In this study, analysis of variance (ANOVA), cluster analysis (CA) and principal component analysis (PCA) were employed in order to evaluate the concentration profile of organic contaminants found in three main river from central Transylvania, Romania. Samples were collected from nine sampling stations, in two different sampling campaigns (wet season and dry season). Water samples were extracted using solid-phase extraction and analyzed using gas chromatography coupled with mass spectrometry (GC/MS). Twelve organic pollutants belonging to different classes were used for further interpretations. ANOVA highlighted compounds which distinguished Olt River from Mures River, and compounds that are influenced by increased river flow from the wet season. CA was applied to group the sampling stations. Three clusters were obtained, according to their organic load. PCA extracted five principal components explaining 87.330% from data set variability. Based on these results, a future monitoring study may be optimized by reducing the sampling points and compounds to those that are representative for each river, thereby reducing costs, without any information loss.

INTRODUCTION

Surface water pollution is an important issue of the natural environment that our society needs to manage nowadays. The main sources of water pollution are represented by anthropogenic activities (industrial, agricultural or domestic) or natural processes (precipitation, erosion or weathering of crustal materials); all these contribute to the change of water natural status, by direct discharges into the fresh water of synthetic contaminants. Rivers play a major role in translocation of municipal, industrial wastewater and agricultural run-off. Municipal and industrial wastewater discharges constitute constant pollution sources, whereas surface runoff might be seasonal phenomenon, largely affected by the weathering process. Seasonal variations in precipitations, surface runoff, interflow and groundwater flow and pumped in/out flows have a strong effect on river discharge and, subsequently, on the concentration of pollutants in river water.

Some of these pollutants are highly toxic for aquatic organisms and human health, even if they are at trace level concentration (ng L−1 or μg L−1). These compounds require monitoring studies to evaluate their chronic impact. It is mandatory to prevent and control river pollution and to have reliable information on water quality for effective management plans. The monitoring programs include frequent water sampling at various sites, followed by determination of the parameter values that are usually characterized by high variability. Consequently, monitoring studies generate a large and complex database of results that are difficult to interpret. To explore the information included in this environmental data, different chemometric methods can be applied. These techniques allow interpretation of the obtained analytical data, while reducing the number of data dimensions without losing important information.

Analytical chemistry provides highly sensitive and selective methods that are very useful tools, demonstrated by the large number of published papers reported in the literature (Loos et al. 2010; Rodil et al. 2010; Varga et al. 2010).The organic pollutants analysis is generally based on two steps: the first step consists of analytes isolation and concentration from water samples, and the second step implies an analytical technique in order to separate and quantify the compounds. The most common extraction techniques are: liquid–liquid extraction, solid-phase extraction (SPE), solid-phase microextraction, single-drop microextraction or stir-bar sorptive extraction coupled with various analytical techniques such as gas chromatography (GC) or liquid chromatography with different detector types (Petrovic et al. 2010; Diodiu et al. 2012; Farre et al. 2012; Kouzayha et al. 2012). Generally, for monitoring studies, the extraction methods employed are as general as possible, in order to allow isolation and identification of a large number of compounds (Gómez et al. 2009; Loos et al. 2009; Marti et al. 2011; Portolés et al. 2011).

The most important European regulation regarding the water quality is the European Water Framework Directive 2000/60/EC (WFD) (European Commission 2000). Being an EU member state, Romania needs to adopt this Directive and start to implement river basin management plans. The most important objectives of directive are ‘to achieve good ecological and chemical status, to protect human health, water supply, natural ecosystems and biodiversity’. Thus, to accomplish this demand an overview of the problem needs to be made and river-specific pollutants and its sources must be identified.

A recent trend in water quality data interpretation is to use chemometric approach, such as cluster analysis (CA) and principal component analysis (PCA). This approach may be very useful for interpreting and establishing correlations for a large number of sampling points and different chemical properties or compounds (Iscen et al. 2008; Akbal et al. 2011; Cieszynska et al. 2011; Ajorlo et al. 2013). This analysis was successfully applied for characterization of different river basins, such as: Nakdong River (Venkatramanan et al. 2014), Tigris River (Alhassan et al. 2014), Langat River (Osman et al. 2012), Lis River (Vieira et al. 2012) and Three Gorges area of China (Zhao et al. 2011), Haraz River Basin (Pejman et al. 2009) or Dalio River (Zhang et al. 2009). The results of these studies highlighted the main pollution sources of each river, or managed to characterize the ecological status of the water body. CA was used for interpretation of the spring water collected from Swietokrzyski National Park, in Poland, that proved to be very useful for finding homogeneity groups among data obtained from chemical analysis (Michalik 2008).

However, chemometric methods have not been widely used for interpreting organic pollutants data. Few studies reported the use of chemometric techniques to assess the distribution of human pharmaceutical in surface water (Al-Odaini et al. 2012). As far as the authors know, in our country there are no studies reporting the association between highly sensitive analytical techniques, such as GC/MS and chemometric tools, for assessment the river water quality, concerning the organic pollution. The aims of this work are to apply chemometric techniques in order to evaluate the effect of seasonal variation upon organic compounds concentrations and to extract those parameters that are most representative for each river basin. A future monitoring river plan can be optimized based on obtained results. Furthermore, the identification of the main components can be useful for finding the best removal methods, in order to eliminate or to decrease the pollutants' negative impact upon aquatic life and human health.

MATERIALS AND METHODS

Study area and sample collection

The study was focused on the river from central part of Transylvania, Romania. The main rivers subjected to this screening study were Olt, Cibin (tributary of Olt) and Mures. Olt is one of the most important rivers in our country, having 615 km length and catchment area of 24,050 km2, crossing countries like, Harghita, Covasna, Brasov, Sibiu, Valcea and Olt. It flows into the Danube River at Turnu Magurele. One of its main right tributary is the Cibin River, which crosses the Sibiu city, having a total length of 82 km. Another important river in Transylvania is Mures, having a total length of 761 km with a river basin of 28,310 km2. It crosses important cities like Targu-Mures, Alba-Iulia, Deva and Arad. It flows into the Tisza River at Szeged, Hungary.

Two sampling campaigns were performed in two different seasons, July (dry season – d) and October (wet season – w), respectively. Nine key points were set for sampling along three rivers, as follows: two points on Cibin River, (Cristian and downstream Sibiu), three points on Olt River, (Bradu, Fagaras and Voila-downstream Fagaras) and four points on Mures River (Targu-Mures, Cipau-downstream Targu-Mures, Cuci and Ludus). The map with sampling sites is presented in Figure 1.
Figure 1

Location of sampling points from Olt, Mures and Cibin rivers.

Figure 1

Location of sampling points from Olt, Mures and Cibin rivers.

The acronyms used for all sampling sites, from both seasons, are summarized in Table 1.

Table 1

List of abbreviations used for sampling points

River Sampling point and season Acronym 
Olt Bradu-July BRDd 
Fagaras-July FAGd 
Voila-July VOILAd 
Bradu-October BRDw 
Fagaras-October FAGw 
Voila-October VOILAw 
Mures Targu-Mures-July TGMd 
Cipau-July CIPd 
Cuci-July CUCId 
Ludus-July LUDd 
Targu-Mures-October TGMw 
Cipau-October CIPw 
Cuci-October CUCIw 
Ludus-October LUDw 
Cibin Cristian-July CRSd 
Sibiu-July SIBd 
Cristian-October CRSw 
Sibiu-October SIBw 
River Sampling point and season Acronym 
Olt Bradu-July BRDd 
Fagaras-July FAGd 
Voila-July VOILAd 
Bradu-October BRDw 
Fagaras-October FAGw 
Voila-October VOILAw 
Mures Targu-Mures-July TGMd 
Cipau-July CIPd 
Cuci-July CUCId 
Ludus-July LUDd 
Targu-Mures-October TGMw 
Cipau-October CIPw 
Cuci-October CUCIw 
Ludus-October LUDw 
Cibin Cristian-July CRSd 
Sibiu-July SIBd 
Cristian-October CRSw 
Sibiu-October SIBw 

For a good preservation, the samples were collected in dark brown bottles and 1 mL of HCl 2 N were added to each sample after sampling. The method used for sampling was the grab technique. The collected samples (1 L) were stored in the dark, transported to the laboratory within 48 h and subjected to the filtration, extraction and analysis as soon as possible.

Chemicals and reagents

The mobile phase filtration apparatus and the SPE workstation were purchased from Supelco, USA. The used reagents were: acetonitrile, dichloromethane and water (Merck, Germany), isooctane (J. T. Baker, The Netherlands) and PCB-30 (Fluka, Germany). The Oasis HLB cartridges used for SPE extraction were purchased from Waters, USA. For samples concentration a rotary evaporator RVO 400 was used, from Ingos, Czech Republic. Analyses were performed using a gas chromatograph coupled with a mass spectrometer as a detector from Thermo Scientific, USA.

Sample extraction and analysis

Five hundred milliliters of water samples were filtered with a vacuum system using Teflon filters with 47 mm diameter. Samples were extracted using SPE using OASIS HLB cartridges. Briefly, cartridges were conditioned with 3 mL dichloromethane, 3 mL acetonitrile and 3 mL high performance liquid chromatography (HPLC) water. Samples were passed about 2 mL min−1 flow rate under vacuum. The impurities were removed within the washing step, with 3 ml H2O HPLC grade. The cartridges were left to dry at vacuum for 50 min. Finally, the elution was made with 3 mL of dichloromethane and 3 mL mixture of dichloromethane and acetonitrile (1:1, v/v). The eluents were transferred in a conical flask, evaporated using a rotary evaporator and reconstituted by adding about 1 mL of isooctane and 2 μL were injected in GC/MS system. All samples were run in duplicate. A double blank sample was extracted, in order to eliminate the possible interferences derived from handling and SPE procedures. No significant interferences were observed.

Samples were analyzed using a GC/MS method adapted from another reported paper (Moldovan 2006). For quantitative evaluation, the internal standard method was used (internal standard used was PCB-30, 1,000 ng L−1 in heptane). The GC/MS analysis was carried out by operating in electron ionization (EI) mode at 70 eV. The source temperature and the injector temperature were set at 250 °C. A capillary column DB-5MS of 30 m × 0.25 mm i.d. was used and the MS transfer line was at 300 °C. The temperature program was as follows: initial temperature 90 °C for 1 min, increased with 10 °C min−1 up to 120 °C, increased with 3.5 °C min−1 up to 190 °C and increased again with 4 °C min−1 until 300 °C and maintained here for 8 min. These three temperature gradients are necessary for a good separation of compounds and for coelutions elimination. Helium was the carrier gas at 1.5 mL min−1 flow. The qualitative analysis was made by comparing the obtained mass spectra with those from the NIST library and with mass spectra found in the studied literature. The quantitative evaluation was established based on internal standard area and every compound was correlated to this area.

The acronyms used for statistical interpretation corresponding to the sampling stations and organic pollutants are presented in Table 2.

Table 2

List of abbreviations used for statistical interpretation

Organic pollutant Acronym 
2,4-Di-tert-butylphenol DBP 
2,4-Di-tert-butyl-6-nitrophenol DNTP 
2,6-Di-tert-butyl-4-nitrophenol DBNP 
2,4,6-Triisopropylphenol TPP 
Butylated hydroxytoluene BHA 
Methyl hydrojasmonate MHJ 
Alpha methyl ionone AMI 
Galaxolide GALA 
Tonalide TONA 
Benzophenone BZP 
2-Ethylhexyl-4-metoxycinnamate PRM 
2-Ethyltrans-4-metoxycinnamate EHMC 
Organic pollutant Acronym 
2,4-Di-tert-butylphenol DBP 
2,4-Di-tert-butyl-6-nitrophenol DNTP 
2,6-Di-tert-butyl-4-nitrophenol DBNP 
2,4,6-Triisopropylphenol TPP 
Butylated hydroxytoluene BHA 
Methyl hydrojasmonate MHJ 
Alpha methyl ionone AMI 
Galaxolide GALA 
Tonalide TONA 
Benzophenone BZP 
2-Ethylhexyl-4-metoxycinnamate PRM 
2-Ethyltrans-4-metoxycinnamate EHMC 

Chemometric interpretation

All statistical modeling was carried out using IBM SPSS Statistics 20 software. For examining the range of analyzed organic compounds, some descriptive statistics were employed (mean, maximum, minimum, and standard deviation). ANOVA – analysis of variance (comparison between groups) – is a statistical method which was used for revealing the differences between two or more means from distinct groups. The first comparison was made between two groups, namely the wet and the dry seasons, respectively, and the second comparison was made between three groups represented by each investigated river basin. Pearson correlation was used for revealing the relationship and interdependency between analyzed organic compounds. Positive correlations means that the two variables increase or decrease together, meanwhile negative correlation means that one variable decreases while the other one increases (Gazzaz et al. 2012).

CA is a common method used for grouping variables or objects into clusters, based on similarity within a class and dissimilarity between different classes, after a predetermined criterion. CA analysis was applied to detect multivariate similarities between sampling sites in different sampling points for different sampling periods. Euclidean distance (similarity measure) was used to compute the distance and the applied clustering procedure was the Ward's Method (Salah et al. 2012).

PCA is another powerful technique used for dimension reduction, which provides information on the most representative parameter, with a minimum loss of initial information. PCA reduces the contribution of less significant variables and generates a new group of variables known as principal components/factors (PCs). These PCs are uncorrelated and appear in decreasing order of importance. An important aspect of PCA is the generation of eigenvalues which give a measure of the significance of the components. The components with the highest eigenvalues have the biggest contribution. The best rotation method is widely believed to be Varimax. After Varimax rotation, each original variable tends to be associated with one factor and each factor represents only a small number of variables. Correlation of principal components and original variables is given by loadings. The main purpose of this analysis is to identify possible sources of investigated organic pollutants in river water samples.

RESULTS AND DISCUSSION

The minimum, maximum, mean and the standard deviation were made on concentration obtained in each sampling point. The obtained data are summarized in Table 3 exposed below. The means and standard deviations were low for some pollutants (DTNP, DBNP, TONA and AMI), but other pollutants exhibited large values (BZP, DBP, EHMC, MHJ and GALA). The maximum values were found for benzophenone (BZP). So, the concentrations pattern distribution follows the decreasing order BZP > DBP > EHMC > MHJ > GALA > PRM > TPP > AMI > BHA > TONA > DBNP > DBP.

Table 3

Descriptive statistics for organic contaminants in water (ng L−1) from three river basins (n = 18)

Organic contaminant Minimum Maximum Mean Standard deviation 
DBP <LOQ 708.40 324.78 159.63 
DTNP <LOQ 58.90 5.37 16.06 
DBNP <LOQ 33.00 7.36 12.40 
BHA <LOQ 82.20 23.22 29.60 
TPP <LOQ 358.60 60.82 81.83 
MHJ <LOQ 371.60 160.40 119.84 
AMI <LOQ 441.90 52.1 112.72 
GALA <LOQ 807.30 159.66 176.40 
TONA <LOQ 89.10 22.33 21.14 
BZP <LOQ 2896.20 1087.91 854.73 
PRM <LOQ 432.20 91.60 115.32 
EHMC <LOQ 949.10 197.41 317.96 
Organic contaminant Minimum Maximum Mean Standard deviation 
DBP <LOQ 708.40 324.78 159.63 
DTNP <LOQ 58.90 5.37 16.06 
DBNP <LOQ 33.00 7.36 12.40 
BHA <LOQ 82.20 23.22 29.60 
TPP <LOQ 358.60 60.82 81.83 
MHJ <LOQ 371.60 160.40 119.84 
AMI <LOQ 441.90 52.1 112.72 
GALA <LOQ 807.30 159.66 176.40 
TONA <LOQ 89.10 22.33 21.14 
BZP <LOQ 2896.20 1087.91 854.73 
PRM <LOQ 432.20 91.60 115.32 
EHMC <LOQ 949.10 197.41 317.96 

One way ANOVA (at 5% level of significance) was used to highlight the compounds which concentrations were influenced by increased river flow during the wet season. Only two compounds resulted to be influenced by the increased flow from the wet season (p value lower than 0.05), namely MHJ (0.043) and BZP (0.001). ANOVA was applied for finding out the pollutants that can differentiate the three investigated river. DBNP (0.016) and BHA (0.004) presented higher concentrations in the Olt River than in the Mures River. Unfortunately, the Cibin River could not be differentiated by any organic compounds. These comparisons between rivers were made using Tukey post hoc test.

The dendrogram presented in Figure 2 was obtained by using Ward's method of clustering with Euclidean distances as a measure of similarity between clusters. The sampling points were grouped in three statistical clusters. The sampling points belonging to the same cluster had similar characteristics. This means that, for a rapid screening, it is not necessary to analyze all samples from one cluster because only one sample may be representative for the entire cluster. This site represents the reference of the spatial assessment of water quality. Cluster 1 grouped the sampling points CIPw, CUCIw, TGMw, BRDw, FAGw, VOILAw, SIBw and these are the less polluted sites, due to increased river flow from the rainy season. Cluster 2 comprised the most polluted point overall and all were collected in July (the dry season) namely CRSd, SIBd, CUCId, LUDd, TGMd, BRDd. The Cluster 3 grouped VOILAd, CRSw, CIPd, FAGd and LUDw and these corresponds to moderately polluted sites. The most important contribution to this high degree of pollution is represented by direct discharges into fresh river water of domestic wastewaters and by leaching waters from agricultural landfill. Another significant cause for the high degree of pollution is represented by undersize and old municipal wastewater treatment plants of small towns, like Voila (VOILAd) and Fagaras (FAGd). Results obtained from CA analysis evidenced the most polluted sampling points in terms of organic pollutants. The most discriminating pollutants were revealed by applying ANOVA on variable obtained after CA. Only two components were statistically significant, namely BZP (p = 0.01) and AMI (p = 0.058; even if it is a little over 0.05 it was considered). A future monitoring program might be optimized, based on these results, because samples can be collected only from significant points, thus reducing the time spent with collection and analysis, as well as the costs. Similar results were obtained in previously reported studies by other researchers, when CA was employed for grouping the sampling sites on Potrero de los Funes River, in Argentina (Gónzalez et al. 2011). The used procedure generated three groups of sampling points, because the sites belonging to the groups have similar characteristics and natural background sources types.
Figure 2

Dendrogram of sampling sites according to organic contaminants analyzed in water samples.

Figure 2

Dendrogram of sampling sites according to organic contaminants analyzed in water samples.

The PCA was run on dataset which comprises concentrations for each organic pollutant in every sampling point, in both sampling campaigns. The key parameters employed for this analysis were: the extraction method used was principal component and only components with eigenvalues greater than 1 were retained; rotation was made by Varimax method with Kaiser normalization.

The correlation matrix obtained is presented in Table 4. Values greater than 0.5 represent a statistically significant correlation (p < 0.05) between organic compounds and are shown bold in the table. The positive correlations were observed between DBP, TPP and GALA, DNTP was correlated with DBNP, TPP was correlated with GALA and TONA, GALA with TONA and the last positive correlation appears between PRM and EHMC. No statistically significant negative correlations were observed. These correlations between various analysed organic pollutants might appear due to common sources of pollution.

Table 4

Correlation matrix for organic compounds analyzed in river basins from Transylvania

  DBP DTNP DBNP BHA TPP MHJ AMI GALA TONA BZP PRM EHMC 
DBP 1.000            
DTNP 0.253 1.000           
DBNP 0.144 0.685 1.000          
BHA 0.047 −0.009 0.430 1.000         
TPP 0.721 0.103 0.112 −0.241 1.000        
MHJ 0.217 −0.103 −0.073 0.240 −0.223 1.000       
AMI 0.248 −0.164 −0.168 −0.078 0.406 0.073 1.000      
GALA 0.700 −0.169 −0.172 0.040 0.821 −0.040 0.449 1.000     
TONA 0.288 −0.135 −0.145 −0.083 0.645 −0.332 0.316 0.728 1.000    
BZP 0.374 0.487 0.308 −0.297 0.141 0.471 −0.110 −0.124 −0.292 1.000   
PRM 0.038 −0.211 0.018 −0.039 −0.086 0.324 −0.095 −0.216 −0.371 0.313 1.000  
EHMC −0.114 −0.189 0.296 0.232 −0.092 0.131 −0.170 −0.184 −0.205 0.210 0.815 1.000 
  DBP DTNP DBNP BHA TPP MHJ AMI GALA TONA BZP PRM EHMC 
DBP 1.000            
DTNP 0.253 1.000           
DBNP 0.144 0.685 1.000          
BHA 0.047 −0.009 0.430 1.000         
TPP 0.721 0.103 0.112 −0.241 1.000        
MHJ 0.217 −0.103 −0.073 0.240 −0.223 1.000       
AMI 0.248 −0.164 −0.168 −0.078 0.406 0.073 1.000      
GALA 0.700 −0.169 −0.172 0.040 0.821 −0.040 0.449 1.000     
TONA 0.288 −0.135 −0.145 −0.083 0.645 −0.332 0.316 0.728 1.000    
BZP 0.374 0.487 0.308 −0.297 0.141 0.471 −0.110 −0.124 −0.292 1.000   
PRM 0.038 −0.211 0.018 −0.039 −0.086 0.324 −0.095 −0.216 −0.371 0.313 1.000  
EHMC −0.114 −0.189 0.296 0.232 −0.092 0.131 −0.170 −0.184 −0.205 0.210 0.815 1.000 

Correlation is significant at the 0.01 level (two-tailed).

Bold values correspond to statistically significant correlations.

By running PCA on the data set, five principal components were retained encountering a total variance of 87.330% from total variance. The graphical representation of every principal component against its eigenvalues is presented in Figure 3. From this graphical representation it can be observed that after the fifth component the slope is bigger and the first five components have the most significant contribution, namely: PC1 have 28.655%, PC2 has 20.346%, PC3 has 16.007%, PC4 has 12.178% and PC5 has 10.144% from the total variance of dataset.
Figure 3

Scree plot for each component and its eigenvalues.

Figure 3

Scree plot for each component and its eigenvalues.

According to Table 5, the highest loadings from principal component 1 are given by DBP, TPP, AMI, GALA and TONA. This PC contains the largest number of representative parameters and can indicate the effect of direct discharges of wastewater effluents, mainly from industry or without any preliminary treatment.

Table 5

Loadings of organic contaminants after Varimax rotation (five components extracted) for water samples collected from 18 sampling points (values >0.5 are bold)

Organic parameter Principal component 1 Principal component 2 Principal component 3 Principal component 4 Principal component 5 
DBP 0.788 0.317 0.008 0.351 −0.022 
DTNP −0.046 0.919 −0.235 0.037 −0.099 
DBNP −0.028 0.859 0.187 −0.092 0.373 
BHA −0.036 0.105 0.062 0.110 0.978 
TPP 0.916 0.192 0.050 −0.171 −0.211 
MHJ −0.023 −0.099 0.118 0.935 0.157 
AMI 0.551 −0.286 −0.143 0.145 0.005 
GALA 0.945 −0.145 −0.103 −0.041 0.092 
TONA 0.721 −0.153 −0.175 −0.444 0.022 
BZP 0.043 0.559 0.244 0.583 −0.451 
PRM −0.102 −0.080 0.909 0.262 −0.123 
EHMC −0.105 0.050 0.958 −0.018 0.172 
Eigenvalue 3.439 2.442 1.921 1.461 1.217 
Explained variance (%) 28.655 20.346 16.007 12.178 10.144 
Cumulative variance (%) 28.655 49.002 65.009 77.187 87.330 
Organic parameter Principal component 1 Principal component 2 Principal component 3 Principal component 4 Principal component 5 
DBP 0.788 0.317 0.008 0.351 −0.022 
DTNP −0.046 0.919 −0.235 0.037 −0.099 
DBNP −0.028 0.859 0.187 −0.092 0.373 
BHA −0.036 0.105 0.062 0.110 0.978 
TPP 0.916 0.192 0.050 −0.171 −0.211 
MHJ −0.023 −0.099 0.118 0.935 0.157 
AMI 0.551 −0.286 −0.143 0.145 0.005 
GALA 0.945 −0.145 −0.103 −0.041 0.092 
TONA 0.721 −0.153 −0.175 −0.444 0.022 
BZP 0.043 0.559 0.244 0.583 −0.451 
PRM −0.102 −0.080 0.909 0.262 −0.123 
EHMC −0.105 0.050 0.958 −0.018 0.172 
Eigenvalue 3.439 2.442 1.921 1.461 1.217 
Explained variance (%) 28.655 20.346 16.007 12.178 10.144 
Cumulative variance (%) 28.655 49.002 65.009 77.187 87.330 

Extraction method: Principal component analysis.

Rotation method: Varimax with Kaiser normalization.

Principal components 2 and 3 has loadings of DNTP, DBNP, BZP and PRM and EHMC. Nitroaromatic compounds are antioxidant derivatives known to be carcinogenic for aquatic organism and their major uses are oil industry for improving resistance to oxidation process. The main uses of PRM and EHMC are as UV filter in sunscreen products and UV stabilizers for plastic products. This component can be ascribed to point pollution sources represented here by direct input from recreational activities. Principal component 4 has loading in MHJ and BZP and the last principal component 5 has loadings of BHA. MHJ is a fragrance ingredient used in many fragrance mixtures. It may be found in fragrances used in decorative cosmetics, fine fragrances, shampoos, toilet soaps or household cleaners and detergents. The fact that BZP had high loadings in two PCs can suggest that this compound had two different sources of pollution. BHA is a preservative found very often in many food products, in order to prevent the oxidation of fats and oils. It can also be used in rubber, petroleum products and wax food packaging.

In order to find the relations between sampling sites and analyzed compounds, score and loadings plot for the first two principal components extracted were examined.

For classification of water samples and analyzed organic pollutants, score and loadings plots for the first two principal components extracted were examined. The plots are presented in Figures 4 and 5, respectively. The loading plot indicates the similarities and correlations between compounds. Thus, the compounds with small loadings which appeared near the origin (BHA, EHMC, PRM and MHJ) had only a little influence on data structure, whereas the compounds with high loadings represent those elements with the greatest influence on the grouping and separation of water samples. A close relation was observed between the concentrations of: DBP, TPP, GALA, TONA and AMI (highest loadings of the first PC), and between DNTP, DBNP and BZP (highest loadings of the second PC). When examining the score plot, it can be stated that water samples collected from three distinct rivers are quite similar. The two points from the Olt River, which appear detached from the rest of the samples, correspond to BRDd and FAGd, and presented large values for PC2 (3.072 and 1.847, respectively). Also, one sample from the Mures River, LUDw, presented a large value for PC1 (3.638). These results are in accordance with those obtained from CA, where BRDd appeared in Cluster 2 (highest level of pollution), while FAGd and LUDo are in Cluster 3 (medium pollution). The detachment of these points from the rest might be because it presents the maximum level for some analyzed pollutants. For example, LUDo has the maximum level of DNTP, DBNP and BZP.
Figure 4

Loadings plot of organic pollutants.

Figure 4

Loadings plot of organic pollutants.

Figure 5

Score plot of sampling points.

Figure 5

Score plot of sampling points.

CONCLUSIONS

The aim of this study was to evaluate the main water pollution sources of rivers from central Transylvania and to characterize the spatial distribution of sampling points. Environmental analytical chemistry generates a large amount of data, and multivariate statistical techniques become very helpful for interpretation and for underlying hidden information. In this work, chemometric techniques (descriptive statistics, ANOVA, CA and PCA) were successfully applied to evaluate the variation of organic pollutants in three distinct rivers and proved to be very useful tool for rendering large data sets. ANOVA highlighted the elements that can distinguish the Mures from the Olt River, namely DBNP and BZP, while the Cibin River could not be distinguished by any characteristic compound. Furthermore, the only compounds that were influenced by the increased river flow from the wet season are MHJ and BZP. CA reduced the sampling points at three main clusters, according to similarities among sampling points. Based on this information, a future monitoring study may be optimized by reducing the sampling points to those that are representative for each river and thereby reducing costs, without any information loss. PCA managed to reduce the data set to five principal components, explaining a total variance of 87.330%, each having specific loadings. These results also evidenced the major sources of organic pollutants which are mainly represented by anthropogenic activities (domestic, point sources) and by industry, with negative impact upon surface water quality.

ACKNOWLEDGEMENT

Financial support was provided by National Authority for Scientific Research and Innovation – ANCSI, Core Programme, Project PN16-30 02 04.

REFERENCES

REFERENCES
Ajorlo
M.
Abdullah
R. B.
Yusoff
M. K.
Halim
R. A.
Hanif
A. H.
Willms
W. D.
Ebrahimian
M.
2013
Multivariate statistical techniques for the assessment of seasonal variations in surface water quality of pasture ecosystems
.
Environmental Monitoring and Assessment
185
,
8649
8658
.
Akbal
F.
Gürel
L.
Bahadır
T.
Güler
I.
Bakan
G.
Büyükgüngör
H.
2011
Multivariate statistical techniques for the assessment of surface water quality at the Mid-Black Sea coast of Turkey
.
Water Air Soil Pollution
216
,
21
37
.
Alhassan
H. I.
Basim
S. A.
Shahla
A. Q.
2014
Application of multivariate statistical techniques in the surface water quality assessment of Tigris river at Baghdad stretch, Iraq
.
Journal of Babylon University
22
,
450
462
.
Al-Odaini
N. A.
Zakaria
M. P.
Zali
M. A.
Juahir
J.
Yaziz
M. I.
Surif
S.
2012
Application of chemometrics in understanding the spatial distribution of pharmaceuticals in surface water
.
Environmental Monitoring and Assessment
184
,
6753
6748
.
Cieszynska
M.
Wesolowski
M.
Bartoszewicz
M.
Michalska
M.
2011
Assessment of river water quality in the South Baltic coast by multivariate techniques
.
Central European Journal of Chemistry
9
,
265
274
.
Diodiu
R.
Roba
C.
Beldean-Galea
S.
Cosma
C.
2012
PAHs content determination in water and sludge samples from Somesul Mic River
.
ProEnvironment
5
,
79
84
.
European Commission
2000
Directive 2000/60/EC of the European Parliament and of the Council of 23 October 2000
.
Gazzaz
N. M.
Yusoff
M. K.
Ramli
M. F.
Aris
A. Z.
Juahir
H.
2012
Characterization of spatial patterns in river water quality using chemometric pattern recognition techniques
.
Marine Pollution Bulletin
64
,
688
698
.
Gómez
M. J.
Gómez-Ramos
M. M.
Agüera
A.
Mezcua
M.
Herrera
S.
Fernández-Alba
A. R.
2009
A new gas-chromatography/mass spectrometry method for the simultaneous analysis of target and non-target organic contaminants in water
.
Journal of Chromatography A
1216
,
4071
4082
.
Gónzalez
S. O.
Almeida
C. A.
Quintar
S.
Mallea
M. A.
Gónzalez
P. S.
2011
Application of multivariate statistical techniques to evaluate organic pollution on a river in Argentina
.
Revista de Ambiente & Agua–An Interdisciplinary Journal of Applied Science
3
,
27
42
.
Iscen
C. F.
Emiroglu
O.
Ilhan
S.
Arslan
N.
Yilmaz
V.
Ahiska
S.
2008
Application of multivariate statistical techniques in the assessment of surface water quality in Uluabat Lake, Turkey
.
Environmental Monitoring and Assessment
144
,
269
276
.
Kouzayha
A.
Rabaa
A. R.
Iskandarani
M. A.
Beh
D.
Budzinski
H.
Jaber
F.
2012
Multiresidue method for determination of 67 pesticides in water samples using solid-phase extraction with centrifugation and gas-chromatography-mass spectrometry
.
American Journal of Analytical Chemistry
3
,
257
265
.
Loos
R.
Gawlik
B. M.
Locoro
G.
Rimaviciute
E.
Contini
S.
Bidoglio
G.
2009
EU-wide survey of polar organic persistent pollutants in European river waters
.
Environmental Pollution
157
,
561
568
.
Marti
M.
Aguado
D.
Segovia-Martínez
L.
Bouzas
A.
Seco
A.
2011
Occurrence of priority pollutants in WWTP effluents and Mediterranean coastal waters of Spain
.
Marine Pollution Bulletin
62
,
615
625
.
Michalik
A.
2008
The use of chemical and cluster analysis for studying spring water quality in Świętokrzyski National Park
.
Polish Journal of Environmental Studies
17
,
357
362
.
Osman
R.
Saim
N.
Juahir
H.
Abdullah
M. P.
2012
Chemometric application in identifying sources of organic contaminants in Langat river basin
.
Environmental Monitoring and Assessment
184
,
1001
1014
.
Pejman
A. J.
Bidhendi
G. R.
Karbassi
A. R.
Mehrdadi
N.
Bidhendi
M. E.
2009
Evaluation of spatial and seasonal variations in surface water quality using multivariate statistical techniques
.
International Journal of Environmental Science and Technology
6
,
467
476
.
Petrovic
M.
Farre
M.
Lopez de Alda
M.
Perez
S.
Postigo
C.
Köck
M.
Radjenovic
J.
Gros
M.
Barcelo
D.
2010
Recent trends in the liquid chromatography-mass spectrometry analysis of organic contaminants in environmental samples
.
Journal of Chromatography A
1217
,
4004
4017
.
Rodil
R.
Quitana
J. B.
Basaglia
G.
Pietrogrande
M. C.
Cela
R.
2010
Determination of synthetic phenolic antioxidants and their metabolites in water samples by downscale solid-phase extraction, sialylation and gas chromatography-mass spectrometry
.
Journal of Chromatography A
1218
,
303
311
.
Salah
E. A. M.
Turki
A. M.
Othman
E. M. A.
2012
Assessment of water quality of Euphrates river using cluster analysis
.
Journal of Environmental Protection
3
,
1629
1633
.
Venkatramanan
S.
Chung
S. Y.
Lee
S. Y.
Park
N.
2014
Assessment of river water quality via environmetric multivariate statistical tools and water quality index: a case study of Nakdong river basin, Korea
.
Carpathian Journal of Earth and Environmental Sciences
9
,
125
132
.
Vieira
J. S.
Pires
J. C. M.
Martins
F. G.
Vilar
V. J. P.
Boaventura
R. A. R.
Botelho
C. M. S.
2012
Surface water quality assessment of Lis river using multivariate statistical methods
.
Water Air Soil Pollution
223
,
5549
5561
.