Environmental Monitoring and Assessment

, Volume 185, Issue 7, pp 5951–5964

Soil heavy metal contamination in an industrial area: analysis of the data collected during a decade

Authors

    • IMAA—Istituto di Metodologie per l’ Analisi Ambientale CNR
  • Rosa Caggiano
    • IMAA—Istituto di Metodologie per l’ Analisi Ambientale CNR
  • Maria Macchiato
    • Dipartimento di Scienze FisicheUniversità Federico II
  • Maria Ragosta
    • Scuola di IngegneriaUniversità degli Studi della Basilicata e CINFAI UdR Basilicata
  • Serena Sabia
    • IMAA—Istituto di Metodologie per l’ Analisi Ambientale CNR
Article

DOI: 10.1007/s10661-012-2997-y

Cite this article as:
D’Emilio, M., Caggiano, R., Macchiato, M. et al. Environ Monit Assess (2013) 185: 5951. doi:10.1007/s10661-012-2997-y

Abstract

Soil contamination by heavy metals has become a serious problem mainly because, above certain concentrations, all metals have adverse effects on human health. In particular, the accumulation of heavy metals in agricultural soils leads to elevated uptake by crops and affects food quality and safety. In this paper, we present the results of a study carried out over a decade for evaluating the impact of a new industrial settlement in an area geared to agriculture and livestock and far from urban sites. We focus our study on the bioavailable fraction of Cd, Co, Cr, Cu, Fe, Mn, Ni, Pb and Zn in soil samples. Heavy metal concentrations in soil are analysed with both univariate and multivariate statistical procedures. The main goal of this paper is the development of a statistical procedure, based on a mix of multivariate analysis, able to compare field surveys carried out during different years and to characterize spatial and temporal changes in soil heavy metals concentrations.

Keywords

Soil contaminationHeavy metalIndustrial activitiesCluster analysis

Introduction

Metals are natural components of soil, rock, air, water and living organisms (Besada et al. 2011). Heavy metal concentrations in soils are associated with complex biological and geochemical cycles; they are influenced by anthropogenic factors such as agricultural practices, industrial activities, waste treatments and vehicular traffic (Buccolieri et al. 2010; Al-Khashman 2004; Ramos-Miras et al. 2011; Smith 2009; Ajmone-Marsan and Biasioli 2010; Fabietti et al. 2010). The presence of these pollutants in soils with different types of land use may induce different impacts on public health, so the study of their concentrations in soils characterized by different land uses is a relevant topic (Xia et al. 2011; Chen et al. 2010).

Some metals, like Cu, Mn and Zn, are essential to plant metabolism in trace amounts, but soils can act as sinks for these pollutants leading to phytotoxicity at high concentrations (Brunetti et al. 2009; Nagajyoti et al. 2010; Wu et al. 2010). In the long term, the accumulation of heavy metals in soil could affect the quality of the agricultural soils and could cause the transfer of toxic elements to the human diet as a result of increased crop uptake or soil ingestion by grazing livestock (Poggio et al. 2009; Reis et al. 2012).

Heavy metals are present in various forms in soil. Different forms have different mobility and phytoextractability. Generally, plant uptake of heavy metals is correlated with extractable forms of the metals rather than with the total metal content in the soil. In particular, when metals are present in bioavailable forms, they have the potential to become toxic (D’Amore et al. 2005). Thus, the bioavailable soil metal fraction is the primary concern in the study of metal toxicity (Lamb et al. 2009; Abdu et al. 2012). Unfortunately, the meaning of the term “bioavailable” is not univocal and depends on the context: Bioavailability may represent the fraction of a chemical accessible to an organism for absorption, i.e. the rate at which a substance is absorbed into a living system, or a measure of the potential to cause a toxic effect (Semenzin et al. 2007; Smith 2009). For this reason, there are not a unique protocol and method accepted by the entire scientific community for the evaluation of the bioavailability of chemical elements (Agbenin and Welp 2012; Harmsen et al. 2005). In this work, we consider the bioavailable concentration of heavy metals as the fraction of the pollutant accessible to an organism for absorption, and we measure it by means of a protocol defined by Italian law during 1992 (AAVV 1992), i.e. 1 year before we started our study.

In this paper, we present the results of a study carried out over a decade (1993–2004), concerning the characterization of spatial–temporal patterns of metal bioavailable fraction aimed to evaluate the impact of a new industrial settlement in an area previously geared to agriculture and livestock and far from urban settlements. Understanding the behaviour of heavy metals in a heterogeneous system, such as a soil, is imperative to characterize their mobility and to prevent contaminations of the food chain. Unfortunately, there are few studies comparing different field surveys and suggesting a useful method for the comparison of data sets coming from field surveys carried out in the same area during several years. The final goal of this paper is to propose a methodology based on a mix of different statistical multivariate analysis, currently used for studying soil contamination (Xu and Tao 2004; Micò et al. 2006; Akbar Jan et al. 2010), aimed to define the spatial patterns of measured heavy metals and to compare their spatial and temporal distributions.

Materials and methods

The study area

S. Nicola di Melfi area is the most important industrial zone of Basilicata region, southern Italy, with an extension of about 10 km2. From a geographical point of view, S. Nicola di Melfi industrial area is a part of Vulture basin, a valley along the Ofanto River in the north of Basilicata region. It is surrounded by low hills varying from 137 to 508 m a.s.l. There are two urban sites, Melfi (13,000 inhabitants) and Lavello (16,000 inhabitants), about 10 km far from the industrial area. This is a rather unpolluted zone where the main land use is agriculture, wheat being the most spread cultivation, and where many milk cow farming are present.

The meteorological features of the area are typical of inner and mountainous regions. The annual average temperature is 13 °C, the rainfalls are more frequent in winter and the prevailing wind direction is N-NW (D’Emilio et al. 2010).

In this area, from 1991 to 1993, the FIAT-SATA automotive plant was built up. It has more than 7,000 employees, and it was scheduled for producing 450,000 vehicles a year. Since October 2000, a waste pre-treatment plant and an integrated platform for incineration (SATA FENICE project), combining waste management and electric energy production, are operative. The incinerator has two ovens: a rotating oven in which 20,000 t a year of toxic and harmful waste produced by SATA and 14,400 t a year of toxic waste coming out from Basilicata region are collected and a grid oven in which 30,000 t of municipal solid waste coming from the entire region is collected.

In the same industrial area, there are about 30 plants linked to FIAT-SATA production. Starting from 1994, an important food industry (Barilla) covering an area of 51,000 m2 with 350 employers is also operative. Other small industries are also present in the area. All these industrial activities determine a high volume of vehicular traffic in the surrounding areas.

Georeferenced sampling grid

The sampling grid is composed by 110 georeferenced sampling points set along eight transects (TrA, TrB, TrC, TrD, TrE, TrF, TrG, TrH) which converged towards SATA plants. It was georeferenced by using a global positioning system mapping and a geographical information system data collection/maintenance system Trimble geoexplorer. In Fig. 1, we show the overlay of the sampling grid on the topographical map of the industrial area. The distance between two points of the same transect is 500 m for sampling points located at a distance up to 4 km from SATA industrial settlement and of 1 km for sampling points located at a distance comprised from 4 to 8 km from SATA industrial settlement. As a whole, the sampling grid covers an area of about 200 km2. The sampling points take into account the different geological features (alluvial, sandy, clayey soils) and the different land uses (industrial, agricultural).
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig1_HTML.gif
Fig. 1

Overlay of the georeferenced sampling grid and topographical map of S. Nicola di Melfi industrial area. Different transects (TrA, TrB,…,TrH) are indicated by different bullets

Sampling and analytical protocols

During autumn from 1993 to 2004, we carried out seven field surveys (fs): fs1 in 1993, fs2 in 1994, fs3 in 1995, fs4 in 1996, fs5 in 2000, fs6 in 2002 and fs7 in 2004. For each sampling points, we selected a small square area (1 × 1 m), located at least 5 m from the roadside. At the angles of the square, we collected four soil samples at a deep of about 5 cm. The four samples were mixed in a polyethene bag. The samples were dried in an oven at a temperature of 50 °C. Then they were grounded and sieved with a 2-mm sieve.

Because of a standard method for assessing bioavailability of metals in terrestrial ecosystems has not been universally established (Agbenin and Welp 2012), we followed the Italian law (AA. VV. D.M. 11.05 1992) and used an ethylenediaminetetraacetic acid-extractable metal procedure for soil digestion. For evaluating the bioavailable fraction of nine heavy metals (Cd, Co, Cr, Cu, Fe, Mn, Ni, Pb and Zn), we used the AAS technique.

Data analysis

In order to point out temporal changes in metal concentrations and variations of the spatial patterns, we applied univariate and multivariate statistical procedures.

Univariate analysis

A preliminary univariate study was performed taking into account all the samples and all the metals; we analysed a dataset composed by about 6,200 concentration values. The explorative statistical parameters: mean value (m), median (md), standard deviation (sd), minimum and maximum value (min–max) were calculated for each metal and for each field survey. For comparing values measured in the same sampling point and for highlighting the presence of trends in concentration values, for each metal we plotted the data collected during the different field surveys.

Multivariate analysis

In our multivariate procedure, we analysed seven data matrices [sampling points (objects) × metal concentrations (descriptors)]: M1 [110 × 9], M2 [110 × 8], M3 [87 × 9], M4 [87 × 9], M5 [107 × 9], M6 [103 × 9], M7 [99 × 9], one for each field survey. In particular, we carried out a double classification procedure: a cluster analysis for descriptors and a cluster analysis for objects.

In the first classification procedure (descriptor clustering), for each data matrix (M1,…,M7), we determined the descriptor association matrix evaluating, for each couple of variables, the Pearson’s r coefficient. For clustering, we applied the complete linkage algorithm and we plotted the results by means of a dendrogram. This classification procedure pointed out the presence of systematic correlation among measured elements and was useful for detecting the evolution of the correlation patterns. In this case, the statistical significance of the clusters was evaluated using the statistical test for r-values.

In the second classification procedure, (object clustering), for each data matrix (M1,…,M7), we calculated the sampling point association matrix using the Euclidean distance. For clustering, we applied the complete linkage algorithm and an automated procedure for clusters identification requiring that the number of identified clusters was \( 3\leqslant {N_{\mathrm{clu}}}\leqslant 5 \). The analysis of the clusters was carried out by means of the centroid analysis. Centroids are endogenous indices defined as follows: For each descriptor (metal concentration), the centroid \( V_{j,n}^h \) is the mean value of n-th descriptor calculated only on the objects (sampling points) included in the j-th cluster of the h-th data matrix . To normalize the values and to simplify the result interpretation, it was useful to introduce the centroid percentage index defined as
$$ \left( {I_{j,n}^h} \right)\%=\left( {\frac{{V_{j,n}^h-V_n^h}}{{V_n^h}}} \right)\% $$
in which \( V_{j,n}^h \) is the centroid of n-th descriptor in the j-th cluster of the h-th data matrix (h = 1,…,7) and \( V_n^h \) is the mean value of n-th descriptor calculated on all the objects included in h-th data matrix. In particular, in order to identify significant variations between centroids and mean values, we fixed a threshold value, \( \left| {\left( {{I_{\mathrm{thr}}}} \right)\%} \right|=25\% \) and we underlined all the cases in which this threshold is exceeded. All the statistical analysis was performed by SPSS© software.

Results

Univariate analysis

The univariate parameters are shown in Table 1. In the first field survey, the mean values of heavy metal concentrations follow the sequence: \( \mathrm{Mn}\gg \mathrm{Fe}\gg \mathrm{Pb}\gg \mathrm{Co}\approx \mathrm{Cu}\approx \mathrm{Ni}\approx \mathrm{Zn}\gg \mathrm{Cd}>\mathrm{Cr} \). This sequence remains quite unchanged during all the field surveys.
Table 1

Parameters of explorative statistical analysis, concentrations are expressed in parts per million DW

 

M1 (1993)

M2 (1994)

M3 (1995)

M4 (1996)

M5 (2000)

M6 (2002)

M7 (2004)

n(hm)

Cd

m

0.18

 

0.33

0.29

0.80

0.28

0.85

 

md

0.19

 

0.30

0.27

0.72

0.26

0.83

 

sd

0.06

 

0.18

0.09

0.29

0.10

0.54

 

min–max

0.01–0.31

 

0.07–0.82

0.16–0.74

0.31–1.74

0.11–0.62

0.15–2.58

 

n

110

 

87

87

107

103

99

593

Co

m

4

3

2

3

2

3

4.5

 

md

4

2.5

1

2.5

1.5

3

4.5

 

sd

2

1

1

1

1

1

1

 

min–max

0.7–9.8

0.5–6.3

0.8–5.3

1.2–7.1

0.2–5.2

1.3–5.0

2.9–6.2

 

n

110

110

87

87

107

103

99

703

Cr

m

0.07

0.18

0.23

0.23

0.17

0.18

0.35

 

md

0.06

0.17

0.21

0.22

0.14

0.14

0.31

 

sd

0.03

0.08

0.10

0.10

0.10

0.13

0.23

 

min–max

0.02–0.16

0.05–0.49

0.06–0.45

0.06–0.50

0.04–0.43

0.02–0.77

0.05–1.47

 

n

110

110

87

87

107

103

99

703

Cu

m

3

6

5

6

4

9.5

6

 

md

3

5

5

5.5

3

6

5.5

 

sd

2

4

3.5

2.5

3

8

3

 

min–max

0.2–14.2

2.0–35.2

1.2–29.2

2.7–22.0

1.3–23.0

2.7–56.0

2.2–17.0

 

n

110

110

87

87

104

102

99

699

Fe

m

81

82

101

127

138

135

120

 

md

64

71

84

100

106

102

100

 

sd

56

43

53

62

85

81

63

 

min–max

16–285

23–205

37–256

35–315

32–459

38–424

44–310

 

n

110

110

87

87

107

103

99

703

Mn

m

219

236

250

282

318

298

298

 

md

183

193

206

260

289

289

304

 

sd

119

159

142

121

155

115

110

 

min-max

11–627

35–787

33–784

41–624

48–968

58–643

93–954

 

n

110

110

87

87

107

103

99

703

Ni

m

3

6

3

4

2

3

4

 

md

3

5

3

4

2

3

3

 

sd

1

4

1

1

1

2

3

 

min–max

0.3–8.0

1.2–13.3

1.3–7.6

2.9–6.4

0.6–6.1

1.2–15.7

1.0–15.1

 

n

110

110

87

87

107

103

99

703

Pb

m

10

12

16

15

7

8

11

 

md

10

12

16

15

6

8

11

 

sd

4

3

4

3

2

3

3

 

min–max

0.2–25.1

6.3–19.8

6.2–30.4

9.8–18.9

1.0–14.8

2.0–18.0

4.9–28.3

 

n

110

110

87

87

107

103

99

703

Zn

m

2

3

3

2.5

2

4.5

3

 

md

2

2

2

2

1.5

3

2

 

sd

3

3

2

1

1

5

2

 

min–max

0.1–15.4

1.3–21.0

1.4–13.5

1.0–7.4

0.7–5.9

0.7–32.1

1.1–13.3

 

n

110

110

87

87

107

103

99

703

 

n(fs)

990

880

783

783

960

926

891

 
        

ntot

6,213

m mean value, md median, sd standard deviation, minmax range, n number of measurements, n(fs) total number of measurements carried out during each field survey, n(hm) total number of concentration measurements of each metal, ntot total number of measurements

Mn and Fe are the dominant elements and their concentrations remain substantially unchanged during the investigated period (Figs. 2 and 3). This behaviour suggests that the two metals have a natural origin and they are characteristic of the studied area (Peris et al. 2008).
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig2_HTML.gif
Fig. 2

Mn concentration trends measured in the sampling points during the seven field surveys

https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig3_HTML.gif
Fig. 3

Fe concentration trends measured in the sampling points during the seven field surveys

For Pb, it is possible to note that soil concentrations show a constant increase up to 1996 followed by a high decrease in fs5 (year 2000) and a new increase from this year on (Table 1 and Fig. 4). This behaviour can be explained as follows: Starting from 1994, there was a progressive substitution of leaded gasoline with green petrol. The reduction of lead level in gasoline produced a significant reduction in the presence of this metal in the environment due to traffic emissions (Granero and Domingo 2002). But in the investigated area, starting from October 2000, a waste pre-treatment plant and an integrated platform for incineration had been operative. The presence of this new anthropogenic activity caused a new gradual increase in lead concentrations.
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig4_HTML.gif
Fig. 4

Pb concentration trends measured in the sampling points during the seven field surveys

In Figs. 5 and 6, Cu and Zn concentrations are shown: They are characterized by the occurrence of many peaks, particularly evident during 2002. Co (Fig. 7) shows a very irregular behaviour for the whole investigated period as well as Ni, showing, in addition, very high peaks during fs2 (Fig. 8).
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig5_HTML.gif
Fig. 5

Cu concentration trends measured in the sampling points during the seven field surveys

https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig6_HTML.gif
Fig. 6

Zinc concentration trends measured in the sampling points during the seven field surveys

https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig7_HTML.gif
Fig. 7

Co concentration trends measured in the sampling points during the seven field surveys

https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig8_HTML.gif
Fig. 8

Ni concentration trends measured in the sampling points during the seven field surveys

Cd shows an increase in the concentrations measured during 2000 and 2004 field survey (Fig. 9). Cr shows a low increase only for the concentrations measured during 2004 (Fig. 10). This increase is more evident only in specific zones of the investigated area.
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig9_HTML.gif
Fig. 9

Cd concentration trends measured in the sampling points during the seven field surveys

https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig10_HTML.gif
Fig. 10

Cr concentration trends measured in the sampling points during the seven field surveys

Multivariate analysis: descriptors clustering

Regarding the identification of metal clusters, we analysed all the seven matrices (M1,…,M7). The results are shown in Fig. 11. For each field survey, heavy metals are grouped into statistically significant clusters according to r-values (p = 5 %).
https://static-content.springer.com/image/art%3A10.1007%2Fs10661-012-2997-y/MediaObjects/10661_2012_2997_Fig11_HTML.gif
Fig. 11

Dendrogram of heavy metals measured in S. Nicola di Melfi industrial area soils during seven field surveys. The nine measured heavy metals are grouped into statistically significant clusters (p = 5%)

We note that (Fig. 11) in all the seven classifications, the couple (Mn, Cu) is always statistically significant, except for 2002. (Mn, Cu) represents the only cluster characterizing the entire area during the investigated period. This cluster is ascribable to the presence of many agricultural activities such as grape growing or horticulture (Li et al. 2007). For the other metals, it is not possible to put in evidence clusters characterizing the area during the whole investigated period. Only for the last years (from 1996 to 2004) we note that the cluster (Fe, Cr) is statistically significant, and it may be ascribed to the automotive production. For the other metals, we can suppose that the presence of different clusters for each examined year is due to the complexity of the investigated area and the variety of the pollution sources.

Multivariate analysis: objects clustering

In order to present and to discuss the results of the classification procedure, we analyse the clusters in terms of centroids as shown in Tables 2, 3 and 4. In Table 2, we summarize the clusters obtained for each data matrix. In Tables 3 and 4, centroids and I% values are shown, respectively.
Table 2

Results of objects clustering procedure: For each data matrix, identified clusters are listed; in parentheses, the number of sampling points included in the cluster is indicated

 

Cluster

    

I.E.

M1 (1993)

\( \mathrm{Clu}_1^1\left( {12} \right) \)

\( \mathrm{Clu}_2^1\left( {54} \right) \)

\( \mathrm{Clu}_3^1\left( {44} \right) \)

   

M2 (1994)

\( \mathrm{Clu}_1^2\left( {15} \right) \)

\( \mathrm{Clu}_2^2\left( {33} \right) \)

\( \mathrm{Clu}_3^2\left( {31} \right) \)

\( \mathrm{Clu}_4^2\left( {31} \right) \)

  

M3 (1995)

\( \mathrm{Clu}_1^3\left( {38} \right) \)

\( \mathrm{Clu}_2^3\left( {30} \right) \)

\( \mathrm{Clu}_3^3\left( {19} \right) \)

   

M4 (1996)

\( \mathrm{Clu}_1^4\left( {12} \right) \)

\( \mathrm{Clu}_2^4\left( {29} \right) \)

\( \mathrm{Clu}_3^4\left( {31} \right) \)

\( \mathrm{Clu}_4^4\left( {15} \right) \)

  

M5 (2000)

\( \mathrm{Clu}_1^5\left( {55} \right) \)

\( \mathrm{Clu}_2^5\left( {11} \right) \)

\( \mathrm{Clu}_3^5\left( {40} \right) \)

  

19ths.p.

M6 (2002)

\( \mathrm{Clu}_1^6\left( {18} \right) \)

\( \mathrm{Clu}_2^6(6) \)

\( \mathrm{Clu}_3^6\left( {20} \right) \)

\( \mathrm{Clu}_4^6\left( {18} \right) \)

\( \mathrm{Clu}_5^6\left( {41} \right) \)

 

M7 (2004)

\( \mathrm{Clu}_1^7\left( {33} \right) \)

\( \mathrm{Clu}_2^7\left( {50} \right) \)

\( \mathrm{Clu}_3^7\left( {12} \right) \)

\( \mathrm{Clu}_4^7(4) \)

  

I.E. isolated element, s.p. sampling point

Table 3

Centroids, values are expressed in parts per million DW

 

Cd

Co

Cr

Cu

Fe

Mn

Ni

Pb

Zn

M1 (1993)

\( V_{1n}^1 \)

0.15

3

0.05

6

44

459

3

9

2

\( V_{2n}^1 \)

0.20

4

0.08

2

76

123

3

11

3

\( V_{3n}^1 \)

0.18

4

0.06

4

97

270

3

9

2

 

\( V_n^1 \)

0.18

4

0.07

3

81

219

3

10

2

M2 (1994)

\( V_{1n}^2 \)

 

3

0.10

8

73

563

3

10

2

\( V_{2n}^2 \)

 

2

0.13

7

56

291

4

11

3

\( V_{3n}^2 \)

 

3

0.22

4

62

93

8

14

3

\( V_{4n}^2 \)

 

3

0.21

5

134

161

8

13

3

 

\( V_n^2 \)

 

3

0.18

6

82

236

6

12

3

M3 (1995)

\( V_{1n}^3 \)

0.34

1

0.27

4

79

143

3

16

2

\( V_{2n}^3 \)

0.38

2

0.23

5

143

243

3

16

3

\( V_{3n}^3 \)

0.28

3

0.18

8

80

473

3

13

3

 

\( V_n^3 \)

0.33

2

0.23

5

101

250

3

16

3

M4 (1996)

\( V_{1n}^4 \)

0.28

3

0.22

8

88

524

5

14

2.2

\( V_{2n}^4 \)

0.31

2

0.24

6

123

168

4

15

2.8

\( V_{3n}^4 \)

0.26

3

0.21

6

98

301

4

14

2.3

\( V_{4n}^4 \)

0.34

2

0.27

6

226

271

4

15

2.6

 

\( V_n^4 \)

0.29

3

0.23

6

127

282

4

15

2.5

M5 (2000)

\( V_{1n}^5 \)

0.76

1

0.20

3

139

205

2

7

2

\( V_{2n}^5 \)

0.82

1

0.30

3

312

339

2

7

2

\( V_{3n}^5 \)

0.85

3

0.09

6

84

456

3

6

2

\( C_n^5\left( {19} \right)* \)

0.99

4

0.06

3

98

968

5

7

1

 

\( V_n^5 \)

0.80

2

0.17

4

138

318

2

7

2

M6 (2002)

\( V_{1n}^6 \)

0.30

3

0.11

11

86

488

4

7

4

\( V_{2n}^6 \)

0.22

4

0.35

9

306

360

4

9

6

\( V_{3n}^6 \)

0.23

3

0.23

6

107

162

2

10

4

\( V_{4n}^6 \)

0.23

3

0.25

5

244

239

3

8

4

\( V_{5n}^6 \)

0.31

3

0.14

13

98

297

4

8

5

 

\( V_n^6 \)

0.28

3

0.18

9.5

135

298

3

8

4.5

M7 (2004)

\( V_{1n}^7 \)

1.07

4.4

0.47

5

130

206

2

12

3

\( V_{2n}^7 \)

0.55

4.3

0.23

7

86

329

5

10

3

\( V_{3n}^7 \)

1.35

5.1

0.57

4

251

299

2

13

3

\( V_{4n}^7 \)

1.13

5.2

0.20

9

83

668

5

11

5

 

\( V_n^7 \)

0.85

4.5

0.35

6

120

298

4

11

3

\( V_{j,n}^h \) centroid of n-th descriptor in the j-th cluster of the h-th data matrix (h 1,…,7), \( V_n^h \) mean value of n-th descriptor calculated on all the objects included in h-th data matrix (these values coincide with mean values shown in Table 1); \( C_n^5\left( {19} \right) \) concentrations of heavy metals measured in 19th sampling point during the fifth field survey (year 2000)

Table 4

Centroid percentage index

 

Cd

Co

Cr

Cu

Fe

Mn

Ni

Pb

Zn

M1 (1993)

\( \left( {I_{1n}^1} \right)\% \)

−19

−31

−22

99

−46

110

0

−5

−18

\( \left( {I_{2n}^1} \right)\% \)

9

4

10

−27

−6

−44

1

11

27

\( \left( {I_{3n}^1} \right)\% \)

−1

9

−11

41

20

23

0

−12

20

M2 (1994)

\( \left( {I_{1n}^2} \right)\% \)

 

14

−44

40

−11

139

−45

−17

−28

\( \left( {I_{2n}^2} \right)\% \)

 

−24

−27

16

−32

24

−39

−9

2

\( \left( {I_{3n}^2} \right)\% \)

 

−10

25

−33

−24

−61

26

15

13

\( \left( {I_{4n}^2} \right)\% \)

 

5

17

−14

63

−32

36

7

−10

M3 (1995)

\( \left( {I_{1n}^3} \right)\% \)

3

−36

20

−12

−22

−43

12

1

−22

\( \left( {I_{2n}^3} \right)\% \)

16

−24

−1

−1

42

−3

12

2

−10

\( \left( {I_{3n}^3} \right)\% \)

−16

67

−22

61

−21

89

14

−16

9

M4 (1996)

\( \left( {I_{1n}^4} \right)\% \)

−5

−3

−5

27

−31

86

16

−9

−11

\( \left( {I_{2n}^4} \right)\% \)

6

−23

6

−8

−3

−40

9

2

14

\( \left( {I_{3n}^4} \right)\% \)

−11

3

−8

2

−23

7

12

−5

−7

\( \left( {I_{4n}^4} \right)\% \)

19

−19

18

−7

78

−4

10

−2

5

M5 (2000)

\( \left( {I_{1n}^5} \right)\% \)

−5

−47

17

−30

1

−36

−25

2

−20

\( \left( {I_{2n}^5} \right)\% \)

3

−27

79

−19

126

7

8

3

17

\( \left( {I_{3n}^5} \right)\% \)

6

66

−48

50

−39

43

65

−21

−12

M6 (2002)

\( \left( {I_{1n}^6} \right)\% \)

6

14

−41

19

−36

64

25

−11

−6

\( \left( {I_{2n}^6} \right)\% \)

−21

29

92

−6

127

21

36

12

35

\( \left( {I_{3n}^6} \right)\% \)

−17

−1

27

−42

−20

−46

−23

24

−6

\( \left( {I_{4n}^6} \right)\% \)

−17

15

38

−44

81

−20

−12

−2

−18

\( \left( {I_{5n}^6} \right)\% \)

12

−5

−23

33

−28

0

29

−5

9

M7 (2004)

\( \left( {I_{1n}^7} \right)\% \)

26

−1

35

−20

9

−31

−41

8

0

\( \left( {I_{2n}^7} \right)\% \)

−35

−4

−33

22

−29

11

18

−5

−3

\( \left( {I_{3n}^7} \right)\% \)

59

13

64

−29

109

0

−45

15

−8

\( \left( {I_{4n}^7} \right)\% \)

33

15

−42

54

−31

124

32

0

80

The italicized values indicate the cases in which the threshold value is exceeded

\( \left( {I_{j,n}^h} \right)\% \) centroid percentage index of n-th descriptor in the j-th cluster of the h-th data matrix (h = 1,…,7)

At first, we may note that the metals play different roles in the cluster characterization. In particular, for all the clusters in each field survey, Pb centroids are similar to annual mean values (Table 3). For this metal, I% values are always lower than the fixed threshold (Table 4). In terms of cluster interpretation, this result indicates that Pb is not a metal cluster characterizing. Its variations that show statistical significance in the univariate data analysis are specific of this element and have no relationship with spatial–temporal variations of the other metals. Pb behaviour may be explained taking into account that, as we underlined in “Univariate analysis” section, in the investigated period, there was the progressive substitution of leaded gasoline with green petrol, according with the Italian normative (from 1 January 2002, there was the final ban of leaded gasoline). But this generalized decrease occurs contemporaneously to the increase of anthropogenic activities due to the start up of the incinerator.

Regarding the behaviour of the other metals in centroid analysis, we note the similarities for Co and Zn: for Co \( \left| {(I)\%} \right|\geqslant \left| {\left( {{I_{\mathrm{thr}}}} \right)\%} \right| \) in only six cases and for Zn \( \left| {(I)\%} \right|\geqslant \left| {({I_{\mathrm{thr}}})\%} \right| \) in only four cases (Table 4). Also in this case, we are led to consider Co and Zn no-characterizing metals.

On the contrary, all the other metals are cluster-characterizing and it is possible to distinguish two different sub-groups:
  1. (a)

    Cu, Fe and Mn centroids are significantly different from annual mean values in all the field surveys showing high values of I% index. For these three metals, the threshold value is exceeded in 54 % of examined cases for Cu and in the 65 % of examined cases for Fe and Mn (Table 4). This is in agreement with the result obtained by means of the descriptors clustering classification, in which the couple (Cu, Mn) is statistically significant.

     
  2. (b)

    Cr and Ni show high values of index I% mainly in the field surveys carried out from 2000 to 2004 (92 and 67 % of examined cases for Cr and Ni, respectively; Table 4). This result suggests that, only in the last years, the heavy metals showing the lower concentrations in the sequence \( \mathrm{Mn}\gg \mathrm{Fe}\gg \mathrm{Pb}\gg \mathrm{Co}\approx \mathrm{Cu}\approx \mathrm{Ni}\approx \mathrm{Zn}\gg \mathrm{Cd}>\mathrm{Cr} \) (discussed in “Univariate analysis” section) play an important role in the characterization of the correlation structure. A similar behaviour may be supposed also for Cd concentrations even if for this metal \( \left| {(I)\%} \right|\geqslant \left| {({I_{\mathrm{thr}}})\%} \right| \) only in the last field survey (Table 4).

     
In the following step, we discuss the results taking into account the sign of the index I% and the cluster spatial patterns.

In all the field surveys, it possible to individuate a cluster in which Cu and Mn show very high positive values of index I%, while Fe and Cr show negative values of index I% (\( \mathrm{Clu}_1^1 \), \( \mathrm{Clu}_1^2 \), \( \mathrm{Clu}_3^3 \), \( \mathrm{Clu}_1^4 \), \( \mathrm{Clu}_3^5 \), \( \mathrm{Clu}_1^6 \), \( \mathrm{Clu}_4^7 \)). Furthermore, we may note that, in the last three field surveys (from 2000 to 2004), these clusters are also characterized by positive I% values for Cd, Co and Ni, indicating that the levels of bioavailable fraction of these metals are higher than the mean values. These seven clusters include always sampling points localized in TrA and in the top part of the TrB (Fig. 1). In particular, this last result indicates the potential dangerous impact of fertilizer and fungicide commonly used for wheat crops that are common in this sub area.

The centroid analysis puts in evidence another group of clusters characterized by very high positive I% values for Fe and Cr and negative I% values for other characterizing metals (\( \mathrm{Clu}_2^1 \), \( \mathrm{Clu}_4^2 \), \( \mathrm{Clu}_2^3 \), \( \mathrm{Clu}_4^4 \), \( \mathrm{Clu}_2^5 \), \( \mathrm{Clu}_2^6 \), \( \mathrm{Clu}_4^6 \), \( \mathrm{Clu}_1^7 \), \( \mathrm{Clu}_3^7 \)). For this group of clusters, the other metals (Cd, Co, Pb and Zn) show a specific behaviour for each cluster. In particular, except few cases related to fs4 and fs6, the clusters are characterized by I% positive values for Cd, Co and Pb. The sampling points included in these clusters are spread on all around the investigated area. It is possible to point out some sub-areas: the central zone of the transect C, the bottom part of the transect D and parts of the transects F, G and H (Fig. 1). The first sub-area is located near the SATA settlement and its parking area. Comparing these results with those obtained using a pollutant diffusion model, applied in this area and based on stack emissions (Caggiano et al. 2002), we note that the maximum deposition probability is near the SATA settlement and its parking area. On the contrary, the other two sub-areas in TrC and TrD are characterized by land use with positive ecological valence. The bottom part of transect D for instance is a woody area, typical of the Vulture basin, rich of chestnut trees, turkey oaks, beeches and many shrubby species. Finally, the three sub-zones in transects F, G and H are localized in a zone where agricultural and farming activities are very diffuse. All these results allow us to conclude that the classification procedure we present on this paper is able to put in evidence not only polluted and/or degraded areas but also areas in which anthropogenic pressure may cause a significant increase of pollutant concentrations.

Conclusions

In this paper, we presented the results of a study carried out over a decade concerning the bioavailable fractions of heavy metals in superficial soil samples, collected in an industrial area of southern Italy. The investigated site, the area surrounding the industrial settlements of S. Nicola di Melfi, is part of an unpolluted zone, where the main land use is agriculture. During the investigated period, the entire zone was subject to strong anthropogenic pressure due to the increase of industrial activities. The scope of this paper was to implement a statistical method able to compare different monitoring field surveys and aimed at evaluating changes in spatial and temporal distributions of heavy metals.

We applied a mix of univariate and multivariate procedure. The main results can be summarized as follows: All the examined elements showed intrinsic fluctuations and effects of external influences; the exclusive use of univariate statistical analysis and of a unique classification procedure (metal clustering) did not allow to establish a well-defined trend in metal levels. The combined use of a double classification procedure, with the centroid analysis, allowed to characterize spatial–temporal variations of the correlation structure, taking into account the relationships among descriptors and homogeneous groups of sampling points (clusters). The results confirmed the effectiveness of this mix of data analysis techniques in the cases in which the experimental data sets reflect biosystem complexity with a high degree of uncertainty.

Copyright information

© Springer Science+Business Media Dordrecht 2012