Ms

Ma

b

a

ARRAA

KCMlOWSS

1

NtwhEcnFaLtlNaataaa

(

0d

Analytica Chimica Acta 657 (2010) 19–27

Contents lists available at ScienceDirect

Analytica Chimica Acta

journa l homepage: www.e lsev ier .com/ locate /aca

ultivariate curve resolution of organic pollution patterns in the Ebro Riverurface water–groundwater–sediment–soil system

arta Terradoa,∗, Damià Barcelóa,b, Romà Taulera

Institute of Environmental Assessment and Water Research (IDAEA), Spanish Council for Scientific Research (CSIC), Jordi Girona 18-26, 08034 Barcelona, Catalunya, SpainCatalan Institute for Water Research (ICRA), Parc Científic i Tecnològic de la Universitat de Girona, Edifici Jaume Casademont, Pic de Peguera 15, 17003 Girona, Catalunya, Spain

r t i c l e i n f o

rticle history:eceived 30 June 2009eceived in revised form 8 October 2009ccepted 13 October 2009vailable online 20 October 2009

a b s t r a c t

Multivariate curve resolution alternating least squares (MCR-ALS) is shown to be a powerful chemo-metric method for the analysis of environmental monitoring data sets. It allows for the investigation,resolution, identification, and description of pollution patterns distributed over a particular geographicalarea, time and environmental compartment. An integrated interpretation of the main features charac-terizing pollution patterns of organic contaminants affecting the Ebro River basin (Catalonia, NE Spain)is attempted using the results obtained by MCR-ALS analysis of surface water, groundwater, sediment

eywords:hemometricsultivariate curve resolution alternating

east squaresrganic pollutionater

and soil data sets obtained in a 3-year extensive monitoring study. Agricultural practices were identifiedas the main source of surface and groundwater diffuse pollution, while sediments and soils appearedmostly polluted by a contamination pattern mainly loaded by polycyclic aromatic hydrocarbons (PAHs)of possible pyrolitic origin. Additionally, a third pollution pattern related to past and ongoing industrialactivities was detected to be principally stored in the sediment compartment. Geographical and temporal

lution

edimentoil

distributions of these pol

. Introduction

Diffuse and point pollution in the Ebro River basin (Catalunya,E Spain) caused by pressure factors such as agriculture, indus-

ry and human sewage, is an issue of great concern, since togetherith changes in climatic conditions and land use practices canave large scale adverse impacts over the quality of the basin. Thebro River has a drainage area of approximately 85,000 km2, dis-harging into the Mediterranean Sea, and comprising territories ofine autonomous communities, from Cantabria to Catalunya (seeig. 1A). Population is distributed in a very heterogeneous wayll over the basin, with the major cities being Pamplona, Lleida,ogrono, Vitoria and Zaragoza, which concentrate around 45% of theotal population of the territory. It is the most important irrigatedand in Spain, crops varying with relief and climate (see Fig. 1A).orthernmost regions of Cantabria, País Vasco, Navarra and La Riojare dominated by woodland and pasture (in La Rioja vineyards arelso important). Cereal crops dominate the central regions, in par-

icular Aragón, Castilla y León and Castilla la Mancha. This part islso the most significant for industrial plants such as biomass cropsnd oilseed rape. In the southernmost regions of Castilla La Manchand Catalunya, dry fruit trees and vineyards increase in significance,

∗ Corresponding author. Tel.: +34 934006169; fax: +34 932045904.E-mail addresses: mtcqam@idaea.csic.es (M. Terrado), dbcqam@idaea.csic.es

D. Barceló), rtaqam@idaea.csic.es (R. Tauler).

003-2670/$ – see front matter © 2009 Elsevier B.V. All rights reserved.oi:10.1016/j.aca.2009.10.026

sources are given.© 2009 Elsevier B.V. All rights reserved.

while the Ebro River delta supports a well-developed rice farmingactivity. Diffuse pollution originated by pesticides application inthe basin has been studied elsewhere [1–3]. A historical pollutionfrom chemical plants manufacturing solvents and chlorinated pes-ticides in the southern part of the river basin is also well known[4]. Automobile, textile, food and wood industry as well as miningactivities are important in the northern part.

Multivariate curve resolution alternating least squares (MCR-ALS) is a powerful chemometric method with an increasingapplication for the analysis of environmental monitoring data sets.It has been recently validated for the identification of environ-mental pollution patterns in surface water [5]. This last studywas intended to model pollution in surface water of the EbroRiver delta (a smaller area of around 300 km2), during the maingrowing-season of the rice crop. In the present work, an integratedanalysis of four different environmental compartments – surfacewater, groundwater, sediment and soil – applying MCR-ALS is pro-posed. The concentration of four different families of compounds– organochlorinated compounds (OCs), polycyclic aromatic hydro-carbons (PAHs), pesticides, and alkylphenols (APs) – were analysedduring six different sampling campaigns (from year 2004 to 2006)at various locations distributed within the entire Ebro River basin.

Surface water and groundwater were sampled twice a year, inspring and fall, whereas sediment and soil were only sampled in fall.Sampling was carried out within the framework of the EuropeanIntegrated Project AquaTerra (6th EU RTD Framework Programme),which aims to provide the scientific basis for an improved river

20 M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27

F nomoa erra P

bssdp

2

pwbscwdcawgf

ig. 1. Maps of the Ebro River basin. (A) Location of the basin and division in autogricultural and industrial activities; (B) spatial location of samples from the AquaT

asin management through a better understanding of the river-ediment–soil–groundwater system as a whole [6]. Results allowedummarizing the whole data set as well as the identification andescription of the nature and distribution of the most significantatterns of pollution.

. Materials and methods

Due to their particular physico-chemical properties, every com-ound was not detected in all compartments. Thus, polar pesticidesere usually detected in aquatic compartments, while hydrocar-

ons and organochlorinated compounds were mainly detected inediment and soil. Consequently, the use of different variables (con-entrations of chemical compounds) in the different compartmentsill make difficult to achieve a complete view of the pollutantsynamics. Only alkylphenols were detected in all environmental

ompartments. Concentration units were given in �g L−1 of waternd in �g kg−1 of sediment/soil. The sampling network for surfaceater and sediment coincided geographically, and the same for

roundwater and soil (see Fig. 1B for samples location and Table 1or samples identification).

us communities with the identification of the main urban centers and the generalroject surveillance monitoring.

This section is divided in two different parts: Section 2.1 describ-ing the arrangement of data matrices, and Section 2.2 describing themultivariate method used for their analysis.

2.1. Arrangement of data matrices

Six different individual data matrices (one per sampling cam-paign: sw1, sw2,. . ., sw6) were obtained for surface water, withrows (samples) and columns (variables) coinciding in everydata matrix. These individual data matrices were column-wiseappended, one on top of each other, keeping the same number ofcolumns and originating a new augmented surface water (SW) datamatrix containing 138 samples in total (23 samples analysed in eachof the 6 sampling campaigns) (see Fig. 2). This data arrangementcan be written in a concise way using MATLAB notation program-ming language as [sw1;sw2;. . .;sw6], where swk (k = 1, 2,. . ., 6)corresponds to the different surface water individual matrices from

each sampling campaign, and the semicolon “;” notation is usedto indicate that the different data matrices are column-wise con-catenated, and that they are supposed to share the same vectorspace. Likewise, SW matrix was divided in two new submatri-ces SW1 and SW2, which were obtained from individual matrix

M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27 21

Table 1Identification of samples.

Surface water/sediment Groundwater/soil

R0: Ebro in Reinosa (Cantabria) G1: Cascante (Navarra)R1: Ebro in Miranda de Ebro (Burgos) G2: Cascante (Navarra)T2: Zadorra in Audinaka (Álava) G3: Monteagudo (Navarra)T3: Zadorra in Villodas (Álava) G4: Monteagudo (Navarra)R4: Ebro in Haro (La Rioja) G5: Maleján (Zaragoza)T5: Najerilla in Nájera (La Rioja) G6: Maleján (Zaragoza)R6: Ebro in Logrono (La Rioja) G7: Ainzón (Zaragoza)T7: Ega in Estella (Navarra) G8: Sobradiel (Zaragoza)R7: Ebro in Tudela (Navarra) G9: Alfamén (Zaragoza)T8: Araquil in Alsasua (Navarra) G10: Movera (Zaragoza)T9: Arga in Puente la Reina (Navarra) G11: Mollerussa (Lleida)T10: Jalón in Grisén (Zaragoza) G12: Tornabous (Lleida)T11: Huerva in Zaragoza (Zaragoza) G13: Urunuela (La Rioja)T12: Gállego in Caldearenas (Huesca) G14: San Adrián (Navarra)T13: Gállego in San Mateo de Gállego (Zaragoza) G15: Andosilla (Navarra)R14: Ebro in Presa da Pina (Zaragoza) GA1: Villanueva de Gállego (Zaragoza)R15: Ebro in Sástago (Zaragoza) GA2: Villanueva de Gállego (Zaragoza)T15: Cinca in Alcolea de Cinca (Huesca) GA3: Villanueva de Gállego (Zaragoza)T16: Segre in Torres de Segre (Lleida)

cmpss“tctbm

Fmwstc

R17: Ebro in Flix (Tarragona)R18: Ebro in Tortosa (Tarragona)R19: Ebro in Amposta (Tarragona)R20: Ebro in Delta del Ebro (Tarragona)

oncatenation (see Fig. 2). SW1 data matrix contained the dataatrices sampled in summer (23 samples analysed in 3 sam-

ling campaigns: sw1, sw4 and sw5), while SW2 contained thoseampled in fall (23 samples analysed in 3 sampling campaigns:w2, sw4, sw6). Samples were numbered from North to South,R” indicating “Ebro River” and “T” indicating “tributary”. Four-

een variables were measured in every sample: organophosphateompounds (diazinon, dimethoate, ethion and tributylphosphate),riazines (atrazine, desethylatrazine, simazine, terbutryn and ter-uthylazine), an anilide (propanil), chloroacetanilides (alachlor andetolachlor) and APs (octylphenol and nonylphenol).

ig. 2. Arrangement of the different analysed column-wise concatenated dataatrices. In capitals, the augmented data matrices: SW, surface water; GW, ground-ater; SE, sediment; SO, soil; [SW;GW], surface and groundwater; and [SE;SO],

ediment and soil. In lower case letter, the individual data matrices correspondingo each sampling campaign (i.e. sw1, data matrix corresponding to the first samplingampaign for surface water).

Six different individual data matrices were obtained for ground-water (gw1, gw2,. . ., gw6) but, in this case, rows (samples) werenot common for all data matrices. The new groundwater (GW)data matrix obtained after individual matrix concatenation con-tained 92 samples in total (see Fig. 2). In this case, the number ofsamples for the different sampling campaigns was not coincident(10, 16, 17, 15, 17, and 17 locations were sampled, from 1st to 6thcampaigns, respectively). Seven variables (all of them detected insurface water as well) were measured in every groundwater sam-ple: an organophosphate compound (tributylphosphate), triazines(atrazine, desethylatrazine, simazine and terbuthylazine) and APs(octylphenol and nonylphenol).

Three different individual data matrices (one per samplingcampaign: se1, se2, se3) were obtained for sediment, with rowsand columns coinciding in every data matrix. The new sediment(SE) data matrix obtained after individual matrix concatena-tion contained 69 samples in total (23 samples analysed in 3sampling campaigns, coinciding with surface water campaignssampled in fall) (see Fig. 2). Twenty-seven variables were measuredin every sample: PAHs (naphthalene, acenaphtylene, acenapht-ene, fluorene, phenanthrene, anthracene, fluoranthene, pyrene,benzo(a)anthracene and benzo(g,h,i)perylene), APs (octylphenoland nonylphenol), an organophosphate compound (tributylphos-phate) and OCs (�-HCH, hexachlorobenzene, 2,4-DDE, 4,4-DDE,2,4-DDD, 4,4-DDD, 2,4-DDT and 4,4-DDT).

Finally, three additional individual data matrices were obtainedfor soil (so1, so2, so3), in this case with the same number of sam-ples (rows) each of them. The new soil data matrix (SO) obtainedafter individual matrix concatenation contained 36 samples in total(12 samples analysed in 3 sampling campaigns) (see Fig. 2). Fif-teen variables (all of them detected in sediments as well) weremeasured in every sample: PAHs (acenaphtylene, phenanthrene,anthracene, fluoranthene, pyrene, benzo(a)anthracene, chrysene,benzo(b)fluoranthene, benzo(a)pyrene, indeno(1,2,3-cd)pyrene,dibenzo(a,h)anthracene and benzo(g,h,i)perylene), an organophos-phate compound (tributylphosphate) and an OC (4,4′-DDE).

Matrices SW and GW were then column-wise appended giv-

ing the new matrix [SW;GW] and the same for matrices SE andSO, giving [SE;SO] matrix (see Fig. 2). Only variables coincidingin both compartments (SW and GW on one side, and SE and SOon the other side) are included in the matrix concatenation. Non-coincident variables are not considered in this new arrangement of

2 Chim

ciats1o

2

wpLEu(vdstowbaam[mcs

fipdtmmtt

o

D

YcststredtcotmwactTct

2 M. Terrado et al. / Analytica

ompartmental matrices. Matrix [SW;GW] contained 230 samplesn total (138 surface water samples and 92 groundwater samples)nd 7 variables were measured in every sample (coinciding withhe ones measured in groundwater). Matrix [SE;SO] contained 105amples in total (69 sediment samples and 36 soil samples), and5 variables were measured in every sample (coinciding with thenes measured in soils).

.2. Multivariate data analysis

Data analysis was performed using the MATLAB (The Math-orks, MA, USA, 2004) numerical computing, visualisation androgramming environment, PLS Toolbox 4.0 (Eigenvector Researchtd., Manson, WA, USA) and MCR-ALS Toolbox (www.ub.edu/mcr).stimation of missing data was done by PCA imputation [7]. Val-es below the detection limit (<LOD) were replaced by their halfLOD/2) [8]. Variables in these data matrices were scaled to unitariance since their magnitudes differed considerably. When theifferent appended data matrices corresponding to the differentampling campaigns for each compartment are scaled together,he visualization of the temporal or compartmental variationsf the identified contamination patterns is enhanced. However,hen scaling is applied over each of the individual data matrices

efore their concatenation, the temporal or compartmental vari-tions (scale changes among individual matrices) remain hiddennd a larger relevance is given to geographical differences [3]. Foratrices combining two different environmental compartments

SW;GW] and [SE;SO], scaling was also applied together over theatrices of the different individual sampling campaigns in each

ompartment in order to allow visualisation of compartmentalcale variations as well.

Principal component analysis (PCA) [9] was used in this workor exploratory analysis purposes only to identify the main trendsn the data sets and to estimate the number of main contaminationatterns. This method has been already applied for environmentalata analysis in other works performed within the framework ofhe AquaTerra Project [3,10,11]. A thoroughly application of the

ultivariate curve resolution alternating least squares (MCR-ALS)ethod [5,12,13] is presented here for analysis and summary of

he large environmental monitoring data tables obtained duringhe AquaTerra Project.

As PCA, multivariate resolution methods (MCR-ALS) are basedn a bilinear model which performs the data matrix decomposition:

aug = XaugYT + Eaug (1)

T is usually referred as the matrix of loadings defining, in thease of analyzing environmental data tables, the chemical compo-ition of the resolved environmental patterns. In the analysis ofhe column-wise augmented data matrices, this chemical compo-ition (loadings) of the resolved contamination patterns is forcedo be the same for the different analysed campaigns. Xaug cor-esponds to the matrix of scores, which in the case of analyzingnvironmental data tables contains the geographical and temporalistribution of the resolved patterns (spatial distribution of pat-erns is different in each campaign). Note, however, that in thisolumn-wise matrix augmentation arrangement, the distributionf these contamination patterns (scores) is allowed to change forhe different campaigns (individual data matrices). Eaug is the aug-

ented matrix of error, containing all the parts of the data variancehich are not explained by the model. The bilinear model in Eq. (1)

ssumes that the major sources of the experimental data variance

an be explained by a small number of components defining thewo reduced-size factor matrices (scores and loadings) in Eq. (1).he model described by this equation assumes that the measuredoncentration of a contaminant (variable) in a particular sample ishe sum of a reduced number of contributions of this contaminant

ica Acta 657 (2010) 19–27

(variable) from different sources. It is therefore a mixture analy-sis problem with unknown sources which have to be estimatedfrom the analysis. Since the solution of Eq. (1) is ambiguous, thematrix decomposition in this equation has to be performed undersome constraints. In the case of PCA, the matrix decompositionis performed under orthogonal constraints, loadings normaliza-tion and maximum explained variance for the successive extractedcomponents. Under these constraints, PCA provides unique solu-tions [9]. However, these solutions are abstract linear combinationof the true experimental variance sources and, although they arevery useful for data exploration and summary, in many cases theycan be difficult to interpret in environmental terms. Instead, thematrix bilinear decomposition performed by MCR-ALS uses softernatural constraints giving loading and score profiles more easilyinterpretable and more reasonable from an environmental pointof view [5,14]. Constraints used in this work during the MCR-ALSbilinear matrix decomposition were non-negativity, normalisationof loadings to equal length and, if possible, the trilinearity con-straint [15]. This last constraint can only be applied when the samesamples are analysed in the different sampling campaigns, whichimplies that the data set can be arranged in a regular data cube orthree-way data structure (for more details about three-way dataanalysis methods, see Smilde et al. [16]). The application of thisconstraint forces the score (samples) geographical profiles to bethe same for the different individual data matrices obtained in thedifferent campaigns simultaneously analysed in the column-wiseaugmented data matrix. The advantage of applying a trilinear modelfor the analysis of an environmental data set is that the geograph-ical and temporal information appearing intermixed in the scoresmatrix in the bilinear model of the augmented column-wise datamatrix of Eq. (1), can be now recovered separately and withoutambiguities. This simplifies data interpretation in a considerableway since not only the composition of the identified contamina-tion patterns YT is unique during the different sampling campaigns,but also the spatial distribution of these contamination patterns Xwill follow the same trend, only changing in a factor scale Z definingthe temporal variability among sampling campaigns. Nevertheless,a trilinear model is a highly demanding requirement which is rarelytotally fulfilled in environmental data sets. On the other hand, whenthe trilinearity constraint is not applied, the solutions obtained byMCR-ALS may have some degree of ambiguity, depending on thelocal rank structure of the augmented data matrix [12]. MCR-ALSwas applied in this work for the analysis of the augmented datamatrices SW1, SW2 and SE, as well as for the combined [SW;GW]and [SE;SO] augmented data matrices.

3. Results and discussion

This section is divided in four parts describing the resultsobtained by the application of MCR-ALS to the analysis of thedifferent environmental data matrices corresponding to variouscompartments: Section 3.1, for surface water (SW1 and SW2augmented matrices); Section 3.2, for surface and groundwater([SW;GW] augmented matrix); Section 3.3, for sediment (SE aug-mented matrix) and Section 3.4, for sediment and soil ([SE;SO]augmented matrix).

3.1. Surface water (SW1 and SW2 data matrices)

MCR-ALS analysis of surface water data matrices was performed

separately: on one hand, the augmented matrix SW1 (69 sam-ples × 14 variables), containing data monitored during summercampaigns and, on the other hand, the augmented matrix SW2 (69samples × 14 variables), corresponding to fall campaigns. For thesetwo augmented data matrices, five different patterns of contami-

M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27 23

F ce wai ine; 6a

neOVttsEsomsooipiiiv

abcpbpf

t(inmaapic

ig. 3. Composition of the identified patterns of contamination (loadings) in surfadentification: 1, diazinon; 2, dimethoate; 3, ethion; 4, tributylphosphate; 5, atrazlachlor; 12, metolachlor; 13, octylphenol; 14, nonylphenol.

ation were identified (some of them coinciding in both seasons)xplaining around 75% of the total data variance in both cases.btained composition profiles (loadings) are displayed in Fig. 3.ariables (14 organic compounds) are identified with a number in

he x-axis. Variables with higher loadings are directly labelled inhe plot for clarity. In the y-axis, the relative contribution of everycaled variable to the identified contamination pattern is given.ven though the study was made for the whole river basin in botheasons, a division of the area in three sub-regions (the upper courser river source, the middle course, and the lower course or riverouth), was suggested from the obtained results. Thus, in order to

implify the environmental interpretation of data, the discussionf results is presented in Fig. 3 according to this division of the areaf study. As MCR loadings do overlap (they are not orthogonal liken PCA), the sum of the variances explained by each separate com-onent exceeds the variance explained by the whole model. An

ndication of the relative variance explained by each components also given in Fig. 3. Contamination patterns have been arbitrar-ly ordered, without taking into account the amount of explainedariance.

In the upper course, a contamination pattern associated togricultural practices, mostly described by the contribution of ter-utryn and terbuthylazine, was identified for both, summer and fallampaigns (see upper course patterns in Fig. 3). This contaminationattern was also identified in fall, presenting as well, a little contri-ution of diazinon. A second industrial and/or urban contaminationattern, mostly loaded by octylphenol, was detected in this courserom the analysis of fall campaigns.

In the middle course, a contamination pattern described byhe contribution of atrazine, alachlor, metolachlor and octylphenolthis last variable presenting a lower loading than in the patterndentified in the upper course, mostly loaded only by nonylphe-ol), was obtained from the analysis of summer campaigns (seeiddle course patterns in Fig. 3). This component was associ-

ted to agricultural practices taking place in the central region. Inddition, another contamination pattern loaded by tributylphos-hate and nonylphenol, which are compounds often related to

ndustry and urbanisation, was identified in both, summer and fallampaigns.

ter of the Ebro River in summer and fall campaigns from 2004 to 2006. Variables, desethylatrazine; 7, simazine; 8, terbutryn; 9, terbuthylazine; 10, propanil; 11,

In the lower course, four agricultural contamination patternswere resolved, two of them in summer and two in fall. Theywere all characterized by the presence of triazines in their chem-ical composition (see lower course patterns in Fig. 3). Variablesdiazinon and the group of chloroacetanilides also contributedto define the agricultural contamination pattern resolved inthe lower course in both seasons. However, while in summercampaigns diazinon appeared in one of the identified patternsand chloroacetanilides in the other, both compounds appearedtogether in the same contamination pattern in fall campaigns. Nei-ther industrial nor urban contamination was detected in surfacewater samples analysed in the lower course of the Ebro Riverbasin.

As it was already studied elsewhere [2,17], triazines resulted tobe the most intensively applied pesticides over the Ebro River basin,and also the most ubiquitously found. They are mainly used for fruitbearings, vineyards and olive trees. Normally, the period of pesti-cides application in the Ebro basin corresponds to the months ofMay to September [2]. A contamination pattern loaded by triazinestogether with alachlor and metolachlor was identified in summer inthe central region of the basin, in agreement with their applicationover cereals (wheat) and industrial crops (sun-flower). However,the highest contribution of the triazines pattern was detected forsample T16, around Lleida (a well-known agricultural area), andthis high level was maintained until reaching the Ebro delta. Con-tamination patterns loaded by terbutryn and terbuthylazine wereresolved in the upper part of the basin. These compounds are fre-quently applied over vineyards [2]. As their period of applicationcan vary depending on the climate, they were mostly detected insummer or in fall, depending on the year of sampling and the differ-ent sub-regions of the basin. In the upper course, a contaminationpattern related to industrial and urban activities was resolved aswell, which was mostly loaded by alkylphenols. These compoundsare released in the environment by direct urban or industrial input

or via sewage treatment plants effluents [18]. The same type ofcontamination due to industry and urbanisation was also resolvedin some locations in the middle course of the river, mainly aroundZaragoza, a highly populated city with a considerable industrialactivity.

24 M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27

F ce ans om 1s3

3d

Mvnivpppaa3aettfactlpc[atTAccmwcitoJMtst

ig. 4. Composition of the identified patterns of contamination (loadings) in surfaamples (scores) from year 2004–2006. Samples ordered for both compartments frand 5 sampled in summer and 2, 4 and 6 sampled in fall.

.2. Surface and groundwater simultaneous analysis ([SW;GW]ata matrix)

Two main patterns of contamination were resolved in theCR-ALS analysis of the [SW;GW] data matrix (230 samples × 7

ariables). Composition profiles (loadings) of the resolved compo-ents are shown in Fig. 4 (plots on the left). Variables are identified

n the x-axis. In the y-axis, the relative contribution of every scaledariable to the identified contamination pattern is given. Tem-oral and spatial sample distribution profiles of contaminationatterns (scores) are displayed in Fig. 4 (plots on the right). Sam-les are represented in the x-axis for the two compartments, GWnd SW, successively ordered from the 1st to the 6th campaignnd, within each campaign, from North-West to South-East. 1st,rd and 5th campaigns were sampled in summer while 2nd, 4thnd 6th were sampled in fall. In the y-axis, the contribution ofvery resolved contamination pattern (described in the loadings)o each of the analysed samples is displayed. A contamination pat-ern attributed to agricultural practices was clearly distinguishedrom another associated to an urban and/or industrial origin. Thegricultural contamination pattern was mainly described by theontribution of triazines (atrazine, desethylatrazine, simazine anderbuthylazine). This contamination was generally found at higherevels in groundwater than in surface water (see the upper scorelot in Fig. 4). Triazines have been classified as lixiviable pesti-ides according to the groundwater ubiquity score (GUS) index19], with a medium mobility and persistence which makes themble to pollute both surface and groundwater if there is rain afterheir application on crops or if irrigation networks exist in the area.hus, soil becomes a potential source for groundwater pollution.s displayed in the upper score plots in Fig. 4, the agriculturalontamination of triazines presented different tendencies in theirontribution to samples depending on the environmental compart-ent being analysed. While a rather constant temporal behavioras detected in groundwater during all the analysed sampling

ampaigns, the tendency for surface water was different. Contam-nation by triazines in surface water was always higher in summerhan in fall campaigns. This evidenced that triazines were appliedver crops in the Ebro River basin during the months of May and

une (assuming a break through time of a few weeks or months).

oreover, surface water presents a more sensitive response to pes-icides application than groundwater. Levels of triazines detected inurface water were higher during summer campaigns (after themo be applied), in contrast to its detection in fall campaigns, when

d groundwater of the Ebro River basin and patterns contribution to the analysedt to 6th sampling campaigns and, for each campaign, from NW to SE. Campaigns 1,

levels appeared to be lower. The highest contribution of triazinescontamination in surface water was located downstream from thecentral part of the river basin. Besides, as surface water flows fasterthan groundwater (which travels only a few centimeters per year),contaminants are more rapidly transported. In groundwater, levelsof triazines contamination were quite constant during the wholesampling period, not indicating a punctual but an accumulativecontamination effect. A contamination pattern coming from indus-trial and urban activities was resolved, which was mostly loaded bytributylphosphate, octylphenol and nonylphenol compounds (seethe lower loading plot in Fig. 4). These substances are normallylinked to industrial and household applications being, most of thetimes, directly discharged into the river. As their octanol–waterpartition coefficient (Kow, indicating the ratio of the concentrationof a chemical in octanol and in water at equilibrium) is high, theypresent a low capacity to be dissolved in water and therefore, theyare difficult to lixiviate. In contrast to agricultural contamination,the contamination originated by urban and industrial practices wasfound at slightly higher levels in surface water than in groundwa-ter (see the lower score plot in Fig. 4). Its levels were lower in year2006 for both of the analysed compartments. No seasonal tendencywas identified for this pollution pattern, neither in surface nor ingroundwater.

3.3. Sediment (SE matrix)

Three main patterns of contamination were resolved by MCR-ALS analysis of the augmented SE data matrix (69 samples × 27variables). Fig. 5 shows the geographical representation of resultsover the map of the Ebro River basin. In this figure, the chemicalcomposition (loadings) of the contamination patterns is given inbars, and the importance of the contribution of each pattern onevery sample (scores) is represented by dots presenting a differentcolour and size according to their relative importance. One of theresolved patterns of contamination was loaded by the contributionof most of the polycyclic aromatic hydrocarbons (PAHs) analysedin the sediment compartment. As no trilinearity constraint wasapplied over this component during the MCR-ALS analysis, the spa-tial distribution of PAHs contamination resulted to be different for

the three analysed campaigns. It is for this reason that three dif-ferent plots are given in Fig. 5 to display PAHs spatial distribution(from 2004 to 2006). A contamination pattern loaded by the groupof organochlorinated compounds (OCs) was resolved as well, whichpresented an important contribution of hexachlorobenzene, and

M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27 25

Fig. 5. Composition and spatial distribution of the main patterns of contamination identified in sediment of the Ebro River basin from year 2004 to 2006. Bigger dotsrepresenting higher levels of pattern contribution than smaller dots. Different spatial distribution in time for the PAHs pattern (three graphs) and constant distribution intime for the other two identified patterns (one graph). Variables identification: 1, naphthalene; 2, acenaphtylene; 3, acenaphtene; 4, fluorene; 5, phenanthrene; 6, anthracene;7 nthend l; 19, t4

ahadnocoTimd

f[aeatpt2tw

, fluoranthene; 8, pyrene; 9, benzo(a)anthracene; 10, chrysene; 11, benzo(b)fluoraibenzo(a,h)anthracene; 16, benzo(g,h,i)perylene; 17, octylphenol; 18, nonylpheno,4-DDD; 26, 2,4-DDT; 27, 4,4-DDT.

last resolved pattern of contamination was characterized by aigh contribution of alkylphenols (APs) and, in a less importantmount, by those PAHs having a higher number of rings. For theisplay of these two last contamination patterns, only one plot wasecessary (see Fig. 5), since the trilinearity constraint was appliedver them during the MCR-ALS analysis. This constraint forces theomponents to adopt the same distribution profile (the same shapef the scores plot) during the three different analysed campaigns.he selection of the components over which to apply the trilinear-ty constraint was performed allowing for the obtaining of a good

odel fit which, at the same time, allowed a good environmentalata interpretation.

PAHs are widespread environmental contaminants resultingrom combustion, discharge of fossil fuels and automobile exhausts20]. As they are hydrophobic substances, they are stronglydsorbed to the organic fraction of sediments and soils. A differ-nt spatial distribution of PAHs was obtained for each of the threenalysed years. However, the upper course of the Ebro River washe most affected area by this contamination during the whole

eriod of study. In Fig. 5, larger dots represent higher contribu-ions of this PAHs contamination pattern than smaller dots. In year004 (upper map on the left of Fig. 5), samples R0 (the closest tohe river source) and T8 (an industrial place located in Navarra)ere the most affected sites by PAHs contamination. Due to its

e; 12, benzo(k)fluoranthene; 13, benzo(a)pyrene; 14, indeno(1,2,3-cd)pyrene; 15,ributylphosphate; 20, �-HCH; 21, HCB; 22, 2,4-DDE; 23, 4,4-DDE; 24, 2,4-DDD; 25,

location, R0 was not expected to be a polluted site but a ratherclean site. Apart from these locations, other sites in the middlecourse (close to the cities of Zaragoza and Lleida), were also iden-tified as specially affected by this type of contamination in years2005 and 2006 (central and lower maps on the left of Fig. 5). Asratios obtained from the contribution of different PAHs can helpto recognize the origin of contamination [20], the ratios phenan-threne/anthracene (Phe/Ant) and fluoranthene/pyrene (Flu/Pyr) inthe identified pattern of PAHs contamination were calculated withthe aim to deduce their possible source in the Ebro River basin. Inthis case, the loadings obtained in the MCR-ALS resolution of thePAHs pattern of contamination were used instead of raw concen-trations to calculate the ratio. Since data were scaled previously toperform the analysis, loadings were then unscaled to eliminate theeffect of the scaling transformation on variable proportions (apply-ing the inverse operation used in scaling, that is, multiplying everyscaled variable concentration by the standard deviation of the val-ues of this variable without scaling; see more details about thisoperation in Terrado et al. [5]). According to Benlahcen et al. [20],

ratios of Phe/Ant under 10 and ratios of Flu/Pyr over 1 are indicatorsof pyrolitic origin (combustion processes) in contrast to petrogenicorigin. Results obtained in this way in the present study suggestthat PAHs contamination identified in the sediment compartmentwas mostly of pyrolitic origin.

26 M. Terrado et al. / Analytica Chimica Acta 657 (2010) 19–27

Fig. 6. Composition of the identified patterns of contamination (loadings) in sediments and soils of the Ebro River basin and patterns contribution to the analysed samples( 1st toc ; 6, be1 enzo(g

iupRahpRp

u(cZfa

3m

ACsaegtrmayp

scores) in fall from year 2004–2006. Samples ordered for both compartments fromation: 1, acenaphtylene; 2, phenanthrene; 3, anthracene; 4, fluoranthene; 5, pyrene0, benzo(a)pyrene; 11, indeno(1,2,3-cd)pyrene; 12, dibenzo(a,h)anthracene; 13, b

The distribution of the OCs contamination pattern is presentedn the upper map on the right of Fig. 5. These substances weresed as pesticides in the past, but their application is banned atresent. OCs were specially detected in the lower course of the Ebroiver basin. Sample R17 in Flix, Tarragona, was the one presentinghigher contribution of this contamination pattern, since this areaas a historical well-known contamination problem caused by OCsroduced as sub-products of the local industry. Samples R19 and20, located R17 downstream and very close to the Ebro River delta,resented a high OCs contamination as well.

The APs contamination pattern, which is related to industry andrbanisation, was widely distributed over the whole river basinlower map on the right of Fig. 5). The highest contribution of thisontamination was mainly detected close to the big cities such asaragoza, Lleida and Vitoria, among others, where the main sourcesor this type of contamination (urbanisation, industry and wastew-ter treatment plants) coexist.

.4. Sediment and soil simultaneous analysis ([SE;SO] dataatrix)

Three main patterns of contamination were resolved by MCR-LS analysis of [SE;SO] data matrix (105 samples × 15 variables).omposition profiles (loadings) of the resolved components arehown in Fig. 6 (plots on the left). Variables are identified withnumber in the x-axis. In the y-axis, the relative contribution of

very scaled variable to the identified contamination pattern isiven. Temporal and spatial sample distribution profiles of the con-amination patterns (scores) are represented in Fig. 6 (plots on the

ight). In the x-axis, samples are identified for the two compart-ents, SE and SO, successively ordered from 1st to 3rd campaign

nd, within each campaign, form North-West to South-East. The-axis displays the contribution of every resolved contaminationattern to samples.

3rd sampling campaigns and, for each campaign, from NW to SE. Variables identifi-nzo(a)anthracene; 7, chrysene; 8, benzo(b)fluoranthene; 9, benzo(k)fluoranthene;,h,i)perylene; 14, tributylphosphate; 15, 4,4-DDE.

Two of the resolved contamination patterns were mostly loadedby PAHs. One of them was loaded by PAHs of 3–5 rings (the major-ity of analysed PAHs) while the other had contribution of only theheaviest ones (5–6 rings), presenting hardly any contribution ofPAHs with less than 5 rings. The third resolved contamination pat-tern in sediments and soils was mostly loaded by tributylphosphate(TBP), 4,4-DDE (substances which did not appear in the previousidentified patterns) and some PAHs with 4–5 rings.

The pattern of contamination described by lighter PAHs pre-sented a higher contribution over sediment samples than over soilsamples (see score plots in Fig. 6). Even though from observing Fig. 6this contamination pattern seemed not to present any contributionin soil, it was only an effect caused by the simultaneous scaling ofboth compartments. The contribution of PAHs contamination wasmuch more reduced in soil samples (but also detected). Ratios ofdifferent PAHs suggested, as well, a pyrolitic instead of a petro-genic origin for this contamination pattern. As it was already foundin the sediment compartment, samples R14 (close to Zaragoza) andR0 (close to the river source) presented an important contributionof the PAHs contamination pattern. No clear temporal tendencywas observed for the lighter PAHs in sediments (see the upper ploton the right of Fig. 6). Soils presented higher levels of the contam-ination pattern defined by heavier PAHs (especially in the thirdsampling campaign). Samples GA1 and GA2 (close to Zaragoza), andG14 (close to Navarra, in the upper course) were the most affectedones. From previous studies [21], a higher contribution of heavierPAHs indicates a local pollution in contrast to the pollution origi-nated by atmospheric transport, occurring mostly for the lightestPAHs. Soil samples in this study were taken from agricultural soils,characterized by intense vineyard and corn productions. As dis-

cussed in Hildebrandt et al. [22], burning of plants is a commonpractice in Spain to eliminate plant residues and it is also used asfertilizer, being a potential source of PAHs in agricultural soils. TBPand DDE contribution (specially important in the third identifiedpattern of contamination) was mainly detected in the lower course

Chim

op1mat

4

sbtiMctsrfisivtp

ctbmasedts

A

PcAt

R

[

[

[

[

[

[

[

[

[

[

[

M. Terrado et al. / Analytica

f the river, at higher levels in sediments than in soils. DDE is therincipal metabolite of DDT, which was used in Spain during the980s as insecticide, presenting a large persistence in the environ-ent. Again, as in the sediment compartment, R17 was identified

s the sample with the highest levels of this type of organic con-amination.

. Conclusions

Agricultural practices have been identified as the main respon-ible source of surface and groundwater pollution in the Ebro Riverasin. A main contamination source of triazines was detected forhe whole period of study, from 2004 to 2006, since they arentensively used and distributed all over the territory of the basin.

oreover, the burning of weed and plant residues, which is aommon practice in the area, constitutes a second different con-amination source loaded by PAHs, mostly affecting sediment andoil compartments. In general terms, sediment and groundwatereflect the accumulated historical contamination, providing a goodngerprint of past activities in the area, while surface water andoil reflect the contamination from more recent practices. Contam-nation by alkylphenols coming from industry and urbanization oria sewage treatment plants, was also an important source of pollu-ion affecting mainly sediment and aquatic compartments but notresenting a significant impact over soil.

Since not all the contaminants affecting the Ebro River basinan be measured in all environmental compartments at the sameime (they tend to be accumulated in one or another compartmentut not simultaneously in all of them), different relevant compart-ents (i.e. surface water, groundwater, sediment, soil, atmosphere,

erosol, vegetation and biota) should be analysed. However, theimultaneous analysis of several compartments increases consid-rably the total volume of analysed data. In this sense, chemometricata analysis methods constitute very useful and recommendedools for the summary and comprehension of large volume dataets obtained in environmental monitoring studies.

cknowledgements

This work was supported by the European Union FP6 Integratedroject AquaTerra (GOCE 505428) and the Spanish Ministry of Edu-ation and Science (CTQ2006-15052). Dr. Alain Hildebrandt and Dr.licia Navarro are gratefully thanked for providing the data used in

his study and for their valuable explanations on sampling.

eferences

[1] S. Manosa, R. Mateo, R. Guitart, A review of the effects of agricultural and indus-trial contamination on the Ebro delta biota and wildlife, Environ. Monit. Assess.71 (2001) 187–205.

[

[

ica Acta 657 (2010) 19–27 27

[2] A. Claver, P. Ormad, L. Rodríguez, J.L. Ovelleiro, Study of the presence of pesti-cides in surface waters in the Ebro river basin (Spain), Chemosphere 64 (2006)1437–1443.

[3] M. Terrado, M. Kuster, D. Raldúa, M.J. López de Alda, D. Barceló, R. Tauler, Use ofchemometric and geostatistical methods to evaluate pesticide pollution in theirrigation and drainage channels of the Ebro River delta during the rice-growingseason, Anal. Bioanal. Chem. 387 (2007) 1479–1488.

[4] S. Lacorte, D. Raldúa, E. Martínez, A. Navarro, S. Díez, J. Bayona, D. Barceló, Pilotsurvey of a broad range of priority pollutants in sediment and fish from theEbro River basin (NE Spain), Environ. Pollut. 40 (2006) 471–482.

[5] M. Terrado, D. Barceló, R. Tauler, Quality assessment of the multivariate curveresolution alternating least squares (MCR-ALS) method for the investigation ofenvironmental pollution patterns, Environ. Sci. Technol. 43 (2009) 5321–5326.

[6] J.A.C. Barth, P. Grathwohl, H.J. Fowler, A. Bellin, M.H. Gerzabek, G.J. Lair, D.Barceló, M. Petrovic, A. Navarro, Ph. Négrel, E. Petelet-Giraud, D. Darmendrail,H. Rijnaarts, A. Langenhoff, J. de Weert, A. Slob, E. Frank, A. Gutierrez, R. Kret-zschmar, T. Gocht, D. Steidle, F. Garrido, K.C. Jones, S. Meijer, C. Moeckel, A.Marsman, G. Klaver, T. Vogel, C. Bürger, O. Kolditz, H.P. Broers, N. Baran, J.Joziasse, W. von Tümpling, P. Van Gaans, C. Merly, P. Chapman, S. Brouyere,J. Batlle Aguilar, Ph. Orban, N. Tas, H. Smidt, Mobility, turnover and storage ofpollutants in soils, sediments and waters: achievements and results of the EUProject AquaTerra. A review, Agron. Sustain. Dev. 28 (2008).

[7] B. Walczak, D.L. Massart, Dealing with missing data. Part I and II, Chem. Intell.Lab. Syst. 58 (2001) 15–42.

[8] I.M. Farnham, A.K. Singh, K.J. Stetzenbach, K.H. Johannesson, Treatment ofnon-detects in multivariate analysis of groundwater geochemistry data, Chem.Intell. Lab. Syst. 60 (2002) 265–281.

[9] I.T. Jolliffe, Principal Component Analysis, Springer-Verlag, New York, 2002.10] A. Hildebrandt, M. Guillamón, S. Lacorte, R. Tauler, D. Barceló, Impact of pes-

ticides used in agriculture and vineyards to surface and groundwater quality(North Spain), Water Res. 42 (2008) 3315–3326.

11] A. Navarro, R. Tauler, S. Lacorte, D. Barceló, Occurrence and transport of pesti-cides and alkylphenols in water samples along the Ebro River basin, J. Hydrol.,2009, doi:10.1016/j.hydrol.2009.06.039.

12] R. Tauler, A. Smilde, B. Kowalski, Selectivity, local rank, three-way data analysisand ambiguity in multivariate curve resolution, J. Chemom. 9 (1995) 31–58.

13] A. De Juan, R. Tauler, Comparison of three-way resolution methods for non-trilinear chemical data sets, J. Chemom. 15 (2001) 749–772.

14] M. Felipe-Sotelo, L. Gustems, I. Hernández, M. Terrado, R. Tauler, Investigationof geographical and temporal distribution of tropospheric ozone in Catalo-nia (North-East Spain) during the period 2000–2004 using multivariate dataanalysis methods, Atmos. Environ. 40 (2006) 7421–7436.

15] R. Tauler, I. Marqués, E. Casassas, Multivariate curve resolution applied tothree-way trilinear data: study of a spectrofluorimetric acid–base tritrationof salicylic acid at three excitation wavelengths, J. Chemom. 12 (1998) 55–75.

16] A. Smilde, R. Bro, P. Geladi, Multi-way Analysis, John Wiley & Sons, England,2004.

17] A. Hildebrandt, S. Lacorte, D. Barceló, Assessment of priority pesticides, degra-dation products, and pesticide adjuvants in groundwaters and top soils fromagricultural areas of the Ebro river basin, Anal. Bioanal. Chem. 387 (2007)1459–1468.

18] R. Céspedes, S. Lacorte, A. Ginebreda, D. Barceló, Occurrence and fate ofalkylphenols and alkylphenol ethoxylates in sewage treatment plants andimpact on receiving waters along the Ter River (Catalonia, NE Spain), Environ.Pollut. 153 (2008) 384–392.

19] D.I. Gustafson, Groundwater ubiquity index score: a simple method for assess-ing pesticide leachability, Environ. Toxicol. Chem. 8 (1989) 339–357.

20] K.T. Benlahcen, A. Chaoui, H. Budzinski, J. Bellocq, P.H. Garrigues, Distribu-tion and sources of polycyclic aromatic hydrocarbons in some Mediterranean

coastal sediments, Mar. Pollut. Bull. 34 (1997) 298–305.

21] P. Fernández, J.O. Grimalt, On the global distribution of persistent organic pol-lutants, Chimia 57 (2003) 514–521.

22] A. Hildebrandt, S. Lacorte, D. Barceló, Occurrence and fate of organochlorinatedpesticides and PAH in agricultural soils from the Ebro River basin, Arch. Environ.Contam. Toxicol. 57 (2009) 247–255.