3.1 Descriptive statistics of target pollutants
According to the principles in Sect. 2.3, Cu, phenanthrene, and toluene, which are HMs, SVOC and VOC pollutants, respectively, were selected as the target pollutants. The auxiliary pollutants were determined based on the results of the correlation analysis, which revealed a significant correlation with the target pollutants at the 0.01 level. The auxiliary pollutants for Cu are Cd, As, Pb and Ni, for phenanthrene are fluoranthene, pyrene, chrysene and benzaanthracene, for toluene are benzene, styrene and m/p-xylene. The descriptive statistics of these target pollutants, such as their minimum values (Min), maximum values (Max), CVs and Rh/sum values, are shown in Fig. 2.
3.2 Degree of improvement in the prediction accuracy of the spatial distribution trend of the target pollutants in combining with their various related pollutants
The prediction accuracies of the three target pollutants in combination with their various related pollutants are shown as Fig. 3. When combined with the related pollutants, the prediction error which is indicated by RMSE of the spatial distribution of the target pollutants obtained using single variable cokriging is lower. For Cu combined with As, the reduction in the RMSE was approximately 37%. For toluene combined with benzene, the reduction in the RMSE was 76.8%. For phenanthrene combined with four related pollutants, the reduction in the RMSE was approximately 79%.
However, the reduction in the RMSE for Cu combined with Ni and Pb, toluene combined with M/p-xylene, and phenanthrene combined with benzaanthracene were lower. Therefore, Ni and Pb were not used as auxiliary pollutants for Cu in the multivariable cokriging; M/p-xylene was not used for toluene, and benzaanthracene was not used for phenanthrene. The reductions in RMSEs of Cu, toluene, and phenanthrene using multivariable cokriging were 68.4%, 81.6%, and 81.2%, respectively. Moreover, the RMSEs calculated for the multivariable cokriging were lower than those obtained by only combining one related pollutants.
3.3 The prediction accuracies of the high value areas of the three target pollutants combined with their related pollutants
3.2.1 General distribution trend of the interpolation results
The contour lines in Fig. 4 were extracted from the raster layer predicted using the OK method. The region in Fig. 4 with the dense contour distribution is the area with the high concentration data predicted using the OK method. The contour interval is the same in Fig. 4(I–VI) was same. This is also true for Fig. 4(a–f) and Fig. 4(i–v).
Overall, the spatial distributions of Cu, toluene, and phenanthrene predicted using the OK interpolation and using the related pollutants are similar. The correlation coefficients between the raster layers interpolated using the OK method for Cu with and without the related pollutants are all greater than 0.7; while the correlation coefficients of the raster layers for phenanthrene and toluene are all greater than 0.9. Among them, the spatial transition in the Cu content predicted using the related pollutants (Fig. 4(II–VI)) is relatively gentle compared with that predicted without the related pollutants (Fig. 4(I)). Furthermore, the long axis direction of the Cu and phenanthrene distributions are both northwest-southeast, while the higher concentration of toluene are located in the northern part of the study area.
3.2.2 Prediction accuracy of the high value area
In the process of the investigation and remediation of a contaminated site, the recognition accuracy of the high pollution area is the main concern. In this study, the phenanthrene contents of 5.08% of the sample points exceeded the screening level for the soil environmental risk assessment of sites (DB11T811-2011) (40 mg/kg) (DB11/T 811 2011), while none of the Cu and toluene content of the samples sites exceed their corresponding screening values. In order to analyze the differences in the high value region prediction results of the interpolation methods with different combinations of the related pollutants, the 93rd percentile (80 mg/kg) of the Cu content and the 75th percentile (9 mg/kg) of the toluene content, which make the recognition area easy to identify and analyze, were taken as the boundary of the high value region. The boundaries of the high value region are denoted by the labeled red lines in Fig. 4.
According to Fig. 4, the positions exceeding 40 mg/kg, 80 mg/kg, and 9 mg/kg of Cu, toluene, and phenanthrene, respectively, predicted with and without using the related pollutants are the same, but the area is different. The area enclosed by the Cu contour without related pollutants is the smallest, and the area for the combination of Cu and Pb is the largest. In addition, the area for a combination of Cu with Cd and As falls between those for only Cu and for Cu and Pb. For phenanthrene, the area for the combination of fluoranthene, pyrene, and chrysene is the smallest compared with Fig. 4(a–e). For toluene, the area for the combination of toluene and styrene is the smallest, and the area for the combination of toluene and benzene is the largest. In addition, the area for the combination of toluene with benzene and styrene falls between the areas of the other two combinations.
The prediction accuracy of the area of the different combinations, i.e., the prediction accuracy of the high value area, was judged based on the number of sampling points with values greater than the standard pollutant content values inside and outside the high value area. According to the statistical results, of the 118 sampling points, 9 samples had Cu contents greater than 80 mg/kg, 6 samples had phenanthrene contents greater than 40 mg/kg, and 29 samples had toluene contents greater than 9 mg/kg.
According to Table 1, all 9 of the samples with Cu contents of greater than 80 mg/kg were identified using cokriging and a combination of Cu and Pb, and using a combination of Cu and both Cd and As. However, the high pollution area predicted using the combination of Cu and both Cd and As was smaller than that predicted using the combination of Cu and Pb (Fig. 4(III–VI)). Therefore, the locations with Cu and both Cd and As, and the prediction area of the high value region is small, which lowers the of the remediation. Thus, the prediction results are ideal.
For phenanthrene, although all of the points with phenanthrene contents of greater than 40 mg/kg were accurately identified, the points not exceeding 40 mg/kg were also included in the high value area. The error recognition rate of the prediction results of the combination of phenanthrene with benzaanthracene, chrysene, pyrene, and fluoranthene was approxinately 50%, while that of the combination of phenanthrene and fluoranthene, pyrene, and chrysene was only about 33% (Table 1). In addition, the prediction area of the high value region is the smallest (Fig. 4(a–f)). Therefore, the prediction result for the combination of phenanthrene with fluoranthene, pyrene, and chrysene is more accurate, and the repair cost is the lowest.
The error recognition rates for toluene were all 60%, except for the combination of toluene with benzene and styrene (67%) (Table 1). However, the sample points with toluene contents of greater than 9 mg/kg were distributed in the high value area predicted in the central region as shown in Fig. 4(i–v), which is the region that was predicted accurately. The areas of the regions that were predicted accurately in Fig. 4(i–v) are 23485 m2, 22488 m2, 21278 m2, 23680 m2, and 20876 m2, respectively. Therefore, the prediction result of toluene combined with benzene and styrene was better than the prediction result of toluene combined with other related pollutants.
Table 1
Statistical results for the number of sample points with values greater than the standard pollutant content values in the high value area predicted using different combinations of related pollutants
Target pollutant
|
Interpolation combination
|
Number of sample points in the high value area
|
Number of sample points with values greater than the standard pollutant content values in the high value area
|
The total number of sample points with values greater than the standard pollutant content values for 118 sample points
|
Cu
|
Cu
|
8
|
8
|
9
|
Cu-Ni
|
8
|
8
|
9
|
Cu-Pb
|
9
|
9
|
9
|
Cu-As
|
8
|
8
|
9
|
Cu-Cd
|
8
|
8
|
9
|
Cu-Cd-As
|
9
|
9
|
9
|
Ph
|
Ph
|
12
|
5
|
6
|
Ph-Benza
|
12
|
6
|
6
|
Ph-Ch
|
12
|
6
|
6
|
Ph-Py
|
13
|
6
|
6
|
Ph-Fl
|
12
|
6
|
6
|
Ph-Fl-Py-Ch
|
9
|
6
|
6
|
To
|
To
|
5
|
2
|
2
|
To_M/P
|
5
|
2
|
2
|
To_St
|
5
|
2
|
2
|
To-Benze
|
6
|
2
|
2
|
To-Benze-St
|
5
|
2
|
2
|
Ph: phenanthrene; Benza: benzaanthracene; Ch; chrysene; Py: pyrene; Fl: fluoranthene; |
To: toluene; M/p: m/p-xylene; St: styrene; Benze: benzene.