be concentrations of heavy metal(loid)s (Cr, Ni, Cu, Zn, As, Cd, Pb, Sb, and
combined with source identification models, such as PCA (Fei et al., Mn) in digested solution were detected using inductively coupled
2020; Tao et al., 2020; Wu et al., 2021). The factor detection in Geo­ plasma-mass spectrometry (ICP-MS, Thermo Scientific, Waltham, MA,
detector can quantify the interpretation of factors from the perspective USA). The Hg content was determined using a DMA-80 mercury meter
of spatial heterogeneity (Wang et al., 2016; Wang and Xu, 2017). With (Wu et al., 2021; Zhang et al., 2021b). All of the vessels were soaked in
the help of the geo-detection results, the factors identified by the re­ 30 % HNO3 for 24 h before digestion and rinsed cleanly with ultra-pure
ceptor model can be further clarified. The PCA-MLRD model can water.
quantify the contribution of potential pollution sources in the regression The field survey and chemical analysis of soil and dust samples
equation. However, the combination of Geodetector and PCA-MLRD strictly followed the field and indoor work specifications (GB/T
models has been rarely studied. 36197–2018) (SAC, 2018). The reference materials (GSS-1 and GSS-5)
In this study, a multi-functional industrial park in Central Anhui from the Center of National Standard Reference Material of China,
Province, China, was selected as an example to locate the sources of blank samples, and three duplicates were used to ensure the reliability of
pollutants by combining Geodetector and PCA-MLRD. In this method, the analysis. The analysis errors of each element were less than ± 10 %.
the relationship between the specific pollution source and the pollutant The recovery percentages were within ± 10 %. And the correlation co­
concentration was expressed in an equation. Geodetector was intro­ efficient of the standard curve is greater than 0.999. Milli-Q deionized
duced in the screening of the potential sources of pollution to quantify water (18 MΩ⋅cm− 1) from Merck Millipore, USA, was used for complete
the possibilities, reducing the subjectivity of relying on PCA alone. analytical and cleaning purposes.

2
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

2.3. Potential ecological risk evaluation 2.4. PCA-Geodetector-MLRD

A new potential ecological risk assessment method called the NIRI 2.4.1. Principal component analysis (PCA)
was developed by Men et al. (2020) based on the nemerow integrated The element content was analyzed using the PCA method, and the
pollution index (NIPI) and potential ecological risk index (RI) (Hakan­ principal component factor scores and factor loadings of soil and dust
son, 1980; Huang et al., 2021; Roy et al., 2019; Luo et al., 2022). were obtained. The maximal rotation of variance method was used in
The advantage of this method over NIPI and RI was that it takes into this work. The concentration of the element was used as the initial value
account the effects of both the toxicity factor and the number of ele­ for PCA, and factor extraction with eigenvalues > 1 after varimax
ments. In this study, NIRI was used to assess the ecological risk of ten rotation was used for source identification. Factor loading was adopted
elements. The calculation equations are as follows: to determine the type and range of potential sources. The factor score
was used to determine the principal component factor with the factor
ci
Eir = T ir × , (1) detection in Geodetector. In this study, the spatial distribution of factor
cib
scores was obtained in conjunction with geographic information science
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (GIS). The scores were divided into five levels (Zones 1–5) on the map.
)2 ( )2 /
NIRI =
(
[ Eir max + Eir average ] 2 , (2) Zones 4 and 5 are the high-scoring zones, and the main source targets
were screened within these two zones.

where NIRI: Comprehensive ecological risk index of various heavy metal 2.4.2. Geodetector modeling
(loid)s in soil or dust. A common submodule, namely, the factor detector, was used in this
Eir : Potential ecological risk coefficient of the heavy metal(loid)s study (Wang et al., 2010, 2016). The factor detector uses the q value to
element i in soil or dust. measure the degree of influence of a certain factor (X) on dependent
T ir : Toxicity coefficient of heavy metal(loid)s element i. variable Y. Here, factors X1, …, Xn represented the distance from the
ci : Heavy metal(loid) element i content. sampling point to the potential source (e.g., river).
cib : Calculated required reference values (select local soil background ∑L 2
h=1 Nh σ h
values for comparison) of heavy metal(loid)s element i. qx = 1 − , (3)
N σ2
NIRI: Comprehensive ecological risk index of various heavy metal
(loid)s in soil or dust. where qx is an index value used to measure the spatial association
The toxicity coefficients of Cr, Ni, Cu, Zn, Hg, As, Pb, Cd, Mn, and Sb between X and Y; h = 1, 2, 3,., L, where L is the number of strata
are 2, 5, 5, 1, 40, 10, 5, 30, 1, and 10, respectively (Hakanson, 1980; (subregions or subclasses) of factor X; and N and Nh are the number of
Zhang et al., 2019; Huang et al., 2021). For the local background soil, we samples in the study area and in each stratum h, respectively. Symbols σ2
referred to the study of Li et al. (2012) and Wang et al. (2019). The and σ2h represent the variances of Y in the entire region and in each
background values of Cr, Ni, Cu, Zn, Hg, As, Pb, Cd, Mn, and Sb are 80.0, stratum (h), respectively. The value of qx∈[0,1] indicates how much Y is
34.0, 34.0, 79.0, 0.06, 13.8, 29.0 0.23, 598, and 1.00 mg/kg, respec­ influenced by X. The larger the value of qx is, the stronger the spatial
tively. According to Men et al. (2020), the evaluation criteria for the risk association between X and Y is (Tao et al., 2020).
level were shown in Table S1. In the factor detection of this research, Y is the score of each factor
obtained by the PCA of soil and dust, and X is the distance of the po­
tential sources determined initially. The type and location of potential
sources were determined (Section 2.4.1) based on the spatial distribu­
tion of factor scores. Geodetector was obtained from the website htt

Fig. 1. Study area and location of sampling points.

3
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

p://geodetector.cn/. 2.5. Data analysis

2.4.3. Multiple linear regression of distance (MLRD) Source apportionment and the corresponding uncertainty analysis
Using PCA to identify potential sources, the source contribution can were also performed using the PMF model (Version 5.0) (USEPA, 2014).
be calculated based on the heavy metal(loid) content and the MLR of the The calculation method of the PMF model can be seen in the supple­
distance from the source. The content of heavy metal(loid)s theoretically mentary material. The data processing in this study was completed on
decreases with increasing distance from the source. Considering this, the Excel 2016 and SPSS 24. Graphs were created in Origin 2021 and ArcGIS
distance from the sampling point to the emission source can be used to 10.6. The shortest distance between the sampling point and the source
quantify their contribution. Here we propose three possible regression used for analysis in Geodetector and MLRD was calculated by the
models. These three models are common functional forms. At the same analysis tools of ArcGIS 10.6 (Huang et al., 2018).
time, it refers to the data transformation attempts proposed in the
existing source resolution research (Huang et al., 2018; Yang et al., 3. Results and discussion
2020). As the independent variable increases, the dependent variable
gradually decreases. Since the specific relationship between the source 3.1. Descriptive statistics of heavy metal(loid)s in soil and dust
and heavy metals was not very clear. The final source apportionment
model will be determined based on the validation results. The following Table 1 presented the descriptive statistical results of the heavy
three regression models were used in this study: metal(loid) content of soil and dust. The contents of 10 heavy metal
∑p (loid)s varied greatly. The maximum concentration can be 100 times the
1
Cik = Bin , (4) minimum value, whereas the average heavy metal(loid)s content was
n=1 Dnk
higher than the local background value. The As content exceeded the
∑p screening value (MEE, 2018).
Model2 : Cik = Bin exp( − Dnk ), (5)
n=1 The variation coefficients of the soil’s Cd, Zn, Cu, Mn, As, and Pb
∑p were 1.84, 1.69, 1.55, 1.46, 1.27, and 1.04, respectively, indicating that
Model3 : Cik = n=1
Bin 0.5Dnk , (6) these elements fluctuated within a wide range and exhibited a strong
spatial heterogeneity. The fluctuation range of most of the dust metal
where Cik is the content of element i in sampling site k, Dnk is the distance (loid)s in this region, such as Ni, As, and Pb, was near that in soil. The
from sampling site k to source n, Bin is the regression coefficient matrix metal(loid) elements in the dust with a strong spatial heterogeneity were
of distance Dnk , and n is the number of sources, p is the number of Hg, Cu, As, and Cr.
pollution sources by the regression. The results of other studies near the same area (industrial park) were
MLR uses a stepwise regression method to avoid collinearity, 80 % of compared with those of the current study. The maximum Cu, Zn, and Cd
the data was used as the training set for the construction of three models, contents in this study were higher than those (340.43, 859.25, and
and 20 % of the data was used as the validation set. The R2 and the RMSE 10.20 mg/kg, respectively) obtained by Wang et al. (2019). The Hg
were used as the metrics to evaluate the models. The evaluation results content in this study (0.01–0.13 mg/kg) was similar to that obtained by
served as the basis for selecting the best models for soil and dust. Ac­ Wang et al. (2019) (0.03–0.17 mg/kg). The differences may be related to
cording to Huang et al. (2018) and Wang et al. (2020a), R2 > 0.5 and the dispersed sampling sites of Wang et al. (2019). In their study, 29 soil
p ≤ 0.01 were used to filter all element models. Then, the source results sampling sites covered most of the entire city, with approximately six
were compared with the results of the PMF model. A detailed flow chart sampling sites near the mining area. By contrast, all the soil samples in
can be seen in Fig. 2. this study were obtained adjacent to the industrial park. Shen et al.

Fig. 2. Flowchart of PCA-Geodetector-MLRD model.

4
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Table 1
Descriptive statistics of heavy metal(loid)s in soil and dust.
Cr Ni Cu Zn Hg As Pb Cd Mn Sb

Soil Minimum (mg/kg) 47.86 13.24 43.70 70.93 0.01 12.21 20.95 0.09 105.40 0.01
Maximum(mg/kg) 200.32 61.28 1371.53 2785.99 0.13 228.11 363.59 23.20 12,520.00 11.88
Mean(mg/kg) 88.92 31.28 249.70 431.86 0.04 45.49 97.35 3.09 1707.14 3.08
SDa 27.27 12.63 385.89 730.81 0.03 57.81 100.97 5.70 2487.80 2.66
CVb 0.31 0.40 1.55 1.69 0.76 1.27 1.04 1.84 1.46 0.86
Skewness 1.90 0.46 2.03 2.27 1.30 2.06 1.49 2.07 2.79 1.18
Kurtosis 4.96 -0.31 2.55 3.81 0.55 2.86 1.04 3.24 8.14 1.38
Dust Minimum (mg/kg) 35.62 13.24 51.70 83.12 0.02 4.35 23.90 0.09 585.74 0.72
Maximum (mg/kg) 955.92 72.09 1822.38 895.62 3.41 279.39 279.79 3.82 2348.77 21.00
Mean (mg/kg) 138.60 33.41 228.50 350.63 0.27 63.18 72.93 1.21 1258.74 5.70
SD 133.35 14.16 263.66 200.06 0.60 63.41 40.40 0.93 411.21 4.43
CV 0.96 0.42 1.15 0.57 2.23 1.00 0.55 0.77 0.33 0.78
Skewness 5.14 0.61 4.97 0.79 4.57 2.12 2.95 0.83 0.82 2.13
Kurtosis 31.03 -0.33 29.31 0.18 21.16 4.43 14.00 -0.02 0.32 5.00
Risk screening value (mg/kg) 200.00 900.00 18,000.00 – 38.00 60.00 800.00 65.00 180
Soil background value(mg/kg) 80.00 34.00 34.00 79.00 0.06 13.80 29.00 0.23 598.00 1.00

Risk screening value refer to MEE (2018); soil background value refer Li et al. (2012) and Zhang, . et al. (2019)
a
SD means standard deviation
b
CV means coefficient of variation

(2019) studied the heavy metal(loid)s in the farmland around a mining the soil sampling site) were higher than those in soil. They found that the
area and found that the average concentration of Cd in more than 82 % difference was mainly caused by the transportation and sedimentation
of the soil samples exceeded the screening value (0.3 mg⋅kg− 1) in Chi­ of dust. The differences in ecological risks between soil and dust indi­
nese Environmental Quality Standard for Soils (SAC, 2018). And the cated the requirement for accurate source identification.
concentrations of all heavy metal(loid)s in the topsoils were higher than
those of the subsoils.
3.3. Identification of the potential sources of pollution

3.2. Characteristic of ecological risks 3.3.1. Identification of principal components by PCA


Based on the 10 heavy metal(loid)s, two principal component factors
For Cu and Cd, some samples showed a high or very -high risk were screened for soil, and three principal component factors were
ecological index (Fig. 3(a), Table S1), while for other elements, the screened for dust (Table 2). In the soil, Factor 1 explained the majority of
potential ecological risk index (Eir ) was generally low (Fig. 3a). Cd was a the nine heavy metal(loid)s, namely, Cr, Cu, Zn, Hg, As, Pb, Cd, Mn, and
high-risk pollutant in several regions of China (Wu et al., 2022). Sb, with coefficients of 0.82, 0.96, 0.97, 0.87, 0.97, 0.92, 0.96, 0.94, and
Smelting activities and inappropriate agricultural measures were the 0.79, respectively. However, among them, the coefficients of Cr, Hg, and
major sources of Cd pollution (Zhang et al., 2012a; 2012b; Han et al., Sb were slightly lower than those of other elements. Taking Cr as an
2021; Wu et al., 2022). The reference value of Cd in the study area was example, its factor 2 coefficient was 0.43. This shows that factor 2 also
0.23 mg/kg (Li et al., 2012), while the Cd content in the soil was up to has a certain explanatory effect on Cr. The results for Hg and Sb were
23.20 mg/kg. This directly leaded to the higher ecological risk of Cd. similar to those for Cr. This result indicated that the heavy metal(loid)
And the high content of Cd was speculated to be highly related to local elements in the soil were mainly from one source (Factor 1), but some of
copper mining and smelting activities (Wang et al., 2004). The risk them also were influenced by Factor 2.
levels of most heavy metal(loid)s in the dust samples were higher than in Section 3.1 showed that the background values of soil heavy metal
the soil (Fig. 3b). This trend was obvious for Hg, showing a significantly (loid)s in the study area were not high. And the main local industry was
higher risk in the dust samples than in the soil samples. The numbers of copper mining and processing. This industry may contribute to the
sampling points with As in the low- and medium-risk range were rela­ diffusion and pollution of heavy metal(loid)s from atmospheric depo­
tively large. The comprehensive potential ecological risk index (NIRI) sition, transport vehicles, and rivers (Xu et al., 2013; Shen et al., 2019).
showed the overall characteristics of the soil and dust elements content Therefore, Factor 1 was presumed to be an anthropogenic source,
(Fig. 3c). The NIRI of 1/3 of the points of the soil was above the medium associated with smelting activities. And the exact type of it needed to be
risk (Table S1). A few points have a high NIRI, reaching 2155.21. The determined in the next step. As seen in Table 2, Factor 2 explained the
proportion of the dust points above the medium risk was high, reaching source of Ni (0.94). Several studies concluded that the Ni in soil mainly
1/2, but the highest NIRI was lower than that of soil, which was originated from the soil parent material (Guan et al., 2019; Wang et al.,
1618.80. Few points have a high dust NIRI. 2020b; Liu et al., 2021). Here, factor 2 may represent natural sources.
Several heavy metal(loid)s exhibited high-risk levels in the dust Combined with GIS technology, the spatial distribution map of soil
(represented by Hg) and soil (represented by Cd). The apparent differ­ element principal component factors 1 and 2 was obtained (Fig. 4(a) and
ence between the risks exhibited by soil and dust indicated that the (b)). Based on the factor scores, the interpolation area could be divided
sources of heavy metal(loid)s in dust and soil might be different. into five zones. In Fig. 4(a), the sources related to factor 1 were most
Gabarrón et al. (2017b; 2017a) conducted several detailed studies on the likely to be distributed in Zones 4 and 5 (high scores). A comparison of
heavy metals in soil and road dust. They found that heavy metals in road the site survey results showed that the area was near the river, and no
dust have a higher health risk than those in soil. The sources of heavy other obvious source of pollution was identified. Sun (2013) and Kuang
metals included the leather industry, smelters, and fertilizer industries. et al. (2018) have found that the heavy metal(loid)s in the river water
Chang et al. (2009) found that soil and road dust have different sources were obviously high and the highest concentrations of Cu, Zn, Cd, and As
of heavy metal(loid)s, and the content of heavy metal(loid)s in the dust can reach 5204.30, 2368.90, 222.50, and 310.20 μg/L. Hence, the river
was higher than that in soil. Another study (Long et al., 2021) from a city was first included as potential sources of soil pollution.
in Southwestern China showed that the heavy metal(loid)s in dust (the The study area was an industrial park, mainly for copper processing,
dust from road guardrails, air conditioners, and residents’ balconies near and the mines outside the park were mainly for copper production

5
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Table 2
Rotating component matrix obtained by PCA.
Component of soil Component of dust

Factor 1 Factor 2 Factor 1 Factor 2 Factor 3

Cr 0.828 0.43 0.188 0.79 -0.015


Ni -0.152 0.938 -0.183 0.741 0.125
Cu 0.96 -0.197 0.285 0.006 0.895
Zn 0.969 -0.073 0.848 0.116 0.363
Hg 0.873 -0.288 -0.213 0.016 0.911
As 0.969 -0.164 0.617 0.646 -0.016
Pb 0.918 -0.253 0.836 0.008 -0.046
Cd 0.956 -0.162 0.828 0.075 -0.043
Mn 0.935 0.16 0.682 0.393 -0.046
Sb 0.783 -0.318 0.44 0.686 -0.104

Elements As and Sb were explained by Factors 1 and 2. The sources of As


and Sb were also closely related to human activities (Lv, 2019).
Fig. 4(c), (d), and (e) were the interpolation results of the principal
component factors of the dust elements. The zones with a high Factor 1
score were scattered and distributed in the south, southwest, and north
of the study region. Within these high-scoring areas, two targets were
identified: mining (southwest) and community (south). However, no
more obvious targets were found in the high-scoring area in the north.
The interpolation results of Factor 2 show that the high-scoring areas
(Zones 4 and 5) were located near the main roads. Fig. 4(e) showed an
obvious target, that was, the metal plant, in the high score area (Zone 4).
Therefore, combining the soil and dust results, the PCA approach
helped obtain the potential source range, including five targets, namely,
river, mining area, community, main road, and metal plant.
The possibility of the above sources was determined through existing
research and professional knowledge, which might be subjective. Ac­
curate judgement was difficult when multiple factors have strong in­
terpretations. Therefore, a quantitative method (Geodetector) was
further introduced to reveal the explanatory power of specific sources.
On the basis of the above discussion, five specific targets were summa­
rized as the next source identification, namely, metal plant, river, main
road, mining area, and community. The shortest distances from the five
targets to the sampling points were set to X, and the PCA factor scores
were set to Y.

3.3.2. Spatial detection of pollution sources by Geodetector


The factor detection in Geodetector matched the above potential
source range and principal component factors from the perspective of
spatial heterogeneity, verifying the accuracy of the principal component
analysis results. Fig. 5 presented the result of factor detection. In the
matching results of soil Factor 1, river, metal plant, and community have
a significant (p < 0.05) explanatory power. That was, the spatial dis­
tribution of the three was similar to soil Factor 1. According to the
distribution map of the principal component factors (Fig. 4), when the
point was near the river, the factor score showed an increasing trend, but
when the point was near the metal plant and the community, the factor
score was low. This meant that the river matched the actual meaning of
the source of Factor 1. Soil Factor 2 was previously judged to be a nat­
ural source, and the factor detection results showed that only the mining
Fig. 3. Ecological risks of heavy metals in soil and dust.
area and soil Factor 2 had a certain matching relationship among the
five targets.
(Wang et al., 2019; Wu et al., 2021). The explanatory factors corre­ The spatial distribution of dust Factor 1 was scattered, and the target
sponding to the 10 heavy metal(loid)s in dust were more scattered than of the main road has a certain explanatory effect on dust Factor 1 (q =
those in soil. Cr and Ni were mainly explained by factor 2 (the co­ 0.23, p < 0.05). This finding was consistent with the principal compo­
efficients were 0.79 and 0.74, respectively). Cu (0.89) and Hg (0.91) nent results in Section 3.3.1. The detection results of dust Factor 2
were mainly explained by Factor 3. The source of Cu may be highly showed that many targets have a similar spatial distribution with dust
related to local copper mining and processing. Factor 2. In addition to the main road, metal plant, community, and
Factor 1 of the dust elements explains most of the sources of Zn, Pb, river have a significant explanatory power (p < 0.05). The geographical
Cd, and Mn (Table 2). Pb was considered to be an important indicator of location of the first three overlapped with the high score of dust Factor 2,
the transportation source because the lead in gasoline was one of the but the river was located in the low score of dust Factor 2. The reason
main sources of the lead in soil (Cui et al., 2018; Zhang et al., 2021a). why rivers have a certain explanatory role was similar to soil Factor 2

6
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Fig. 4. Spatial distribution of principal component factors of soil (a and b) and dust elements (c, d, and e).

above. The detection results of dust Factor 3 did not show an obvious result will also provide a basis for the next step of model building.
target, and the only target with a significant explanatory power
(p < 0.05), the river, was supposed to explain the region with a low dust
Factor 3, which does not belong to the source range. The detection re­ 3.4. Sources apportionment by the establishment of the regression model
sults of dust Factor 3 must be further revealed. In general, the factor based on distance
detection in Geodetector basically verified the judgment results of the
principal component analysis in Section 3.3.1. That was, the potential 3.4.1. Comparison and screening of model performance
source range composed of five targets (metal plant, river, main road, Three regression methods were used for the source apportionment of
mining area, and community) has significant explanatory power. This soil and dust elements. The regression coefficient (R2) of the 10 elements
in the soil and dust training sets was shown in Fig. 6. The fitting result of

7
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Fig. 5. Detection results (q value) of principal component factors types using Geodetector. The shortest distances from the five targets (metal plant, river, main road,
mining area, and community) to the sampling points are set to X, and the PCA factor scores are set to Y.

the three regression methods was different among different elements. respectively, and the fitting coefficient of Model 3 was slightly higher
For example, the fitting effect of Model 2 of Zn (R2 =0.622) was higher than those of the other two models. Overall, most elements have an R2
than that of Models 1 and 2 (R2 =0.614, 0.581). However, the R2 of the greater than 0.5. Similarly, the training set regression coefficient R2 of
three Hg models (Models 1, 2, and 3) were 0.751, 0.763, and 0.778, Pb, Zn, and Cu in dust exceeds 0.6. The R2 of Mn was higher than those

Fig. 6. Comparison of three models for training set of soil (a) and dust (b) elements.

8
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

of the other elements. In terms of soil and dust, although the three can explain the source of most elements (Huang et al., 2018; Wang et al.,
regression methods were different, one model shows no absolute 2020a), and the highest was 0.791 (Pb).
advantage over the other two. The model to be selected for subsequent Regression equations were also developed for the 10 heavy metal
source apportionment should also be evaluated by the validation results. (loid)s in the dust. Droad, Dmine, Dmetal, and Dcommunity represent the
Table 3 shows the validation results of soil and dust. R2 and RMSE distance from the sampling site to the main road, the mine, the metal
were used to measure the regression effects of the three methods. For plant, and the community, respectively. An obvious feature was that the
soil, the R2 values of Cu, Zn, Hg, As, Pb, Cd, Mn, and Sb obtained by sources of most elements in dust were different from those in soil, and
Model 1 were higher than those obtained by Models 2 and 3. The RMSE the source amount was higher. The main road, community, and mine
values of these elements were lower than those of the other two models. were the primary sources of heavy metal(loid)s in dust, with the main
For example, the R2 of As Model 1 was 0.80, which was higher than road acting as the main source of eight heavy metal(loid)s, including Cu,
those of Models 2 (0.71) and 3 (0.57). The results show that the fitting Zn, As, Pb, Cd, Sb, and As. Community plays an important role in
method of model 1 has more advantages than those of Models 2 and 3. explaining the source of Cr, As, and Mn in the dust (Table S5).
The fitting effect of the Cr and Ni validation set differed greatly from that The results of the heavy metal(loid)s source analysis for soil and dust
of the training set, indicating that the two elements may be over-fitting reveal that dust was a different and more diverse source of heavy metal
to a certain extent. In Table 3, the R2 of soil Cr and Ni was lower than (loid)s than soil. The community represents the living behavior of the
that of other elements. The most important reason may be that their inhabitants of the study area, suggesting that heavy metal(loid)s affect
sources were from natural sources. The results of the PCA (Table 2) human health in such a multifunctional park and human living behavior
indicated that the natural source (soil factor 2) explained most of the Ni may contribute to the heavy metal(loid)s in soil and dust.
and part of the Cr sources. This was consistent with the results of several In addition, the metal plant was not identified as a major source of
existing studies that elements grouped with Cr and Ni were often soil heavy metals in the regression model. This differed from previous
considered to be associated with natural sources (Cai et al., 2019; Wang studies in which metal plants caused serious contamination in the sur­
et al., 2020b; Liu et al., 2021). rounding area (Roels et al., 1978; Izydorczyk et al., 2021). A major
For dust, the R2 of Ni, Cu, Zn, Mn, and Sb of Model 3 was higher than reason is that this plant is mainly engaged in metal-processing instead of
those of Models 1 and 2. The R2 of Model 1 was high in Hg and Cd, and metal-smelting. The mining and smelting of copper ore in the study area
the R2 of Model 2 was high in Cr, As, and Pb. Accordingly, the regression was carried out in the mine area while the metal plant in the park mainly
equation (RMSE) value of the dust element also presented a similar performs post-processing. Compared to mining and smelting,
trend. Therefore, based on the validation results, the soil Model 1 metal-processing plant usually presented a lower, heavy metal pollution
method was selected, while the dust Model 3 method was selected for and risk (Hutter et al., 2016b; Klasson et al., 2016).
subsequent source analysis. Also, in this study, the shortest distances were chosen in both the
Geodetector and MLRD. Considering that the line source of the road
3.4.2. Sources of elements identified by the regression models affects the heavy metal content through atmospheric deposition, this
The capability to quantitatively identify pollution sources was one of process was affected by the wind direction (Lv, 2019; Zhang et al.,
the most important features of distance-based regression models 2019). And when calculating the distance between the road and the
compared with the PMF model. Fig. 7 showed the regression model for receiving point, the distance was not calculated according to the wind
heavy metal(loid)s in soil and dust (Table S4-5). direction. This may add uncertainty to the results.
Fig. 7 showed the sources of soil and dust elements obtained by PCA-
Geodetector-MLRD. The river and mining areas were the main factors 3.5. Comparison of the established PCA-Geodetector-MLRD with the
that explain the source of heavy metal(loid)s in soil. The river was the commonly used PMF in terms of source identification
most likely sources of pollution by Cu, Zn, Hg, As, Pb, Cd, and Mn. The
mining area was the most important source of Sb and explains part of the 3.5.1. PMF model results
sources of Cr, Ni, and Mn. R2 was used to measure the degree of Fig. 8 showed the results of the PMF model. Fig. 8(a) showed the
explanation of the variables using the regression equation. The R2 values analytical results of soil. The explanatory power of Factor 1 on Cr, Ni,
of Cu, Zn, Hg, As, Pb, Mn, and Sb exceed 0.5, implying that the model Hg, and Mn was evidently stronger than that on the other elements. The
Cr and Ni in soil mainly originated from parent materials (Jiang et al.,
Table 3 2017; Liu et al., 2021). Factor 1 was regarded as the parent material of
Validation results of soil and dust elements regression models. soil formation. Factor 2 has a certain explanatory effect on all metals,
except Sb (Zhuo et al., 2019). Zhang et al. (2019) believed that Sb was
R2 RMSE
more likely to come from industrial activities. Here, Factor 2 was also
model 1 model 2 model 3 model 1 model 2 model 3 speculated to be an industrial source. Moreover, the specific source
Soil Cr 0.02 0.14 0.16 17.177 20.894 18.234 categories should be further studied. Factor 3 has a similar contribution
Ni 0.06 0.05 0.07 15.271 15.717 14.231 to Cr, Ni, Cu, and As, that was, approximately 40 %. The Cr and Ni in soil
Cu 0.78 0.63 0.50 194.409 226.523 241.881
were considered to be mainly from natural sources (Jiang et al., 2017;
Zn 0.78 0.65 0.46 143.902 285.916 208.427
Hg 0.69 0.18 0.17 0.014 0.023 0.022 Liu et al., 2021). Meanwhile, Cu and As were considered to be from
As 0.80 0.71 0.57 13.041 22.318 20.609 anthropogenic sources (Fei et al., 2020). The type of Factor 3 was un­
Pb 0.75 0.63 0.55 41.817 49.557 55.277 clear. Factor 4 strongly explained Cu, Zn, As, Pb, Cd, and Mn (62.1 %,
Cd 0.76 0.53 0.44 2.670 2.956 3.180 53.9 %, 37.4 %, 54.3 %, 72.2 %, and 35.4 %, respectively). Agricultural
Mn 0.71 0.19 0.48 538.040 752.680 640.201
Sb 0.25 0.01 0.21 1.755 2.195 2.642
activities, such as phosphate fertilizers, pesticides, organic fertilizers,
Dust Cr 0.04 0.40 0.17 64.838 42.472 58.724 and sewage irrigation, usually provided large amounts of Cd (Shao et al.,
Ni 0.33 0.61 0.68 11.740 10.393 9.155 2016; Fei et al., 2020). Multiple studies using the PMF model for heavy
Cu 0.57 0.69 0.70 126.601 138.350 124.671 metal(loid) source analysis showed that the factors with a strong
Zn 0.33 0.77 0.75 71.073 54.293 43.169
explanatory effect on Cd, Cu, and As elements were likely to originate
Hg 0.58 0.51 0.50 0.007 0.041 0.027
As 0.36 0.58 0.57 9.857 13.445 11.828 from agricultural activities (Fei et al., 2020; Liu et al., 2021). In this
Pb 0.50 0.77 0.74 6.087 15.027 9.081 study, the industrial park used to be an agricultural land in the early
Cd 0.46 0.22 0.21 0.147 0.315 0.235 stage. Therefore, Factor 4 may represent agricultural activities.
Mn 0.77 0.68 0.86 182.800 254.455 140.290 Fig. 8(b) shows the analytical results of heavy metal(loid)s sources in
Sb 0.59 0.53 0.74 0.751 1.135 0.747
the dust. Factor 1 explains some of the sources of Ni, Mn, and Sb.

9
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Fig. 7. R2 is determined by the coefficient of determination from the regression equation for each element.
(a) Sources of soil and dust elements obtained by PCA-Geodetector-MLRD. (b) (a): the sources of soil elements; (b): the sources of dust elements.

Previous studies (Filella et al., 2009; Mandal et al., 2021) pointed to likely to represent the atmospheric deposition caused by human activ­
natural factors, such as rock weathering, and anthropogenic factors, ities. Factor 3 contributed another part of the source of Ni and Hg. Ni
such as mining and smelting activities, vehicle emissions, and agricul­ was also considered to have a partial relationship with natural sources.
tural fertilizers, as potential sources of Sb. The contribution of Factor 1 According to the above analysis, Hg may come from human activities.
to other elements was also low, so Factor 1 may represent a natural Therefore, the type of Factor 3 was unclear. Factor 4 mainly explains the
source. The explanatory power of Factor 2 was obviously stronger than sources of Cu and Zn. Local copper mining and smelting activities may
that of Factor 1, which plays a major role in explaining the sources of Zn, result in high levels of Cu in the dust. Factor 4 may be related to mineral
Hg, Pb, Cd, Mn, and Sb. The contribution ratio reached 25 %, 53 %, 50 smelting activities. Factor 5 explains the main sources of Cr, As, and Cd
%, 45 %, 35 %, and 54 %. Hg was considered to be closely related to as well as Cu and Zn. As and Cd were important indicators used to judge
atmospheric deposition (Zhang et al., 2019). Pb was often regarded as an anthropogenic sources (Yoon et al., 2022). Agricultural activities, such
important indicator of traffic source (Lv, 2019). Therefore, Factor 2 was as the application of pesticides and sewage irrigation, often provide

10
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Fig. 8. Source analysis results of soil (a) and dust (b) heavy metal(loid)s by the PMF model.

large amounts of Cd (Lv, 2019). As was associated with mining activ­ The field investigations in this area showed that the soil in the industrial
ities, the use of phosphate fertilizers, the herbicides and pesticides in park of the study area was used as agricultural land in the early stage,
agricultural areas (Yoon et al., 2022). In this study, anthropogenic and this area was near a river flowing upstream through the mining area
sources that can discharge these substances included early agricultural (Wang et al., 2019). Sun (2013) combined isotope and scanning electron
activities (such as sewage irrigation) and wastewater and exhaust microscopy to test the heavy metals content in the river and found that
emissions from mining and smelting activities. Thus, Factor 5 may more than 55 % of the points were seriously polluted. Another local
represent agricultural activity. For the uncertainty of the result, Qtrue study (Jiang et al., 2018) showed that the Cd in the soil was mainly from
and Qrobust are two sets of Q data used in the PMF model (Wang et al., mining and irrigation with contaminated water. Xu et al. (2013) and
2021b). The number of factors was determined by the lowest Qrobust. The Shen et al. (2019) found that local mining activities were the main
error estimate contributed by the PMF model source was determined by reason for the high content of heavy metals in soil and river. Therefore, it
the DISP and BS methods (Figs. S1–4). The errors from the base run were is suggested that one of the pollution sources for the soil heavy metal
all within the interquartile range (25th–75th) of the BS and DISP runs, (loid)s pollution is irrigation by river water carrying heavy metal(loid)s.
indicating that the fitting errors of the PMF base run were all within the Multiple factors were identified by the PMF method. Unlike PCA-
acceptable range (Paatero et al., 2003; Yang et al., 2021b). Geodetector-MLRD, these factors need to be further defined and speci­
fied (Fig. 9). For instance, Factor 4 of soil heavy metal(loid) identified by
3.5.2. Comparison of source identification results by the PCA-Geodetector- the PMF model, which has a strong interpretation for Cu, Zn, As, Pb, and
MLRD and PMF model Cd, has a high probability of being an anthropogenic source. According
The heavy metal(loid) sources identified by the PCA-Geodetector- to Shao et al. (2016) and Fei et al. (2020), agricultural activities may
MLRD and PMF models were compared (Fig. 9). According to the re­ provide large amounts of Cd. And multiple studies suggested that the
sults of the PCA-Geodetector-MLRD model, the river was the most factors contributing to Cd, Cu, and As were mainly related to agricul­
important pollution source of heavy metal(loid)s in soil, showing a tural activities (Fei et al., 2020; Liu et al., 2021). Combined with the
strong explanatory power for the sources of Cu, Zn, Hg, As, Pb, and Cd. field survey results of the park, this source could be an agricultural

11
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

Fig. 9. Comparison of source identification results between PCA-Geodetector-MLRD and PMF model.

activity (Liu et al., 2021). 4. Conclusion


The comparison above showed that there were differences in the
sources obtained by the two methods. The main reason was that Geo­ In this study, the PCA-Geodetector-MLRD method was used for
detector and MLRD can express the potential targets and locations cor­ source identification. The PCA-MLRD approach uses a quantitative
responding to the source types identified by PCA. Therefore, the PCA- rather than just a qualitative approach to identify sources, and the
Geodetector-MLRD method was considered to have certain advantages combination with Geodetector further reduces the subjectivity in
in source location identification over the PMF model. screening potential sources. The source identified by this method was
While, the PCA-Geodetector-MLRD method also has shortcomings, more specific than the PMF method. The combination of Geodetector
such as the inability to quantitatively analyze some nonpoint sources, and PCA-MLRD was applied to the multifunctional industrial park,
including the influence of soil parent materials and atmospheric depo­ enabling the source identification and localization of heavy metal(loid)
sition. The study attempted to convert nonpoint sources to point source sources. The PCA-Geodetector-MLRD method used in this study can
treatment. A mining area and a metal plant were identified as targets to provide new information for source analysis methods, which will also
represent partial atmospheric subsidence. Furthermore, the living provide a foundation for the subsequent development of more complete
behavior of residents in the industrial park may also be a nonpoint and accurate source identification methods.
source. The distance to the community was selected to represent this Further, the risk levels of soil differ at various levels. Most heavy
factor. The PCA results can indicate the influence of natural sources metal(loid)s, such as Hg and As, show a higher concentration in dust
(such as soil parent material), but the importance of this factor did not than in soil. However, Cd and Pb show a higher concentration in soil
receive attention in the subsequent research steps. This shortcoming of than in dust. The main sources were also different. The heavy metal
PCA-Geodetector-MLRD should be addressed in future work. (loid)s in soil mainly come from the river, whereas the metal(loid)s in
In general, for the risk management of a certain area, the source dust were associated with the road and the mine. The difference between
might be difficult to manage when only the type of the source was soil and dust indicates that the heavy metal(loid) sources were complex
known (the location was uncertain). Because some areas may have inside and outside the park. The current study can provide useful in­
several sources that belong to the same category. For example, the formation for environmental management.
mining area and metal plant in this study were all industrial activities.
The sources obtained by the PMF model were difficult to use for further Environmental implication
source location. But the PCA-Geodetector-MLRD method can be used
with distance to quantify whether a possible source (with known loca­ The existing source analysis methods rely on subjective judgments
tion) was the main source factor. This approach has strong operability when concretizing the main factors. This study proposed a modified
and an important value in source analysis and risk management for soil method for the combined use of Geodetector and PCA-MLRD. Factor
pollution in a complicated environment. detection provides quantitative tools for determining principal compo­
nent factors and screening potential pollution sources. The main

12
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

pollution sources of specific elements were further identified by Hexi Corridor, Northwest China. Chemosphere 193, 189–197. https://doi.org/
10.1016/j.chemosphere.2017.10.151.
regression equation. And the combination of soil and dust indicated that
Guan, Q., Zhao, R., Pan, N., Wang, F., Yang, Y., Luo, H., 2019. Source apportionment of
soil heavy metals are closely associated with rivers, whereas dust is more heavy metals in farmland soil of Wuwei, China: comparison of three receptor
associated with roads. models. J. Clean. Prod. 237, 117792 https://doi.org/10.1016/j.
jclepro.2019.117792.
Hakanson, L., 1980. An ecological risk index for aquatic pollution-control - a
CRediT authorship contribution statement sedimentological approach. Water Res. 14, 975–1001. https://doi.org/10.1016/
0043-1354(80)90143-8.
Han, Q., Liu, Y., Feng, X., Mao, P., Sun, A., Wang, M., Wang, M., 2021. Pollution effect
Weibin Zeng: Methodology, Investigation, Writing – original draft,
assessment of industrial activities on potentially toxic metal distribution in
Visualization. Xiaoming Wan: Conceptualization, Writing - review & windowsill dust and surface soil in central China. Sci. Total Environ. 759, 144023
editing, Lingqing Wang: Investigation, Resources. Gaoquan Gu: https://doi.org/10.1016/j.scitotenv.2020.144023.
Writing – review & editing, Visualization. Mei Lei: Resources, Super­ Hu, Y.N., He, K.L., Sun, Z.H., Chen, G., Cheng, H.F., 2020. Quantitative source
apportionment of heavy metal(loid)s in the agricultural soils of an industrializing
vision. Tongbin Chen: Resources, Supervision. region and associated model uncertainty. J. Hazard. Mater. 391, 122244 https://doi.
org/10.1016/j.jhazmat.2020.122244.
Huang, J., Wu, Y., Sun, J., Li, X., Geng, X., Zhao, M., Sun, T., Fan, Z., 2021. Health risk
Declaration of Competing Interest assessment of heavy metal(loid)s in park soils of the largest megacity in China by
using Monte Carlo simulation coupled with positive matrix factorization model.
J. Hazard. Mater. 415, 125629 https://doi.org/10.1016/j.jhazmat.2021.125629.
The authors declare that they have no known competing financial Huang, Y., Deng, M., Wu, S., Japenga, J., Li, T., Yang, X., He, Z., 2018. A modified
interests or personal relationships that could have appeared to influence receptor model for source apportionment of heavy metal pollution in soil. J. Hazard.
the work reported in this paper. Mater. 354, 161–169. https://doi.org/10.1016/j.jhazmat.2018.05.006.
Hutter, H.-P., Wallner, P., Moshammer, H., Marsh, G., 2016b. Dust and cobalt levels in
the austrian tungsten industry: workplace and human biomonitoring data. Int. J.
Data Availability Environ. Res. Public Health 13. https://doi.org/10.3390/ijerph13090931.
Izydorczyk, G., Mikula, K., Skrzypczak, D., Moustakas, K., Witek-Krowiak, A.,
Chojnacka, K., 2021. Potential environmental pollution from copper metallurgy and
Data will be made available on request.
methods of management. Environ. Res. 197, 111050 https://doi.org/10.1016/j.
envres.2021.111050.
Acknowledgment Jia, X., Hu, B., Marchant, B.P., Zhou, L., Shi, Z., Zhu, Y., 2019. A methodological
framework for identifying potential sources of soil heavy metal pollution based on
machine learning: a case study in the Yangtze Delta, China. Environ. Pollut. 250,
This work was supported by the National Key Research and Devel­ 601–609. https://doi.org/10.1016/j.envpol.2019.04.047.
opment Program of China (Grant No. 2018YFC1800302) and the Inno­ Jiang, H., Nie, J., Fan, S., Yin, G., Ma, Y., Ma, T., Wang, Q., Shi, R., 2018. Integrating GIS
to determine the spatial distribution of principal components and the sources of
vation Academy for Green Manufacture of the Chinese Academy of
heavy metals in farmland soils near mining area in Tongling, China. Fresenius
Sciences (Grant No. IAGM-2019-A16-5). Environ. Bull. 27, 2662–2670.
Jiang, Y.X., Chao, S.H., Liu, J.W., Yang, Y., Chen, Y.J., Zhang, A.C., Cao, H.B., 2017.
Source apportionment and health risk assessment of heavy metals in soil for a
Appendix A. Supporting information township in Jiangsu Province, China. Chemosphere 168, 1658–1668. https://doi.
org/10.1016/j.chemosphere.2016.11.088.
Supplementary data associated with this article can be found in the Klasson, M., Bryngelsson, I.L., Pettersson, C., Husby, B., Arvidsson, H., Westberg, H.,
2016. Occupational exposure to cobalt and tungsten in the swedish hard metal
online version at doi:10.1016/j.jhazmat.2022.129468.
industry: air concentrations of particle mass, number, and surface Area. Ann. Occup.
Hyg. 60, 684–699. https://doi.org/10.1093/annhyg/mew023.
References Kuang, Y., Fang, F.M., Li, Y.B., Lin, Y.S., Yao, Y.R., Wu, M.H., Wu, H.J., Wang, Y., 2018.
Concentrations and pollution assessment of mercury in farmland soil of Xinqiao
Mining Area of Tongling, Anhui, China (in Chinese). Ying yong sheng tai xue bao =
Cai, L., Wang, Q., Wen, H., Luo, J., Wang, S., 2019. Heavy metals in agricultural soils
J. Appl. Ecol. 29, 2746–2752.
from a typical township in Guangdong Province, China: Occurrences and spatial
Li, R., Pan, C., Chen, J., Jiang, Y., Ding, G., 2012. Heavy metal contamination and health
distribution. Ecotoxicol. Environ. Saf. 168, 184–191. https://doi.org/10.1016/j.
risk assessment for urban topsoil and dust in Tongling City (in Chinese). China
ecoenv.2018.10.092.
Environ. Sci. 32, 2261–2270.
Chang, S.-H., Wang, K.-S., Chang, H.-F., Ni, W.-W., Wu, B.-J., Wong, R.-H., Lee, H.-S.,
Li, W., Wu, J., Zhou, C., Nsabimana, A., 2021. Groundwater pollution source
2009. Comparison of source identification of metals in road-dust and soil. Soil
identification and apportionment using PMF and PCA-APCS-MLR receptor models in
Sediment Contam. Int. J. 18, 669–683. https://doi.org/10.1080/
Tongchuan City, China. Arch. Environ. Contam. Toxicol. 81, 397–413. https://doi.
15320380903085691.
org/10.1007/s00244-021-00877-5.
Chen, X., Lu, X., 2018. Contamination characteristics and source apportionment of heavy
Liu, G., Shi, Y., Guo, G., Zhao, L., Niu, J., Zhang, C., 2020. Soil pollution characteristics
metals in topsoil from an area in Xi’an city, China. Ecotoxicol. Environ. Saf. 151,
and systemic environmental risk assessment of a large-scale arsenic slag
153–160. https://doi.org/10.1016/j.ecoenv.2018.01.010.
contaminated site. J. Clean. Prod. 251, 119721 https://doi.org/10.1016/j.
Cui, Z., Wang, Y., Zhao, N., Yu, R., Xu, G., Yu, Y., 2018. Spatial distribution and risk
jclepro.2019.119721.
assessment of heavy metals in paddy soils of yongshuyu irrigation area from
Liu, H., Zhang, Y., Yang, J., Wang, H., Li, Y., Shi, Y., Li, D., Holm, P.E., Ou, Q., Hu, W.,
Songhua River Basin, Northeast China. Chin. Geogr. Sci. 28, 797–809. https://doi.
2021. Quantitative source apportionment, risk assessment and distribution of heavy
org/10.1007/s11769-018-0991-1.
metals in agricultural soils from southern Shandong Peninsula of China. Sci. Total
Deng, M.H., Zhu, Y.W., Shao, K., Zhang, Q., Ye, G.H., Shen, J., 2020. Metals source
Environ. 767, 144879 https://doi.org/10.1016/j.scitotenv.2020.144879.
apportionment in farmland soil and the prediction of metal transfer in the soil-rice-
Long, Z., Zhu, H., Bing, H., Tian, X., Wang, Z., Wang, X., Wu, Y., 2021. Contamination,
human chain. J. Environ. Manag. 260. https://doi.org/10.1016/j.
sources and health risk of heavy metals in soil and dust from different functional
jenvman.2020.110092.
areas in an industrial city of Panzhihua City, Southwest China. J. Hazard. Mater.
Dhaliwal, S.S., Singh, J., Taneja, P.K., Mandal, A., 2020. Remediation techniques for
420, 126638 https://doi.org/10.1016/j.jhazmat.2021.126638.
removal of heavy metals from the soil contaminated through different sources: a
Luo, H., Wang, Q., Guan, Q., Ma, Y., Ni, F., Yang, E., Zhang, J., 2022. Heavy metal
review. Environ. Sci. Pollut. Res. 27, 1319–1333. https://doi.org/10.1007/s11356-
pollution levels, source apportionment and risk assessment in dust storms in key
019-06967-1.
cities in Northwest China. J. Hazard. Mater. 422, 126878 https://doi.org/10.1016/j.
Fei, X., Lou, Z., Xiao, R., Ren, Z., Lv, X., 2020. Contamination assessment and source
jhazmat.2021.126878.
apportionment of heavy metals in agricultural soil through the synthesis of PMF and
Lv, J.S., 2019. Multivariate receptor models and robust geostatistics to estimate source
GeoDetector models. Sci. Total Environ. 747, 141293 https://doi.org/10.1016/j.
apportionment of heavy metals in soils. Environ. Pollut. 244, 72–83. https://doi.org/
scitotenv.2020.141293.
10.1016/j.envpol.2018.09.147.
Filella, M., Williams, P.A., Belzile, N., 2009. Antimony in the environment: knowns and
Mandal, S.K., Ray, R., González, A.G., Pokrovsky, O.S., Jana, T.K., 2021. Antimony
unknowns. Environ. Chem. 6, 95–105. https://doi.org/10.1071/EN09007.
uptake by mangroves and its environmental fate in the Sundarbans, India. Estuar.
Gabarrón, M., Faz, A., Acosta, J.A., 2017a. Effect of different industrial activities on
Coast. Shelf Sci. 248, 106923 https://doi.org/10.1016/j.ecss.2020.106923.
heavy metal concentrations and chemical distribution in topsoil and road dust.
MEE, Ministry of Ecology and Environment of the People’s Republic of China, 2018. Soil
Environ. Earth Sci. 76, 129. https://doi.org/10.1007/s12665-017-6449-4.
environmental quality Risk control standard for soil contamination of development
Gabarrón, M., Faz, A., Acosta, J.A., 2017b. Soil or dust for health risk assessment studies
land. 〈https://www.mee.gov.cn/ywgz/fgbz/bz/bzwb/trhj/201807/W02019062
in urban environment. Arch. Environ. Contam. Toxicol. 73, 442–455. https://doi.
6596188930731.pdf〉.
org/10.1007/s00244-017-0413-x.
Men, C., Liu, R., Xu, L., Wang, Q., Guo, L., Miao, Y., Shen, Z., 2020. Source-specific
Guan, Q., Wang, F., Xu, C., Pan, N., Lin, J., Zhao, R., Yang, Y., Luo, H., 2018. Source
ecological risk analysis and critical source identification of heavy metals in road dust
apportionment of heavy metals in agricultural soil based on PMF: a case study in

13
W. Zeng et al. Journal of Hazardous Materials 438 (2022) 129468

in Beijing, China. J. Hazard. Mater. 388, 121763 https://doi.org/10.1016/j. Wang, Y., Guo, G., Zhang, D., Lei, M., 2021b. An integrated method for source
jhazmat.2019.121763. apportionment of heavy metal(loid)s in agricultural soils and model uncertainty
Paatero, P., Hopke, P.K., Hoppenstock, J., Eberly, S.I., 2003. Advanced factor analysis of analysis. Environ. Pollut. 276, 116666 https://doi.org/10.1016/j.
spatial distributions of PM2.5 in the Eastern United States. Environ. Sci. Technol. 37, envpol.2021.116666.
2460–2476. https://doi.org/10.1021/es0261978. Wang, Y., Zhang, L., Wang, J., Lv, J., 2020b. Identifying quantitative sources and spatial
Paatero, P., Tapper, U., 1994. Positive matrix factorization: a non-negative factor model distributions of potentially toxic elements in soils by using three receptor models and
with optimal utilization of error estimates of data values. Environmetrics 5, sequential indicator simulation. Chemosphere 242, 125266. https://doi.org/
111–126. https://doi.org/10.1002/env.3170050203. 10.1016/j.chemosphere.2019.125266.
Parra, S., Bravo, M.A., Quiroz, W., Moreno, T., Karanasiou, A., Font, O., Vidal, V., Wu, Q., Hu, W., Wang, H., Liu, P., Wang, X., Huang, B., 2021. Spatial distribution,
Cereceda-Balic, F., 2014. Source apportionment for contaminated soils using ecological risk and sources of heavy metals in soils from a typical economic
multivariate statistical methods. Chemom. Intell. Lab. Syst. 138, 127–132. https:// development area, Southeastern China. Sci. Total Environ. 780, 146557 https://doi.
doi.org/10.1016/j.chemolab.2014.08.003. org/10.1016/j.scitotenv.2021.146557.
Pekey, H., Doğan, G., 2013. Application of positive matrix factorisation for the source Wu, Y., Li, X., Yu, L., Wang, T., Wang, J., Liu, T., 2022. Review of soil heavy metal
apportionment of heavy metals in sediments: a comparison with a previous factor pollution in China: spatial distribution, primary sources, and remediation
analysis study. Microchem. J. 106, 233–237. https://doi.org/10.1016/j. alternatives. Resour., Conserv. Recycl. 181, 106261 https://doi.org/10.1016/j.
microc.2012.07.007. resconrec.2022.106261.
Pyatt, F.B., Amos, D., Grattan, J.P., Pyatt, A.J., Terrell-Nield, C.E., 2002. Invertebrates of Wu, Z., Zhang, L., Xia, T., Jia, X., Li, H., Wang, S., 2020. Quantitative assessment of
ancient heavy metal spoil and smelting tip sites in southern Jordan: their distribution human health risks based on Soil heavy metals and PAHs sources: take a polluted
and use as bioindicators of metalliferous pollution derived from ancient sources. industrial site of Beijing as an example (in Chinese). Huanjing Kexue 41, 4180–4196.
J. Arid Environ. 52, 53–62. https://doi.org/10.1006/jare.2002.0982. https://doi.org/10.13227/j.hjkx.201910152.
Qiao, P., Dong, N., Yang, S., Gou, Y., 2021. Quantitative analysis of the main sources of Xu, D., Zhou, P., Zhan, J., Gao, Y., Dou, C., Sun, Q., 2013. Assessment of trace metal
pollutants in the soils around key areas based on the positive matrix factorization bioavailability in garden soils and health risks via consumption of vegetables in the
method. Environ. Pollut. 273, 116518 https://doi.org/10.1016/j. vicinity of Tongling mining area, China. Ecotoxicol. Environ. Saf. 90, 103–111.
envpol.2021.116518. https://doi.org/10.1016/j.ecoenv.2012.12.018.
Roels, H.A., Buchet, J.-P., Lauwerys, R., Bruaux, P., Claeys-Thoreau, F., Lafontaine, A., Yang, X., Yang, Y., Wan, Y., Wu, R., Feng, D., Li, K., 2021a. Source identification and
van Overschelde, J., Verduyn, G., 1978. Lead and cadmium absorption among comprehensive apportionment of the accumulation of soil heavy metals by
children near a nonferrous metal plant: a follow-up study of a test case. Environ. Res. integrating pollution landscapes, pathways, and receptors. Sci. Total Environ. 786,
15, 290–308. https://doi.org/10.1016/0013-9351(78)90105-6. 147436 https://doi.org/10.1016/j.scitotenv.2021.147436.
Roy, S., Gupta, S.K., Prakash, J., Habib, G., Baudh, K., Nasr, M., 2019. Ecological and Yang, Y., Christakos, G., Guo, M.W., Xiao, L., Huang, W., 2017. Space-time quantitative
human health risk assessment of heavy metal contamination in road dust in the source apportionment of soil heavy metal concentration increments. Environ. Pollut.
National Capital Territory (NCT) of Delhi, India. Environ. Sci. Pollut. Res. 26, 223, 560–566. https://doi.org/10.1016/j.envpol.2017.01.058.
30413–30425. https://doi.org/10.1007/s11356-019-06216-5. Yang, Y., Yang, X., He, M.J., Christakos, G., 2020. Beyond mere pollution source
SAC, Standardization Administration of China, 2018. Soil quality-Guidance on sampling identification: determination of land covers emitting soil heavy metals by combining
techniques (GB/T 36197–2018). Beijing, China. PCA/APCS, GeoDetector and GIS analysis. Catena 185. https://doi.org/10.1016/j.
Safari, Y., Delavar, M.-A., 2019. The influence of soil pollution by heavy metals on the catena.2019.104297.
land suitability for irrigated wheat farming in Zanjan region, northwest Iran. Arab. J. Yang, S., Taylor, D., Yang, D., He, M., Liu, X., Xu, J., 2021b. A synthesis framework using
Geosci. 12, 21. https://doi.org/10.1007/s12517-018-4190-2. machine learning and spatial bivariate analysis to identify drivers and hotspots of
Shao, D., Zhan, Y., Zhou, W., Zhu, L., 2016. Current status and temporal trend of heavy heavy metal pollution of agricultural soils. Environ. Pollut. 287, 117611 https://doi.
metals in farmland soil of the Yangtze River Delta Region: Field survey and meta- org/10.1016/j.envpol.2021.117611.
analysis. Environ. Pollut. 219, 329–336. https://doi.org/10.1016/j. Yoon, S.J., Hong, S., Lee, C., Lee, J., Kim, T., Lee, J., Kim, B., Noh, J., Kwon, B.-O.,
envpol.2016.10.023. Khim, J.S., 2022. 10 years long-term assessment on characterizing spatiotemporal
Shen, Z., Xu, D., Li, L., Wang, J., Shi, X., 2019. Ecological and health risks of heavy metal trend and source apportionment of metal(loid)s in terrestrial soils along the west
on farmland soils of mining areas around Tongling City, Anhui, China. Environ. Sci. coast of South Korea. Sci. Total Environ. 826, 154214 https://doi.org/10.1016/j.
Pollut. Res. 26, 15698–15709. https://doi.org/10.1007/s11356-019-04463-0. scitotenv.2022.154214.
Sun, Y., 2013. Pollution characteristics and source analysis of heavy metal elements in Zhang, H., Cai, A., Wang, X., Wang, L., Wang, Q., Wu, X., Ma, Y., 2021a. Risk assessment
Xingqiao River (in Chinese). Hefei University of Technology. and source apportionment of heavy metals in soils from Handan City. Appl. Sci. 11,
Tao, H., Liao, X., Li, Y., Xu, C., Zhu, G., Cassidy, D.P., 2020. Quantifying influences of 9615. https://doi.org/10.3390/app11209615.
interacting anthropogenic-natural factors on trace element accumulation and Zhang, M., Li, X., Yang, R., Wang, J., Ai, Y., Gao, Y., Zhang, Y., Zhang, X., Yan, X., Liu, B.,
pollution risk in karst soil. Sci. Total Environ. 721, 137770 https://doi.org/10.1016/ Yu, H., 2019. Multipotential toxic metals accumulated in urban soil and street dust
j.scitotenv.2020.137770. from Xining City, NW China: spatial occurrences, sources, and health risks. Arch.
USEPA, U.S.Environmental Protection Agency, 2014. EPA positive matrix factorization Environ. Contam. Toxicol. 76, 308–330. https://doi.org/10.1007/s00244-018-
(PMF) 5.0 fundamentals and user guide. 00592-8.
Wang, C., Shen, Z., Li, X., Luo, C., Chen, Y., Yang, H., 2004. Heavy metal contamination Zhang, Z., Song, X., Wang, Q., Lu, X., 2012. Cd and Pb contents in soil, plants, and
of agricultural soils and stream sediments near a copper mine in Tongling, People’s grasshoppers along a pollution gradient in Huludao City, Northeast China. Biol.
Republic of China. Bull. Environ. Contam. Toxicol. 73, 862–869. https://doi.org/ Trace. Elem. Res. 145, 403–410. https://doi.org/10.1007/s12011-011-9199-2.
10.1007/s00128-004-0506-x. Zhang, M., Wang, X., Liu, C., Lu, J., Qin, Y., Mo, Y., Xiao, P., Liu, Y., 2020. Quantitative
Wang, G., Li, Y., Wang, J., Jia, Z., Zhou, Y., Zhou, S., Xie, X., 2020a. A modified receptor source identification and apportionment of heavy metals under two different land
model for historical source apportionment of polycyclic aromatic hydrocarbons in use types: comparison of two receptor models APCS-MLR and PMF. Environ. Sci.
sediment. Sci. Total Environ. 702, 134931 https://doi.org/10.1016/j. Pollut. Res. 27, 42996–43010. https://doi.org/10.1007/s11356-020-10234-z.
scitotenv.2019.134931. Zhang, H., Yao, Q., Zhu, Y., Fan, S., He, P., 2012. Review of source identification
Wang, J., Xu, C., 2017. Geodetector: principle and prospective (in Chinese). Acta Geogr. methodologies for heavy metals in solid waste. Chin. Sci. Bull. 58, 162–168. https://
Sin. 72, 116–134. https://doi.org/10.11821/dlxb201701010. doi.org/10.1007/s11434-012-5531-2.
Wang, J., Su, J., Li, Z., Liu, B., Cheng, G., Jiang, Y., Li, Y., Zhou, S., Yuan, W., 2019. Zhang, H., Zhang, F., Song, J., Tan, M.L., Kung, H.-T., Johnson, V.C., 2021b. Pollutant
Source apportionment of heavy metal and their health risks in soil-dustfall-plant source, ecological and human health risks assessment of heavy metals in soils from
system nearby a typical non-ferrous metal mining area of Tongling, Eastern China. coal mining areas in Xinjiang, China. Environ. Res., 111702 https://doi.org/
Environ. Pollut. 254, 113089 https://doi.org/10.1016/j.envpol.2019.113089. 10.1016/j.envres.2021.111702.
Wang, J., Yu, J., Gong, Y., Wu, L., Yu, Z., Wang, J., Gao, R., Liu, W., 2021a. Pollution Zhao, F.-J., Ma, Y., Zhu, Y.-G., Tang, Z., McGrath, S.P., 2015. Soil contamination in
characteristics, sources and health risk of metals in urban dust from different China: current status and mitigation strategies. Environ. Sci. Technol. 49, 750–759.
functional areas in Nanjing, China. Environ. Res. 201, 111607 https://doi.org/ https://doi.org/10.1021/es5047099.
10.1016/j.envres.2021.111607. Zhuo, H., Fu, S., Liu, H., Song, H., Ren, L., 2019. Soil heavy metal contamination and
Wang, J.F., Li, X.H., Christakos, G., Liao, Y.L., Zhang, T., Gu, X., Zheng, X.Y., 2010. health risk assessment associated with development zones in Shandong, China.
Geographical detectors-based health risk assessment and its application in the neural Environ. Sci. Pollut. Res. 26, 30016–30028. https://doi.org/10.1007/s11356-019-
tube defects study of the Heshun Region, China. Int. J. Geogr. Inf. Sci. 24, 107–127. 05979-1.
https://doi.org/10.1080/13658810802443457. Zhang, Z., Song, X., Wang, Q., Lu, X., 2012. Cd and Pb contents in soil, plants, and
Wang, J.F., Zhang, T.L., Fu, B.J., 2016. A measure of spatial stratified heterogeneity. grasshoppers along a pollution gradient in Huludao City, Northeast China. Biol.
Ecol. Indic. 67, 250–256. https://doi.org/10.1016/j.ecolind.2016.02.052. Trace Elem. Res. 145, 403–410. https://doi.org/10.1007/s12011-011-9199-2.

14

You might also like