3.1. Content characteristics of heavy metals in soil
As Showed in Table 1, the values of the soil pH in the study area ranged from 3.2 to 8.1, and the average Cd, Hg, As, Pb, Cr, Cu, Ni, and Zn contents in the topsoil were 0.46, 0.14, 9.66, 31.2, 127, 95.6, 76.0, and 158mg/kg, respectively. Compared with the risk screening values in the Environmental Quality Standard for Soils (GB 15618 − 2018) (MEE, 2018), the order of the excess ratios of the heavy metals in the study area was Cu (80%) > Cd (56%) > Ni (32%) > Zn (8%) > Hg, As, Pb, and Cr (0%). This indicates that Cu and Cd were the main pollutants. Compared with the background values of Chongqing soils (Li and Jia, 2018), the average contents of Cd, Hg, As, Pb, Cr, Cu, Ni, and Zn in the topsoil were 4.2, 2.38, 1.93, 1.20, 1.58, 3.68, 2.37, and 1.98 times the background values, respectively. This indicates that heavy metals likely accumulated in the topsoil.
The coefficient of variation (CV) reflects the degree of dispersion of the variables and can be applied to evaluate the impact of human activities on element accumulation (Yang et al., 2021). The CV values of heavy metals in the topsoil, in order from highest to lowest, were Cd (51.7%) > Zn (39.7%) > Hg (38.1%) > As (37.7%) > Ni (27.9%) > Cu (24.0%) > Pb (16.4%) > Cr (12.7%). The results showed that heavy metals were affected to different degrees by human activities, as shown in Fig. 2. In general, the soil pH was the lowest in the topsoil, and the concentration increased with soil depth. The order of the Cd, Hg, As, Pb, Zn, and Ni contents in the profile was as follows: surface layer (0–20cm) > middle layer (20–60cm) > bottom layer (60–100cm). The Cr and Cu contents were basically the same in all three soil layers in each profile.
Given the topsoil (Table 1) and soil profile data (Fig. 2), acidification and heavy metal accumulation occurred in the topsoil. According to previous studies (Atafar et al., 2010; Jiang et al., 2014; Yang et al., 2016), fertilizer application, acid atmospheric deposition, gangue accumulation, and leaching might decrease the pH and increase the content of heavy metals in the topsoil.
Table 1
Statistical characteristics of the heavy metal concentration in the soil (units: mg·kg− 1)
|
|
pH
|
Cd
|
Hg
|
As
|
Pb
|
Cr
|
Cu
|
Ni
|
Zn
|
Soil (n = 25)
|
Min
|
3.2
|
0.11
|
0.08
|
4.60
|
22.5
|
102
|
50.1
|
38.9
|
98.5
|
Max
|
8.1
|
1.07
|
0.30
|
17.1
|
39.8
|
150
|
140
|
113
|
248
|
Avg
|
6.2
|
0.46
|
0.14
|
9.66
|
31.2
|
127
|
95.6
|
76.0
|
158
|
CV/%
|
22.6
|
51.7
|
38.1
|
37.7
|
16.4
|
12.7
|
24.0
|
27.9
|
39.8
|
Risk screening values
|
pH ≤ 5.5
|
0.3
|
1.3
|
40
|
70
|
150
|
50
|
60
|
200
|
5.5 < pH ≤ 6.5
|
0.3
|
1.8
|
40
|
90
|
150
|
50
|
70
|
200
|
6.5 < pH ≤ 7.5
|
0.3
|
2.4
|
30
|
120
|
200
|
100
|
100
|
250
|
pH > 7.5
|
0.6
|
3.4
|
25
|
170
|
250
|
100
|
190
|
300
|
Background values
|
0.11
|
0.06
|
5
|
26
|
80
|
26
|
32
|
80
|
3.3. Correlation analysis of heavy metals in soil
The Pearson correlation coefficient matrix for the heavy metals is illustrated in Fig. 4. Significant positive correlations were found among Cd, Hg, As, and Pb levels (r = 0.45–0.70, P < 0.01); significant positive correlations were also found between Cr and Cu levels (r = 0.47, P < 0.05). A significant positive correlation was observed between Ni and Zn (r = 0.88, P < 0.01). The high correlations between heavy metal concentrations suggest that the corresponding metals share similarities in terms of their contamination levels and pollution sources (Li et al., 2009). The results showed that Cd, Pb, Zn, and Cu; Cr and Cu; and Ni and Zn were grouped together and might have had the same or similar pollution sources. The results of the correlation analysis are consistent with the characteristics of the spatial distribution.
3.4. Source apportionment of heavy metals in soil
3.4.1. Source apportionment by the APCA-MLR
The Kaiser-Meyer-Olkin measure was 0.573 (> 0.5), and Bartlett’s test of sphericity was significant at 0.01, which indicated that the PCA results were useful (Šprajc et al., 2019). Combined with the results of the correlation analysis, the number of principal components (PCs) was set to three. Generally, the factor loadings were divided, with “strong,” “moderate,” and “weak” referring to absolute loading values of 0.75–1, 0.75–0.5, and 0.5–0.3, respectively (Liu et al., 2003). As shown in Fig. 5 and Table S1, three main PCs were derived, accounting for 80.7% of the total variance. PC1 had strong positive loadings for Cd (0.827) and Hg (0.828) and moderate positive loadings for As (0.743) and Pb (0.731), accounting for 39.1% of the total variance. PC2 had strong positive loadings for Ni (0.958) and Zn (0.961), explaining 26.2% of the total variance. PC3 had high positive loadings for Cr (0.830) and Cu (0.830), accounting for 15.4% of the total variance.
Based on the results of PCA, it was necessary to convert the data using APCS-MLR. The average contribution rates of heavy metals corresponding to each source are shown in Fig. 6. The pollution source represented by Factor 1 (F1) was the main factor leading to the enrichment of Cd, Hg, As, and Pb in the soil, with contribution rates of 74.6%, 79.4%, 69.1%, and 67.2%, respectively. Correlation analysis and the spatial distribution confirmed that these elements came from the same source and were clearly influenced by the gangue heap in the study area. Some studies suggest that heavy metals in gangue, mainly in the form of sulfides, are released during sulfide oxidation and accumulate on the surface of the soil through migration (Szczepanska and Twardowska, 1999; Hua et al., 2018; Sun et al., 2020). Therefore, it was presumed that F1 was mainly affected by the gangue heap accumulation during mining exploitation.
F2 had the largest contamination sources for Ni and Zn, with contribution rates of 74.6% and 81.8%, respectively. Correlation analysis and the spatial distribution confirmed that these elements came from the same source and were not primarily influenced by the gangue heap in the study area. Some studies have suggested that large quantities of wastewater and sludge containing heavy metals are produced by the electroplating industry, which might lead to heavy metal pollution (e.g., Zn, Ni, Cu, and Cr) in the surrounding soil (Liu et al., 2011; Xiao et al., 2019). The high content of Ni and Zn in the study area is close to that in the urban area, which may be affected by electroplating emissions. Furthermore, the long-term use of organic (natural) fertilizers, inorganic (synthetic) fertilizers, and pesticides might cause the accumulation of heavy metals such as Ni, Cu, Zn, Pb, and Cd in the soil (Zaccone et al., 2010; Wang et al., 2017; Alengebawy et al., 2021). Therefore, it was presumed that PC2 was mainly affected by industrial activities in nearby urban areas and agricultural activities in the study area.
F3 had the largest contamination sources for Cr and Cu, with contribution rates of 58.9% and 69.4%, respectively. The average contents of Cr and Cu were similar in all three soil layers, and the CVs of Cr and Cu in the topsoil samples were relatively low (Fig. 2). This indicates that the soil layer was less disturbed, which might have been affected by natural factors. However, the average contents of Cr and Cu in the topsoil were 1.58 and 3.68 times the background values, respectively, indicating that Cr and Cu might also be affected by anthropogenic factors. Combined with the result of F2, this shows that Cr and Cu were indeed affected by anthropogenic activities, such as industrial and agricultural activities, with contribution rates of 21.8% and 21.2%, respectively. Several recent studies have shown that Cr and Cu are associated with both natural and anthropogenic sources, which vary by location (Davis et al., 2009; Wu et al., 2018; Liu et al., 2020; Wu et al., 2021), consistent with the results of this study. Therefore, it was presumed that F3 was mainly affected by natural factors.
APCS-MLR not only qualitatively determines the load of each heavy metal to each pollution source but also quantitatively determines the contribution rate (CoR) of each heavy metal at each sampling point (Jin et al., 2019). This was combined with IDW interpolation using ArcGIS 10.2 for visualization (Fig. 7). Cd and Hg were primarily affected by F1 (with CoRs of 60–100%), and they were secondarily affected by F2 (CoR < 40%). As and Pb were primarily affected by F1 (CoR = 40–80%) and then F3 (CoR = 20–40%). Cr and Cu were primarily affected by F3 (CoR = 40–80%) and thenF2 (CoR < 40%). Ni and Zn were primarily affected by F2 (CoR = 60–100%) and were secondarily affected by F1 (CoR < 40%). Understanding the CoR of each sampling point supports the precise prevention and control of soil pollution.
3.4.2. Positive matrix factorization (PMF)
PMF was conducted to better identify and quantify the sources of heavy metals in the study area. To ensure the rationality of the analysis, it was necessary to determine the minimum Q to control the residual matrix E (Zhang et al., 2020). The number of factors for the model was initially set to 2–5, the start seed number was randomly obtained, and the system was run 20 times to obtain the best solution. The results show that the PMF model confined four factors, which resulted in the lowest Q value (Q(true) = Q(robust) = 174.9), where the residuals were between − 3 and 3. This shows that the model results are credible (Table S2).
The results show the average contribution rate of each source to its heavy metals (Figs. 8 and S1). F1 in the PMF analysis had the largest contamination sources for Cd, Hg, As, and Pb, with CoRs of 69.7%, 60.7%, 57.4%, and 41.9%, respectively. Both F1 in PMF and F1 in APCS-MLR showed that the CoR was highest for Cd, Hg, As, and Pb. Therefore, it was presumed that the F1 in PMF was the same as that in APCS-MLR, having been primarily affected by the gangue heap accumulation.F4in the PMF analysis had the largest contamination sources for Cr and Cu, with CoRs of 39.7% and 50.2%, respectively. Both F4 in PMF and F3 in APCS-MLR showed that the contribution rate was highest for Cr and Cu; therefore, it was presumed that F4 in PMF was the same as F3 in APCS-MLR, both of which were mainly affected by natural factors. F2 in PMF had the largest Ni and Zn contamination sources, with CoRs of 44.5% and 40.1%, respectively. The electroplating industry generates a large volume of wastewater with high concentrations of heavy metals owing to metal surface cleaning, rinsing, and water leakage (i.e., Ni, Mn, and Zn), posing a high risk to the environment (Hang et al., 2009); it was presumed that F2 was mainly affected by industrial activities. The long-term use of fertilizers and pesticides may cause the accumulation of heavy metals in the soil (Zaccone et al., 2010; Wang et al., 2017; Alengebawy et al., 2021), and it was presumed that F3 was primarily affected by agricultural activities. The sum of the contribution rates of F2 and F3 to Ni was 64.4%, and the sum of the contribution rates of F2 and F3 to Zn was 62.6%. Therefore, it was presumed that F2 and F3 in PMF were the same as F2 in APCS-MLR, both of which were primarily affected by industrial and agricultural activities.
When using the PMF model to quantify the heavy metal sources, the results may still be uncertain owing to random errors, rotation ambiguity, and other factors caused by the model itself (methods for estimating uncertainty in factor analytic solutions).To evaluate this uncertainty, two methods (DISP, displacement of factor elements; BS, bootstrap) proposed by the PMF5.0 model were used to estimate the error of the source contribution obtained by the model (methods for estimating uncertainty in factor analytic solutions) (Figs.S2 and S3).The DISP can be used as a preliminary screening method for model robustness and reliability (Steven et al., 2015). In this study, the percentage decrease in Q is less than 1%, indicating that the results of DISP are acceptable, but an obvious exchange phenomenon is found among various factors, which means that the results are affected by rotation ambiguity to a certain extent (Steven et al., 2015; Wu et al., 2020). In addition, 100 BS runs and 42 random seeds were set to test the BS method. All BS factors were successively assigned to basic factors ranging from 1–4 according to the mapping principle, and more than 85% of the factors were resampled, indicating that the model had good robustness (Md and Warren, 2017). The variability in the source contributions obtained using the DISP and BS methods is shown in Figs. S4 and S5, demonstrating considerable uncertainty in the source profile, such as in Cd, Hg, and Ni in F1 and As, Pb, and Cr in F2, which are not within the interquartile range in the BS analysis and may be subject to random errors. In addition, the interval ratio between DISP and BS can be used as a basis for determining whether the uncertainty of each heavy metal in the study area is significant (Ram et al., 2022). When the value is close to 2, there is significant uncertainty. In this study, Cr, Cu, and Ni in F1; Cd, Hg, and As in F2; Cd, Cu, and Ni in F3; and Cd, Ni, and Zn in F4 all demonstrated significant uncertainty (Fig.S4).
3.4.3. Comparing the source apportionment results between APCS-MLR and PMF
In general, the closer the R2 is to 1, the better the interpretation of the dependent variable by the independent variable in the regression analysis (Duan et al., 2020), as shown in Table 2 (Figs.S5 and S6). Thus, the APCS-MLR and PMF models fit well, and the results of the source apportionment were reliable.The R2 value between APCS-MLR and PMF indicated that the R2 values of As, Pb, and Cr in APCS-MLR were higher than those in PMF, while the other elements were opposite, suggesting that PMF was more suitable for this study area.
Table 2
The R2 values of the APCS-MLR and PMF method
R2
|
Cd
|
Hg
|
As
|
Pb
|
Cr
|
Cu
|
Ni
|
Zn
|
APCS-MLR
|
0.778
|
0.744
|
0.783
|
0.709
|
0.793
|
0.740
|
0.963
|
0.946
|
PMF
|
0.999
|
0.774
|
0.776
|
0.527
|
0.525
|
0.826
|
0.980
|
0.953
|
The results of source apportionment were consistent for both the APCS-MLR and PMF (Fig. 9). Mining pollution sources were industrial, agricultural, and natural, and the CoRs were 2.7%, 10.2%, and 7.5%, respectively. Comparison of heavy metal elements by APCS-MLR and PMF (Figs. 6 and 8) showed that Cd, Hg, As, and Pb were mainly affected by the gangue heap accumulation. Cr and Cu were primarily affected by natural factors; Ni and Zn were primarily affected by industrial and agricultural activities. However, the contribution rate of each element differed according to theoretical principles. The APCS-MLR model extracts the factor load matrix and factor score matrix based on eigen value analysis, and it usually selects a cumulative variance contribution greater than 75% as the main component factor. The PMF model provides a non-negative constraint for the factorization matrix, forcing all the values in the solution profiles and contributions to be positive. Therefore, PMF can provide more accurate source identification than the PCA-MLR method can (Reff et al., 2007; Duan et al., 2020). However, the APCA-MLR model can quantitatively determine the contribution rate of each heavy metal at each sampling point (Jin et al., 2019) and can be combined with GIS for visualization; the PMF cannot quantitatively determine the contribution rate of each heavy metal at each sampling point based on PMF 5.0 (US EPA 2014). Some scholars believe that source analysis with only one receptor model could obtain an ex parte and imprecise evaluation of contamination sources, leading to disputes about the reliability of the model results (Zhang et al., 2020). The combined application of two receptor models can therefore overcome this limitation.