Measurement of Trace Element Distributions in Soils and Sediments Using Sequential Leach Data and a Non-specific Extraction System With Chemometric Data Processing†

Measurement of Trace Element Distributions in Soils andSediments Using Sequential Leach Data and aNon-specific Extraction System With Chemometric DataProcessing†

Mark R. Cave and Joanna WraggAnalytical and Regional Geochemistry Group, British Geological Survey, Keyworth, Nottingham,UK NG12 5GG

A chemometric mixture resolution procedure suitable fordetermining the number and composition ofphysico-chemical components in data derived from soilleachates is described. The procedure is used to determinethe number of components in sequential leachate dataobtained for a NIST certified soil (SRM 2710) using awidely employed leaching scheme. The resulting datashow that the sequential leaching media are not specificfor their designated target fractions and that erroneousidentification of fractions occurs. A scoping study inwhich a new non-specific extraction method is tested isdescribed. The experimental design varies theconcentration of nitric acid, the reaction time and theratio of sample to extractant. The resulting solutions wereanalysed by ICP-AES for major and trace metals and thedata obtained from 34 experiments subjected to thechemometric resolution procedure. Four components areidentified and the effects of the three variables on eachcomponent are modelled using multiple linear regression,allowing the conditions which favour dissolution of eachcomponent to be identified. Calculated elementcompositions of the components identified in thenon-specific extraction trial are compared with thoseidentified in the sequential extraction data. Significantcorrelations between the two sets of components are notedand tentative identification of the source of thecomponents is made. In particular, there is evidence thatthe Tessier method extracts both Fe and Mn oxidessimultaneously, whereas the non-specific method hasresolved the Fe and Mn oxides as separate entities.

Keywords: Sequential leaching; chemometric mixtureresolution; soils analysis; trace element partitioning;inductively coupled plasma

The determination of potentially toxic inorganic substances(e.g., heavy metals) in soils and sediments is an important toolfor monitoring environmental pollution. Although the totalconcentrations of these potentially toxic substances providesbroad evidence for possible contamination, it has been recog-nised that quantification of the chemical forms of metals in soilsis essential for estimating the mobility and bioavailability of themetals in the environment.1,2 Metals in soils may be present inseveral different geochemical phases that act as reservoirs orsinks of trace elements in the environment.3,4 The chemicalphases considered to be important are divided up into a series ofbroad categories usually consisting of: exchangeable; specifi-cally adsorbed; carbonate; Fe and Mn-oxides; organic matter;mineral lattice. To obtain information on the distributions of

trace metals between these soil/sediment phases, a number ofworkers5–8 have developed extraction schemes in which phasesare selectively dissolved with carefully chosen reagents. Bysubsequent chemical analysis of the extraction media theconcentration of trace elements associated with the target phasecan be determined. The method of Tessier et al.5 has beenwidely adopted in a number of applications.9–15

Despite the widespread use of these selective extractionmethods and the insight they have given to understanding thegeochemical processes governing trace metal distributions,selective extraction techniques have been demonstrated to havea number of weaknesses,16–20 the two most important being:

1 The so-called ‘selective extraction’ reagents are not specificfor one mineral phase; therefore the associated analysis is not atrue representation of the amount of trace elements from a singlephase.2 The design of the selective extraction schemes leads to amethodological definition of the distribution of trace elementsbetween solid phases which may not reflect the actualdistribution within the test samples.

A recent study by Cave and Harmon21 investigated the traceelements associated with the iron oxide phase of red-bedsandstones and showed that chemometric processing of the ironoxide phase data from related samples could identify thepresence of more than one component being mobilised by theso-called ‘selective extractive reagent’. This work confirmedthe limitations noted earlier and also suggests an alternativeapproach to the study of speciation that should overcome someof the problems of the traditional methods.

If the chemical composition of the products of individualextraction steps within a sequential extraction procedure areconsidered to be mixtures of the physico-chemical componentsof the soil or sediment, each containing different proportions ofeach component, then a similar procedure to that described byCave and Harmon21 could be used to identify and quantify thesecomponents. Further, if this chemometric mixture resolution isviable for the sequential extraction data, a new extractionmethod could be applied in which a relatively simple non-specific reagent is used to extract the different physico-chemicalphases from the target soil or sediment. The resulting solutionwould be made up of a mixture of different proportions of eachphysico-chemical phase. By producing a series of these mixedphase solutions with different proportions of each phasepresent, chemometric methods could be used resolve thecomposition of each phase. The method used to produce thesolutions containing the different proportions of each phasecould be: (1) time series extractions, because different phasesshould dissolve at different rates; (2) variation of rock/extractant ratio; (3) a series of different extractant concentra-tions.

The main advantages of this approach would be thesimplicity of the extraction procedures, as only one extractant

† Presented at Geoanalysis 97: 3rd International Conference on the Analysis ofGeological and Environmental Materials, Vail, CO, USA, June 1–5, 1997.

Analyst, November 1997, Vol. 122 (1211–1221) 1211

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online / Journal Homepage / Table of Contents for this issue

would be required, and the partitioning of trace metals betweenthe different phases would not be methodologically defined.

In this study, a mixture resolution approach to interpreting atraditional sequential extraction scheme was carried out and afeasibility study on the use of a simple single extractionapproach combined with chemometric data processing isreported.

Experimental

Test Material

To test the new methodologies, a well characterised materialwas required. At present there are very few, if any, SRMscertified for leaching purposes. Li et al.,15 however, havecarried out a rigourous Tessier style5 sequential extraction

Fig. 1 Relationship between the leachate concentration matrix, the matrix containing the physico-chemical component compositions and the matrixcontaining the proportions of each physico-chemical component in each leachate solution.

Table 1 Central composite experimental design for the non-specific extraction trial and total solids extracted for each of the 4 identified components

Levels (see below for actualvalues) Total extracted solids/mg kg21

Sample :Nitric acid extractant Component Component Component Component

Replicate Time concentration ratio 1 2 3 4

1 21 21 21 8689 4485 7749 31051 21 1 1 7438 7279 10151 28301 1 21 1 6191 690 10839 36291 1 1 21 4704 16634 15746 37171 0 0 0 6653 4456 11153 36961 21 21 1 6495 854 6815 39111 21 1 21 7792 8601 9464 28421 1 21 21 7442 8359 13292 32001 1 1 1 4258 13686 15624 36561 0 0 0 6490 4442 10981 37651 21.67 0 0 7524 1113 4404 37121 1.67 0 0 5844 5277 13003 38591 0 21.67 0 9 0 138 39051 0 1.67 0 5448 13904 15187 34301 0 0 21.67 7851 10618 11185 38261 0 0 1.67 6242 191 10126 38511 0 0 0 6497 4333 10669 36332 21 21 21 8757 4446 7663 29992 21 1 1 7939 7331 10138 28962 1 21 1 6302 831 11248 37652 1 1 21 4432 16333 15328 39942 0 0 0 6690 4186 10862 35502 21 21 1 6747 838 6680 37682 21 1 21 8184 8363 9250 26242 1 21 21 6974 8326 13364 33142 1 1 1 4256 12983 15337 36702 0 0 0 6543 4233 10862 35382 21.67 0 0 7101 1030 4089 34872 1.67 0 0 5853 5226 13053 38522 0 21.67 0 2 0 138 38822 0 1.67 0 5754 12406 13979 29562 0 0 21.67 7982 10169 10921 36032 0 0 1.67 7498 0 11116 50132 0 0 0 6745 4242 10947 3493

NitricLevel Time acid/m Sample : extractant ratio

21.67 0.5 0.01 0.0121 18.3 0.21 0.05

0 44.75 0.50 0.121 71.79 0.80 0.201.67 90.00 1.0 0.25

1212 Analyst, November 1997, Vol. 122

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

procedure on an NIST certified soil (SRM 2710) in which theylooked at extending the number of trace metals under study bymodifying the method of Tessier et al.5 This particular SRM ishighly contaminated soil from pasture land along Silver BowCreek in the Butte, Montana, area. Li et al.15 noted that thissample was unusual in that it had a high level of heavy metalsin the exchangeable fraction, which would make it a suitablematerial for studying the mobility and bioavailabilty of metalsin contaminated soils. Because of the availability of data and thenature of the sample, the published data15 for this soil was usedfor the mixture resolution exercise and a sample of the soil wasused for the non-specific extraction trial.

Chemometric Mixture Resolution Procedure

Chemometric strategies for mixture resolution have been usedwidely in analytical chemistry22,23 and have also been used forstudying environmental data sets.21,24,25 The procedure devel-oped here is a combination of the methods of Thurston andSpengler,25 Gamp et al.26 and Cave and Harmon.21

This method is based on the assumption that the sample (soilor sediment) is made up of a number of physico-chemicalcomponents each of which has its own chemical composition(e.g., carbonate component, iron oxide component). By leach-ing the sample, under certain conditions, a proportion of these

components is leached into solution. The concentration of anelement in a particular leach solution can be described as alinear sum of the amounts leached from each physico-chemicalsources present, such that:

E En n

n

n c

tot ==

=

∑ α1

(1)

where Etot is the total concentration of an element E in a givenleach solution, Ec the concentration of element E in componentn, ac the proportion of element E leached from component n,and c the number of components.

In this study there is more than one element being consideredand a number of leaches have been carried out. In this instance,eqn. (1) can be expressed in matrix form which is shownpictorially in Fig. 1 showing the dimensions of each matrix witha description of the contents of each matrix.

In order to be able to tell which elements are associated withwhich each physico-chemical component it is necessary to findmatrices B and C given matrix A.

The first stage in determining B and C is to use principalcomponent analysis (PCA) of matrix A to estimate the numberof physico-chemical components (c) present and to give a firstestimate of the proportions of each component in each leach(i.e., matrix B). This was carried out in a similar manner to thatproposed by Thurston and Spengler25 but required a number ofmodifications to make the method suitable to the leachatecomposition data. Firstly, PCA is normally carried out on amatrix with the compositional data (in this case elementalconcentrations) in columns and the different samples in rows (asshown by matrix A in Fig. 1) using a pre-scaling procedure inwhich each column is scaled by subtracting the column meanand dividing by the column standard deviation. This ensuresthat all elements whether present at high or low concentrationcontribute equally to the PCA model. In the case of the leachatedata, however, the total amount of material in each leachate canvary considerably (for example, in the Tessier method differ-ences between the amount leached in the exchangeable and theresidual fractions can vary by 1–2 orders of magnitude) andscaling down the element columns followed by PCA wouldhave produced a model which was dominated by the leachatesamples in which the largest amount of material was extracted.

Table 2 Eigenvalues and percentage variance explained for the Varimaxrotated principal components for the Tessier and non-specific extractiondata

Extraction method

Tessier method Non-specificPrincipal

component Variance Varianceno. Eigenvalue (%) Eigenvalue (%)

1 1.5 30.6 16.3 47.92 1.2 24.9 5.1 15.13 1.1 21.2 3.8 11.24 1.0 20.6 6.9 20.35 0.13 2.7 0.14 0.416 — — 0.07 0.20

Table 3 Varimax rotated scores the Tessier and non-specific extraction data (numbers in bold indicate elements with the highest score within a givenPC)

Tessier method Non-specific method

Principal component Principal component

Element 1 2 3 4 Element 1 2 3 4

Al 0.15 3.26 20.67 20.49 Al 0.78 20.98 20.69 1.26Ca 20.48 0.24 20.75 3.07 Ba 20.74 20.48 20.32 20.34Cd 20.64 20.51 20.40 20.51 Ca 20.20 2.60 1.64 20.18Co 20.64 20.51 20.39 20.53 Cd 20.79 20.48 20.31 20.29Cu 1.67 20.53 20.34 20.67 Cu 20.46 20.77 1.63 1.34Fe 20.31 1.05 2.53 20.79 Fe 3.03 20.52 20.37 21.78K 0.19 0.74 20.57 0.63 K 20.17 0.87 0.09 20.95

Mn 20.01 20.53 2.20 1.07 Mg 20.30 0.07 20.46 20.33Ni 20.64 20.51 20.39 20.53 Mn 1.06 0.82 21.07 2.56P 20.58 20.48 20.34 20.54 Na 20.66 0.30 20.35 20.63

Pb 2.95 20.45 20.19 20.05 Ni 20.80 20.49 20.32 20.29Sr 20.63 20.50 20.40 20.46 P 20.76 20.49 20.28 20.32Ti 20.64 20.34 20.41 20.52 Pb 0.97 21.28 2.53 0.23V 20.64 20.51 20.39 20.53 Si 0.07 0.10 20.88 20.18Zn 0.24 20.39 0.52 0.83 Ti 20.77 20.49 20.33 20.31

V 20.79 20.48 20.32 20.30Zn 0.51 1.68 20.18 0.49

Analyst, November 1997, Vol. 122 1213

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

This could lead to an underestimate of the number of physico-chemical components being leached. It was therefore decided totranspose matrix A (AT) and scale the matrix over each leachate,therefore making each leachate contribute equally to the PCAmodel.

PCA of the AT matrix was carried out followed by Varimax27

rotation. According to Thurston and Spengler,25 this procedureprovides abstract PC score and loadings matrices which shouldbe closely related to the true proportion and componentconcentration matrices (matrices B and C, respectively, inFig. 1). The number of physico-chemical components (c)present was found by the number of PCs with eigenvaluesgreater than one after Varimax rotation.25

The next stage is to use the abstract PCs derived from thePCA of AT to provide a first approximation of the proportions ofmatrix B. Thurston and Spengler25 used a procedure where theyderived absolute PC scores which could be used as the firstestimate. This procedure was not successful for the leachatedata which gave proportions with negative values. This step wastherefore replaced by that used by Cave and Harmon21 in whicheach of the significant (c) PCs within the Varimax rotated scoresmatrix was examined. The elements which have the highestscores within each of these PCs are assumed to be the mosthighly correlated to the rotated PCs and hence to the truephysico-chemical components. The concentration of theseelements in each leachate should be linearly related to thecolumns in matrix B. Regressing the concentrations of each ofthese identified elements against the total extractable solids foreach leachate solution, using multiple linear regression (MLR),gives estimates of the coefficients which convert the elementconcentrations into physico-chemical component mass con-tributions (in mg kg21) for each leachate solution, allowing afirst approximation of B to be calculated.

In the final stage matrix A is scaled so that each elementconcentration is expressed as a fraction of the total extracted

mass for each leach solution (AA). Similarly, the first approxi-mation of B is scaled so that each mass contribution is expressedas a fraction of the total extracted mass for each leach solution(BA). Using the pseudoinverse calculation22 ( in other words anMLR regression of BA against AA) a first estimate of CA (thescaled physico-chemical component concentration matrix) canbe calculated. The first estimate of CA is further refined bysetting any negative values to 0 and by scaling each row so thatthe relative element contributions for each component adds upto 1. Using the pseudoinverse calculation with the refined CAand matrix AA a second approximation for BA is calculated. Thesecond approximation for BA is refined by setting negativevalues to 0 and by scaling each column so that the relative masscontribution for each leachate adds up to 1. Further estimates ofBA and CA are iteratively calculated using this procedure until nofurther improvement in their values is obtained. This follows themethod first used by Gamp et al.26 and more recently by Caveand Harmon.21 Finally the refined matrices BA and CA are re-scaled to absolute concentration values.

The PCA, Varimax rotation and MLR calculations werecarried out using Statistica (Version 5.1) and the iterativepseudo-inverse calculation routine was written in MathCad Plus(Version 6.0).

Non-specific Extraction Procedure

A central composite experimental design procedure was used toinvestigate the effects of three variables (time, nitric acidconcentration and sample to extractant ratio) on the chemicalcomposition of the resulting acid extracts from SRM 2710.Using this formal experimental design, the relative effects andinteractions of each parameter could be measured and MLRmodelling of the response surface was carried out.

The five levels used for each variable and the experimentaldesign are shown in Table 1. Each measurement was carried out

Fig. 2 Relative proportions of the each resolved component in the sequential extraction data.

1214 Analyst, November 1997, Vol. 122

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

in duplicate resulting in a total of 34 test solutions for analysis.The extractions were carried out in 30 ml polycarbonate screwtop tubes which were mixed on an ‘end-over-end’ rotatingshaker. All experiments were carried out in an air conditionedlaboratory with the temperature nominally maintained at20 °C.

The experimental design matrix and the analysis of the resultswas carried out using the experimental design module ofStatistica Version (5.1).

Analysis of the Acid Leachate

The 34 leachate solutions generated by the experimental designprocedure were analysed by ICP-AES for Al, Ba, Ca, Cd, Cu,Fe, K, Mg, Mn, Na, Ni, P, Pb, Si, Ti, V and Zn. Theinstrumentation and wavelengths used have been previouslydescribed.21

Results and Discussion

Sequential Leach Data

In addition to the average concentration of each elementextracted in the sequential extraction scheme described by Li etal.,15 uncertainty measurements for each determinand were alsosupplied. In order to include this information in the chemo-metric data processing, an additional nine sets of data weregenerated with random amounts of uncertainty, within thereported limits, and were added to the average values. Theoriginal data matrix of 15 elements (columns) and 5 leachates(rows) was combined with the additional nine data sets toproduce a single matrix of 150 columns and 5 rows. This newmatrix was then processed as a single data matrix using themixture resolution procedure described in the experimentalsection.

The eigenvalues for the Varimax rotated PCA model for theTessier extraction data are shown in Table 2. This clearly showsthat there are four components with eigenvalues greater thanone and a step change in eigenvalue between the fourth and fifthPC. In addition, Table 3 shows the scores for the first foursignificant PCs identifying the elements Pb, Al, Fe and Ca ashaving the highest scores. Using these element concentrationprofiles and four components as being significant, the composi-tion of each component was calculated using the chemometricprocedure previously described. The relative contributions of

each identified component for the ten sets of data were thenrecombined by taking an average. The standard deviation of thedata was used to give a measure of the uncertainty for therelative proportion of any component in a given extractionstep. Fig. 2 shows a plot of the calculated average relativeproportions of each component in each of the 5 sequentialleaches with error bars representing twice the standard deviationon each value (n = 10). Fig. 3 shows the chemical compositionsof each of the four component represented as a pie diagram.

Component 1 makes up a significant proportion of the firstfour extracts with particularly high levels in the designatedcarbonate and organic/sulfide extracts. This component is madeup principally from the heavy metals Pb, Cu and Zn. The originof this component is not clear but may be related to an organicrich component or possibly a fine clay fraction. The reasons forthis will be discussed later.

Component 2 only appears in the last two extracts and is morethan 50% Al. This can be interpreted as the alumino-silicatematrix of the soil.

Component 3 appears predominantly in the Fe/Mn oxideextracted fraction although the designated organic/sulfide andresidual fractions contain approximately 10% of this componentwhich is made up predominantly from Mn and Fe. Thisindicates an Fe/Mn oxide component.

Component 4 appears predominantly in the designatedexchangeable and carbonate fractions and is dominated by Ca,Mn, Zn, K and Pb. The high proportion of this component in thefirst fraction indicates this is the easily exchangeable fraction.

Although components 2–4 appear predominantly in a singleextract and are consistent with the designation of the extract inwhich they occur, significant proportions of each component‘spill’ over into preceding or subsequent fractions. Component1, however, is predominant in two designated fractions and doesnot clearly fit into either the carbonate or the organic/sulfidedesignations.

If the data processing method used is valid, the problems ofnon-specificity of extraction reagents have been demonstrated.In addition, possible mis-identification of extracted fractionshas also been shown.

By multiplying the proportion of each component by itschemical composition and re-scaling to the total extracted solidsfor a given fraction, an analogous table to that previouslyproduced15 can be calculated (Table 4). In this instance,however, it is not the methodologically defined fractions thatare reported but the composition of each of the resolved

Table 4 Resolved component compositions for NIST 2710 in mg g21 with ±2 s (n = 10)

Component

1 2 3 4 S CTV

Element Value Error Value Error Value Error Value Error Value Error Value Error

Al 322 65.2 63 700 5600 608 227 0.00 0 64 600 5600 64 400 800Ca 0.00 0 10 000 883 0.00 0 2510 1330 12 500 1600 12 500 300Cd 1.89 0.38 0.00 0 3.06 1.14 14.8 7.86 19.8 7.95 21.8 0.2Co 1.56 0.31 2.24 0.2 3.22 1.2 0.51 0.27 7.53 1.28 (10)Cu 2020 408 586 51.5 515 192 0.00 0 3120 454 2950 130Fe 71.8 14.5 22 700 1990 10 600 3950 0.00 0 33 400 4420 33 800 1000K 379 76.6 17 000 1494 0.00 0 893 473 18 300 1570 21 100 1100

Mn 751 152 0.00 0 9860 3680 1240 657 11 900 3740 10 100 400Ni 0.99 0.2 8.65 0.76 0.13 0.05 0.08 0.04 9.85 0.79 14.3 1P 65.5 13.2 791 69.6 227 84.8 0 0 1080 111 1060 150

Pb 3040 615 15.4 1.35 1210 452 523 277 4790 812 5532 80Sr 6.76 1.37 263 23.1 7.87 2.94 48.6 25.8 326 34.8 240Ti 0.00 0 2530 222 0.00 0 10.8 5.72 2540 222 2830 100V 3.33 0.67 53.3 4.69 17.9 6.69 0.00 0 74.5 8.2 76.6 2.3Zn 789 160 1055 92.7 3760 1410 1030 547 6630 1520 6952 91

* S represents the sum of the four components. † CTV is the certified total concentration.

Analyst, November 1997, Vol. 122 1215

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

components and their associated uncertainties that is given. Thisis a useful check to show that at the end of the processingprocedure the total amount of each element extracted is stillcomparable with the certified values for the soil.

Non-specific Extraction Trial

The requirements for the non-specific extraction experimentwere: (i) to produce a series of leachate solutions containingvarying proportions of the different physico-chemical com-ponents of the soil, to allow the chemometric procedure toidentify and quantify each component; and (ii) to carry out theexperiments in such a way that analysis of data would allow theeffects and interactions of the three variables (acid concentra-tion, sample to extractant ratio and time) on the dissolution ofthe physico-chemical components of the soil to be studied.

In order to achieve these objectives a formal experimentaldesign was required. The central composite design was chosenas being more suitable than a two level design as it allowscurvature of effects to be investigated and the results can bemore readily visualised as 3 dimensional surface plots. Thedesign outlined in Table 1 was chosen from a menu of designsgiven within the Statistica software package.

The total extracted solids value for each experiment is shownin Table 1 and the chemical compositions of each leachatesolution are shown in Table 5. The eigenvalues for the Varimaxrotated PCA model for this data are shown in Table 2. Thisclearly shows that there are four components with eigenvaluesgreater than 1 and a step change in eigenvalues between thefourth and fifth PC. In addition, Table 3 shows the scores for thefirst four significant PCs identifying the elements Fe, Ca, Pb and

Mn as having the highest scores. Using these elementconcentration profiles and four components as being significantthe composition of each component was calculated using thechemometric procedure previously described. The total ex-tracted solids in each sample due to each component weremodelled separately using a standard MLR method. The totalextracted solids for a given component for each sample isregressed against the main effects of acid concentration (A), thetime (T), the sample to extractant ratio (S) and all of their linearinteractions (AT, AS and TS) and the quadratic effects (A2, T2

and S2). The model takes the form:

Ec = k + x1A + x2T + x3S + x4AT + x5AS + x6TS + x7A2 +x8T2 + x9S2 (2)

where Ec is the total extracted solids for component c; k aconstant term; and x1···x9 the linear regression coefficients. Thek term and the x1···x9 coefficients are calculated by the MLRalgorithm.

For each component model a regression coefficient (R2) iscalculated that gives a measure of how well the model fits thedata (R2 = 1 is a perfect fit).

Initially, all main effects, interactions and quadratic effectsare used to form the MLR model. This model is examined andeffects which are shown to be insignificant (only t valuessignificant at the 95% confidence interval were retained), areremoved one by one (lowest t value first) until the modelconsists only of effects that are significant. The final MLRmodels for each component are shown in Table 6. As anadditional check for the significance of each effect ANOVAwas also carried out on the total extractable solids for each

Table 5 Chemical composition of each leach solution obtained using the experimental design shown in Table 1 (results in mg kg21 and in the same orderas the experimental design in Table 1)

Al Ca Fe Mg P K Si Na Ti Ba Cd Cu Mn Ni V Zn Pb

1932 3346 2744 537 51.6 1134 1217 288 8.29 41.9 19.0 2447 3433 2.12 11.0 2725 40912517 3422 4293 622 61.7 1128 849 300 17.2 34.5 19.4 2535 4442 2.02 15.2 3146 42942066 3214 1003 591 11.1 868 953 290 1.01 7.38 18.4 2274 4618 1.82 6.07 3128 22994633 3374 8085 1466 66.3 1809 2482 385 78.3 125.0 18.5 2579 6153 4.15 20.1 4343 51802454 3387 2753 656 35.8 1060 1443 305 3.33 21.0 18.8 2471 4746 2.67 10.0 3324 32681537 3149 1071 456 16.2 842 730 270 1.01 8.42 18.0 2206 3084 2.33 6.07 2425 22532590 3265 4585 652 67.3 1316 1750 292 33.2 94.0 18.6 2453 4031 2.21 14.8 2900 46333092 3423 4381 849 48.9 1269 2378 343 7.37 49.4 18.8 2506 5480 2.77 14.2 3791 46404048 3479 7395 1197 60.7 1439 764 354 30.4 27.8 18.8 2574 6557 3.34 17.9 4652 46062439 3362 2732 651 35.0 1082 1432 309 3.33 20.6 18.9 2444 4685 2.50 9.92 3269 3185993 3133 1253 380 31.6 889 351 243 5.83 16.7 18.3 2209 2314 1.83 7.66 2108 2798

2721 3423 3148 764 31.7 1087 1465 333 2.50 20.8 18.2 2394 5584 2.50 12.4 3690 328720.0 1152 0.49 188 4.2 436 234 234 0.83 2.09 6.3 175 729 1.17 4.17 839 25.4

4013 3516 7346 1129 70.0 1590 1200 433 39.2 76.2 19.3 2657 6318 3.33 18.8 4447 50933197 3422 5069 912 67.7 1588 3299 367 35.9 193.2 18.7 2594 4484 2.99 15.3 3303 49101928 3284 823 544 8.00 784 848 307 4.00 5.91 18.0 2337 4351 4.40 2.40 3044 21172341 3308 2673 629 3.4 1051 1406 298 3.33 21.5 18.4 2373 4599 2.83 9.91 3181 32131906 3314 2710 533 52.5 1125 1235 283 7.37 40.8 19.2 2414 3395 2.39 10.6 2684 41312651 3536 4303 639 67.8 1067 869 293 18.2 35.0 20.4 2569 4320 2.33 15.5 3379 45202125 3326 1094 617 10.1 898 969 313 1.01 7.23 18.4 2323 4835 2.33 6.27 3213 23874479 3398 7963 1447 67.3 1810 2425 380 77.4 123.0 17.9 2515 6115 3.87 20.1 4226 50172348 3289 2597 648 33.3 1073 1378 307 3.33 22.5 18.8 2395 4644 2.50 10.6 3244 32751519 3153 1061 451 17.2 830 736 259 2.02 8.3 18.2 2204 3019 1.92 5.77 2385 23622551 3234 4464 640 66.3 1328 1716 290 33.2 92.1 18.1 2465 3907 2.86 14.7 2856 47443026 3369 4374 845 47.0 1260 2354 347 7.37 50.4 17.9 2461 5574 2.85 14.2 3766 44623857 3472 7088 1133 58.7 1365 749 346 30.4 27.0 18.4 2535 6523 3.64 18.0 4552 44702341 3263 2618 640 34.1 1056 1371 305 3.33 22.3 18.5 2369 4651 2.33 10.6 3236 3236925 2925 1147 352 30.8 880 337 237 5.00 17.7 17.1 2046 2169 1.67 7.17 1953 2656

2718 3412 3120 761 30.0 1098 1484 333 2.50 20.9 18.2 2403 5598 2.50 12.7 3696 327620.8 1147 0.97 188 0.83 429 227 233 0.83 2.08 6.3 170 724 0.75 4.17 843 26.1

3708 3297 6560 1035 65.8 1502 1142 335 37.5 73.4 18.2 2486 5764 3.83 18.2 4128 49233103 3349 4866 901 66.1 1560 3161 354 36.0 195.0 18.8 2541 4370 2.20 15.0 3235 49022176 4092 758 696 4.0 924 1004 307 4.00 6.45 22.4 2708 4960 2.00 4.40 3488 24692374 3309 2627 652 30.8 1056 1392 301 3.33 21.6 18.8 2387 4678 2.42 10.3 3243 3322

1216 Analyst, November 1997, Vol. 122

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

Fig. 3 Chemical compositions of the four resolved components found in the sequential extract data.

Table 6 Multiple linear regression models for the four components in the non-specific extraction trial data*

Regressioncoefficient SE† t value p 295% CL‡ +95% CL‡ R2

VariableComponent 1—

k 2915.6539 776.9068 3.752901 0.000749 1328.999 4502.309 0.548A 18974.789 3399.415 5.581781 4.51E-06 12032.26 25917.32A2 214660.055 3137.452 2 4.672599 5.86E-05 221067.59 28252.523AT 264.284855 18.04548 2 3.562379 0.001251 2101.1387 227.43106

Component 2—

k 10723.18 1731.21 6.194038 1.27E-06 7171.031 14275.33 0.919A 212730.264 4479.041 2 2.842185 0.008428 221920.5 23540.031A2 14723.282 3623.775 4.062968 0.000375 7287.91 22158.65S 299468.687 18487.88 2 5.38021 1.1E-05 2137402.7 261534.69S2 186638.66 60949.61 3.06218 0.004931 61580.39 311696.9AS 39170.263 17685.89 2.214775 0.035402 2881.821 75458.71AT 140.48036 20.36927 6.896682 2.07E-07 98.68607 182.2746

Component 3—

k 2336.1316 1219.135 2 0.275713 0.784659 22825.937 2153.674 0.788A 18251.044 4489.926 4.064887 0.000319 9081.391 27420.7A2 29202.4231 4267.909 2 2.15619 0.039211 217918.66 2486.1903T 99.33633 13.53768 7.337764 3.58E-08 71.68869 126.984

Component 4—

k 3288.867 145.6462 22.58121 2.2E-20 2991.418 3586.316 0.488A2 21206.386 271.995 24.435324 0.000114 21761.874 2650.8981S 2371.3722 877.2696 2.703128 0.0112 579.7487 4162.996

AT 16.166035 4.259985 3.794857 0.000669 7.465985 24.86608* Data significant at the 95% confidence limit (i.e., p < 0.05). † SE, standard error. ‡ CL, confidence limit.

Analyst, November 1997, Vol. 122 1217

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

component with the ‘F’ statistic being used to test forsignificance. The ANOVA tables for the significance the factorsis summarised in Table 7.

Each MLR model is plotted as a surface plot of the two mostsignificant factors against total extracted solids (Fig. 4) and thechemical composition of each component is given in piediagrams in a similar manner to the sequential leach data(Fig. 5).

Component 1 is predominantly made up of Pb, Ca and Cu andwith smaller amounts of Fe, K, Zn, Fe, Si and Al. Table 6 showsthat the most important factor controlling its dissolution is acidconcentration (significant A and A2 coefficients) with a smallbut significant interaction effect between time and acidconcentration. The ANOVA analysis confirms these findings(A2 and T effects significant at the 95% confidence level). The

surface plot (Fig. 4) shows that the optimum acid concentrationfor dissolution is approximately 0.3–0.7 m with highestconcentrations at very short reaction times. This suggests thiscomponent dissolves very quickly as long as there is areasonable acid concentration, but that its concentration de-creases slightly with time possibly indicating a re-adsorptioneffect. Such behaviour and chemical composition could reflectthe dissolution of very fine particulate or clayey material.

Component 2 is predominantly Fe, the MLR analysis showsthe most important factors controlling its dissolution are sampleto extractant ratio and acid concentration (significant A, A2, S,S2 and AS coefficients) with an additional time and acidconcentration interaction effect. The ANOVA analysis con-firms this (significant A, A2, S, S2, T, AS and AT effects). Itsdissolution is favoured by low sample to extractant ratios, high

Table 7 ANOVA tables for the significance of the factors in the non-specific extraction trial data*

Sums of Degrees Mean squareVariable squares (SS) of freedom value F ratio p

Component 1—

A 3931968 1 3931968 2.121374 0.158215A2 2.7E + 07 1 2.7E + 07 14.7554 0.00079S 4913781 1 4913781 2.651081 0.116538S2 7796433 1 7796433 4.206329 0.051344T 1.8E + 07 1 1.8E + 07 9.95551 0.00428T2 1477269 1 1477269 0.797016 0.380852AS 967137.7 1 967137.7 0.52179 0.477059AT 6455185 1 6455185 3.482699 0.074281ST 243634.1 1 243634.1 0.131445 0.720111

Error 44484014 24 1853501Total SS 1.28E + 08 33

Component 2—

A 4.2E + 08 1 4.2E + 08 177.125 1.4E-12A2 3.9E + 07 1 3.9E + 07 16.3547 0.00047S 1.6E + 08 1 1.6E + 08 66.9972 2.1E-08S2 2.3E + 07 1 2.3E + 07 9.46522 0.00517T 9E + 07 1 93 + 07 37.4768 2.5E-06T2 271800.5 1 271800.5 0.113712 0.738888AS 1.2E + 07 1 1.2E + 07 5.0374 0.03429AT 2.6E + 07 1 2.6E + 07 11.0809 0.00281ST 8025314 1 8025314 3.357519 0.079336

Error 57366030 24 2390251Total SS 8.18E + 08 33

Component 3—

A 1.9E + 08 1 1.9E + 08 51.682 2E-07A2 12767880 1 12767880 3.440736 0.075933S 1848968 1 1848968 0.498267 0.487059S2 7284935 1 7284935 1.963172 0.173973T 1.9E + 08 1 1.9E + 08 51.5757 2E-07T2 1996555 1 1996555 0.538039 0.470352AS 2793707 1 2793707 0.752859 0.394166AT 558913 1 558913 0.150618 0.701367ST 1164110 1 1164110 0.313709 0.580604

Error 89059172 24 3710799Total SS 5.05E + 08 33

Component 4—

A 566192 1 566192 5.46793 0.02803A2 343451.7 1 343451.7 3.316845 0.081063S 755136 1 755136 7.29263 0.01249S2 58748.29 1 58748.29 0.567355 0.458638T 849040 1 849040 8.1995 0.00856T2 83393.03 1 83393.03 0.805358 0.378411AS 401611.7 1 401611.7 3.878518 0.060556AT 877586 1 877586 8.47518 0.00766ST 101024.5 1 101024.5 0.975633 0.333134

Error 2485145 24 103547.7Total SS 6669169 33

* Bold values indicate correlations significant at the 95% confidence limit (i.e., p < 0.05).

1218 Analyst, November 1997, Vol. 122

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

acid concentrations (Fig. 4) and longer dissolution times. Thiscomponent is probably derived from iron oxide dissolution.

Component 3 is predominantly Mn: the MLR and ANOVAanalysis shows the most important factors controlling itsdissolution are acid concentration and time (significant A, A2

and T coefficients for the MLR and significant A, A2 and Teffects for the ANOVA). Its dissolution is favoured by longerreaction times and high acid concentrations (Fig. 4). Thiscomponent is probably derived from Mn oxide dissolution.

Component 4 is predominantly made up of Ca, Zn and Mn:the MLR analysis shows most important factors controlling itsdissolution are acid concentration and ratio (significant A2 andR coefficients) with an additional time and acid concentrationinteraction effect. The ANOVA analysis confirms this butshows T to be a significant effect on its own (significant A, S,and T effects). Its dissolution is favoured by low acidconcentration and high sample to extractant ratio (Fig. 4). Thecomposition of this component does not intuitively point to itsorigin but the mild conditions which favour its dissolution

suggests that this is an easily extractable component, possiblythe exchangeable fraction.

Comparison of the Two Extraction Methods

From a practical point of view, the non-specific extractionmethod has a number of analytical advantages. The simple nitricacid leaching solution does not cause analytical matrixproblems and is likely to have lower blank values than thosefound in the Tessier extraction scheme. In addition, it allows Mgand Na to be determined; these are masked by the extractionmedia used in the Tessier method.

Comparison of the results of the two chemometric data setsreveals a number of distinct similarities between the two sets ofcomponents. Table 8 shows the correlation between thechemical compositions of the components identified in eachextraction method data set.

The compositions of component 1 from both methods aresignificantly correlated. This fraction is dominated by the

Fig. 4 Surface plots of the MLR models of the four components identified in the non-specific extraction trial.

Analyst, November 1997, Vol. 122 1219

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

metals Cu, Pb, Mn and Zn but also has significant quantities ofAl and K. This component could be a fine clay material whichadsorbs heavy metals. Alternatively, this could be an organicmaterial. Further work is required to identify the source of thisextracted fraction.

Component 2 from the Tessier method shows no significantcorrelation with the non-specific method. This is not surprisingas this is the silicate matrix component which is unlikely to beattacked by the relatively mild dissolution conditions of thenon-specific extraction method.

Component 3 from the Tessier method has a low butsignificant correlation with both components 2 and 3 of the non-specific method. This component is predominant in the Fe/Mnoxide designated fraction and the components identified in thenon-specific method are dominated by Fe and Mn respectively.The sum of the compositions of components 2 and 3 in the non-

specific method give a high and significant correlation withcomponent 3 of the Tessier method. This suggests that theTessier method extracts both Fe and Mn oxides simultaneously,whereas the non-specific method has resolved the Fe and Mnoxides as separate entities.

The composition of component 4 from both methods issignificantly correlated. The predominance of this component inthe designated exchangeable fraction in the Tessier scheme andthe fact that it is extracted under very mild extraction conditionssuggests that this is the exchangeable fraction.

Conclusions

The application of a chemometric mixture resolution procedureto a well established sequential leach method and to a new non-specific leach procedure has produced data that are geochem-

Fig. 5 Chemical compositions of the four resolved components found in the non-specific extraction trial data.

Table 8 Correlation coefficients between the component compositions found in the Tessier sequential leach data and the non-specific extraction trialdata*

Tessier method

Component 1 Component 2 Component 3 Component 4Non-specific

method Correlation p value Correlation p value Correlation p value Correlation p value

Component 1 0.7933 0.001 20.1003 0.745 20.1186 0.7 0.3552 0.234Component 2 0.2117 0.487 0.3268 0.276 0.5664 0.044 20.1752 0.567Component 3 0.0961 0.755 0.1702 0.578 0.5831 0.036 0.4069 0.168Component 4 20.0368 0.905 20.1186 0.7 0.1768 0.563 0.9471 < 0.0001Component 2 + 3 0.2422 0.425 0.3891 0.189 0.8759 < 0.0001 0.1398 0.649

* Effects in bold are significant at the 95% confidence limit.

1220 Analyst, November 1997, Vol. 122

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online

ically consistent with the material being studied. It has revealeda certain lack of specificity in the Tessier method for somephases and has been shown to be a potentially powerful methodfor studying the fate of heavy metals in soils and sediments.

The non-specific extraction trial scoping study has demon-strated considerable promise. The results are comparable withthe data independently obtained by the Tessier scheme15 andsuggest that the new method has more flexibility and selectivityin identifying the presence of different physico-chemicalcomponents within a soil material and the trace elementsassociated with it. The method has considerable potential forapplication to environmental pollution studies and to geochem-ical exploration work.

This paper is published with the approval of Director, BritishGeological Survey (NERC).

References

1 Leschber, R., Davis, R. D., and L’Hermite, P., Chemical Methods forAssessing Bio-available Metals in Sludges and Soils, Elsevier,London, 1985.

2 Broekaert, J. A. C., Guçer, S., and Adams, F., Metal Speciation in theEnvironment, Springer, Berlin, 1990.

3 Jenne, E. A., in Proceedings of the Symposium on Molybdenum in theEnvironment, ed. Chappell, W., and Peterson, S. K., Marcel Dekker,New York, 1977, pp. 425–552.

4 Kramer, J. R., and Allen, H. E., Metal Speciation: Theory, Analysisand Application, Lewis, Chelsea, Michigan, 1988.

5 Tessier, A., Campbell, P. G. C., and Bisson, M., Anal. Chem., 1979,51, 844.

6 Breward, N., and Peachey, D., Sci. Total Environ., 1983, 29, 155.7 Lake, D. L., Kirk, P. W. W., and Lester, J. N., J. Environ. Qual., 1984,

13, 175.

8 Schuman, L. M., Soil Sci., 1985, 140, 11.9 Harrison, R. M., Laxen D. P. H., and Wilson, S. J., Environ. Sci.

Technol., 1981, 15, 1378.10 Hickey, M. G., and Kitterick, J. A., J. Environ. Qual., 1984, 13,

372.11 Forstner, U., in Chemical Methods for Assessing Bio-available

Metals in Sludges and Soils, ed., Leschber, R., Davis, R. D., andL’Hermite, P., Elsevier, London, 1985, pp. 1–31.

12 Xian, X., Environ. Pollut., 1989, 57, 127.13 Clevinger, T. E., Water, Air Soil Pollut., 1990, 50, 241.14 Li, X., Coles, B. J., Ramsey, M. H., and Thornton, I., Chem. Geol.,

1995, 124, 109.15 Li, X., Coles, B. J., Ramsey, M. H., and Thornton, I., Analyst, 1995,

120, 1415.16 Jouanneau, J. M., Latouche, C., and Pautrizel, F., Environ. Technol.

Lett., 1983, 4, 509.17 Tipping, E., Hetherington, N. B., Hilton, J., Thompson, D. W.,

Bowles, E., and Hamilton-Taylor, J., Anal. Chem., 1985, 57, 1944.18 Khebonian, C., and Bauer, C., Anal. Chem., 1987, 59, 1417.19 Sholkovitz, E. R., Chem. Geol., 1989, 77, 47.20 Bermond, A., Environ. Technol., 1992, 23, 1175.21 Cave, M. R., and Harmon K., Analyst, 1997, 122, 501.22 Malinowski, E. R., Factor Analysis in Chemistry, Wiley, New York,

2nd edn., 1991.23 Brereton, R. G., Analyst, 1995, 120, 2313.24 Hopke, P. K., in Chemometrics in Environmental Chemistry—

Applications, ed. Einax, J., Springer, Berlin, 1995, vol. 2, part H,pp. 47–86.

25 Thurston, G. D., and Spengler, J. D., Atmos. Env., 1985, 19, 9.26 Gamp, H., Maeder, M., Meyer, C. J., and Zuberbuhler, A. D.,

Talanta, 1985, 32, 1133.27 Kaiser, H. F., Psychometrika, 1958, 23, 187.

Paper 7/05163HReceived July 18, 1997

Accepted October 8, 1997

Analyst, November 1997, Vol. 122 1221

Publ

ishe

d on

01

Janu

ary

1997

. Dow

nloa

ded

on 1

4/04

/201

5 12

:59:

54.

View Article Online