Chemometric analysis of ecological toxicants in petrochemical and industrial environments Richard Olawoyin a,⇑ , Brenden Heidrich b , Samuel Oyewole c , Oladapo T. Okareh d , Charles W. McGlothlin a a Oakland University, School of Health Sciences, Rochester, MI 48309, USA b Irradiation Services and Operations, Radiation Science & Engineering Center, The Pennsylvania State University, USA c U.S. Chemical Safety & Hazard Investigation Board, Washington D.C. Metro Area, USA d Division of Environmental Health Sciences, College of Medicine, University of Ibadan, Ibadan, Nigeria highlights Comprehensive review of chemometrics techniques in environmental assessments. Chemometrics methods were used to assess the study area – the Niger Delta. Identification of pollution sources and distributions were done using the PCA. PCA minimized and orthogonolize the multivariable datasets from samples. Accurate results obtained can enhance appropriate mitigating measure designs. article info Article history: Received 4 February 2014 Received in revised form 14 March 2014 Accepted 23 March 2014 Handling Editor: Tamara S. Galloway Keywords: Chemometrics Toxicants Principal component analysis Pollutants Environment Variogram abstract The application of chemometrics in the assessment of toxicants, such as heavy metals (HMs) and polycyclic aromatic hydrocarbons (PAHs) potentially derived from petrochemical activities in the microenvironment, is vital in providing safeguards for human health of children and adults residing around petrochemical industrial regions. Several multivariate statistical methods are used in geosciences and environmental protection studies to classify, identify and group prevalent pollutants with regard to exhibited trends. Chemometrics can be applied for toxicant source identification, estimation of contam- inants contributions to the toxicity of sites of interest, the assessment of the integral risk index of an area and provision of mitigating measures that limit or eliminate the contaminants identified. In this study, the principal component analysis (PCA) was used for dimensionality reduction of both organic and inorganic substances data in the environment, which are potentially hazardous. The high molecular weight (HMW) PAHs correlated positively with stronger impact on the model than the lower molecular weight (LMW) PAHs, the total petroleum hydrocarbons (TPHs), PAHs and BTEX correlate positively in the F1 vs F2 plot indicating similar source contributions of these pollutants in the environmental material. Cu, Cr, Cd, Fe, Zn and Pb all show positive correlation in the same space indicating similar source of contamination. Analytical processes involving environmental assessment data obtained in the Niger Delta area of Nigeria, confirmed the usefulness of chemometrics for comprehensive ecological evaluation. Ó 2014 Elsevier Ltd. All rights reserved. 1. Introduction Environmental media are contaminated by pollutants via the deposition of suspended particulate matter (Morillo et al., 2007), contaminated sediments deposition (Zhou et al., 2007) and also through underground water circulation (Duruibe et al., 2007). The contaminants deposited in the soil, sediment, air or water can migrate to the human body through inhalation in their natural state when re-suspended as particulate matters, as soil particles or as adsorbates on dust (Shi et al., 2008). Several studies have investigated environmental contamina- tions of soils in and around cities with children’s playgrounds (Chen et al., 2005; Vidovic et al., 2005; De Miguel et al., 2006). The major routes of exposure of humans to environmental toxi- cants are through inhalation, ingestion and dermal contact. Several studies have identified accidental ingestion of soil as the route with the highest tendency of human exposure (especially children) to http://dx.doi.org/10.1016/j.chemosphere.2014.03.107 0045-6535/Ó 2014 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. Tel.: +1 248 364 8653. E-mail address: [email protected](R. Olawoyin). Chemosphere 112 (2014) 114–119 Contents lists available at ScienceDirect Chemosphere journal homepage: www.elsevier.com/locate/chemosphere
Richard Olawoyin a,⇑, Brenden Heidrich b, Samuel Oyewole c, Oladapo T. Okareh d, Charles W. McGlothlin a
a Oakland University, School of Health Sciences, Rochester, MI 48309, USAb Irradiation Services and Operations, Radiation Science & Engineering Center, The Pennsylvania State University, USAc U.S. Chemical Safety & Hazard Investigation Board, Washington D.C. Metro Area, USAd Division of Environmental Health Sciences, College of Medicine, University of Ibadan, Ibadan, Nigeria
h i g h l i g h t s
� Comprehensive review of chemometrics techniques in environmental assessments.� Chemometrics methods were used to assess the study area – the Niger Delta.� Identification of pollution sources and distributions were done using the PCA.� PCA minimized and orthogonolize the multivariable datasets from samples.� Accurate results obtained can enhance appropriate mitigating measure designs.
a r t i c l e i n f o
Article history:Received 4 February 2014Received in revised form 14 March 2014Accepted 23 March 2014
The application of chemometrics in the assessment of toxicants, such as heavy metals (HMs) andpolycyclic aromatic hydrocarbons (PAHs) potentially derived from petrochemical activities in themicroenvironment, is vital in providing safeguards for human health of children and adults residingaround petrochemical industrial regions. Several multivariate statistical methods are used in geosciencesand environmental protection studies to classify, identify and group prevalent pollutants with regard toexhibited trends. Chemometrics can be applied for toxicant source identification, estimation of contam-inants contributions to the toxicity of sites of interest, the assessment of the integral risk index of an areaand provision of mitigating measures that limit or eliminate the contaminants identified. In this study,the principal component analysis (PCA) was used for dimensionality reduction of both organic andinorganic substances data in the environment, which are potentially hazardous. The high molecularweight (HMW) PAHs correlated positively with stronger impact on the model than the lower molecularweight (LMW) PAHs, the total petroleum hydrocarbons (TPHs), PAHs and BTEX correlate positively in theF1 vs F2 plot indicating similar source contributions of these pollutants in the environmental material.Cu, Cr, Cd, Fe, Zn and Pb all show positive correlation in the same space indicating similar source ofcontamination. Analytical processes involving environmental assessment data obtained in the NigerDelta area of Nigeria, confirmed the usefulness of chemometrics for comprehensive ecological evaluation.
� 2014 Elsevier Ltd. All rights reserved.
1. Introduction
Environmental media are contaminated by pollutants via thedeposition of suspended particulate matter (Morillo et al., 2007),contaminated sediments deposition (Zhou et al., 2007) and alsothrough underground water circulation (Duruibe et al., 2007).The contaminants deposited in the soil, sediment, air or water
can migrate to the human body through inhalation in their naturalstate when re-suspended as particulate matters, as soil particles oras adsorbates on dust (Shi et al., 2008).
Several studies have investigated environmental contamina-tions of soils in and around cities with children’s playgrounds(Chen et al., 2005; Vidovic et al., 2005; De Miguel et al., 2006).The major routes of exposure of humans to environmental toxi-cants are through inhalation, ingestion and dermal contact. Severalstudies have identified accidental ingestion of soil as the route withthe highest tendency of human exposure (especially children) to
R. Olawoyin et al. / Chemosphere 112 (2014) 114–119 115
contaminants (Ljung et al., 2006; Shi et al., 2008). Volatile toxicantsare easily inhaled than ingested but chances of ingesting other tox-icants such as lead (Ljung et al., 2006) are higher than beinginhaled.
The use of multivariate analytical techniques have gained moresteam in recent years but univariate statistical analyses remainuseful to a wide range of other environmental scientists. Signifi-cant use of chemometrics methods are available, which includesource apportionment methods (Zhang et al., 2008; Chen et al.,2009), principal component analysis (PCA) and linear discriminantanalysis (LDA) (Seilaff and Einax, 2007). Multivariate analyticaltechniques give more details on the data structure collected fromenvironmental media than univariate methods since multipleparameters are measured during soil, sediment, water and air qual-ity analysis. In soil analytical studies, a variety of geostatisticaltools with a wide range of applications exist (Kanevski et al., 2004).
Attempts have been made by researchers to analyze; polycyclicaromatic hydrocarbons (PAHs) Table 1, polychlorinated napthenes(PCNs) and heavy metals (HMs) contaminations levels in environ-mental media chemometrics techniques (Bostrom et al., 2002;Mielke et al., 2004; Jarosinska et al., 2006). Although these sub-stances are ubiquitous in the environment (Zhang et al., 2006),but there are concerns when the levels are significantly higher thananticipated background concentrations.
1.1. Pollutants and associated features
1.1.1. Fate and effectThe contamination levels of PAHs, PCNs and HMs are elated in
environmental media due to contributions from anthropogenicactivities such as petrochemical industrial activities (Morilloet al., 2007). The effects of these toxicants on human health rangefrom mild to severe depending on the concentration and exposureof either an individual toxicant or a mixture of different organicand inorganic toxicants.
2. Methodology
2.1. Principal component analysis (PCA)
The PCA method has improved considerably in recent times, thePCA method has been noted for its applicability as a classic statis-tical and diagnostic tool (Diamantaras and Kung, 1996), whichhelps to analyze covariance structures embedded in any multivar-iate numerical datasets, detect outliers, validate and interpret
Table 1PAHs characteristics in sampled media.
Compound Abbreviation Benzene ring
Naphthalene Nap 22-Methylnaphthalene 2MNap 2Acenaphthylene Acy 2Acenaphthene Ace 2Fluorene Flu 3Phenanthrene Phe 3Anthracene Ant 3Fluoranthene Flr 4Pyrene Pyr 4Benz[a]anthracene BaA 4Chrysene Chr 4Benzo[b]fluoranthene BbF 4Benzo[k]fluoranthene BkF 4Benzo[a]pyrene BaP 5Dibenz[a,h]anthracene DahA 5Benzo[ghi]perylene BghiP 6Indeno(1,2,3-cd)pyrene InP 6Benzene, Toluene, Ethylene and Xylene BTEX
datasets, and for visualization. Simplified explanation of the PCAmethod is presented in published studies (Shaw, 2003). However,the specific mathematical manipulations behind the PCA methodare presented in other literatures (Jolliffe, 2002; Jackson, 2003).The fundamentals of PCA involves the transformation of a set ofmultivariate data containing analytical constituents (variables)into a new orthogonal set by allocating total variance to uncorre-lated variables (principal components – PCs) using the correlationmatrix, whereby the individual variable represent the linear com-bination of the initial data variables. The PCs are in decreasingorder based on factor loading, having the PCs with the largestvariance occupying the first PC (PC1) and this follows successivelyto the PC with the smallest variance (PCn). Eq. (1) shows themathematical expression for the PC computation with respect tothe original variables.
PC1 ¼ a11x1 þ a12x2 þ . . .þ a1nxn
PCn ¼Xn
j¼1
a1jxjð1Þ
where a1j is the eigenvectors obtained from the correlation matrix,and xj is the input variables.
The PCA method is performed sequentially, first by informationextraction in the input space (with n-dimensions) to determine thedirections of which the input variables xj display the most substan-tial variability. The PC coefficients and the eigenvaluesðki > 0; i ¼ 1; . . . ;nÞ for the correlation matrix ðC ¼ E xxT
� �Þ with
respect to their eigenvectors (ei > 0, i = 1, . . ., n) make up what iscalled the loadings are then calculated. This gives a new set ofvariables that explain the variability in the original dataset; thefirst PCs retain a greater proportion of the total variance, conse-quently leading to effective and practical dimensionality reductionexercise. This is an effective means of describing sources, distribu-tions and concentrations of pollutants in environmental materialsby displaying multivariate patterns in the original datasets inlower dimensions. However a rational amount of PCs are necessaryto be chosen for the avoidance of losing meaning information andpoor predictability (Jolliffe, 2002).
3. Principal component analysis (PCA) results
The datasets used in this study include soil, sediment and waterquality assessment data collected at different locations in the NigerDelta area of Nigeria (Olawoyin et al., 2012; 2013). The datainclude physic-chemical, organic and inorganic variables. ThePCA was carried out using SAS� version 9.3 and the XLSTAT. Vari-max rotation was applied to simplify the factor interpretation byreducing the total number of variables that exhibit high loadingsper factor (Everitt and Dunn, 1992). For all the pollutant variablesanalyzed the correlation matrix was built with equal weights(Chatfield and Collins, 1980).
3.1. Organic substances
The summary statistics for the soil and water PAHs, andsediment physico-chemical valid data (Olawoyin et al., 2013)calculated are presented in Table 2.
The interpretation of PCs in this study comprises of the evalua-tion of the relationship between the loadings and contaminationsources from petrochemical activities. PCs with variables with highloadings depict greater importance from the contaminationsources, whereas lower loadings point to lower importance withregards the sources of these contaminations. The PCA in this studysuggest a robust solution for the dimensionality reduction of vari-ables based on the score calculation. For the soil PAHs, majority ofthe pollutants correlation show similar trends in the output space
Table 2Summary Statistics of PAHs.
Variable Nap BghiP 2MNap Acy Ace Flu Phe Ant Flr Pyr BaA Chr BbF BkF BaP DahA InP
116 R. Olawoyin et al. / Chemosphere 112 (2014) 114–119
which indicate common source of pollution. The principal compo-nent matrix have similar values indicating strong correlation val-ues between the variables, this matrix showed strong correlationin water heavy molecular weight (HMW) PAHs, while Table 3showed strong correlation between the PAHs and BTEX variablesfor the sediment parameters. The eigenvalues were chosen for 7principal components based on variability calculated using valuesof the eigenvector and factor loadings for the soil parameters(Table 3). The computed eigenvalues for the soil, water andsediment variables are presented in Table 3.
Table 3 showed that 10 and 9 PCs were chosen for the water andsediment variables respectively. The cut-off eigenvalues ki was setat 1.0 for the PCA. The scree plot for the factor variability (soil). Thepercentage contribution of the 1th–2th components for the soilvariables as represented in Eq. (2) is 89.0%,
X2
i¼1
ki
)( , X7
i¼1
ki
( ) !ð2Þ
this necessitated the selection of the first two parameters as the PCssince there is significant evidence of high enough total variancefrom the percentage contributions. The eigenvalueskiðk3; k4; k5; k6; k7Þ have little contributions to the total structure of
the data under study. The percentage contribution of the 3th–7thcomponents (Eq. (3)) is 11%. This suggests that very little informa-tion (which can be considered negligible) will be lost.
X7
i¼3
ki
( ) X7
i¼1
ki
( ), ! ð3Þ
However, the percentage contribution of the 1th–3th componentsfor the water variables as represented in Eq. (4) is 90.4%,
X3
i¼1
ki
( ) X10
i¼1
ki
( ), ! ð4Þ
which dictated the selection of the first three parameters as the PCsand the eigenvalues kiðk4; k5; k6; k7; k8; k9; k10Þ have minimal contri-butions to the nature of the general data. The percentage contribu-tion of the 4th–10th components as illustrated in Eq. (5) is 9.6%.
X10
i¼4
ki
( ) X10
i¼1
ki
( ), ! ð5Þ
For the sediment parameters, the total percentage contribution ofimportant is 81.1% which represent the 1th–4th components vari-ables as shown in Eq. (6).
Fig. 1. Projection of variables and their correlation in the factor space for soilsamples.
R. Olawoyin et al. / Chemosphere 112 (2014) 114–119 117
X4
i¼1
ki
( ) X9
i¼1
ki
( ), ! ð6Þ
Marginal contributions were derived from eigenvalueskiðk5; k6; k7; k8; k9Þ that are not substantial to the entire data settherefore defining the PCs. Eq. (7) represent the percentage contri-bution of the 5th–9th components which amount to 18.9%.
X9
i¼5
ki
( ) X9
i¼1
ki
( ), ! ð7Þ
It is important to display a graphical representation of the loadingvectors in order to determine the most influential variables interac-tion in the model. Variables are distributed into the four differentquadrants relative to the PCs as shown in Fig. 1.
Appearance of variables on the same quadrant indicatespositive correlation between them, if orthogonal; this means thatthe variables are slightly uncorrelated or not correlated, whilethose on opposite sides of the quadrants are significantlynegatively correlated with regards the PC of reference.
Fig. 2. Illustrating the variables c
The impact of any individual variable to the entire PCA modelcan be measured by the distance by how far the variable is fromthe origin. Variables that show longer distances have largerimpacts on the general architecture of the model than variableswith shorter distances, for instance, in the loading plot correspond-ing to the first two PCs in Fig. 1; the following 4 variables; BaP,DahA, BkF, BbF showed clear positive correlation but they havestronger impact on the PCA model than Flr, which also correlatespositively with the 4 variables. Likewise, the 5 variables; BaA,Pyr, Ace, Nap, 2Mnap correlate positively together but with stron-ger impact on the model than Flu and Acy with Phe having the leastimpact on the PCA model. Shown in Fig. 2, it is clear that most ofthe HMW PAHs are positively correlated with stronger impact onthe model than the lower molecular weight (LMW) PAHs. Ingeneral, the water PAHs correlate positively in the PC1 (F1) space.
Total petroleum hydrocarbons (TPHs), PAHs and BTEX correlatepositively in the F1 vs F2 plot indicating similar source contribu-tions of these pollutants in the sediment samples (Fig. 3). Otherphysic-chemical variables such as EC, Salinity and SO4 show nega-tive correlation with the organic pollutants, this shows potentialsof multiple pollutions, which could be due to both crude oil con-tamination and gas flaring events. Fig. 4 illustrates the correlationbetween the observation locations, indicating a nucleated patternfor most of locations. This suggests high similarities between thelocations, but due to locational, geological and concentrationdifferences, some of the locations showed negative correlations.The biplot represented in Fig. 4 depicts the correlation betweenthe variables and how strongly they impact the overall model withrespect to the observations.
There seem to be a consistent trend in the correlation betweenvariables for the soil, water and sediment cases. The plots (Figs. 3and 4) showed positive correlations between the variables, obser-vations and both as well as some areas with negative correlation.The observations that are uncorrelated and placed isolated fromthe rest of the group validly shows areas with no contaminationsamples for quality control, for instance ERS14 and BS45 (Fig. 4).
3.2. Inorganic substances in soils
Soil heavy metals analyzed (Olawoyin et al., 2012) were treatedwith the varimax rotation and PCA method. The analysis of PCs
orrelation for water samples.
Fig. 3. Illustrating the variables correlation in the sediment.
Fig. 4. Correlation circle for soil inorganic variables (F1 vs F2; F1 vs F3).
118 R. Olawoyin et al. / Chemosphere 112 (2014) 114–119
entails the evaluation of the link that exists between the sources ofcontamination and the effective loadings that define the PCs.
Table 4 illustrates the sparse distribution of pollutant correla-tions which partly indicates the extent and variability of the pollu-tants in the soil. A total of 13 PCs were identified in the model butnot all of these contribute largely to the variability in the dataset.
The cut-off eigenvalues ki was set at 1.0 for the PCA also for thesoil inorganic datasets (Table 4). The percentage contribution ofthe 1th–6th components (Eq. (8))
X6
i¼1
ki
( ) X13
i¼1
ki
( ), ! ð8Þ
Table 4Principal component analysis: heavy metals.
is 75.6%, which translates to the fact that only 6 PCs strongly impactthe model due to high enough percentage contributions to the totalvariance. The eigenvalues kiðk7; k8k9; k10; k11; k12; k13Þ contributed24.4% to the total structure of the data studied based, the percent-age contribution of the 7th–13th components can be summarizedas shown in Eq. (9).
X13
i¼7
ki
( ) X13
i¼1
ki
( ), ! ð9Þ
This suggests that relatively few information would be lost but thegeneral structure of the datasets would be preserved using these 6
Fig. 5. PC biplot between F1/F2 and F1/F3 for observation samples and variables (inorganic variables).
R. Olawoyin et al. / Chemosphere 112 (2014) 114–119 119
PCs. The graphical representation of the correlation circle of thedifferent variables is seen in Fig. 5. The most impactful variablesare shown with distances farther away from the axis origin on thecorrelation circle, (i.e. longer radius suggest stronger impacts ofthe inorganic variable to the soil data plotted in the PCs). Cu, Cr,Cd, Fe, Zn and Pb all show positive correlation in the same spaceindicating similar source of contamination (Fig. 5).
The BN location is strongly and positively correlated to most ofthe heavy metals such as Zn, Pb, Fe, Cd and Cr. The strong correla-tion of Pb in this location potentially resulted from heavy equip-ment used for dredging, petrochemical terminal activities etc. Itcan also be observed that positive correlation has been establishedbetween the SO4, PO4 and pH which shows a potential source dif-ferent from the heavy metal source, this might be due to gas flaringin the correlated areas (OD, ER and OG).
4. Summary
The unique capabilities of the PCA model in dimensionalityreduction of multivariate data have increasing made the techniqueideal in determining the aggregated influence of environmentalindicators. As presented in the results of this study, the PCA wasapplied to minimize and orthogonolize the multivariable datasetsfrom three environmental materials; soil, water and sediment withan attempt to interpret the correlation between them. It alsoeliminates the interdependency challenge between the variousindicators since the variable indicators with strong and positivecorrelations were grouped in the same PC.
PAHs such as BaP, DahA, BkF, BbF showed distinct positive linkon the PCA model than Flr. Similarly, BaA, Pyr, Ace, Nap, 2Mnapshowed positive correlation with stronger effect on the model thanFlu and Acy with Phe having the least impact on the PCA model. Forthe inorganic components, the source of Pb was recognizable fromthe model which potentially resulted from heavy equipment usedfor dredging, petrochemical terminal activities such as pipe layingand flow stations constructions. These correlations showed thepotential source different from the heavy metal source, whichcould be from the prevalent gas flaring activities in the correlatedareas (OD, ER and OG).
The use of the PCA as a multivariate tool allows for severalecological variables and assessment data to be incorporated intocategories with substantial importance.
References
Bostrom, C., Gerde, P., Hanberg, A., Jernstrom, B., Johansson, C., et al., 2002. Cancerrisk assessment, indicators and guidelines for polycyclic aromatic hydrocarbonsin ambient air. Environ. Health Perspect. 110 (Supplement 3), 451–488.
Chatfield, C., Collins, A.J., 1980. Introduction to Multivariate Analysis. Chapman andHall, London. 246 pp.
Chen, T.B., Zheng, Y.M., Lei, M., Huang, Z.C., Wu, H.T., Chen, H., Fan, K.K., Yu, K., Wu,X., Tian, Q.Z., 2005. Assessment of heavy metal pollution in surface soils ofurban parks in Beijing. China Chemosphere 60, 542–551.
Chen, T., Liu, X., Li, X., Zhao, K., Zhang, J., Xu, J., Shi, J., Dahlgren, R.A., 2009. Heavymetal sources identification and sampling uncertainty analysis in a field-scalevegetable soil of Hangzhou China. Environ. Pollut. 157, 1003–1010.
De Miguel, E., Iribarren, I., Chacon, E., Ordonez, A., Charlesworth, S., 2006. Risk-based evaluation of the exposure of children to trace elements in playgroundsin Madrid (Spain). Chemosphere 66 (3), 505–513.
Diamantaras, K., Kung, S., 1996. Principal Component Neural Networks – Theoryand Application. Wiley.
Duruibe, J.O., Ogwuegbu, M.O.C., Egwurugwu, J.N., 2007. Heavy metal pollution andhuman biotoxic effects. Int. J. Phys. Sci. 2 (5), 112–118.
Everitt, B.S., Dunn, G., 1992. Applied Multivariate Data Analysis. Oxford UniversityPress, New York. 304 pp.
Jackson, J.E., 2003. A User’s Guide to Principal Components. John Wiley & Sons, NewYork. 569.
Jarosinska, D., Biesiada, M., Miszynska-Graca, M. 2006. Environmental burden ofdisease due to lead in urban children from Silesia Poland Science of the TotalEnvironment the 367, 71–79.
Jolliffe, I.T., 2002. Principal Component Analysis, Second ed. Springer-Verlag, NewYork. 487 pp.
Kanevski, M., Parkin, R., Pozdnukhov, A., Timonin, V., Maignan, M., Demyanov, V.,Canu, S., 2004. Environmental data mining and modeling based on machinelearning algorithms and geostatistics. Environ. Modell. Software 19 (9), 845–855.
Ljung, K., Selinus, O., Otabbong, E., 2006. Metals in soils of children’s urbanenvironments in the small northern European city of Uppsala. Sci. Total Environ.366 (2–3), 749–759.
Mielke, H.W., Berry, K.J., Mielke, P.W., Powell, E.T., Gonzales, C.R., 2004. Multiplemetal accumulation as a factor in learning achievement within various NewOrleans elementary school communities. Environ. Res. 97 (1), 67–75.
Morillo, E., Romero, A.S., Maqueda, C., Madrid, L., Ajmone-Marsan, F., Grcman, H.,Davidson, C.M., Hursthouse, A.S., Villaverde, J., 2007. Soil pollution by PAH’s inurban soils: a comparison of three European cities. J. Environ. Monitor. 9, 1001–1008.
Olawoyin, R., Oyewole, S.A., Grayson, R.L., 2012 Potential risk effect from elevatedlevels of soil heavy metals on human health in the Niger delta Ecotoxicol.Environ. Saf, 85, 1 November 2012 pp. 120–130 <http://dxdoiorg/101016/jecoenv201208004>.
Olawoyin, R., Nieto, A., Grayson, R.L., Hardisty, F., Oyewole, S., 2013. Application ofArtificial Neural Network (ANN) – Self-Organizing Map (SOM) for thecategorization of water, soil and sediment quality in petrochemical regions.Exp. Syst. Appl. 40 (9), 3634–3648.
Shaw, P.J.A. 2003. Multivariate Statistics for the Environmental Sciences, ArnoldLondon, 233 pp.
Shi, G., Chen, Z., Xu, S., Zhang, J., Wang, L., Bi, C., Teng, J., 2008. Potentially toxicmetal contamination of urban soils and roadside dust in Shanghai China.Environ. Pollut. 156, 251–260.
Seilaff, K., Einax, J.W., 2007. The application of multivariate statistical methods forthe evaluation of soil profiles. J. Soils Sediments 7 (1), 45–52.
Vidovic, M., Sadibasic, A., Cupic, S., Lausevic, M., 2005. Cd and Zn in atmosphericdeposit soil wheat and milk. Environ. Res. 97 (1), 26–31.
Zhang, H.B., Luo, Y.M., Wong, M.H., Zhao, Q.G., Zhang, G.L., 2006. Distributions andconcentrations of PAHs in Hong Kong soils. Environ. Pollut. 141, 107–114.
Zhang, W., Zhang, S., Wan, C., Yue, D., Ye, Y., Wang, X., 2008. Source diagnostics ofpolycyclic aromatic hydrocarbons in urban road runoff dust rain and canopythrough fall. Environ. Pollut. 153, 594–601.
Zhou, F., Guo, H., Liu, L., 2007. Quantitative identification and source apportionmentof anthropogenic heavy metals in marine sediment of Hong Kong. Environ.Geology 53 (2), 295–305.
