Department of Chemistry, ICTM, University of Belgrade, Studentski trg 12-16, 11000 Belgrade, Serbia
H I G H L I G H T S
• Chemometric methods were used to analyze the quality of groundwater data sets.• High Mn indicates a strong susceptibility to fertilization on the land surface.• Samples with large amounts of metals may not be suitable for human consumption.• Banat region and a village at the Great Morava River chronically exposed to As.
Article history:Received 22 July 2013Received in revised form 3 September 2013Accepted 3 September 2013Available online xxxx
Editor: Damia Barcelo
Keywords:Ground-water quality of SerbiaCluster analysisFactor analysisDiscriminant analysisGroundwater pollution sources
Various chemometric techniqueswere used to analyze the quality of groundwaterdata sets. Seventeenwater qual-ity parameters: the cations Na, K, Ca, Mg, the anions Cl, SO4, NO3, HCO3 and nine trace elements Pb, As, Mn, Ni, Cu,Cd, Fe, Zn and Cr weremeasured at 66 different key sampling sites in ten representative areas (low land-NorthernAutonomous Province of Serbia, Vojvodina and central Serbia) for the summer period of 2009. HCA grouped thesample sites into four clusters based on the similarities of the characteristics of the groundwater quality. DAshowed two parameters, HCO3 and Zn, affording more than 90% correct assignments in the spatial analysis offour/three different regions in Serbia. Factor analysis was applied on the log-transformed data sets and allowedthe identification of a reduced number of factors with hydrochemical meaning. The results showed severe pollu-tion with Mn, As, NO3, Ni, Pb whereby anthropogenic origin of these contaminants was indicated. The pollutioncomes from both scattered point sources (industrial and urban effluent) and diffuse source agricultural activity.These samplesmaynot be suitable for human consumption; thewater quality belongs to class III/IV (contaminated).The Fe anomalies (7.1 mg/L) in thewater from the Vetrnica site can be attributed to natural sources, such as the dis-solution of rock masses and rock fragments. The serious groundwater contamination with As (25.7–137.8 μg/L) inthe area of Banat (Northern Autonomous Province of Serbia, Vojvodina) and a sample No. 9 at the Great MoravaRiver requires urgent attention.
Water is critical for sustainable development and is indispensable forhuman health and well being. The United Nations has proclaimed theyears of 2005–2015 as the international decade for action on “Waterfor Life” (UN, 2005). The global goal of ensuring that human beingshave access to acceptable quality water and sufficient quantity faces anumber of challenges in the years to come (Barth et al., 2009). To providesafe drinkingwater especially to rural populations, groundwater has beensought as the source inmany developing and under developed countries.Groundwater quality in a region is largely determined by both naturalprocesses (dissolution and precipitation of minerals, groundwater veloci-ty, quality of recharge waters and interaction with other types of wateraquifers) and anthropogenic activities (Andrade et al., 2008). The concen-trations of nitrate and sulfate increase notably as a result of the largeamount of chemical fertilizers used in agriculture (Compton and Boone,2000; Jiang et al., 2008). Various anthropogenic activities under theshadow of urbanization and the industrial development results in efflu-ent disposal, which introduced into the groundwater system highconcentrations of trace metal (Manzoor et al., 2006; Kouras et al.,2007; Papaioannou et al., 2010). A characteristic feature of most oftrace metals is their tendency to form hydrolyzed species in water anda few of them form complex species by combining with inorganic an-ions such as HCO3
−, SO42−, Cl−, and NO3
− (HAsO42−,CdHCO3
+, Pb(OH)+,CdCl3−, Cr(OH)4−). Many of these metals are considered essential forhuman health (Midrar-Ul-Haq et al., 2005), but human consumptionof poor quality water can lead to different kinds of health problems(Dissanayake and Chandrajith, 2010), while the use of poor qualitywater for irrigation reduces crop productivity (Azizullah et al., 2011).According to the WHO organization, about 80% of all the diseases inhuman beings are caused by water.
In the last decades, methods such as hierarchical cluster analysis(HCA), discriminant analysis (DA), factor analysis (FA) have becomeaccepted in identifying variations and sources of groundwater pollution(Reghunath et al., 2002; Manzoor et al., 2006; Mouron et al., 2006;Andrade et al., 2008; Yidana et al., 2008; Papaioannou et al., 2010;Zhang et al., 2012). The Box–Whisker plot is a tool that helps in the iden-tification of the basic statistical features of a large data set through visualdisplays (Bhattacharjee et al., 2005). It is employed to maximize insightinto a data set, uncover underlying structures and detect outliers andanomalies (Statgraphics, 2006).
This study presents necessity and usefulness of multivariate statisti-cal techniques for evaluation and interpretation of large complex datasets with a view to get better information about the water quality anddesign of monitoring network for effective management of waterresources in Serbia. Although the groundwater participates with 70%in water supply of the population, most groundwaters do not have ade-quate monitoring and ecological protection. Besides, existing data havenot been systematized properly and are not available to all participants.
The purpose of this study was to evaluate the importance of waterquality variables (major ions — Na, K, Ca, Mg, Cl, SO4, NO3, HCO3 andtrace elements Pb, As, Mn, Ni, Cu, Cd, Fe, Zn, Cr), throughout the applica-tion of multivariate technique and to distinguish the influence of theimpact of natural processes and anthropogenic input on the composi-tion of groundwater in Serbia. Application of standard methods, carefulstandardization, procedural blankmeasurements, and duplicate samplesreflect themethodologic quality and results obtained in research. There-fore, in this paperwas applied the standardmethods for determining the
parameters APHA et al. (1998) andUS EPA standardmethods (1992) andthere was provided quality of the analytical data.
2. Materials and methods
2.1. Studied sites
Groundwater is the traditional water supply resource in Serbia.Serbia is relatively rich in groundwater reserves. They are however,unevenly distributed across the territory (Komatina, 1998; Nikic et al.,2007; Stojadinovic et al., 2007). The major groundwater reserves areaccumulated in thick Quaternary and Neogene intergranular aquifers(Northern Autonomous Province of Serbia, Vojvodina). This provinceis part of a large flat depression of the Pannonian Basin. Alluvial aquifersof large rivers with variable-thickness layers of 20–60 m (Danube, Sava,Morava and Drina) are particularly important and widely used fordrinking water supply. Karstic aquifers dominate the south-westernand eastern regions of Serbia. Water deficiencies are found in thesouth of Serbia, as well as in a central region of Serbia, Šumadija.
Agriculture has long been themainstay of Serbia's economy. Croplandoccupies nearly two-thirds of the territory of Serbia. Although Serbia isone of the largest food producers on the Balkans, only some 1–2% of itsarable land is irrigated. The principal area of commercial agriculture isthe Vojvodina region and adjacent lowlands south of the Sava andDanube Rivers, including the valley of the Morava River. Almost all themain towns of Serbia that are located in the Basins of the Danube, Sava,Morava, and Drina Rivers utilize water from alluvial aquifers. The under-groundwaters in Serbia are used in the water supply for the human pop-ulation, the water supply of industry, balneotherapy, heating systems(urban areas, industry and agricultural areas with greenhouses) andirrigation (Komatina, 1998; Nikic et al., 2007; Stojadinovic et al., 2007).The total capacity of the existing ground water sources in Serbiaamounted to a total of about 678 million m3 per year or 21.5 m3/s, ofwhich 6.5 m3/s is for Vojvodina and 15 m3/s for Central Serbia(Statistical Yearbook of Serbia, 2006). Groundwater samples werecollected from a variety of aquifers in Serbia (Fig. 1).
2.2. Sampling
The national monitoring network and sampling strategy weredesigned to cover a wide range of determinants at key sites, whichreasonably represent the water quality of the groundwater system.Under the water-quality monitoring program, samples were collectedin summer seasons (between 11th and 14th day) of 2009. In total,66water samples were taken at 10sampling sites (Fig. 1). Groundwatersamples were collected in high-density PVC bottles from an under-ground water aquifer depth range of about 4 to 16 m. All sampleswere collected between 08:00 and 16:00 h local time. The water fromthe wells is used for drinking and for irrigation. The last three wells(8-Backa; 9-Banat and 10-Srem) are located in the northern part ofthe Serbia–Vojvodina region (Fig. 1). The groundwater levels in thewells were recorded, using a water level recorder. Samples werecollected into plastic bottles, in minimally anthropogenically impacted.Before collection of the water samples, the bottles were thoroughlycleaned by rinsingwith HNO3 and deionizedwater. The bottled sampleswere immediately transported to the laboratory and kept in an ice-boxat a temperature of 4 º C for subsequent analysis.
Fig. 1. Location of the study area: The main areas in Serbia: 1 — The Great Morava River, 2— The West Morava River, 3 — The South Morava River, 4 — The Veternica, 5— The Kolubara,6 — The Macva, 7 — The Podunavlje, 8 — Backa, 9 — Banat, 10 — Srem.
935G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
In the laboratory, the samples were filtered throughWhatman filterpaper No.42. The basic parameters (unstable) of the water qualitytemperature, and pH were determined using pH meter, on site. Themain sampling material for the measurements was the groundwatersampled and analyzed according to APHA et al. (1998), and US EPAstandardmethods (1992). All concentrations of the chemical parametersare expressed in mg/L or μg/L (Table 2), except for pH (units).
Replicate samples were collected immediately after the routinesamples in the field using the same collection method and equipment.All difference measured in the concentrations between the replicatepairs were well within the precision of the method for all measuredparameters. Analysis of blank samples did not show any inherent biasin themethod of analysis for the elements andmajor ions. The accuracyof methods was checked using standard solutions. The analytical errorswere in the range of 5–10% (Domenico and Schwartz, 1990).
2.3. Chemometric methods
Chemometric methods, such as hierarchical cluster analysis (HCA),discriminant analysis (DA), and factor analysis (FA, using principalcomponent analysis, PCA) were used (Singh et al., 2004, 2005).
HCA is an unsupervised pattern detection method that partitionsall cases into smaller groups or clusters of relatively similar cases thatare dissimilar to the other groups. Q-modeHCA was applied on thestandardized log-transformed data set to classify the data into spatialassociations. Squared Euclidean distances measures were chosen tomeasure similarity/dissimilarity among the variables while the Ward'slinkage method was chosen to link initial clusters resulting from theinitial clustering steps. The combined use of squared Euclidean distancesas a similarity/dissimilarity measure and theWard's method as a linkagealgorithm was observed to produce very reliable clusters in HCA.
DA is a method of analyzing dependence that is a special case ofcanonical correlation used to analyze dependence. One of its objectives
is to discriminate between two or more groups in terms of the means ofthe discriminating variables. DA is performed on log-transformed datawithout affecting the results and comparability with other chemometricmethods and constructs a discriminant function for each group as follows:
nf Gið Þ ¼ ki þ
Xwijpij
j ¼ 1
where i is the number of groups (G); ki is a constant inherent to eachgroup; n is the number of parameters used to classify a set of data into agiven group; and wij is the weight coefficient assigned by DA to a givenparameter (pj). DA was performed on the data set based on two differentmodes, i.e., standard and stepwise to construct the best discriminantfunctions (DFs) to confirm the clusters determined by means of HCAand evaluate the spatial variations in the groundwater quality. The higheris the DFs value the more important is the variable. The monitoring sites(spatial) were the grouping variable, and the measured parameterswere the independent variables.
PCA, as one method of extracting the eigenvectors, extractseigenvalues and eigenvectors from the covariance matrix of originalvariables to produce new orthogonal variables known as varifactors(VFs) through varimax rotation, which are linear combinations of theoriginal variables. VFs provide information on the most meaningfulparameters that describe the whole data set, allowing data reductionwith minimum loss of original information.
3. Results
The samples were analyzed for 17 parameters, which includedbicarbonate (HCO3
936 G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
(Zn), cadmium (Cd), lead (Pb), chromium (Cr) and arsenic (As). All thewater quality parameters are expressed in mg/L (major cations, anions,Fe and Mn) or μg/L (trace elements). Mean pH values varied between6.9 and 7.4. There was a relatively uniform temperature regime inmost groundwater samples (12.6 to 15.3 °C).
The quality of the analytical data, was ensured through carefulstandardization, procedural blank measurements, and duplicate sam-ples. The laboratory also participates in a regular national programon analytical quality control. The basic statistics of the data set on thewater quality of the aquifers are summarized in Table 1. The statisticalsummaries of the parameters analyzed using the Box and Whiskerplots in Fig. 2. The mean values of the World Health Organization(WHO, 2006), European Parliament and Council of the EuropeanUnion, 2006 and Water Act and Regulations on the Monitoring ofWater Quality introduced by the Government of the Republic of Serbia(Water Act) for water are presented in Table 2. Arsenic (μg/L) versusmanganese (mg/L) concentrations in the groundwater of Serbia isshown in Fig. 3. The results obtained using the Q-HCA mode are
shown in Fig. 4. The DA data of the investigated groundwaters arepresented in Tables 3 and 4, and Fig. 5, while the results of the correla-tions and PCA among the investigated groundwaters are presented inTable 5.
4. Discussion
4.1. Statistical Screening of water quality data
Cations including sodium (Na+), potassium (K+), calcium (Ca+2),magnesium (Mg+2) and the anions such as nitrates (NO3
−), bicarbonates(HCO3
−), sulfates (SO4−2), and chlorides (Cl−) naturally occur in water
and are usually determined in water quality evaluation tests. Waterquality parameters reflect the level of contamination of the waterresources. The statistical summaries of the parameters analyzed usingthe Box and Whisker plots are shown in Fig. 2, while Table 1 presentsthe data for the major hydrochemical parameters analyzed in thisstudy. The chemical quality of groundwater varied drastically among
Fig. 2. Box andWhisker plot showing the variation of chemical constituents in the studiedgroundwater samples of Serbia.
937G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
the different sampling sites. Bicarbonate and zinc showed the highestvariance in the distribution of their concentrations-large spatial varia-tions. Fig. 2 also shows that the order of the relative abundance ofthe major cations in groundwater was Ca N Mg N Na N K (mg/L basis)while that of the anions was HCO3 N SO4 N Cl N NO3. It is obviousthat the major chemical parameters displayed wide differences in theirvariance. However, these ions may in higher concentrations render thewater unfit for living organisms, humans.
The sodium values are varied between 2.6 mg/L and 295 mg/L(Table 1). The highNa concentrations in themajority of samples suggesta strong water–aquifer interaction related to direct cation exchangebetween the groundwater and the clay fraction of the aquifer materialand/or pollution from wastewater, septic tanks (Wayland et al., 2003).Like sodium, potassium (K+) was also usually found above the qualitystandards. The WHO recommended standard is 12 mg/L (Table 2).Analyzing the data of different sites, the average level of K+ was
Table 2Range in values of water quality variables in waters and WHO (2006), EC (2006) andSerbian Standard for drinking water (unit: Fe andMn inmg/L and trace elements in μg/L).
Parameters Range Risk baseddrinking watercriteria (WHO)
Data in bold are concentrations in the samples that exceeded the WHO and EC Directive2006/118/EC recommended values for drinking water.
documented at 4.4 mg/L but its concentration in some samples reached38.7 mg/L above the standard of 12 mg/L. Potassium is a nutrientwhicheither occurs naturally in local groundwater, or can arise from man-made sources. Agricultural fertilizers are the most common source ofnutrients in local groundwaters. Other sources include animal andhuman waste. High concentrations of Mg were found in the analyzedsamples (110 mg/L; Table 1). Magnesium in groundwater often comesfrom minerals and/or from pollution sources-contaminated due to theapplication of chemical fertilizers or domestic effluents (Mohamedet al., 2003; Jiang et al., 2009).
The groundwater samples contained significantly high contents ofHCO3 (the average concentrations was 497 mg/L, while the maximumcontent was 988 mg/L) in most region, while sulfates and chloridesdid not exceed the standard limits in the examined samples (Tables 1and 2). The high content of bicarbonate in the Serbian groundwater isclosely related to natural dissolution of soil and rock (Dimitrijevic,1995). Among the nitrogenous anion compounds, nitrates are impor-tant from the health point of view (Lalehzari et al., 2013). There weregreat variations in the nitrate level in the groundwater of Serbia. TheNO3 concentration ranged between 0.02 and 42.45 mg/L in differentsites for individual samples (Table 1). Thus, the results indicated thatthe groundwater of the Great, West and South Morava Rivers andsome part of the Macva region of Serbia are severely polluted dueto anthropogenic activities. Considering the health effect of NO3, theWHO andWater Act have set a guideline for the maximum permissiblelevels in drinking water (Tables 1 and 2). The pollution comes probablyfrom diffuse sources-agricultural activity. A trend in the spatial varia-tions in the groundwater nitrate levelwas observed among the differentSerbian study sites, possibly due to the difference in fertilizer applica-tion rate, crop rotation, irrigation, soil texture and local pedo-climaticvariability. The influence of agricultural activities on the water qualityis difficult to quantify due primarily the non-standardized access toagricultural production (crops, the amount and type of fertilizersand pesticides, cultural practices, and cultivation of system) and theheterogeneity of the conditions at the actual locations (soil properties(e.g. ability to retain water), hydrological conditions (rainfall) andslope). About 85% of the land under crops is privately owned.
The concentrations of Ca, Cl, SO4, Zn, Cu, Cd, and Cr in the sampleswere below the WHO and Serbian Water Act recommended limits.The evaluation of metal–metal correlation coefficients (significant atthe 0.01 level) pertaining to the data for the samples showed that theconcentrations of Ni and Pb mutually depended on each other andwith Cd, possibly because of the fact that they originated from a singlesource. The heavymetals Ni and Pb are known to have a number of neg-ative impacts on human health, such as DNA damage, cancer and dam-age of the central nervous system (Stohs and Bargchi, 1995). As the Niand Pb levels exceeded the WHO guidelines in only two samples(Krusevac, at the West Morava River and the village sample at theGreat Morava River), these heavy metals should actually not be consid-ered as having a high impact on the disease burden of people living inSerbia, except at the mentioned locations.
The concentrations of As, Mn and Fe in groundwater were not signif-icantly correlated (p N 0.05), suggesting that the contamination sourceof As, Mn, Fe was different in the wells. The well depth and pH of thegroundwater (6.9–7.4) were not related to the As concentration in thegroundwater (data not shown).
Manganese concentrations above the WHO guideline, Table 2, werepresent in 11% of the samples: hence, manganese has to be consideredas the second most important groundwater contaminant at site-1 TheGreat Morava River, site-3 The South Morava River and sites 5 and 6,the Kolubara and Macva region of Serbia. Regions of uncontaminatedwells are only present in the Banat and Srem in the Vojvodina Province,the central part of Serbia, Podunavlje and along theWest Morava River.The obtained results revealed that 23% of the studied wells are contam-inated with As and/or Mn. Many groundwater wells have low As levelsbut highMn levels because arsenic is lessmobilized under Mn reducing
Fig. 3. Arsenic (μg/L) versus manganese (mg/L) concentration in groundwater of Serbia.
938 G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
conditions (Fig. 3). Other samples showed the opposite relation: lowMn and high As concentrations or low Mn and low As concentrations.A combination of high arsenic and/or manganese was also reported ina regional study in Bangladesh and in The Mekong Delta (Cheng et al.,2004; Buschmann et al., 2008), where only 11%; i.e. 37% and 16%i.e. 50% of 629/352 samples met the WHO guidelines for arsenic andmanganese, respectively. High arsenic concentrations in groundwaterare commonly correlated with high HCO3 concentrations (Anawaret al., 2002). The increased HCO3
− concentrations are usually associatedwith reducing conditions, under which arsenic takes the form of arse-nite, which is less strongly sorbed than arsenate at pH 7 and for concen-trations b1 μM As (Appelo et al., 2002). The concentration of arsenic inmost samples from Banat (25.7–137.8 μg/L) and sample at the GreatMorava River (21.7 μg/L) falls outside of the WHO recommended max-imum of 10 μg/L in drinkingwater. This suggests that a large proportionof Serbian population from the Banat region and village at GreatMoravais chronically exposed to As.
Quaternary sedimentary aquifers within the Pannonian Basin con-tain high concentrations of naturally occurring arsenic. In the water–sediment system, arsenic is specially enriched in clay fraction of thesoil (10 mg As/kg) (Jovanovic et al., 2011). Although arsenic is aknown water pollutant in Vojvodina for many decades, public watersupply systems lack technical and financial resources for its removal.Knowing that arsenic causes skin, bladder, and internal cancers, its pres-ence in water represents a major public health topic. Other sources ofarsenic include some human activities, such as the use of arsenicalpesticides, burning of fossil fuels. Considering predominant agriculturalactivity in Vojvodina, arsenic exposure frompesticides andoccupationalexposure will be assessed as further important risk factors.
In the nearest future, Serbia will implement many European UnionDirectives regarding water quality and will have to find means to treatthis problem (Jovanovic et al., 2011).
The most common sources of iron and manganese in groundwaterare naturally occurring, for example from the weathering of iron-and manganese-bearing minerals and rocks. Industrial effluent, sewageand landfill leachates may also contribute iron and manganese to localgroundwater (CCME, 1995; Rosen, 2001). The groundwater froma loca-tion at the Veternica site contains a maximum iron concentration of7.1 mg/L, but this area is characterized by the absence of industrialactivities and the Fe anomalies can be attributed to natural sources,such as the dissolution of rock masses and rock fragments.
The highest pollution by tracemetals Pb (15.8 μg/L), Ni (41.11 μg/L),and cadmium (0.8 μg/L) was determined in Krusevac City. Krusevac
belong to theWestMorava River Basin, which is exposed to large uncon-trolled metal inputs from anthropogenic sources (Sakan et al., 2011).
The results of the chemical analyses (Table 1) showed that thegroundwater in Serbia at the other tested locations is generally of achemical quality suitable for domestic or agricultural use. The waterquality is of class II.
Mn, Fe, As and nitrates are the key factors impairing the quality ofthe groundwater of Serbia. The water quality belongs to class III/IV(contaminated).
4.2. Spatial similarities and grouping of sample sites (cluster analysis anddiscriminant analysis)
The initial exploratory approach involved the use of hierarchicalCluster Analysis (HCA) on the log transformed data set of the 66 sam-pling sites. Spatial HCA identified similar monitoring sites, and, in thiscase, produced a dendrogram grouping the 66 sampling sites into fourmain groups. The classifications of the studied sampling sites varied,because the sites in these groups had similar features and natural back-ground, and were affected by more or less similar sources. Cluster 1(Sites 2,6 9–10), cluster 2 (Sites 5, 7), cluster 3 (Sites 1, 8) and cluster4 (Sites 3–4) correspond to a relatively low pollution, moderate pollu-tion regions and very high pollution regions, respectively. An observa-tion of the dendrogram reveals similarities among the four clusters atprogressively higher linkage distances. Clusters 1 and 2 are linked at ashorter distance and are together linked to Cluster 3 and especiallywith Cluster 4 at a higher distance. Cluster 4 is the most dissimilaramong all four clusters as it is connected to the other three clusters ata high linkage distances (Fig. 4).
The water analysis results indicated that most samples classified asCluster 1 were contaminated by Mn (1.04–2.03 mg/L), indicating astrong susceptibility to fertilization of the land surface. The concentra-tions of the other chemical parameters were within the acceptablelimits for drinking water. Cluster 2 was associated with the highestaverage NO3 and Ni concentrations. Clusters 3 and 4 were associatedwith the highest average concentrations of Na, K, Mg, HCO3, Fe, Zn, Cr,Cd and As. These groups contained the groundwater samples with theworst hydrochemical quality. Groups 1 and 2 occupied most of thestudy area. The other groups occupied smaller areas along the SouthandWestMorava Rivers (group 4), the Veternica site and the northeast-ern part of Serbia (Banat) or the center of the study area.
Spatial variations in groundwater quality were evaluated usingDiscriminant Analysis (DA), with the clusters based on the spatial
Data in bold are concentrations in the samples that exceeded the WHO and EC Directive2006/118/EC recommended values for drinking water.
Fig. 4. The dendrogram obtained applying the Q mode hierarchical cluster analysis ingroundwater of Serbia.
939G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
HCA. The objective of the DAwas to test the significance of discriminantfunctions and determine the most significant variables associated withthe differences among the HCA-groups. The spatial DA was performed,using the log-transformed data set of 17 parameters, after classificationinto four major HCA-groups obtained from the spatial HCA. The siteswere the dependent variables and the measured parameters were theindependent variables. Wilks' lambda and the Chi-square for eachdiscriminant function varied from 0.143 to 0.704 and from 19.11 to208.5, respectively, at p b 0.001, suggesting that the spatial DA wascredible and effective. The discriminant functions (DFs) and classifica-tion matrices (CMs) obtained from the standard and stepwise modesof the DA are shown in Tables 3 and 4, respectively. In the stepwisemode, the variables were added step by step, beginning with the mostsignificant, until no significant changes were obtained; the standardDAmode constructed DFs containing all parameters (Table 3). The stan-dard mode DA, using 17 discriminant variables yielded CMs correctlyassigning 97.07% of the cases. However, in the stepwise mode, the DAproduced CMswith 90.09% correct assignments using only two discrim-inant parameters (Table 4). These resultswere similar to those obtainedusing the standard mode, but with significantly fewer parameters(only two). Thus, the spatial DA suggests that HCO3 and Zn (results asBox and Whisker plots) were the most significant parameters for dis-criminating between the four C-groups and accounted for most of theexpected spatial variation in the groundwater quality. Based on theabove results, HCA provided a classification of groundwater qualitythat aided in designing an optimal spatialmonitoring planwith a sharp-ly reduced number of monitoring sites and corresponding costs.
Fig. 5. Plot of discriminant functions showing four (a) and three (b) clusters with error bars.
940 G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
Stepwise DA was proven to be a useful tool in recognizing the discrim-inant parameters in spatial variations of groundwater quality. It isessential to strengthen the monitoring accuracy of HCO3
− and Zn toclearly identify variations in the future. The plot between the first two
Table 5Principal component loadings of 17 chemical variables in the groundwater samples of Serbia.(PCA/FA for complete data set; n = 66).
discriminant factors (Fig. 5a) resulted in the four groups. This indicatedtowards small differences in the water qualities in the C-1 and C-2groups (new C1*) and suggested that the data set could be dividedinto three groups. Application of DA on the new data set comprised ofthree groups, resulted into two discriminant functions; one withp b 0.001, Wilks' lambda of 0.075 and the other function with 0.353.The total variance contained in these discriminant functions was 95.5%.Fig. 5b shows clearly the separated three clusters in C-1*, C-2 and C-3.
4.3. Factor analysis
The PCA/FA results comprising the loadings, eigenvalues and vari-ance are summarized in Table 5. The six factors explain 71.54% of thetotal variance for the log-transformed data. The parameters with load-ings whose absolute value is more than 0.65 are considered significant.This means that the six-factor model explains the variability of almostall variables and can be used to indicate the dominant hydrochemicalprocesses controlling the groundwater composition, without losing anysignificant characteristics.
Factor, F1, which explained 17.89% of the total variance had strongpositive loadings on Ca, Cl and SO4, which could be interpreted as themineral component of the groundwater (natural sources). The nextfactor, F2, which explained 17.3% of the total variance has strong positiveloading on NO3, Pb and Ni. Factor F2, represents ‘anthropogenic-toxic’pollution sources from industrial effluents and agricultural activities. Itis clear that the samples of Krusevac City, has the largest values of factorF2. Group C2 possesses scores in factor F2, but low scores in other factors.
941G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
Factor, F3 has high loadings for Na, As and HCO3, and represents 11% ofthe total variance. Significant positive correlation were found betweenAs, Na and HCO3 in the groundwater (p b 0.01), reflecting that theseions had a similar origin. Factor 3 values were mainly observed in thearea of the Morava River and in the Autonomous Province of Serbia,Vojvodina-Banat, which is characterized by cultivated areas andintensive agricultural practices. Quaternary sedimentary aquiferswithin the Pannonian Basin contain high concentrations of naturallyoccurring arsenic (Jovanovic et al., 2011). Contamination can also becaused by anthropogenic activities such as the application of agricul-tural chemicals (pesticide) (Shomar et al., 2005). This suggests thatF3 is associated with natural processes and/or agricultural inputs tothe groundwater. Groups 3 and 4 had similar scores for factor F3. Fac-tor F4,representing 9.2% of the total variance, records high loadingsfor Fe and Zn, and was mainly associated with natural mineral pollu-tion including soil weathering (Simeonov et al., 2003; Singh et al.,2005; Papaioannou et al., 2010). Only the sample from location No.32, the Veternica site, exceeded the maximum values of Fe(7.1 mg/L) and also had high contents of zinc (757 μg/L). However,the values of the Zn concentrations were much lower than the per-missible limit of 5 mg/L. The fifth PC was highly loaded with Mn(8.6% of the total variance) and finally the sixth factor was correlatedwith Cr (7.5% of the total variance) and is likely to represent “soilleaching” processes (Simeonov et al., 2003). Group 1 possessedscores in factor F5. Manganese showed a strong correlation withSO4, and a weak correlation with NO3, while chromium does not cor-related with any of the other water quality parameter. F5 and F6 areconsidered as representing anthropogenic pollution.
The results from spatial FA suggested that most of the variationsis explained by the set of soluble salts (natural) and anthropogenicpollutants.
5. Conclusions
Chemometric methods were successively applied to evaluate thespatial variations in groundwater quality and source identifications atsites in Serbia indicating that the different methods are effective andharmonious with each other.
The results suggested that most of the variations are explained bythe set of soluble salts (natural) and anthropogenic pollutants. The pol-lution comes from both scattered point sources (industrial and urbaneffluent) and diffuse sources-agricultural activity.
The measured contents of metals in the groundwaters of Serbia indi-cate the presence of severe pollution. With 35% of the studied wellsbeing contaminated by Mn, As and NO3 this is an alarming result, andthey are the key factors impairing the water quality of the groundwaterof Serbia. The arsenic concentrations in most samples in Banat (25.7–137.8 μg/L) and sample at the Great Morava River (21.7 μg/L) exceededof the WHO recommended value of 10 μg/L for drinking water. This sug-gests that a large proportion of the Serbian population in the Banat regionand a village on theGreatMorava River are chronically exposed to As. Theregionwith the highest tracemetal pollution Pb (15.8 μg/L), Ni (41.11 μg/L), and cadmium (0.8 μg/L) is located in Krusevac City, which is exposedto large uncontrolledmetal inputs from anthropogenic sources. The sam-ples containing large amounts of As,Mn, NO3, Fe, Pb, Ni, fall thus into classIII/IV. According to the SerbianWater Act regulations, class III/IV water isof unacceptable quality and requires highest urgency sanitation. The con-centrations of Ca, Cl, SO4, Zn, Cu, Cr and Cd did not exceed the guidelinesfor safe water. The results of the chemical analyses showed that thegroundwater is generally of chemical quality suitable for domestic or ag-ricultural use, at most locations tested; water quality of class II.
Acknowledgments
This research was supported by the Ministry of Science andTechnological Development of the Republic of Serbia, Grant Nos.
172001 and 43007. The authors are grateful to the RepublicHydrometerological Service of Serbia, Belgrade, for providing thesamples.
References
Anawar HM, Akaib J, Mostofac KMG, Safiullahd S, Tareq SM. Arsenic poisoning ingroundwater. Health risk and geochemical sources in Bangladesh. Environ Int 2002;27:597–604.
Andrade E, Palacio HAQ, Souza IH, Leao RA, Guerreiro MJ. Land use effects in groundwatercomposition of an alluvial aquifer (Trussu River, Brazil) by multivariate techniques.Environ Res 2008;106:170–7.
APHA, AWWA, WEF. Standard Methods for the Examination of Water and Wastewater.18th ed. Washington DC: American Public Health Association; 1998.
Appelo CAJ, Van der Weiden MJJ, Tournassat C, Charlet L. Surface complexation of ferrousiron and carbonate on ferrihydrite and themobilization of arsenic. Environ Sci Technol2002;36:3096–103.
Azizullah A, Khattak MNK, Richter P, Hader D-P. Water pollution in Pakistan and itsimpact on public health — a review. Environ Int 2011;37:479–97.
Barth JAC, Grathwohl P, Fowler HJ, Bellin A, GerzabekMH, Lair GJ, et al. Mobility, turnoverand storage of pollutants in soils, sediments and waters: achievements and results ofthe EU project AquaTerra. A review. Agron Sustain Dev 2009;29:161–73.
Bhattacharjee S, Chakravarty S, Maity S, Dureja V, Gupta KK. Metal contents in thegroundwater of Sahebgunj district, Jharkhand, India, with special reference to arsenic.Chemosphere 2005;58:1203–17.
Buschmann J, Berg M, Stengel C, Winkel L, Sampson M, Trang PTK, et al.Contaminantion of drinking water resources in the Mekong delta foodplains:arsenic and other trace metals pose serious health risks to population. EnvironInt 2008;34:756–64.
CCME (Canadian Council of Ministers of the Environment). Protocol for the derivationof Canadian sediment quality guidelines for the protection of aquatic life.CCMEEPC-98E. Prepared by Environment Canada, Guideline Division, Technical Secretariatof the CCME Task Group on Water Quality Guidelines, Ottawa; 1999 [Reprinted inCanadian environmental quality guidelines, Chapter 6, Canadian Council of Ministersof the Environment].
Cheng Z, Zheng Y, Mortlock R, van Geen A. Rapid multi-element analysis of groundwaterby high-resolution inductively coupled plasma mass spectrometry. Anal BioanalChem 2004;379:512–8.
Compton JE, Boone RD. Long-term impacts of agriculture on soil carbon and nitrogen inNew England. For Ecol 2000;81:2314–30.
Dimitrijevic M. Geology of Yugoslavia. Belgrade: Geoinstitute; 1995 [in Serbian].Dissanayake CB, Chandrajith R. Introduction to medical geology. New York, LLC:
Springer-Verlag; 2010.Domenico PA, Schwartz FW. Physical and chemical hydrogeology. New York: John Wiley
& Sons; 1990 [824 pp.].European Commission (EC). Directive 2006/118/EC of the European Parliament and the
Council of 12th of December 2006 on the protection of groundwater against pollutionand deterioration. Off J Eur Union 2006. [L 372/19. 27/12]. vol. XLIII, No. 4.
Jiang Y, Yuan D, Zhang C, Zhang G, He R. Impact of land use change on groundwater qualityin a typical karst watershed of southwest China. Hydrogeol J 2008;16:727–35.
Jiang Y, Wu Y, Groves C, Yuan D, Kambeis P. Natural and anthropogenic factors affectingthe groundwater quality in the Nandongkarstunderground river system in Yunan,China. J Contam Hydrol 2009;109:49–61.
Jovanovic D, Jakovljevic B, Rasic-Milutinovic Z, Paunovic K, Pekovic G, Knezevic T. Arsenicoccurrence in drinking water supply systems in ten municipalities in VojvodinaRegion, Serbia. Environ Res 2011;111:315–8.
Komatina M. Groundwater in rocks of intergranular porosity on the territory ofYugoslavia. Collection of Works: Ground Waters of Yugoslavia—Invisible Resource,Novi Sad; 1998.
Kouras A, Katsoyiannis I, Voutsa D. Distribution of arsenic in groundwater in the area ofChalkidiki, Northern Greece. J Hazard Mater 2007;147:890–9.
Lalehzari R, Tabatabaei SH, Kholghi M. Simulation of nitrate transport and wastewaterseepage in groundwater flow system. Int J Environ Sci Technol 2013. http://dx.doi.org/10.1007/s13762-013-0213-4.
Manzoor S, Shah Munir H, Shaheen N, Khalique A, Jaffar M. Multivariate analysis of tracemetals in textile effluents in relation to soil and groundwater. J Hazard Mater2006;137:31–7.
Midrar-ul-Haq, Khattak RA, Puno HK, Saif MS, Memon KS, Sial NB. Heavy metalsaccumulation in potentially contaminated soils of NWFP. Asian J Plant Sci2005;4:159–63.
Mohamed AA, Mohamed MA, Terao H, Suzuki R, Babiker IS, Ohta K, et al. Naturaldenitrification in the Kakamigahara groundwater basin, Gifu perfecture, centralJapan. Sci Total Environ 2003;307:191–201.
Mouron P, Nemecek T, Scholz R, Weber O. Management influence on environmentalimpacts in an apple production system on Swiss fruit farms: combining life cycleassessment with statistical risk assessment. Agric Ecosyst Environ 2006;114:311–22.
Nikić Z, NadeždićM,Nikolić I. National network of groundwater quality observation in theVelika Morava Alluvium. Regional IWA Conference on Groundwater in Management inthe Danube River Basin and Other Large River Basinis.Proceedings. Belgrade: JaroslavČerni Institute for Development of Water Resources; 2007. p. 485–91.
Papaioannou A, Dovrili E, Rigas N, Plageras P, Rigas I, Kokkora M, et al. Assessment andmodelling of groundwater quality data by environmetric methods in the context ofpublic health. Water Resour Manag 2010;24:3257–78.
942 G. Devic et al. / Science of the Total Environment 468–469 (2014) 933–942
Reghunath R, Murthy TRS, Raghavan BR. The utility of multivariate statistical techniquesin hydrogeochemical studies: an example from Karnataka, India. Water Res 2002;36:2437–42.
Rosen MR. Hydrochemistry of New Zealand's aquifers. In: Rosen MR, White PA, editors.Groundwaters of New Zealand. Wellington: New Zealand hydrological Society Inc.;2001. p. 77–110.
Sakan S, Djordjevic D, Devic G, Relic D, Andjelkovic I, Djuricic J. A study of trace elementcontamination in river sediments in Serbia using microwave-assisted aqua regiadigestion and multivariate statistical analysis. Microchem J 2011;99(2):492–502.
Shomar BH, Muller G, Yahya A. Geochemical features of topsoils in the Gaza Strip: naturaloccurrence and anthropogenic inputs. Environ Res 2005;98:372–82.
Simeonov V, Stratisb JA, Samarac C, Zachariadisb G, Voutsac D, Anthemidis A, et al. Assess-ment of the surface water quality in Northern Greece. Water Res 2003;37:4119–24.
Singh KP, Malik A, Mohan D, Sinha S. Multivariate statistical techniques fort he evaluationof spatial and temporal variations in water quality of Gomti River (India)—a casestudy. Water Res 2004;38:3980–92.
Singh KP, Malik A, Sinha S. Water quality assessment and apportionment of pollutionsources of Gomti river (India) using multivariate statistical techniques—a casestudy. Anal Chim Acta 2005;538:355–74.
Statgraphics. Statgraphics Centurion XV. http://www.statgraphics.com, 2006.
Statistical Yearbook of Serbia. Belgrade: Statistical Office of the Republic of Serbia; 2006.Stohs SJ, Bagchi D. Oxidative mechanisms in the toxicity of metal-ions. Free Radic Biol
Med 1995;18:321–36.Stojadinović D, Stojadinović V, Nikić Z. Alluvial springs in the Great Morava valley and
their sustainable development. Ecologica 2007;14:13–8.United Nations, UN Millennium Project. Investing in development: a practical plan to
achieve the millennium development goals. New York: United Nations; 2005.US-EPA. Methods for analysis of water and wastes; 1992.Water Act (Official Gazette of the Republic of Serbia nos. 46/91, 53/93, 67/93, and 48/94)
and the Regulations on theMonitoring ofWater Quality introduced by theGovernmentof the Republic of Serbia.
Wayland K, Long D, Hyndmann D, Pijanowski B, Woodhams S, Haack K. Identifyingrelationships between baseflow geochemistry and land use with synoptic samplingand R-Mode factor analysis. J Environ Qual 2003;32:180–90.
WHO Guidelines for Drinking-Water Quality. Volume 1 2000 — recommendations. 3rded. Geneva: Word Health Organization; 2006.
Yidana SM, Ophori D, Yakubo BB. A multivariate statistical analysis of surface waterchemistry data — the Ankobra Basin, Ghana. J Environ Manag 2008;86:80–7.
Zhang B, Song X, Zhang Y, Han D, Tang C, Yu Y, et al. Hydrochemical characteristicsand water quality assessment of surface water and groundwater in Songnen plain,Northeast China. Water Res 2012;46(8):2737–47.
