Estimation of uncertainty arising from different soil sampling devices: The use of variogram parameters

Available online at www.sciencedirect.com

Chemosphere 70 (2008) 745–752 www.elsevier.com/locate/chemosphere

Estimation of uncertainty arising from different soil sampling devices: The use of variogram parameters Paolo de Zorzi a,*, Sabrina Barbizzi a, Maria Belli a, Maria Barbina b, Ales Fajgelj c, Radojko Jacimovic d, Zvonka Jeran d, Sandro Menegon b, Alessandra Pati a, Giannantonio Petruzzelli e, Umberto Sansone c, Marcel Van der Perk f a

Agenzia per la Protezione dell’Ambiente e per i Servizi Tecnici (APAT), Servizio Laboratori, Misure ed Attivita` di Campo, Via di Castel Romano, 100-00128 Roma, Italy b Agenzia Regionale per lo Sviluppo Rurale, Via Sabbatini, 5-33050 Pozzuolo del Friuli, Udine, Italy c International Atomic Energy Agency (IAEA), Agency’s Laboratories Seibersdorf, A-1400 Vienna, Austria d Jozˇef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenia e Institute of Ecosystem Study ISE CNR, Area della Ricerca, Via Moruzzi, 1-56124 Pisa, Italy f Department of Physical Geography, Utrecht University, P.O. Box 80115, 3508 TC Utrecht, The Netherlands Received 12 April 2007; received in revised form 12 July 2007; accepted 20 July 2007 Available online 20 September 2007

Abstract In the frame of the international SOILSAMP project, funded and coordinated by the National Environmental Protection Agency of Italy (APAT), uncertainties due to field soil sampling were assessed. Three different sampling devices were applied in an agricultural area using the same sampling protocol. Cr, Sc and Zn mass fractions in the collected soil samples were measured by k0-instrumental neutron activation analysis (k0-INAA). For each element-device combination the experimental variograms were calculated using geostatistical tools. The variogram parameters were used to estimate the standard uncertainty arising from sampling. The sampling component represents the dominant contribution of the measurement uncertainty with a sampling uncertainty to measurement uncertainty ratio ranging between 0.6 and 0.9. The approach based on the use of variogram parameters leads to uncertainty values of the sampling component in agreement with those estimated by replicate sampling approach.  2007 Elsevier Ltd. All rights reserved. Keywords: Sampling devices; Comparative sampling; Geo-statistics; Variography; Replicate approach

1. Introduction The objective of environmental monitoring is to quantify the condition of ecological systems in spatial and temporal differentiation. In this framework, the sampling of the environmental components represents an integral and decisive step, in achieving a reliable picture of the status of the environment under study. It is necessary to be reasonably sure

*

Corresponding author. Fax: +39 065050519. E-mail address: [email protected] (P. de Zorzi).

0045-6535/$ - see front matter  2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.chemosphere.2007.07.068

that the results from the samples represent the different characteristics of the area as closely as possible. The uncertainty associated with the sampling is frequently neglected and in some cases it may be a major contributor to the combined measurement uncertainty and can have important impact on the outcome and conclusions of the investigation. Since the 1950s, various theories on sampling have been developed (Gy, 1979; Pitard, 1993), but only in the last 15 years an increasing effort in the evaluation of uncertainty arising from sampling has involved a considerable number of scientists (Thompson and Ramsey, 1995; Ramsey et al., 1995; Ramsey and Argyraky, 1997;

746

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752

Ramsey, 1997, 1998; de Zorzi et al., 2002; Kurfurst et al., 2004; Minkkinen, 2004). In addition, specific requirements of international standards, such as ISO 17025 (ISO/IEC, 2005), induced international organizations, such as Eurachem, to develop a technical guide on uncertainty evaluation from sampling (Ramsey and Ellison, 2007). In this framework, two main approaches are considered: modelling and empirical approaches. The selection of the most appropriate approach depends on the objective of the sampling, the definition of the measurands of interest, the source of error to be taken into account (random or systematic) and the quantity and quality of the information available. The modelling approach applies mathematical models able to quantify the contribution of the identified uncertainty components (Kurfurst et al., 2004; Minkkinen, 2004). The empirical approach is basically based on experimental design characterized by repeated samplings and analysis (Ramsey, 1998). This last approach encompasses the assessment of uncertainty contributions from four sources, namely random and systematic effects from sampling and analyses, which allows the estimation of repeatability and bias. Myers (1996) described the estimation of the sampling component of variance using variogram parameters in combination with sampling theory. This approach can be classified as an empirical approach, as the identification of the different sources of uncertainty individually is not required. This paper is aimed at the estimation of the uncertainty arising from soil sampling associated with three different sampling devices. The reliability of measurement results may be strictly correlated to some operational aspects of soil sampling. Size, shape and other technical characteristics of the sampling device may affect the representatives of samples collected and its ability to represent the sampling target. The Gy’s sampling theory of particulate material (Pitard, 1993; Gy, 1979) considers the effect of sample characteristics and includes them in the seven categories of sampling errors. In the framework of the SOILSAMP project, aimed at quantifying the uncertainty associated to different soil sampling devices in agricultural, semi-natural and contaminated areas, a one hectare agricultural field located in Pozzuolo del Friuli in North-Eastern part of Italy (Barbizzi et al., 2004) was sampled. Following the same sampling strategy, three different sampling devices commonly used in environmental investigations, but different in terms of operation and of volume/mass of the collected sample, were chosen. For selected elements, experimental variograms were calculated from the analytical data. The results obtained by variography were compared with the replicate sampling approach proposed by Ramsey (Ramsey, 1998) which allows the estimation of the random component of the measurement uncertainty through duplication of sampling and analysis. The replicate approach foresees the replication of a part of the total samples collected (typically 10% and at least eight) while in the variogram approach no additional measurements are required.

2. Materials and methods The agricultural site used is a research field belonging to a public scientific institution (ERSA-Friuli Venezia Giulia). The site was characterised for the main pedo-chemical properties (Barbizzi et al., 2004). The site of 10 000 m2, after a preliminary rolling operation, was subdivided into 100 sub-areas (cells), of 10 m · 10 m. The following sampling techniques were investigated: • Edelman hand auger (20 cm length, 7 cm diameter); • mechanical auger (20 cm length, 10 mm diameter) and • shovel. 2.1. Comparative sampling To calculate the uncertainty from sampling associated with the above three sampling devices, a stratified random sampling scheme was applied on a grid of 10 · 10 m cells. Each cell corresponds to a sampling target, defined as ‘‘the portion of material, at a particular time, that the sample is intended to represent’’ (Ramsey and Ellison, 2007). From each cell, three samples were collected at a minimum distance to each other (approximately 0.5 m) using the three sampling devices. All samples were collected up to 20 cm depth. The estimation of short-range variation and the geo-statistical elaboration were improved collecting, within five randomly selected grid cells, additional samples at a short distance from the initial sampling locations. Comparative sampling was carried out by a single sampler (operator). Latitude and longitude of each sampling point were recorded in the field. Fig. 1 shows the sampling points selected for the comparative sampling. 2.2. Sample preparation The sample preparation from primary sample to test sample included: • mixing and stone hand-picking; • drying; • sieving, homogenising and splitting (sample mass reducing) and • milling. The soil samples, stored in card-board boxes, were dried in an oven fan at 36–40 C, then disaggregated and sieved at 2 mm. The fractions above 2 mm were removed and disregarded in the analytical phase. In order to obtain test samples of suitable mass, the soil samples were reduced by using two riffle dividers with a capacity of 5000 ml and 300 ml. Each test sample of about 200 g was milled by planetary ball-mill with jars and spheres in agate up to a particle size less than 100 lm. At the end of the sample preparation step, 315 test samples were obtained (105 for each sampling device).

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752

747

Fig. 1. Sampling design by three sampling devices in the comparative sampling.

In addition 30 test portions from three independent test samples (10 test portions from each test sample) were subsampled to verify the repeatability of the sub-sampling and measurement procedure. 2.3. Analysis The k0-method of instrumental neutron activation analysis (k0-INAA) was used at the Jozˇef Stefan Institute, Ljubljana, Slovenia, for measurements of Cr, Sc and Zn in the soil samples. More details about k0-INAA and the relevant nuclear data are reported in Jac´imovic´ et al. (2002). Test portions of about 200 mg (one for each test sample) were measured. The test portions were sealed in pure polyethylene ampoules (SPRONK System, Lexmond, The Netherlands) and irradiated for about 20 h in the carousel facility of the TRIGA Mark II reactor, Ljubljana with a thermal neutron flux of 1.0 · 1012 n cm2 s1. A standard

(1.0 mm Al–0.1%Au alloy wire pressed into a disc in diameter of 6 mm and 0.2 mm high) and a sample were stacked together and fixed in the polyethylene ampoule in sandwich form before irradiation. The irradiated samples were subsequently transferred to clean polyethylene vials and counted on calibrated coaxial HPGe detectors connected to a multichannel analyzer (MCA). Each irradiated sample was measured three times: after 4–5, 8–10 and 30 days cooling time. For peak area evaluation the HyperLab (HyperLab, 2002) program was used and for elemental concentrations and effective solid angle calculations a software package KAYZERO/SOLCOI (DSM Research, 2003) was applied. k0INAA quality control was performed by using the reference material IAEA Soil-7. The k0-INAA technique allows achieving high precision levels and requires little or no sample processing before the analysis. A single laboratory was responsible for all the analyses. With this approach the analytical uncertainty was kept as small as possible.

748

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752

2.4. Classical statistics and geo-statistics Before applying geo-statistical analysis, the analytical data were analysed by classical statistics (mean, standard deviation, standard deviation of the mean, coefficient of skewness, kurtosis, Kolmogorov–Smirnov test) to test the normality of data distribution. Levene’s test was also applied to verify the homogeneity of the variances. One-way analysis of variance (ANOVA) with a = 0.05 was applied to verify the differences among sampling devices. Geo-statistics is a branch of applied statistics that quantifies the spatial dependence and the spatial structure of a measured property (Mulla and McBratney, 2000). It is based on the regionalised variable theory by which spatial correlation of some properties can be treated (Matheron, 1963). Commonly, the geo-statistical analysis includes two phases: the first is the spatial modelling (variography) and the second phase is the spatial interpolation (kriging). The variogram describes the spatial correlation between observations and its main parameters are: nugget, sill and range (de Zorzi et al., 2005). The nugget is the discontinuity at the origin of the variogram and represents the variance at distances much shorter than the sampling interval. The range provides the distance beyond which variogram values remain constant and the samples become spatially independent. The sill is the variogram value at distances beyond the range and, generally, it equals or approaches the population variance. The relative nugget is the ratio of nugget variance to the sill. For each element and sampling device, the experimental variogram of the element mass fractions was calculated using Splus 6 for Windows (Kaluzny et al., 1997), from which the above variogram parameters were estimated. 2.5. Calculation of uncertainty from sampling – variography approach The nugget, expressed as variance, basically represents micro-scale variation. In theory, the value of the intercept should be zero at zero distance. However, several factors, such as short-scale differences in the matrix, sampling and analytical variabilities, cause a discontinuity at the origin of the variogram (Isaak and Srivastava, 1989). In the context of the comparative sampling performed here, the nugget value includes: • the variances due to spatial correlation that occur at distances less than the shortest sampling interval; • the variance associated with the soil sampling device and sample preparation (from the primary sample to the test sample); • the variance associated with the analysis, including sub-sampling (from test samples to test portion) and • the first two variance components can be attributed to sampling and the third one to the analytical procedure.

The variance s2nugget can be expressed by the following equation: s2nugget ¼ s2s þ s2a

ð1Þ

where: s2nugget is calculated for each combination elementdevice by fitting the experimental data with a spherical variogram model; s2s is the sampling variance and s2a is the analytical variance. The sampling variance (s2s ) includes the variance associated with the sampling device, the sample preparation and the short distance variations. Analytical variance (s2a ) includes sub-sampling and analysis. This equation is based on the assumption that the two variances (analytical and sampling) are not correlated. The square root of s2nugget is the measurement standard uncertainty umeas. From Eq. (1) s2s is given by s2s ¼ s2nugget  s2a

ð2Þ

The sampling standard uncertainty us, expressed in mg kg1, is given by qffiffiffiffi us ¼ s2s ð3Þ For each element, the relative sampling standard uncertainty us%, calculated to the mean mass fraction of a selected element (n = 105), becomes us us % ¼  100 ð4Þ xmean This relative standard uncertainty represents the repeatability of sampling operation. As soil sampling is always destructive, repeated samples cannot be collected from exactly the same location. Therefore, dispersion of measurement values includes also the variability at short distances from the original sampling point. This implies that, apart from the sampling variation that refers to the variation within the sample, the short-range spatial variation in the close vicinity of the sampling location should also be considered as repeatibility. The relative standard uncertainty may be used to calculate the uncertainty contribution from sampling in the following cases: (a) during new sampling activities in the Pozzuolo area, using one of the investigated sampling devices; (b) sampling (using Edelman auger, mechanical auger or shovel) in a different areas with a comparable range of element mass fractions and physical soil properties. 2.6. Replicate approach and comparison with variography approach The combined standard uncertainty arising from sampling calculated by variography is compared with the results obtained applying the replicate approach to the same set of data. The experimental design adopted by Ramsey (1998) is reported in Fig. 2a. The application of a two-ways robust ANOVA (RANOVA) allows the separation of the variability due to spatial distribution of ana-

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752 SAMPLING TARGET

Sample 1

Analysis 1

Sample 2

Analysis 2

Analysis 1

Analysis 2

Replicate Approach (Balanced design - Two-split levels)

SAMPLING TARGET

Sample 1

Sample 2

Sample 3

Analysis 1

Analysis 1

Analysis 1

Replicate Approach - SOILSAMP (Balanced design - One-split level)

Fig. 2. Design outline of the (original) empirical-duplicate approach and of the adapted approach performed on the comparative sampling dataset. (a) is based on the duplication of sampling and analysis. Sample 1 and Sample 2 are duplicate primary samples taken at one sampling target repeating the same sampling procedure. One test sample is obtained from each primary sample. Analysis 1 and Analysis 2 represent the duplication of the analysis of the test portion taken from each test sample. This scheme is followed for all the sampling targets randomly selected (10%). The (b) considers Sample 1, Sample 2 and Sample 3 replicate primary samples taken at one sampling target (each cell 10 x 10 m) by the same sampling procedure. Analysis 1 represent the single measurement carried out on the test sample resulted from each primary sample. This scheme is applied to 10 cells randomly selected corresponding to 10% of all the sampling targets.

lytes, sampling and measurements. Fig. 2b reports the experimental design applied to the SOILSAMP data. The original experimental design was simplified, because the uncertainty contribution attributable to the measurement process was calculated from the 30 test portions derived from three independent test samples. A one-way RANOVA was applied to calculate us-SOILSAMP expressed in mg kg1. The value obtained by this calculation was then compared with the sampling standard uncertainty us given by Eq. (3). 3. Results The mass fraction means values of Cr, Sc and Zn in soil, obtained from 105 independent measurements for each device, respectively, range in the interval 228–231 mg kg1 (Cr), 8.5–8.6 mg kg1 (Sc) and 90–91 mg kg1 (Zn). The coefficients of variation (CV%) range from 6% to 12%, showing that the area is quite homogeneous for the selected elements. The homogeneity of the investigated area confirmed the findings of previous pedo-chemical studies car-

749

ried out in the same agricultural field for the reference site characterization (Barbizzi et al., 2004). The results of internal quality control tests are in good agreement with the recommended value of the RM IAEA Soil-7 measured, showing that the measurements by k0INAA are under statistical control. The classical statistic tests performed lead to values of skewness close to zero (from 0.75 to 0.4) indicating a distribution that is basically symmetric. Levene’s test for homogeneity of variance and one-way analysis of variance ANOVA at a confidence level of 95% were performed for comparative sampling data. For Cr, Sc and Zn, the tests show a statistically not significant difference among the variances calculated for the three different sampling devices. These results suggest that the statistical population is the same for all data sets. On the basis of this result, it is possible to conclude that the samples collected with the three devices are not significantly different between them. As a consequence the samplings carried out in the same cell with different devices can be considered as triplicate samplings carried out with the same sampling device. Table 1 reports the element mass fraction mean values on 10 replicates (test portions) with their standard deviations and variances. The variances include the analytical variance and the variance due to sub-sampling (taking 10 test portions from a test sample). Values ranging from 1.5% (Sc and Zn) to 4% (Cr) indicate a good repeatability of the sub-sampling and k0-INAA measurements. The variances, reported in Table 1, were averaged to quantify the contribution due to sub-sampling and analytical variance on the overall measurement variance. The average variance is independent of the sampling device used to obtain the three test samples. This variance represents s2a in the equation giving the variance from sampling (Eq. (2)). To study the spatial structure of the element mass fractions, geo-statistical tools were applied to the 105 results obtained for each couple element-device. More details on the application of geo-statistical tools to SOILSAMP data are reported in (Barbizzi et al., 2004). The number of sampling locations (>100) are suitable for accurate application of this statistical technique, requiring at least 30 sample data values (Journel and Huijbregts, 1978). The full dataset includes all the analytical results and latitude and longitude of each sampling point. Fig. 3 shows the fitted variogram for Cr and the hand auger sampling device. Analogue fitted variograms were obtained for each combination element-device. By geo-statistical analysis, the nugget variances observed are generally less than one third of the sill variance, with the exception of Cr with relative nugget higher than 50% of the sill. The ratio of nugget variance to sill variance is regarded as a criterion to classify the spatial dependence of soil properties. Ratio values less than 25% correspond to strong spatial dependence, while between 25% and 75% the variable has moderate spatial dependence. Weak spatial dependence is associated to ratio values greater than 75% (Chang et al., 1998). The moderate spatial dependence

750

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752

Table 1 Repeatability measurements on three independent test samples Element

Soil sample 28RS

Cr Sc Zn

Soil sample 62RS

Soil sample 88RS

28RS + 62RS + 88RS

Mean value (mg kg1)

CV (%)

Variance

Mean value (mg kg1)

CV (%)

Variance

Mean value (mg kg1)

CV (%)

Variance

Average variance

229 ± 5 10.1 ± 0.2 107 ± 3

2.1 1.9 2.8

23 0.04 9

235 ± 8 7.8 ± 0.1 88 ± 1

3.5 1.9 1.5

68 0.02 2

237 ± 10 6.6 ± 0.1 72 ± 1

4.3 1.5 1.6

104 0.01 1

65 0.02 4

Note: Results are reported with their standard deviation (number of independent replicates = 10). The variance is calculated as the square of the standard deviation. 28RS, 62RS and 88RS are the codes of the selected three test samples. 28RS + 62RS + 88RS represents the pooled (averaged) variances of 28RS, 62RS and 88RS soil samples.

Variogram - Chromium (mg/kg) - Auger

0

50

variance 100

150

200

be caused by intrinsic factors, such as soil parent materials and other soil formation factors. The range values show that samples collected at a distance of more than 37 m for Cr (hand auger) and 76 m for Sc (mechanical auger) are not spatially correlated. The sill values are slightly lower than the population variance, except for Zn, for which the population variance is a factor of two larger than the sill. Table 2 shows the overall uncertainty budget for the elements measured by k0-INAA and in Fig. 4 the relative standard uncertainties, calculated by variography, associated with each sampling device are compared. In Table 3 the results of the application of the replicate approach to the SOILSAMP data are compared with the values of the combined standard uncertainty for the three sampling devices calculated by the variography approach. The results for the approach suggested by Ramsey (1998) are obtained selecting randomly 10% of the data (105) for each sampling device and applying one way RANOVA. The comparison shows an agreement between the two empirical approaches.

0

20

60

40 distance (meters)

Fig. 3. Fitted variogram about Chromium – sampling by auger.

observed for Cr could be due to agricultural management practices (fertilization, ploughing). On the contrary, strong spatial dependence, partly observed for Sc and Zn, could

Table 2 Uncertainty budget for chromium, scandium and zinc Chromium

Scandium

Zinc

Auger

Mechanical auger

Shovel

Auger

Mechanical auger

Shovel

Auger

Mechanical auger

Shovel

Element mass fraction mean value (n = 105) xmean (mg kg1)

231 ± 1

231 ± 1

228 ± 2

8.5 ± 0.1

8.6 ± 0.1

8.6 ± 0.1

90 ± 1

91 ± 1

90 ± 1

Nugget variance s2nugget

120

100

122

0.21

0.14

0.07

17.3

15.0

10.4

65

65

65

0.02

0.02

0.02

4.1

4.1

4.1

Analytical variance

s2a

s2s

55

35

57

0.19

0.12

0.05

13.2

10.9

6.3

Sampling standard uncertainty us (mg kg1) (Eq. (3))

7.4

5.9

7.5

0.43

0.34

0.22

3.6

3.3

2.5

Relative sampling uncertainty to the mass fraction mean value us% (%) (Eq. (4))

3.2

2.6

3.3

5.1

4.0

2.6

4.0

3.6

2.8

Measurement standard uncertainty umeas (mg kg1)

11

10

11

0.46

0.37

0.26

4.2

3.9

3.2

Sampling variance

(Eq. (2))

Note: Element mass fraction mean values are reported with their standard deviation of the mean (n = number of independent measurement).

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752 Uncertainty of Sampling

element

Zn Auger Mech. Auger Shovel

Sc

Cr

0.0

1.0

2.0

3.0

4.0

5.0

6.0

u % sampling

Fig. 4. Comparison among the different relative standard uncertainties from sampling (us%) estimated by elements for three sampling devices.

Table 3 Comparison between standard uncertainty from sampling calculated by variogram parameters and applying the replicate approach Empirical approach

Sampling standard uncertainty (mg kg1) (Cr) Sampling standard uncertainty (mg kg1) (Sc) Sampling standard uncertainty (mg kg1) (Zn)

Variography us

Replicate us-SOILSAMP

5.9–7.5

3.7

0.22–0.43

0.20

2.5–3.6

2.8

4. Discussions and conclusions In the variography approach, the relative standard uncertainty attributable to sampling (Eq. (4)) shows values ranging from 2.6% to 5% of the element mass fraction mean value. The us observed values are about twice the analytical standard uncertainty, except for Cr. Also in a homogeneous area, sampling represents the dominant component of the measurement uncertainty. The values of the ratio between us and umeas range from 0.6 to 0.9. For Sc and Zn, the highest values of uncertainties are observed for samples collected by the auger, whereas the lowest values are generally associated with the shovel. In the case of Cr, which is likely largely of anthropogenic origin, a comparable standard uncertainty is obtained for auger and shovel. The behaviour of Zn and Sc can be attributed to the different mass/volume of the samples collected by the three sampling devices. Primary samples of more than 2 kg were collected by shovel, while the mass of samples collected by auger was about 0.7 kg. The primary sample size can explain the lower relative standard uncertainty for elements of natural origin. The larger volume of the samples collected by shovel determines an averaging effect in the field, reducing the variability due to sampling device. The values of the standard uncertainty from sampling estimated by replicate approach are in agreement with

751

results from variography (see Table 3). In the replicate approach sampling contributes more than analysis to the measurement uncertainty, with values of us and umeas ratio ranging from 0.4, for chromium, to 0.8, for scandium and zinc. These ratio values are comparable to those estimated by variography. The application of geo-statistics for the assessment of uncertainty contribution from sampling can be used in investigations where sufficient (>30) sampling points are available (Journel and Huijbregts, 1978). Considering that variography requires the application of models for fitting the experimental data, it is clear that the increase of the experimental data reduces the uncertainty associated with the model selected for fitting. Furthermore, the absence of spatial correlation of the properties analysed may cause difficulties in modelling the variogram. The size of the data sets (n = 105) used in the frame of this work, assures a negligible uncertainty in fitting the experimental data. In both the empirical approaches, only a suitable number of soil data are required without any further specific information on the sampling target. On the other hand, modelling approaches require the knowledge of substantial detailed information on the intrinsic characteristics of the sampling target (chemical and physical parameters, particle shape and size, density, etc.) (Minkkinen, 2004) which are not always readily available. These requirements can represent serious constraints in terms of costs and time. Comparing the two empirical approaches, the duplicate approach requires the replication of samplings and analyses for at least 10% of the samples collected, while variography does not require any additional samplings or measurements. Acknowledgemets This study was carried out in the framework of the international SOILSAMP project, which was funded and coordinated by the Italian Environmental Protection Agency (APAT, Italy). We thank Damiano Centioli (APAT) for his technical support in the soil sampling carried out at the Pozzuolo del Friuli site. References Barbizzi, S., de Zorzi, P., Belli, M., Pati, A., Sansone, U., Stellato, L., Barbina, M., Deluisa, A., Menegon, S., Coletti, V., 2004. Characterisation of reference site for quantifying uncertainties related to soil sampling. Environ. Pollut. 127, 131–135. Chang, Y.H., Scrimshaw, M.D., Emmerson, R.H.C., Lester, J.N., 1998. Geostatistical analysis of sampling uncertainty at the Tollesbury Managed Retreat site in Blackwater Estuary, Essex, UK: kriging and cokriging approach to minimise sampling density. Sci. Total. Environ. 221, 43–57. de Zorzi, P., Barbizzi, S., Belli, M., Ciceri, G., Fajgelj, A., Moore, D., Sansone, U., Van der Perk, M., 2005. Terminology in soil sampling (IUPAC Recommendation 2005). Pure Appl. Chem. 77 (5), 827–841. de Zorzi, P., Belli, M., Barbizzi, S., Menegon, S., Deluisa, A., 2002. A practical approach to assessment of sampling uncertainty. Accred. Qual. Assur. 7 (5), 182–188.

752

P. de Zorzi et al. / Chemosphere 70 (2008) 745–752

DSM Research, 2003. User’s Manual KAYZERO/SOLCOI Version 5a software package, 2003. DSM Research, Geleen (NL), developed at the INW-RUG, Gent (B) and the KFKI, Budapest (H). Gy, P.M., 1979. Sampling of Particulate Materials – Theory and Practice. Elsevier, Amsterdam. HyperLab 2002 System, 2002. Installation and Quick Start Guide. HyperLabs Software, Budapest, Hungary. Isaak, E.H., Srivastava, R.M., 1989. An Introduction to Applied Geostatistics. Oxford University Press, New York. ISO/IEC 17025, 2005. General Requirements for the Competence of Testing and Calibration Laboratories. International Organization for Standardization (ISO), Geneva. Jac´imovic´, R., Lazaru, A., Mihajlovic´, D., Ilic´, R., Stafilov, T., 2002. Determination of major and trace elements in some minerals by k0instrumental neutron activation analysis. J. Radioanal. Nucl. Chem. 253, 427–434. Journel, A.G., Huijbregts, C.J., 1978. Mining Geostatistics. Academic Press, London, UK. Kaluzny, S.P., Vega, S.C., Cardoso, T.P., Shelly, A.A., 1997. S + Spatial Stats User’s Manual for Windows and Unix. Springer, New York. Kurfurst, U., Desaules, A., Rehnert, A., Muntau, H., 2004. Estimation of measurement uncertainty by the budget approach for heavy metal content in soils under different land use. Accred. Qual. Assur. 9, 64–75. Matheron, G., 1963. Principles of geostatistics. Econ. Geol. 58, 1246– 1266. Minkkinen, P., 2004. Practical applications of sampling theory. Chemometr. Intell. Lab. 74, 85–94.

Mulla, D.J., McBratney, A.B., 2000. Soil spatial variability. In: Sumner, M.E. (Ed.), Handbook of Soil Science. CRC Press, Boca Raton, FL, pp. A321–A352. Myers, C.J., 1996. Geostatistical Error Management. Quantifying Uncertainty for Environmental Sampling and Mapping. Van Norstrand Reinhold, New York. Pitard, F.F., 1993. Pierre Gy’s Sampling Theory and Sampling Practice, second ed. CRC Press, Boca Raton. Ramsey, M.H., 1997. Measurement uncertainty arising from sampling: implication for objectives of geoanalysis. Analyst 122, 1255–1260. Ramsey, M.H., 1998. Sampling as a source of measurement uncertainty: techniques for quantification and comparison with analytical sources. J. Anal. Atom. Spectrom. 13, 97–104. Ramsey, M.H., Argyraky, A., 1997. Estimation of measurement uncertainty from field sampling for the classification of contaminated land. Sci. Total Environ. 198, 243–257. Ramsey, M.H., Ellison, S.L.R, 2007. Eurachem/EUROLAB/CITAC/ Nordtest/AMC Guide, Measurement uncertainty arising from sampling: a guide to methods and approaches (Voting draft April 2007, to be published). Ramsey, M.H., Argyraki, A., Thompson, M., 1995. Estimation of sampling bias between different sampling protocols on contaminated land. Analyst 120, 1353–1356. Thompson, M., Ramsey, M.H., 1995. Quality concepts and practices applied to sampling – an exploratory study. Analyst 120, 261–270.