heavy metals in european soils: a geostatistical analysis of the...

Geoderma 148 (2008) 189–199

Contents lists available at ScienceDirect

Geoderma

j ourna l homepage: www.e lsev ie r.com/ locate /geoderma

Heavy metals in European soils: A geostatistical analysis of the FOREGSGeochemical database

Luis Rodríguez Lado a,⁎, Tomislav Hengl b, Hannes I. Reuter a

a European Commission, Directorate General JRC, Institute for Environment and Sustainability, TP 280, Via E. Fermi 1, I-21020 Ispra (VA), Italyb Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands

⁎ Corresponding author. Tel.: +39 0332 789977; fax: +E-mail addresses: [email protected] (L.R. Lad

(T. Hengl), [email protected] (H.I. Reuter).

0016-7061/$ – see front matter © 2008 Elsevier B.V. Aldoi:10.1016/j.geoderma.2008.09.020

a b s t r a c t
a r t i c l e i n f o
Article history:

This paper presents the resu

Received 20 March 2008Received in revised form 16 September 2008Accepted 22 September 2008

Keywords:Soil mappingRegression-krigingMODISNight lights imageGeochemical databasePan-European monitoring

lts of modeling the distribution of eight critical heavy metals (arsenic, cadmium,chromium, copper, mercury, nickel, lead and zinc) in topsoils using 1588 georeferenced samples from theForum of European Geological Surveys Geochemical database (26 European countries). The concentrationswere mapped using regression-kriging (RK) and accuracy of predictions evaluated using the leave-one-outcross validation method. A large number of auxiliary raster maps (topographic indexes, land cover, geology,vegetation indexes, night lights images and earth quake magnitudes) were used to improve the predictions.These were first converted to 36 principal components and then used to explain spatial distribution of heavymetals. The study revealed that this database is suitable for geostatistical analyses: the predictors explainedfrom 21% (Cr) to 35% (Pb) of variability; the residuals showed spatial autocorrelation. The PrincipalComponent Analysis of the mapped heavy metals revealed that the administrative units (NUTS level3) withhighest overall concentrations are: (1) Liege (Arrondissement) (BE), Attiki (GR), Darlington (UK), Coventry(UK), Sunderland (UK), Kozani (GR), Grevena (GR), Hartlepool & Stockton (UK), Huy (BE), Aachen (DE) (As, Cd,Hg and Pb) and (2) central Greece and Liguria region in Italy (Cr, Cu and Ni). The evaluation of the mappingaccuracy showed that the RK models for As, Ni and Pb can be considered satisfactory (prediction accuracy45–52% of total variance), marginally satisfactory for Cr, Cu, Hg and Zn (36–41%), while the model for Cd isunsatisfactorily accurate (30%). The critical elements limiting the mapping accuracy are: (a) the problem ofsporadic high values (hot-spots); and (b) relatively coarse resolution of the input maps. Automation of thegeostatistical mapping and use of auxiliary spatial layers opens a possibility to develop mapping systems thatcan automatically update outputs by including new field observations and higher quality auxiliary maps. Thisapproach also demonstrates the benefits of organizing standardized joint European monitoring projects, incomparison to the merging of several national monitoring projects.

© 2008 Elsevier B.V. All rights reserved.

1. Introduction

The generalized mobilization and dispersion of pollutants fromtheir natural reservoirs to the atmosphere, soil and water is one of themost significant negative impacts of human activities on terrestrialand aquatic ecosystems (Salemaa et al., 2001; Lin et al., 2002; Koptsiket al., 2003). Mining, iron and steel industry, road transport, wasteincineration, and the use of fertilizers and agrochemicals are identifiedas the main human sources of heavy metals in soils and water in thesuperficial ecosystems (Hutton and de Meeûs, 2001; Hansen et al.,2002). In addition, emissions from volcanoes, degassing processes inthe Earth's crust, forest fires or the chemical composition of the parentmaterial can be also important sources of heavy metals in soils (Løkkeet al., 1996; Palumbo et al., 2000).

39 0332 786394.o), [email protected]

l rights reserved.

A large proportion of soils in industrialized countries containhigher levels of several elements and compounds considered aspollutants than the corresponding natural background values in apristine situation (Hijmans et al., 2005). Pollution caused by heavymetals is especially problematic in areas where synergy with othertypes of polluting agents exists. This is true in the case of industrialareas receiving large inputs of acidifying compounds, which createsoptimal conditions for increased mobilization, bioavailability and thustoxicity of the metals stored in soils (Salemaa et al., 2001; Lin et al.,2002; Clemente et al., 2003; Cappuyns et al., 2004; de Vries et al.,2005).

In Europe, mandatory reductions on the annual emissions ofcadmium, lead and mercury to avoid significant adverse effects onecosystems have been put in the Heavy Metals Protocol (UN/ECE,1998). This will have an increasing importance because, according tothe most recent report of the Coordination Center for Effects (Poschet al., 2005), the distribution andmagnitude of the deposition of theseelements puts large areas of European ecosystems at risk both in 2000and 2020. Apart from cadmium, lead and mercury, six additional

mailto:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1016/j.geoderma.2008.09.020http://www.sciencedirect.com/science/journal/00167061

190 L.R. Lado et al. / Geoderma 148 (2008) 189–199

heavy metals (chromium, nickel, copper, zinc, arsenic and selenium)present an increasing threat for human health and the environment.

Recently, the European Commission (EC) has been preparing aproposal for a framework Directive (European Communities, 2006)that sets out common principles for the protection of soils across theEC. From the aspect of soil pollution, three main questions need to beclarified: (i) which threshold values should be used to classify soil aspolluted? (ii) at which locations can high natural background values ofheavy metals be expected? and (iii) which methods should be used toexplain the spatial distribution of these elements in soils of Europe?

Threshold values for soils are difficult to evaluate since the toxicityand bioavailability of heavy metals is not only dependent on the totalcontent in soils but also on many other environmental variables.Although some European countries (Netherlands, UK) developed theirown soil quality standards for some heavy metals, at European levelonly threshold values related to the application of sewage sludge inagricultural soils have been defined (EU Directive 86/278/EC). Thedetermination of natural background values is controversial becausethey can decrease the responsibility of human activities for the overallpollution on soils (Baize and Sterckeman, 2001).

At pan-European scale, it is often difficult to determine thebackground values that would correspond to a pristine situationsince the geochemistry of most of our ecosystems is greatly influencedby a long history of anthropic activities, and even the concept of abackground value is often fuzzily defined (Reimann and de Caritat,2005; Reimann and Garrett, 2005). To improve this situation, in 1993the International Union of Geological Sciences and the InternationalAssociation of Geochemistry and Cosmochemistry (IUGS/IAGC)initiated the Global Geochemical Baselines Programme with an aimto establish the geochemical reference baselines for a number ofelements at a global scale (Salminen, 2006). The European contribu-tion to this International Programme started in 1997 and it has beencarried out by governmental institutions of 26 European countriesunder the advisement of the Forum of European Geological Surveys(FOREGS). The final product of this collaboration was the “Geochem-ical Atlas of Europe” (http://www.gsf.fi/publ/foregsatlas/) thatincludes a database of about three thousand samples for solidmedia: topsoil, floodplain sediment, stream sediment and humus(ranging from 385 samples of sub soil to 852 samples of streamsediment) in addition to a number of maps of the element contents insoils of Europe based on the former database. The term “geochemicalbaseline” in this context is not equal to the “background value” since itrepresents a measurement that does not correspond to a pristinesituation. However, these geochemical baselines can provide aperspective on the present status of pollution of the Europeans soils,and serve as a model for future pan-European monitoring projects.Some recent analyses of the FOREGS database can be seen in Imrieet al. (2008).

So far, several attempts were made to determine the spatialdistribution of the concentrations of heavy metals in European soils.Reimann et al. (2003) created maps of heavy metal contents in soilsusing an Inverse Distance Weighted interpolator on about 740samples from agricultural fields. The Geochemical Atlas of EasternBarents region (Salminen et al., 2004) also includes interpolated mapsof heavy metals from 1358 sampling sites in the northern part ofEurope. Gawlik and Bidoglio (2006) produced maps for Cd, Cr, Cu, Hg,Ni, Pb and Zn in 11 European countries by using empirical relation-ships with soil parent material and land use. The European Environ-mental Agency (2006) merged the sampling points from threedifferent soil databases to create a map of the concentration of leadin topsoils across Europe (http://dataservice.eea.europa.eu).

In principle, the only official maps of heavy metals in soils ofEurope are those presented in the “Geochemical Atlas of Europe”. Thisatlas contains maps for 85 variables for five different media:floodplain sediment, humus, soil, stream sediment and streamwater, all produced using the Alkemia Smooth interpolation on a

6 km grid (de Vos and Tarvainen, 2006). Although this workrepresents an important contribution, we identified several pointsthat could be considered to improve the reliability and accuracy of thefinal maps produced. Firstly, the authors of the FOREGS atlas did notconsider merging the samples taken for different media or runningmore sophisticated geostatistical analyses to generate the maps.Secondly, they did not mask out areas that were not represented in aspecific sample (histosols, water bodies), which limits the further usesof such maps. Moreover, the FOREGS atlas shows multiple pan-European maps of the same variable, which might be confusing fordecision makers and spatial modelers. For example, there are 10 mapsof lead distribution in the book: in subsoil (two laboratorytechniques), in topsoil (two laboratory techniques), in humus, instream water, in stream sediment (two laboratory techniques) and infloodplain sediment (two laboratory techniques). All 10 maps showvalues of the same element (often with different patterns) over thewhole of Europe. Thirdly, because no transformation has been appliedto the original data (most of the heavy metal concentrations show askewed distribution), the resulting patterns were very much influ-enced by locally high values, which questions the overall reliability ofthe produced maps (see e.g. Papritz et al., 2005; Romić et al., 2007 fordiscussion). Fourthly, the interpolationmethod used to build themapsdid not provide an estimate of its uncertainty.

With this background, we decided to test a geostatistical, digitalsoil mapping framework to interpolate the concentration of heavymetals over broad areas and to provide information about theuncertainty of the produced maps. Our assumption was that auxiliarypredictors can be used to improve the detail and accuracy of theexisting FOREGS atlas maps and extend the spatial analysis of thepoint data using state-of-the-art geostatistical techniques. We alsowanted to demonstrate the benefits of producing geoinformationthrough a joint pan-European project and to suggest ways to designsoil monitoring networks for the permanent monitoring of this soilthreat.

2. Materials and methods

2.1. The FOREGS dataset

The point dataset was obtained from the FOREGSwebsite, courtesyof the Association of the Geological Surveys of The European Union.For the purpose of this study, we used only the laboratorymeasurements of extractable heavy metal concentrations (HMC) intop-soil and floodplains determined for As, Cd, Cr, Cu, Ni, Pb and Zn(mg kg−1) by ICP-AES using the Aqua Regia method (ISO, 1995) and forHg (mg kg−1), directly on the soil solid samples using a cold vapourabsorption technique within an Advanced Mercury Analyzer (AMA-254, ALTEC). We did not include the stream sediment measurementsin the analyses because the geochemical processes and accumulationin the water environment are different. At five locations the precisionof the coordinates was not sufficient for separation of samplinglocations. At these locations values for topsoil and floodplain wereaveraged and replaced the original value. The FOREGS database israther extensive (85 variables measured in five media), so we havefocused on mapping a selection of variables in the database. In thenear future, we anticipate that also other variables from this databasewill be mapped with an improved spatial and thematic detail.

The total number of points used in this exercisewas 1588, althoughnot all values were available at all locations, ranging from 17 missingvalues for As, Cd, Cu, Ni, Pb and Zn to 79 missing values for Hg. Theoriginal coordinates of the sampling points have been provided indegrees, rounded to 0.01°, which corresponds to a horizontal precisionof 850 m at 40° North latitude, and transformed to EuropeanTerrestrial Reference System (see www.euref.eu). The reorganizedpoint database and all input and derivedmaps shown in this paper canbe accessed from the http://eusoils.jrc.it website.

http://www.gsf.fi/publ/foregsatlas/http://dataservice.eea.europa.euhttp://www.euref.euhttp://eusoils.jrc.it

191L.R. Lado et al. / Geoderma 148 (2008) 189–199

2.2. The study area

The area of interest corresponds to the 26 European countries thatcontributed their data to the IUGS/IAGC program. It includes a veryhigh diversity of climatic, geologic, edaphic and socio-economicconditions. The Canary Islands were excluded from the mappingproject for computational reasons. All the maps were projected to theLambert Azimuthal Equal Area projection (European TerrestrialReference System 1989) following the recommendations of INSPIRE(Annoni, 2005). Each raster map had the following grid definition:MinX: 2,636,000, MinY: 1,425,000, MaxX: 5,956,000, MaxY: 5,416,000and a grid size of 1 km. We further masked out non-soil surfaces suchas water bodies (rivers, lakes, sea etc.) and permafrost areas.

A consistent Europeanwide water mask, indicating the percentageof water area inside a 1 km pixel, was produced by using the blocksumfunction of the NASA SRTM (Shuttle Radar Topography Mission) V2data set (Rabus et al., 2003), the CORINE (Coordinated Information onthe European Environment) land cover classification (http://dataser-vice.eea.europa.eu), lakes contained in the GISCO (GeographicInformation System of the European Commission) data, waterreflection by the use of Image2000 dataset (de Jager et al., 2006),the Global Self-consistent, Hierarchical, High-resolution ShorelineDatabase (Wessel and Smith, 1996), and the Global Lakes andWetlands Database (Lehner and Doll, 2004). Areas with permanentice cover have been detected using the mean annual EnhancedVegetation Index (EVI) derived from the MODIS (Moderate ResolutionImaging Spectroradiometer) 1 km images obtained for the years 2003and 2004. In this case, ice cover, water bodies and bare rock areasweredetected based on a negative EVI value. The total soil-cover area forthese 26 countries was estimated to be 4,217,241 km2.

2.3. Auxiliary GIS layers

Because the FOREGS data set is covering a large extent, we decidedto go to the effort to prepare a large number of auxiliary raster maps.These include geological and land cover class-type maps, MODIS-based vegetation indices, night lights images, earthquake magnitudesand distance to roads and railroads (Fig. 1).

The map of geology was obtained from Pawlewicz et al. (2003).This map includes a generalized synthesis of themain geologic surfaceoutcrops of bedrock in Europe as well as coarse information on therock types and their geologic age of formation (http://pubs.usgs.gov/).The original legendwas reduced to 10 classes relevant for themappingof heavy metals: (1) Granites, rhyolites and quartzites; (2) Paleozoicschists, phyllites, gneisses and andesites; (3) Shales and sandstones;(4) Mesozoic Ultramafic, basic phyllites, schists, limestones andevaporates; (5) Jurassic, Triassic and Cretaceous calcareous rocks; (6)Cenozoic serpentinites, gabbros and sand deposits; (7) Tertiarybasanites and andesites; (8) Neogene and Paleogene calcareousrocks; (9) Quaternary limestones and basaltic rocks; and (10) OtherUltramafic and undefined rocks.

The CORINE Land Cover 2000 map of Europe (CLC2000), general-ized to a 1 km grid, was used to represent the main classes of landcover. For Switzerland, we used the Corine Land Cover 1990 (CLC1990)since no updated information was available. The CLC1990 classes forthis country were adjusted to those described in the CLC2000 andboth data sets were merged together and aggregated to 1 kmresolution. The original 44 classes were simplified to 8 classes: (1)urban infrastructures; (2) agriculture; (3) forest; (4) natural vegeta-tion; (5) beaches; (6) ice bodies, (7) wetlands and (8) water bodies.

Monthly averaged MODIS images of the EVI at 1 km resolution forthe period 01/01/2004 to 31/12/2006 were obtained from the MODISTerra imagery at the Earth Observing System Data Gateway (http://redhook.gsfc.nasa.gov/~imswww/pub/imswelcome/). Seventeen sin-gle blocks covering the whole study area were mosaicked andreprojected to the Lambert Azimuthal Equal Area (ETRS89) projection.

We performed Principal Component analysis on 19 complete mosaicsand used the first 10 principal components of the EVI images.

The 1 km Digital Elevation Model (DEM) was derived from theSRTM30 V2 data set obtained from the Jet Propulsion Laboratory(http://www2.jpl.nasa.gov/srtm/). The SRTM DEM was used to derivea slope map, the Topographic Wetness Index and total incoming solarinsolation (Böhner et al., 2006).

The cumulative earthquake magnitude (quakem) image wascalculated by using the global Seismology point database (http://earthquake.usgs.gov/eqcenter/). In Europe, there were over 90,000registered earthquakes in the period 1973–1994. We used thelogarithmic measure of the size of an earthquake, which is relatedto the energy released as seismic waves at the focus of an earthquake.The 1 km density grid was generated by using a kernel smoother andsearch radius of 10 km.

The lights at night (nlight) image for the year 2003 was obtainedfrom the Defense Meteorological Satellite Program (http://www.ngdc.noaa.gov/dmsp/), which measures night-time light emanating fromthe earth's surface at 1 km resolution. The lights at night map containsthe lights from cities, towns, and other sites with persistent lighting,including gas flares. Ephemeral events, such as fires have beendiscarded. The background noise was identified and replaced withvalues of zero. This map is a direct estimate of the urbanization leveland is now increasingly used for quantitative estimation of globalsocioe-conomic parameters as well as for human population mapping(Sutton, 1997; Sutton et al., 1997; Doll et al., 2000).

Themap of distances to roads, airports and utility lines (dinfra) wascalculated using the distance operation in ILWIS and the GIS layersfrom the GISCO database of the European Commission (http://eusoils.jrc.it/giscodbm/dbm/). Firstly, distance maps were derived for eachlayer separately and then these three layers were combined toproduce average distance to infrastructure lines. Mean annualtemperature and accumulated precipitation maps were obtainedfrom the very high resolution raster layers created by Hijmans et al.(2005) on a global scale at 1 km grid resolution. The annual potentialevapotranspiration (PET) was calculated from monthly temperaturedata using the method of Thornthwaite (1948). Runoff was calculatedas the difference between annual accumulated precipitation and PET.

We also used the estimated annual deposition and emission ratesof cadmium, lead and mercury in Europe for year 2004 calculatedwithin the European Monitoring Evaluation Programme (http://webdab.emep.int). The original 50 km grids were down-scaled to1 km grids using ordinary kriging.

This gives a total of 36 maps of predictors. Many of the predictorswere intercorrelated and showed interesting correlations, e.g. PC2 ofEVI seems to be highly correlated with the density of traffic (dinfra)and the map of runoff; PC1 of EVI is correlated with the map oftemperature and elevation (DEM) is correlated with some geologicalunits. To minimize multicollinearity, these predictors were convertedto independent principal components (raster maps) in ILWIS GIS.However, due to the heterogeneous nature of the original predictors,they were all previously rescaled to a common scale ranging from 0 to255. Since some predictors show continuous changes and others (landcover, geological units) represent abrupt changes of the values, thefinal principal component shows a hybrid representation of the totalenvironmental conditions.

2.4. Geostatistical analysis

The geostatistical analysis steps used in this paper basically followthe regression-kriging framework previously described in Hengl et al.(2007) and Romić et al. (2007). For more in-depth explanation of theprocedures see also Hengl (2007).

The original variables in all cases showed skewed distributions.Logit transformations with physical limits set at zmin=0 and zmax=106

were performed to make the data suitable for regression and

http://dataservice.eea.europa.euhttp://dataservice.eea.europa.euhttp://pubs.usgs.gov/http://redhook.gsfc.nasa.gov/~imswww/pub/imswelcome/http://redhook.gsfc.nasa.gov/~imswww/pub/imswelcome/http://www2.jpl.nasa.gov/srtm/http://earthquake.usgs.gov/eqcenter/http://earthquake.usgs.gov/eqcenter/http://www.ngdc.noaa.gov/dmsp/http://www.ngdc.noaa.gov/dmsp/http://eusoils.jrc.it/giscodbm/dbm/http://eusoils.jrc.it/giscodbm/dbm/http://webdab.emep.inthttp://webdab.emep.int

Fig. 1. Some auxiliary raster maps used as predictors for estimation of heavy metal concentrations. Geology and CORINE land cover: simplified classes as shown in Section 2.3; MODIS EVI, distance to roads, night light image and SRTM DEM:rescaled values from 0 to 255.

192L.R.Lado

etal./

Geoderm

a148

(2008)189

–199

Fig. 2. The biplot display of the Principal component analysis on eight HMCs (N=1588).

Fig. 3. Variogram models fitted for target variables and their residuals (broken lin


variogram analyses (Hengl et al., 2004; Papritz et al., 2005). Thetransformed variables in the point samples were first correlated withan extensive set of environmental predictors (auxiliary GIS layers at1 km grid resolution). The regression models retained were selectedby stepwise regression (backward selection) using the AkaikeInformation Criterion (AIC). The residuals were tested for normalityusing the standard normality tests (Thode, 2002) and then analysedfor spatial auto-correlation. The predictions were obtained byregression-kriging, which was run in the computing environment R(http://www.r-project.org/) using the gstat V0.9-39 (Pebesma andWesseling, 1998), sp V0.9-14 (Pebesma and Bivand, 2005) and rgdalV0.5-14 packages.

We used block regression-kriging to obtain spatial averaged HMCon a continental scale. All regression analyses were performed on thepoint data, using the information from the 1 km auxiliary grids in thesame locations as ancillary data. Variogram fitting was performed onthe residuals of these regression analyses. The regression equationswere applied on the auxiliary variables at 1 km grid resolution.

e). All fitted automatically using the default settings in gstat. See also Table 1.

http://www.r-project.org/


However, for computational efficiency, the resulting models wereaveraged to 5 km grids and the final predictionswere also generated atthis latter resolution. After mapping all HMCs over the whole area ofinterest, we derived the principal components and the overall (mean)normalized interpolation error map. The normalized interpolationerror is calculated by dividing the interpolation error (universalkriging variance in gstat) by the global variance in the point data(Hengl et al., 2004):

σ �RK ¼σRKσ

This way the interpolation error can be summed for heavy metalconcentrations of different type. The map of overall normalizedinterpolation error can be used to locate additional point samples toimprove the quality of the maps.

3. Results

3.1. Preliminary data screening

An inspection of the samples shows that the observations are wellspread over the area of interest, with a mean shortest distance betweenpoint pairs of 5.6 km. The points are somewhat clustered toward shorterdistances, because the samples in different media have been selected inthe near vicinity. This can even be beneficial for geostatistical analysissince, for the fitting of variograms, we are often more interested in theestimation of the spatial auto-correlation model at short distances.According to our analysis, the inspection density is about 0.4 observa-tions per 1000 km2 (N=1588 points, A=4,217,241 km2), whichcorresponds to an effective scale of about 1:8M, i.e. a cell size of about4.1 km (Hengl, 2006). We eventually decided to work with a relativelycoarse grid (5 km), and to consider downscaling only if the regressionmodels become highly significant. Note also that the FOREGS points donot represent organic soils and soils with permafrost, thus these soilswere also masked out from the analyses.

The Principal Component Analysis of the samples shows thatseveral HMCs are highly (positively) correlated (Fig. 2). This is the caseof Cr and Ni (r=0.85), Cu and Ni (r=0.75), Pb and Zn (r=0.74), Cd and

Table 1Summary input and fitted parameters, and results of cross-validation

Variable As Cd Cr

SamplesMin 2.5 0.010 1Max 410 23.60 2340Median 6 0.20 22SD 19.98 1.115 85.2Median⁎ −12.02 −15.20 −10.75Variance⁎ 0.847 1.127 0.681

Fitted modelsNugget 0.365 0.655 0.333Sill 0.331 0.483 0.411Range (km) 175.5 1,036.8 806.0Adj. R-square 29% 24% 21%Dominant SPCs 4 2, 8, 13 4, 17Nugget (residual) 0.380 0.688 0.327Sill (residual) 0.232 0.123 0.180Range (residual) 101.5 332.3 129.7

EvaluationMap SD (OK) 4.16 0.112 14.60Map SD (RK): 4.35 0.129 14.19RMSPE (OK)⁎ 0.687 0.908 0.641RMSPE (RK)⁎ 0.685 0.885 0.635Variance explained (OK) 44% 27% 39%Variance explained (RK) 45% 30% 41%

Units of measurement (mg kg−1 ); ⁎ — transformed variables; SPC — Soil Predictive CompPrediction Error.

Zn (r=0.69) and Pb and Hg (r=0.69). In summary, the HMCs are wellrepresented using the two components.

The second component clearly differentiates two groups ofelements, As, Cd, Hg, Pb and Zn with positive correlation and Cr, Cuand Ni with negative correlation. Similar groupings were found inother studies on heavy metals (Facchinelli et al., 2001; Micó et al.,2006; Zhang, 2006; Luo et al., 2007) and it is often interpreted asindicator of the source of origin of the elements: anthropogenic for As,Cd, Hg, Pb and Zn, and geogenic for Cr and Ni. The intermediateposition of Cu may indicate a joint contribution from both natural andanthropogenic sources.

3.2. Variogram modeling and regression analyses

Further analysis of the spatial autocorrelation structure of all HMCsshowed that they present distinct spatial autocorrelation patterns(Fig. 3). However, in all cases the nugget variation is significant,ranging from 45% of global variance for As, up to a value of 66% forCd. Assuming that the HMCs are measured with a relatively highprecision, most of the nugget variance can be attributable to the short-range variability of the HMCs. The nugget variance remains large evenafter the regression modeling, so it was decided to increase thesupport size from 1 to 5 km grid cell size to minimize the impact ofshort range variability.

The correlation analyses between HMCs and predictors confirmedthat several environmental variables can be used to explain thedistribution of heavy metal content in soils. Preliminary correlationplots showed several interesting relationships: negative correlationbetween Cd and mean annual biomass; Zn and Pb are positivelycorrelated with the night light image; Hg increases with an increase oftemperature etc. Single predictors explain only a small portion ofvariation and most of the relationships often seem to be non-linear orat least very erratic. The final results of step-wise regression (8 targetHMCs versus 36 principal components derived from a whole bulk ofpredictors) gave eight regression models which were all significant at0.001 probability level. The adjusted R-square ranges from 0.21 for Crup to 0.35 for modeling of Pb (Table 1). Step-wise regression reducedinitial set of predictors (36) by 35–45% of the total number. The most

Cu Hg Ni Pb Zn

1 0.002 1 1.5 4421 4.391 2565 5200 283214 0.0405 16 16 5226.29 0.2052 96.47 174.4 155−11.18 −17.06 −10.98 −11.11 −9.810.705 0.968 0.906 1.021 0.601

0.397 0.596 0.383 0.524 0.3580.217 0.362 0.680 0.427 0.215308.4 550.6 1,244.2 771.8 354.627% 24% 32% 35% 28%2, 4 2, 4, 13 4, 14 2, 4, 13 2, 4, 130.346 0.613 0.401 0.471 0.3480.152 0.178 0.163 0.200 0.09865.7 195.1 178.6 87.0 111.3

6.48 0.0274 20.27 9.76 21.657.21 0.0294 19.13 11.33 24.950.683 0.794 0.662 0.756 0.6440.664 0.787 0.657 0.737 0.61534% 35% 51% 44% 31%37% 36% 52% 47% 37%

onents; RK — regression-kriging; OK — ordinary kriging; RMSPE — Root Mean Square

Fig. 4. Final maps of heavy metal concentrations in topsoil [mg kg−1] interpolated using block regression-kriging (support size=5 km). All maps described in this paper are availableon-line via the http://eusoils.jrc.it website.


http://eusoils.jrc.it

Fig. 5. Highest baseline concentrations of heavy metals in Europe represented using the factors #1 and #2 of the eight maps of interpolated HMCs: (left) factor #1 (47.8% of variance)showing the areas of high concentrations of As, Cd, Hg and Pb, mainly caused by anthropogenic factors, and (right) factor #2 (22.6% of variance) showing the areas of highconcentration of Cr, Cu and Ni.


significant predictors were PC04 (limestones, elevation), PC02(agriculture), PC13 (EVI, infrastructures) and PC14 (earthquakes).Much of variation was left unexplained by our predictors, but theresiduals still showed spatial autocorrelation (Fig. 3).

3.3. Regression-kriging — maps of HMCs

The final transformed maps of HMCs obtained through regression-kriging are shown in Fig. 4. The most inter-correlated maps of HMCswere Cr and Ni (r=0.91), Pb and Zn (r=0.89), Cd and Zn (r=0.79) and

Fig. 6. Overall normalized interpolation error. This shows areas of extrapolation in both geogsurveyors seem to have systematically omitted urban soils, some geological units and some

Hg and Pb (r=0.75). This agrees with the relationships obtainedpreviously in the PCA for the observations (Fig. 2).

High values of Cr and/or Ni are mainly found in central Greece,northern Italy, the central Pyrenees, northern Scandinavia, Slovakiaand Croatia. Our analyses show that there is a strong correlationbetween the contents of Ni and Cr and the magnitude of earthquakes.In some tectonically unstable areas we find spots of ultramafic rocks,polymetallic sulphide minerals as well as sedimentary rocks andlateritic deposits derived from these materials (Kane, 1977; Skarpelis,2006; Meissner and Kern, 2007). The seismic activity is indirectly

raphical and feature space and can be used to collect additional samples. In this case, theland cover classes.


correlated with heavy metal concentrations — such materials providehigh quantities of Ni and Cr to the soils by weathering processes.

For the other heavy metals, the higher concentrations are mainlyfound in Central Europe and are directly related to human activities.Cd, Cu, Hg, Pb, Zn present a high correlation with Agriculture (r=0.7)and with quaternary limestones (r=0.41), where most of theagricultural areas in Central Europe are located. The use of fertilizers,manure and agrochemicals are important sources of these elements inEuropean soils. These heavy metals are also highly correlated with thedistance to infrastructures and to components 1 and 3 of the EVIimages (PCEVI1 and PCEVI3), which depict clearly urban andindustrial areas.

3.4. Validation

The success of the technique was evaluated using the leave-one-out cross validation method, as implemented in the krige.cv methodof gstat (Pebesma and Wesseling, 1998). The algorithm works asfollows: it visits a data point, predicts the value at that location byleaving out the observed value, and proceeds with the next data point.This way each individual point is assessed versus the whole data set.The results of the validation are shown in Table 1. This shows that thecontrast of the maps produced using regression-kriging is in general20–40% higher than for the ordinary kriging (see also further Fig. 7).The Root Mean Square Prediction Error (RMSPE), on the other hand, isalways smaller for regression-kriging, but the relative difference doesnot exceed 5%. The percentage of variability explained by RK is alwayshigher than by OK.

The results of evaluationusing the leave-one-out technique suggestthat the models for As, Ni and Pb can be considered satisfactorily

Fig. 7. A comparison of maps of Pb produced using: (a) auxiliary environmental predict

accurate (prediction accuracy 45–52% of total variance), themodels forCr, Cu, Hg and Zn are only marginally satisfactory (36–41%), while themodel for Cd has lowaccuracy (30%), i.e. it does not seem to carrymuchuseful information about the distribution of this element. These resultsdemonstrate that there is quite some difference in how the HMCs varyin space.

3.5. Factor analysis — the overall concentrations

In all cases the HMCs are positively correlated, so the results offactor analysis on these maps can be used to depict areas where theoverall values are high or low. The dominantly high/low values in allmaps (factor #1) can be seen in Fig. 5. Note that, according to this map,administrative units (NUTS level3) with highest overall concentra-tions are: (1) Liege (Arrondissement) (BE), Attiki (GR), Darlington(UK), Coventry (UK), Sunderland (UK), Kozani (GR), Grevena (GR),Hartlepool & Stockton (UK), Huy (BE), Aachen (DE) (As, Cd, Hg and Pb)and (2) central Greece and Liguria region in Italy (Cr, Cu and Ni). Insome countries (England, Belgium), high values of factor #1 can beconnectedwith the urbanization level, i.e. population density, while inother countries (Spain, Czech republic) there is no obvious relation-ship between the HMCs and the environmental factors. This provesthat inventory of heavy metals in soils and reconstruction of theirsources is complex and can not be achieved by using only few spatialpredictors.

The precision of mapping the eight elements, derived as theaverage normalized prediction error, can be seen in Fig. 6. This mapshows high uncertainty for some specific areas systematically omittedor under-sampled by the national FOREGS survey teams, especiallyurban soils, but also national parks, wetlands and some geological

ors (regression-kriging) and (b) pure geostatistical interpolation (ordinary kriging).


units. We assume that we would be able to produce maps with moreaccurate estimates of the HMCs if a model-based design method hadbeen used instead (see e.g. Brus and Heuvelink (2007)).

4. Discussion and conclusions

Many spatial features like geology or land use influence thedistribution of HMCs in soils and we find it convenient to use them asan input for mapping. This is illustrated with the map in Fig. 7a thatshows how the values of HMCs change due to the changes in relief(mountain chains), geological unit and density of roads. Suchenvironmental patterns are normally completely ignored in themaps that are produced by plain geostatistical techniques such asordinary kriging (Fig. 7b).

The validation of the results shows that regression-kriging is notmuch more accurate than ordinary kriging and that the model for Cdhas low accuracy. Although this questions the use of more complicatedtechniques in the first place, we still believe that there is an addedvalue in running analysis using large number of auxiliarymaps (see alsoPebesma (2006) for an in-depth discussion). Auxiliary predictors cancontribute to understanding if the high values are due to the urbaniza-tion, geological substrata, hydrological shape or if the distribution israndom. All this allows an in-depth analysis of the processes that causethe distribution of HMCs, so that also the remediation policies can beselected appropriately.

There are two possible explanations for why the overall accuracy ofmaps of HMCs is relatively low. Firstly, we discovered that bothregression-kriging and ordinary kriging are not fit to deal with manysporadic high values (hot-spots), which are almost always oversmoothed. We assume that the right approach to dealing with localhot-spots would be to always include the most complete and mostdetailed information that can beused to explain such features (e.g. log ofthe distance to locations of all possible HMC sources). Another solutionto over-smoothing is to, instead of using the predicted values, usegeostatistical simulations of HMCs for spatial modeling purposes. Thesecond critical issue affecting the accuracy of mapping is the relativelycoarse resolution and low accuracy of the input maps. The relativelycoarse scale of the project and consequent large support size obviouslylimits the predictability of features that possibly happen at short-ranges.From the computational point of view, we experience the seriouslimitations of using regression-kriging with large data sets (detailedscales for large areas). Because the raster maps were quite large (36predictive components, each 3991×3320 pixels), the first attempts tointerpolate using the standard universal kriging settings in R (globalmodel, all predictors) resulted inmemory usage problems. Interpolationof the maps was still possible using the gstat stand-alone exeapplication, but it often took between 24 and 38 h on a standard PC.Solving the computational limitations of regression-kriging will remaina task for programmers and geostatisticians to solve (Hengl et al., 2007).

Imrie et al. (2008), in their study of the FOREGS database, alsoreported that the overall distribution of the geochemical elements inEuropean topsoils follows diverse patterns that can be explained bydifferent processes occurring at different spatial scales. They concludethat the geochemical variation at short scales depends mainly on thelocal variation of lithology, land use, weathering processes and organicmatter content, at medium scales the major parent material and theeffects of the recent glaciation can be used to explain those differences,while at large scales the effects of mineralization, pollution, climate anddeposition processes are the main factors influencing the variability inthe spatial distribution of geochemical elements. This is in agreementwith our analyses — the larger autocorrelation ranges have been foundfor Ni and Cr (related to mineralization processes) and for Cd, Pb and Hg(anthropogenic pollution); whereas, smaller ranges were obtained forAs, Cu and Zn. This indicates that local factors have greater influence inthevariabilityof As, CuandZn in soils and, consequently, theseHMCs aremore difficult to map over large areas.

Despite all of these difficulties, the framework presented here hasa number of important advantages over the smooth interpolator(currently used by FOREGS). The geostatistical technique presentedin this paper is flexible and generic. The results can be easilyimproved by adding new observations and/or new auxiliaryinformation. It can produce the best linear predictions of the valuesover the whole area of interest, but also an estimate of the mappingerror (regression-kriging variance), which can be equally importantfor decision making and further spatial modeling. The R script wedeveloped is fully automated so that both step-wise selection of bestregression model, fitting of the variogram, interpolated and export toa GIS environment is done without any/much human interventionand the results can be directly reproduced. This way the analyst canfocus his/her resources on improving the quality of inputs and oninterpreting the final outputs. Further research must be done on theeffects of cross-interactions between independent variables, as wellas on possible non-linear relationships. These points could facilitatea greater understanding of the causes behind the distribution ofHMCs in soils.

Unlike many different databases that are produced by mergingdifferent national survey protocols, our impression was that theanalytical quality of the FOREGS database is rather high. There is awell-defined protocol that includes the use of standardized analyticaltechniques, and a systematic control of the results based on duplicatemeasurements. These quality standards make this database fairlyattractive for geostatistical analysis at pan-European scale. However,the accuracy of the geostatistical models is greatly dependent on thequality of sampling. In the case of the FOREGS database, the samplingdensity corresponds to cell sizes of 1–5 km, which obviously limits itsusage to general scales only. The biggest limitation of the FOREGSdataset seems to be its sampling design which obviously carrieslimitations: sampling points are somehow clustered and severalfeatures (certain land use features and urban soils) have been omittedby the surveyors, which might lead to under or over-estimation overthe whole continent (Fig. 6). We also observed unrealistic low valuesfor carbon content in soils of Fennoscandia, which is because organicsoils have been avoided in the sampling protocol. Organic componentsstrongly bind heavymetals in soils, so that by avoiding the organic soilareas the overall content can be under-estimated.

Finally, we hope that this work will motivate the national agenciesin Europe to contribute additional point data and increase the numberof points in the database up to a few thousand. The Geological Surveypoint database of USA (http://tin.er.usgs.gov/geochem/) has about60,000 points; the Australian mapping agency has used a data set of150,000 points to map main soil physical and chemical properties(Henderson et al., 2005). Also in Europe, there must be thousands andthousands of high quality field-sampled environmental data that iswaiting for integration and further geostatistical analysis.

References

Annoni, A., (Ed.), 2005. European Reference Grids. EUR 21494 EN. EuropeanCommission, DG-JRC, IES, Ispra, Italy.

Baize, D., Sterckeman, T., 2001. Of the necessity of knowledge of the natural pedo-geochemical background content in the evaluation of the contamination of soils bytrace elements. Sci. Total Environ. 264 (1–2), 127–139.

Böhner, J., McCloy, K.R., Strobl, J. (Eds.), 2006. SAGA — Analysis and ModellingApplications. Göttinger Geographische Abhandlungen, Heft, vol. 115. Verlag ErichGoltze GmbH, Gottingen, Germany.

Brus, D.J., Heuvelink, G.B.M., 2007. Optimization of sample patterns for universal krigingof environmental variables. Geoderma 138 (1–2), 86–95.

Cappuyns, V., Swennen, R., Verhulst, J., 2004. Assessment of acid neutralizing capacityand potential mobilisation of trace metals from land-disposed dredged sediments.Sci. Total Environ. 333 (1–3), 233–247.

Clemente, R., Walker, D.J., Roig, A., Bernal, M.P., 2003. Heavy metal bioavailability in asoil affected by mineral sulphides contamination following the mine spillage atAznalcollar (Spain). Biodegradation 14, 199–205.

de Jager, A., Rimavičiūte, E., Haastrup, P., 2006. A water reference for Europe. In:Haastrup, P., Würtz, J. (Eds.), Environmental Data Exchange Network for InlandWater. Elsevier, pp. 259–286.

http://tin.er.usgs.gov/geochem/


de Vos, W., Tarvainen, T., 2006. Geochemical atlas of Europe part 2: interpretation ofgeochemical maps, additional tables, figures, maps and related publications. Vol. 2of Geochemical Atlas of Europe. EuroGeosurveys & Foregs, Espoo, Finland.

de Vries, W., Schütze, G., Lofts, S., Tipping, E., Meili, M., Römkens, P.F.A.M., Groenenberg,J.E., 2005. Calculation of critical loads for cadmium, lead and mercury. BackgroundDocument to a Mapping Manual on Critical Loads of Cadmium, Lead and Mercury.

Doll, C.N.H., Muller, J.P., Elvidge, C.D., 2000. Night-time imagery as a tool for globalmapping of socioeconomic parameters and greenhouse gas emissions. Ambio 29 (3),157–162.

European Communities, 2006. Proposal for a directive of the European Parliament and ofthe Council Establishing a Framework for the Protection of Soil and AmendingDirective 2004/35/EC.COM(2006) 232.2004/35/EC.COM(2006) 232. European Com-mission, Brussels, Belgium.

Facchinelli, A., Sacchi, E., Mallen, L., 2001. Multivariate statistical and GIS-basedapproach to identify heavy metal sources in soils. Environ. Pollut. 114 (3), 313–324.

Gawlik, B.M., Bidoglio, G., 2006. Background values in European soils and sewagesludges part II — contents of trace elements and organic matter in European soils.Results of a JRC-coordinated Study on Background Values. EUR 22265 EN. EuropeanCommission, DG-JRC, IES, Ispra, Italy.

Hansen, E., Lassen, C., Stuer-Lauridsen, F., Kjølholt, J., 2002. Heavy Metals in Waste.Henderson, B.L., Bui, E.N., Moran, C.J., Simon, D.A.P., 2005. Australia-wide predictions of

soil properties using decision trees. Geoderma 124 (3–4), 383–398.Hengl, T., 2006. Finding the right pixel size. Comput. Geosci. 32 (9), 1283–1298.Hengl, T., 2007. A practical guide to geostatistical mapping of environmental variables.

EUR 22904 EN. Office for Official Publications of the European Communities,Luxembourg.

Hengl, T., Heuvelink, G.B.M., Stein, A., 2004. A generic framework for spatial predictionof soil variables based on regression-kriging. Geoderma 120 (1–2), 75–93.

Hengl, T., Heuvelink, G.B.M., Rossiter, D.G., 2007. About regression-kriging: fromequations to case studies. Comput. Geosci.

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolutioninterpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978.

Hutton, M., de Meeûs, C., 2001. Analysis and conclusions from Member States'Assessment of the risk to health and the environment from cadmium in fertilisers.

Imrie, C.E., Korre, A., Munoz-Melendez, G., Thornton, I., Durucan, S., 2008. Application offactorial kriging to the FOREGS European topsoil geochemistry database. Sci. TotalEnviron. 393, 96–110.

ISO, 1995. Soil quality. Extraction of Trace Elements Soluble in Aqua Regia.Kane, M., 1977. Correlation of major eastern earthquake centers with mafic/ultra-mafic

bassement masses.Koptsik, S., Koptsik, G., Livantsova, S., Eruslankina, L., Zhmelkova, T., Vologdina, Z., 2003.

Heavy metals in soils near the nickel smelter: chemistry, spatial variation, andimpacts on plant diversity. J. Environ. Monit. 5 (3), 441–450.

Lehner, B., Doll, P., 2004. Development and validation of a global database of lakes,reservoirs and wetlands. J. Hydrol. 296, 1–22.

Lin, Y.P., Teng, T.P., Chang, T.K., 2002. Multivariate analysis of soil heavy metal pollutionand landscape pattern in Changhua county in Taiwan. Landsc. Urban Plan. 62 (1),19–35.

Løkke, H., Bak, J., Falkengren-Grerup, U., Finlay, R.D., Ilvesniemi, H., Nygaard, P.H., Starr,M., 1996. Critical loads of acidic deposition for forest soils: is the current approachadequate? Ambio 25, 510–516.

Luo, W., Wang, T., Lu, Y., Giesy, J.P., Shi, Y., Zheng, Y., Xing, Y., Wu, G., 2007. Landscapeecology of the Guanting reservoir, Beijing, China: multivariate and geostatisticalanalyses of metals in soils. Environ. Pollut. 146, 567–576.

Meissner, R., Kern, H., 2007. Earthquakes and strength in the laminated lower crust —can they be explained by the “corset model”. Tectonophysics.

Micó, C., Recatalá, L., Peris, M., Sánchez, J., 2006. Assessing heavy metal sources inagricultural soils of an European Mediterranean area by multivariate analysis.Chemosphere 65, 863–872.

Palumbo, B., Angelone, M., Bellanca, A., Dazzi, C., Hauser, S., Neri, R., Wilson, J., 2000.Influence of inheritance and pedogenesis on heavy metal distribution in soils ofSicily, Italy. Geoderma 95 (3–4), 247–266.

Papritz, A., Herzig, C., Borer, F., Bono, R., 2005. Modelling the spatial distribution ofcopper in the soils around ametal smelter in northwestern Switzerland. In: Renard,P., Demougeot-Renard, H., Froidevaux, R. (Eds.), Geostatistics for EnvironmentalApplications: Proceedings of the Fifth European Conference on Geostatistics forEnvironmental Applications. Springer, New York, pp. 343–354.

Pawlewicz, M.J., Steinshouer, D.W., Gautier, D.L., 2003. Map showing geology, oil and gasfields, and geologic provinces of Europe including Turkey. Open-File Report 97-470I. US Geological Survey, Denver, CO.

Pebesma, E.J., 2006. The role of external variables and GIS databases in geostatisticalanalysis. Trans. GIS 10 (4), 615–632.

Pebesma, E.J., Wesseling, C.G., 1998. Gstat: a program for geostatistical modelling,prediction and simulation. Comput. Geosci. 24 (1), 17–31.

Pebesma, E.J., Bivand, R.S., 2005. Classes and methods for spatial data in R. R News 5 (2),9–13.

Posch, M., Slootweg, J., Hettelingh, J.P., 2005. European critical loads and dynamicmodelling: CCE status report 2005.

Rabus, B., Eineder, M., Roth, A., Bamler, R., 2003. The shuttle radar topographymission—a new class of digital elevation models acquired by spaceborne radar. Photogramm.Eng. Remote Sensing 57 (4), 241–262.

Reimann, C., de Caritat, P., 2005. Distinguishing between natural and anthropogenicsources for elements in the environment: regional geochemical surveys versusenrichment factors. Sci. Total Environ. 337 (1–3), 91–107.

Reimann, C., Garrett, R.G., 2005. Geochemical background-concept and reality. Sci. TotalEnviron. 350 (1–3), 12–27.

Reimann, C., Siewers, U., Tarvainen, T., Bityukova, L., Eriksson, J., Giucis, A., Gregor-auskiene, V., Lukashev, V.K., Matinian, N.N., Pasieczna, A., 2003. Agricultural soils innorthern Europe: A geochemical atlas. Geol. Jb. Sonderhefte, vol. SD 5. Hannover,Germany.

Romić, M., Hengl, T., Romíc, D., Husnjak, S., 2007. Representing soil pollution by heavymetals using continuous limitation scores. Comput. Geosci. 33, 1316–1326.

Salemaa, M., Vanha-Majamaa, I., Derome, J., 2001. Understorey vegetation along aheavy-metal pollution gradient in SW Finland. Environ. Pollut. 112 (3), 339–350.

Geochemical atlas of Europe part 1: background information, methodology and maps.In: Salminen, T. (Ed.), Vol. 1 of Geochemical Atlas of Europe. EuroGeosurveys &Foregs, Espoo, Finland.

Salminen, R., Chekushin, V., Tenhola, M., Bogatyrev, I., Fedotova, E., Tomilina, O.,Zhdanova, L., Glavatskikh, S.P., Selenok, I., Gregorauskiene, V., Kashulina, G.,Niskavaara, H., Polischuok, A., Rissanen, K., 2004. Geochemical Atlas of the EasternBarents Region. Elsevier Science, Amsterdam.

Skarpelis, N., 2006. Lateritization processes of ultramafic rocks in Cretaceous times: thefossil weathering crusts of mainland Greece. J. Geochem. Explor. 88 (1–3), 325–328.

Sutton, P., 1997. Modeling population density with night-time satellite imagery and GIS.Comput. Environ. Urban 21 (3), 227–244.

Sutton, P., Roberts, D., Elvidge, C., Meij, H., 1997. A comparison of nighttime satelliteimagery and population density for the continental United States. Photogramm.Eng. Remote Sensing 63, 1303–1313.

Thode Jr., H.C., 2002. Testing for Normality, vol. 164. Stony Brook University, New York.Thornthwaite, C.E., 1948. An approach toward a rational classification of climate. Geogr.

Rev. 38, 55–94.UN/ECE, 07/05/2007, 1998. Protocol to the 1979 convention on long-range transbound-

ary air pollution on heavy metals.Wessel, P., Smith, W.H.F., 1996. A global self-consistent, hierarchical, high-resolution

shoreline database. J. Geophys. Res. 101, 8741–8743.Zhang, C., 2006. Using multivariate analyses and GIS to identify pollutants and their

spatial patterns in urban soils in Galway, Ireland. Environ. Pollut. 142, 501–511.

本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP

图书馆。

图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具

http://www.xuebalib.com/cloud/http://www.xuebalib.com/http://www.xuebalib.com/cloud/http://www.xuebalib.com/http://www.xuebalib.com/vip.htmlhttp://www.xuebalib.com/db.phphttp://www.xuebalib.com/zixun/2014-08-15/44.htmlhttp://www.xuebalib.com/

Heavy metals in European soils: A geostatistical analysis of the FOREGS Geochemical databaseIntroductionMaterials and methodsThe FOREGS datasetThe study areaAuxiliary GIS layersGeostatistical analysis

ResultsPreliminary data screeningVariogram modeling and regression analysesRegression-kriging — maps of HMCsValidationFactor analysis — the overall concentrations

Discussion and conclusionsReferences

heavy metals in european soils: a geostatistical analysis of the...

Documents