art%3A10.1007%2Fs10661-013-3268-2

Improving the sampling strategy of the Joint Danube Survey 3(2013) by means of multivariate statistical techniques appliedon selected physico-chemical and biological data

Carmen Hamchevici & Ion Udrea

Received: 20 November 2012 /Accepted: 18 May 2013 /Published online: 31 May 2013# Springer Science+Business Media Dordrecht 2013

Abstract The concept of basin-wide Joint DanubeSurvey (JDS) was launched by the InternationalCommission for the Protection of the Danube River(ICPDR) as a tool for investigative monitoring underthe Water Framework Directive (WFD), with a fre-quency of 6 years. The first JDS was carried out in2001 and its success in providing key information forcharacterisation of the Danube River Basin District asrequired by WFD lead to the organisation of the sec-ond JDS in 2007, which was the world’s biggest riverresearch expedition in that year. The present paperpresents an approach for improving the survey strate-gy for the next planned survey JDS3 (2013) by meansof several multivariate statistical techniques. In orderto design the optimum structure in terms of parametersand sampling sites, principal component analysis(PCA), factor analysis (FA) and cluster analysis wereapplied on JDS2 data for 13 selected physico-chemical

and one biological element measured in 78 samplingsites located on the main course of the Danube.Results from PCA/FA showed that most of the datasetvariance (above 75 %) was explained by fivevarifactors loaded with 8 out of 14 variables: physical(transparency and total suspended solids), relevantnutrients (N–nitrates and P–orthophosphates), feed-back effects of primary production (pH, alkalinityand dissolved oxygen) and algal biomass. Taking intoaccount the representation of the factor scores givenby FA versus sampling sites and the major groupsgenerated by the clustering procedure, the spatial net-work of the next survey could be carefully tailored,leading to a decreasing of sampling sites by more than30 %. The approach of target oriented sampling strat-egy based on the selected multivariate statistics canprovide a strong reduction in dimensionality of theoriginal data and corresponding costs as well, withoutany loss of information.

Keywords Joint Danube Survey .Multivariatestatistics

Introduction

Following the first institutional framework for commonactivities designed to protect the water environment inthe Danube River basin—the Bucharest Declaration in1985—the Convention on the Protection and SustainableUse of the Danube River (short Danube River Protection

Environ Monit Assess (2013) 185:9495–9507DOI 10.1007/s10661-013-3268-2

C. Hamchevici (*)National Administration “ApeleRomane”, Edgar QuinetStreet, Postal Code 010018 Bucharest, Romaniae-mail: [email protected]

C. Hamchevicie-mail: [email protected]

I. UdreaResearch Center for Environmental Protection and WasteManagement, Bucharest, University of Bucharest,Romania,Sos, Panduri 90-92,sect.5,Bucharest 050663, Romania

Convention, DRPC) was signed in June 1994 in Sofia.With the entry into force of the DRPC, the InternationalCommission for the Protection of the Danube River(ICPDR) was legally established as the main decisionmaking body under the convention. Currently, theICPDR’s water management policy is strongly based onthe implementation of the Water FrameworkDirective—WFD (2000/60/EC). In Annex V, Section1.1.1 (Rivers), WFD clearly specifies that surface waterstatus should be assessed by means of several mainquality elements, among which chemical and physico-chemical elements supporting the biological elements arelisted. Under the “General” category, the followinggroups of variables are mentioned: thermal conditions,oxygenation conditions, salinity, acidification status andnutrient conditions. Even if the mentioned groups ofvariables are annually measured by a basin-wide networkof National Reference Laboratories within theTransnational Monitoring Network launched in 1996 bythe ICPDR (TNMN), the harmonisation of methods andimprovement of analytical data comparability is an on-going process. Although good results in this respect havebeen obtained (annual analytical quality control schemesorganised at the basin level, availability of standard op-erational procedures), further activities have to be carriedout in order to achieve full comparability of monitoringdata among various institutions and riparian countries.Therefore, the investigative monitoring activity at thebasin-wide level was initiated and launched by theICPDR every 6 years, starting from 2001, known asJoint Danube Survey (JDS). The first survey took placein 2001 and its successful results (Woitke et al. 2003;Sommerwerk et al. 2009) lead in 2007 to the organisationof the second JDS, in which the physico-chemical sup-portive indicators to biological quality elements weremeasured among more than 280 individual parametersinvestigated (Micić and Hofmann 2009;Janauer et al.2010; Loos et al. 2010). The next survey (JDS3) is beingorganised in 2013 and the information from previousresults offers a sound basis for designing the futuresurvey strategy in terms of chosen set of parametersand sampling locations. Taking into account that themain strength of JDS-type investigation consists in theuse of a single sampling platform, the application of thesame sampling methods as well as the use of the sameanalytical methods along the entire length of theDanube, at each site samples were collected once, mean-ing that the obtained results have a “snap-shot” character

and should be regarded as partial information on short-term basis, reflecting the existing situation at the time ofthe survey, between 13 August and 26 September 2007.The comparison of these momentary data with the dataobtained within the long-term monitoring programme(TNMN) lead to the conclusion that surveillance datarecorded during the similar time interval were signifi-cantly comparable (Hamchevici and Udrea 2013).

Usually, water quality is being assessed using apattern according to which the water quality is classi-fied into quality classes based on the comparison of ameasured value for a certain parameter with a limit orthreshold value set by a normative legal instrument,aim that frequently requires a complex data matrix(Chapman 1996; Kovalkovski et al. 2006). A relative-ly similar approach was applied on the water qualitydata set obtained from several Danubian studies(Literáthy and László 1999; Slobodník et al. 2005;Hamchevici and Craciun 2008) and in the first inves-tigative monitoring survey carried out at the wide-scale for the Danube River, known as Joint DanubeSurvey 1, in 2001 (Literáthy et al. 2002). Furthermore,in water quality monitoring, it is usual to challenge theproblem of determining whether a variation in theconcentration of measured parameters is caused byan anthropogenic source, by the natural variability(temporal, climatic) or changes in the hydrology con-ditions (Wunderlin et al. 2001; Li and Zhang 2009).Also, it should be determined which parameters arethe most significant to describe such spatial and tem-poral variations and the pollution sources. Therefore,another method which can be adopted to evaluatewater quality changes is the use of multivariate tech-niques, also called chemometrics (Chapman 1996;Mendiguchia et al. 2004), widely used in a variety ofenvironmental applications including the Danubianissues (Birk and Hering 2009). One of the most im-portant advantages of using these statistical techniquesis the possibility to identify the relevant parameters tobe considered in the assessment of water quality and todiscriminate among sampling sites having rather sim-ilar characteristics (Shine et al. 1995; Vega et al. 1998;Yu et al. 1998; Perkins and Underwood 2000; Tauleret al. 2000; Voutsa et al. 2001; Gangopadhyay et al.2001; Bengraine and Marhaba 2003; Ouyang 2005;Astel et al. 2006; Ouyang et al. 2006; Shrestha andKazama 2007; Bouza-Deaño et al. 2008; Kazi et al.2009). There are several common methods in this

9496 Environ Monit Assess (2013) 185:9495–9507

respect, among which principal component analysis(PCA), factor analysis (FA) and cluster analysis (CA)are the most recently used in monitoring studies (Bu etal. 2010; Huang et al. 2010; Razmkhah et al. 2010).

The main objective of this contribution is to inves-tigate the possibility to design the structure of thesampling sites network of the next Joint DanubeSurvey (JDS3-2013) in which the supportive generalphysico-chemical variable to be measured based onthe results obtained from selected multivariate statisti-cal techniques applied on the JDS2 data set.

Materials and methods

Sampling and analytical methods

Water samples designated to the determination of thebiological and general physico-chemical parameterswere collected from the surface, in the middle of theriver, using a black bucket (8 L).

Table 1 shows the parameters determined and ana-lytical techniques used. In situ measurements for watertemperature, dissolved oxygen, pH and electrical con-ductivity were done using WTW multi-probes; Secchidepth and incident PAR-radiation was measured ateach sampling station. Dissolved nutrients forms wereanalysed on-board after the filtration through 0.45-μmpore size membrane filters by colorimetric methodsusing a dual-beam VARIAN spectrophotometer. Watersamples for the determination of total nutrients formswere stored in darkness at 40C and transported undercontrolled conditions to a previously selectedaccredited laboratory (Vituki, Budapest, Hungary).Total Phosphorous was analysed on the filtered(0.45-μm membrane filters) sample (pH=2) by ICP-AES (Perkin Elmer 3200DV). Suspended solids wereimmediately separated by filtering an aliquot, 0.2–1 Lonto pre-combusted and pre-weighed glass fibre filtersfor gravimetric analysis (Greenberg et al. 2005). Thefilters were pre-dried on board in an oven at 35 °C andkept in a desiccator for final analysis in the laboratory.After drying the filters at 105 °C for 4 h, totalsuspended solids (TSS) was calculated from the dif-ference in dry weight between the filter plus residueand the blank filter. The filter was then combusted in amuffle furnace at 500 °C for 4 h. Ash-free dry weightof inorganic suspended solids was determined again

from the difference in weight to the blank filter. Thedifference in the weight before and after ignition isassumed to be roughly equivalent to organic suspendedsolids (OSS). The quantitative samples (100 ml) forphytoplankton counting and sizing were fixed with0.4 ml Utermöhl’s acetic acid Lugol solution, preservedwith a few drops of formalin in brown screw cap glassbottles and stored in a cool dry place. Samples wereprocessed in the laboratory applying the sedimentationtechnique. The number counted was converted intopopulation density (unit cells per litre) and the fresh-weight biomass was converted from total bio-volume(Dokulil and Kaiblinger 2008).

The dataset

The JDS2 collected data from 96 sites along a 2,600-km stretch of the Danube River, 18 of which werelocated in the mouth of major tributaries. The relative-ly uneven distribution of sites along the entire stretchof the Danube River was caused by the pre-definedcriteria applied in sites selection: located justupstream/downstream of an international border, lo-cated upstream of confluence between Danube andmain tributaries, located downstream of the biggestpoint sources, located to control important water uses.This paper takes into account results for (14) selectedbiological and physico-chemical parameters obtainedfrom the 78 monitoring stations located on the maincourse of the Danube River, since the obtained datasuggested that the Danube River remained rather un-affected by the tributaries input (Liška et al. 2008). Forthe results presentation, the splitting of the DanubeRiver course into three sections was applied (Literáthyet al. 2002): upper Danube, from river km 2,600 toriver km 1,880 and sampling stations JDS1–JDS15,middle Danube, from river km 1,869 to river km 1077and sampling stations JDS16–JDS58 and lowerDanube, from river km 1071 to river km 0 and sam-pling stations JDS59–JDS96 (missing numbers wereassigned to the monitoring sites located on the maintributaries of the Danube River, not approached in thispaper). The basic statistics of selected variables aresummarised in Table 1. In the raw dataset, valuesbelow the limit of quantification were replaced by halfof this limit.

All statistical computations were carried out usingSTATISTICA 7.0 trial version.

Environ Monit Assess (2013) 185:9495–9507 9497

Statistical procedures

Principal component analysis and factor analysis

In order to avoid any misclassification coming up fromthe different orders of magnitude of numerical value andvariance in the dataset, the multivariate statistical tech-niques were applied on raw data normalised to zeromean and unit variance (standardised data) (Johnsonand Wichern 2002; Otto 1998; Vega et al. 1998;Helena et al. 2000). Factor analysis that includes prin-cipal components analysis is one of the most powerfuland common techniques used for reducing the dimen-sionality of large sets of data, because only a fewvarifactors/principal components (VFs/PCs) are neededto get a good description of the entire data set variabilitywithout loss of information (Battegazzore and Renoldi1995; Adams 1988; Otto 1998; Vega et al. 1998; Helenaet al. 2000). Besides, VFs usually group the studied vari-ables in accordance with common features. So FA/PCA isparticularly valuable when a chemical, physical or bio-logical interpretation of the data grouped in VFs is possi-ble (Voutsa et al. 2001; Bengraine and Marhaba 2003;Ouyang, Y. et al. 2006).

Cluster analysis

Cluster analysis is a multivariate technique whose pri-mary purpose is to assemble objects based on the char-acteristics they possess. Cluster analysis classifies ob-jects, so that each object is similar to the others in thecluster with respect to a predetermined selection criteri-on. The resulting clusters of objects should then exhibithigh internal (within-cluster) homogeneity and highexternal (between clusters) heterogeneity. Hierarchicalagglomerative clustering is the most common approach,which provides intuitive similarity relationships be-tween any one sample and the entire data set, and istypically illustrated by a dendrogram (tree diagram)(McKenna 2003). The dendrogram provides a summaryof the clustering processes, presenting a picture of thegroups and their proximity, with a strong reduction indimensionality of the original data. In this study, hierar-chical agglomerative CA was performed by means ofthe Ward’s method, using squared Euclidean distancesas a measure of similarity. The spatial variability ofwater quality in the whole river basin was determinedfrom CA, using the linkage distance, reported asDlink/Dmax, which represents the quotient between the

Table 1 General water quality parameters, units, analytical techniques and basic statistics (range, mean and S.D) recorded in 78monitoring stations along the Danube River during Joint Danube Survey 2 (JDS2)

Parameter Unit Abbreviation Analytical technique Min.–Max. Mean±S.D.

Secchi-depth m SD White disc of 25 cm diameter 0.20–1.65 0.89±0.343

Photosynthetic availableradiation

μmol m−2 s−1 PAR 2 π flat Li-Cor sensor 20–3824 600±571.258

Total suspended solids mg L−1 TSS Filtration/drying at 105 °C/weighing

3.90–87.60 19.35±17.934

Organic suspended solids mg L−1 OSS Calculation after loss of ignition 1.20–12.40 3.62±2.128

Dissolved oxygen (absolutepercentage deviation fromoxygen saturation)

% DO% In situ 0.20–60.70 8.51±9.823

pH – pH In situ 7.32–8.10 7.76±0.156

Total alkalinity mmol L−1 Alk Titration 2.30–4.40 2.71±0.315

Electrical conductivityat 20 °C

μS cm−1 EC In situ 346–577 387±35.110

N–ammonium mg L−1 N N–NH4 Spectrophotometry, sodiumdichloroisocyanurate

0.016–0.379 0.041±0.052

N–nitrite mg L−1 N N–NO2 Spectrophotometry, sulphanilicacid and naptyl-ethylenediamine

0.005–0.072 0.020±0.018

N–nitrate mg L−1 N N–NO3 Spectrophotometry, salicylic acid 0.780–3.120 1.661±0.362

P-ortho-phosphates mg L−1 P P-PO4 Spectrophotometry, ascorbic acid 0.010–0.093 0.036±0.017

Total phosphorous mg L−1 P TP ICP-AES 0.060–0.910 0.109±0.102

Phytoplankton biomass mg L−1 BM Derived from bio-volume 0.150–6.960 1.421±1.566


linkage distances for a particular case divided by themaximal linkage distance. The quotient is then multi-plied by 100 as a way to standardise the linkage distancerepresented on the y-axis (Wunderlin et al. 2001;Simeonov et al. 2003; Singh et al. 2004, 2005).

Results and discussions

Spatial pattern of the selected variables

Besides the main output of this paper—selection of thesampling sites located along the main course of theDanube River in which the general physico-chemicalmeasurements are going to be carried out in the nextJDS (2013)—designing of the sampling strategy reliesalso on the general aspect of the spatial distribution ofthe selected parameters. Therefore, a brief descriptionof this distribution is given in the following: physicalvariable photosynthetic available radiation (PAR)which strongly depended on cloudiness conditionshad a relative constant distribution along the most partof the Danube, with higher values observed in theupper reach, where sunny conditions predominated.Secchi depth visibility (SD) directly reflected the totalsuspended solids (TSS) content in the water column:as the TSS concentration increased from upper tolower Danube, the SD decreased accordingly. ThepH and DO % deviation had similar longitudinal pro-files, demonstrating the balanced effect of the primaryproduction and decomposition of organic matter, or, inother words, healthy conditions in this particular eco-system. Conductivity values followed a quasi-constantline from upper to lower stretch, while alkalinity rep-resented the normal range of buffered capacity waters.Nutrients levels showed different patterns related toindividual forms: N–NH4 presented a relatively con-stant profile along the Danube River, with the maxi-mum concentration peak located in the Iron Gatesreservoir backwaters. N–NO2 had a decreasing linein the upper course of the Danube, a uniform profile inthe middle one, followed by a peak in the Iron Gatesreservoir, similar with the N–NH4, but slightly shiftedto the headwater of the reservoir and downstream thedam; since nitrites are the intermediary product in theoxidation/reduction process of nitrogen forms, its var-iation pattern in this area reflected the specific behav-iour of the nitrogen species in a big reservoir. In thelower Danube, N–NO2 decreased probably due to the

dilution effect. A significantly decreasing profile ofN–NO3 concentrations from upper to middle and low-er Danube reaches was noticed, with distinctive vari-ations in the side arms, caused by specific summerpattern in shallow waters (uptake by biological activ-ity and increased organic pollution). A strong decreasein P–PO4 concentrations was observed in the upperDanube; in the middle stretch, a slight increasing linewas noticed, with rather higher values measured in theside arms. A slightly increasing profile was present inthe lower reach, mainly caused by discharges ofmunicipal waste waters containing non-free P deter-gents. Total phosphorous showed a general increas-ing profile from upper to lower Danube. As regardsthe selected biological variable, the biomass devel-opment showed low values in the upper and lowerDanube, while an elevated profile was noticed in themiddle stretch. More detailed description of the spa-tial distribution of physico-chemical and related bio-logical variables measured during JDS2 could befound in Liška et al. (2008).

Since the description of the spatial pattern of themeasured variables offered a scattered distribution, whichrelatively depended on individual variable, the use ofchemometrics applied on the entire selected dataset ismore appropriate for designing of the JDS3 samplingstrategy for general physico-chemical parameters.

Multivariate statistics

The correlation matrix for the 14 selected variables isshown in Table 2. Statistically significant correlations(p<0.05) were found between variables showing causes(nutrient inputs), favourable conditions (PAR), algalresponse of the primary production process (phyto-plankton biomass) and effects of DO % deviation(pH, alkalinity, turbidity, dissolved oxygen content):DO % deviation showed positive correlation with N–NH4 (0.55) and N–NO2 (0.24) and negative with N–NO3 (−0.33). Relatively fair correlations were foundbetween acidification conditions (pH and alkalinity)and nutrients forms: negative coefficients between pHand N–NH4 (−0.39), N–NO2 (−0.45) and P–PO4

(−0.33), alkalinity and N–NO2 (−0.29); positive co-efficients between pH and N–NO3 (0.35), alkalinityand P–PO4 (0.40), while the strongest correlation wasnoticed between the alkalinity and N–NO3 (0.87). Therelationship between the phytoplankton biomass andnutrients input was reflected by negative correlation


coefficients: −0.22 with N–NO2, −0.26 with N–NO3

and −0.28 with P-PO4. Since the concentration of theTSS and its organic component OSS are directly relat-ed to Secchi depth visibility, these two variables werehighly significant correlated (−0.75 and −0.62, respec-tively). Similar correlation pattern was noticed be-tween pH and DO% (−0.52) and algal biomass andOSS (0.51). Weaker correlations were found amongnutrients forms: negative coefficients between N–NO3

with N–NH4 and N–NO2 (−0.30 and −0.37, respec-tively), as result of nitrification process and positivebetween P–PO4 with N–NH4 and N–NO2 (0.33 and0.38, respectively).

The correlation matrix was further used to applyPCA in order to generate new sets of uncorrelatedvariables, the principal components (PCs). The scree-plot of eigenvalues and percentage variances versusnumber of components is shown in Fig. 1a while theprojection of the 14 variables on the first two mainfactors (or axes) plane (F1×F2), explaining 43.16 %of the total variance in the dataset is shown in Fig. 1b.For the first axis (explaining 21.81 % of variance) thehighest negative contribution belongs to N–NO3 andalkalinity (−0.90 and −0.81, respectively). Axis 2 rep-resents 21.35 % of the global data and shows thepositive contribution of Secchi depth (0.75) and thenegative contributions of TSS and OSS (−0.79 and−0.84), respectively. In order to simplify even more

the data structures resulting from PCA, factor analysis(FA) was applied to reduce the contribution of lesssignificant variables. According to the eigenvalue-onecriterion, only eigenvalues greater than 1 are consideredas important ones; this criterion is based on the fact thatthe average eigenvalues of the auto-scaled data is one(Kowalkowski et al. 2006). Therefore, the first five PCswere used for further analysis, explaining a total vari-ance of 75.52 %. By a varimax rotation (raw) of the PCscoming from the original standardised data set, newgroups of variables, called VFs were obtained. If a PCis a linear combination of measured water quality vari-ables, a VF can include unobservable, hypothetical,“latent” variable (Chapman 1996; Otto 1998; Vega etal. 1998; Helena et al. 2000; Wunderlin et al. 2001). Thefactor loadings matrix for the five PCs of the dataset isgiven in Table 3 (factor loadings greater than 0.700 weremarked as significant), from which it can be seen thatthe variables associated with each factor are well loadedand contribute in a very small extent to the other factors.Thus, factor 1 (VF1), which explains 19.3 % of the totalvariance, is related to alkalinity and N–NO3, both in-dicators highly negatively loaded with this factor(−0.923 and −0.871); it could be termed as “nitrates”factor. Factor 2 (VF2), explaining 19.3 % of the vari-ance, positively loaded with the total and organicsuspended solids (0.907 and 0.869, respectively) andnegatively with Secchi depth could be named as

Table 2 Correlations matrix for 14 water quality selected variables

SD PAR TSS OSS DO% pH Alk EC NNH4 NNO2 NNO3 PPO4 TP BM

SD 1.00

PAR −0.14 1.00

TSS −0.75* 0.01 1.00

OSS −0.62* −0.02 0.79* 1.00

DO% 0.24* −0.26* −0.19 −0.12 1.00

pH 0.08 0.16 −0.16 0.02 −0.52* 1.00

Alk 0.10 0.46* −0.25* −0.18 −0.15 0.22 1.00

EC −0.13 0.19 0.07 0.11 0.09 −0.13 0.38* 1.00

NNH4 0.27 −0.15 −0.24* −0.26* 0.55* −0.39* −0.06 −0.00 1.00

NNO2 0.09 0.03 −0.02 −0.17 0.24* −0.45* −0.29* 0.06 0.20 1.00

NNO3 0.06 0.40* −0.13 −0.13 −0.33* 0.35* 0.87* 0.38* −0.30* −0.37* 1.00

PPO4 0.00 0.18 −0.06 −0.10 0.04 −0.33* 0.40* 0.33* 0.33* 0.38* 0.20 1.00

TP −0.22 −0.04 0.13 0.08 −0.10 −0.13 −0.14 0.01 −0.04 −0.06 −0.14 0.15 1.00

BM −0.10 −0.12 0.09 0.51* 0.18 0.05 −0.19 0.08 −0.09 −0.22* −0.26* −0.28* −0.00 1.00

*p<0.05 (correlations are significant)


“turbidity factor”. The third factor (VF3), explaining16.8 % of variance, positively loaded with DO % abso-lute deviation (0.812) and negatively with pH (−0.719)may be taken as “feedback effect”. The fourth factor

(VF4), explaining 12.1 % of variance and positivelyloaded with biomass (0.740) could be assigned to “algalresponse”. Factor 5 (VF5), explaining 7.9 % of varianceis highly negatively loaded with total phosphorous

Fig. 1 Principal Component Analysis results: a Scree-plot of eigenvalues and percentage variances and b Projection of the variables onthe factor-plane (F1×F2)

Table 3 Factor (varifactors) loadings matrix (after varimax rotation) for selected water quality parameters analysed in water samplesfrom the Danube River during JDS2

Variable VF 1a VF 2a VF 3a VF 4a VF 5a

SD 0.012 −0.848 0.128 0.062 0.174

PAR −0.541 0.147 −0.225 −0.395 0.171

TSS 0.131 0.907 −0.072 −0.064 −0.057OSS 0.058 0.869 0.011 0.326 0.021

DO % 0.174 −0.193 0.812 0.145 0.146

pH −0.178 −0.157 −0.719 0.348 0.128

Alk −0.923 −0.178 −0.082 0.008 0.030

EC −0.642 0.252 0.380 0.041 0.011

N-NH4 0.085 −0.328 0.690 −0.100 −0.078N-NNO2 0.250 0.029 0.386 −0.620 0.198

N–NO3 −0.871 −0.107 −0.317 0.036 0.055

P–PO4 −0.468 0.008 0.413 −0.523 −0.272TP 0.075 0.120 −0.007 −0.006 −0.935BM 0.124 0.300 0.242 0.740 0.123

Eigenvalue 2.707928 2.700328 2.353859 1.696925 1.113871

Explained variance 0.193423 0.192881 0.168133 0.121209 0.079562

Significant factor loadings>0.700 are in bolda VF (varifactor)=principal components factor from standardised data after varimax raw rotation


(−0.935) and could be named “Phosphorous factor” andassigned to the primary production under the existing P-based nutrients conditions.

By the representation of the factor scores versusmonitoring sites, the “most relevant” locations couldbe identified. In Fig. 2 (a–e) high scores correspondingto the high influence of the respective factor on thesampling sites are presented. Factor 1 (Fig. 2a) wasstrongly significant to sites JDS1–JDS5, in the upperDanube between river km 2,600 (Upstream Iller) andriver km 2,278 (Niederalteich), for which rather ele-vated N–NO3 and alkalinity values were measured(2.55–3.12 mg L−1 N–NO3 and 3.5–4.4 mmol.L−1).Factor 2 (Fig. 2b) reflected the turbidity conditionsrecorded along several stretches: JDS22–JDS44, in themiddle Danube between river km 1,761 (Iza/\Szony) andriver km 1355 (Dalj) and JDS52–JDS67, between riverkm 1,159 (Upstream Pancevo/Downstream Sava) andriver km 834 (Pristol/Novo SeloHarbour), corresponding

to Secchi depth values in the interval 1.00–1.52 m andbetween 0.90 and 1.65 m. The influence of this factor wasalso visible in the sites JDS16—river km 1,869(Bratislava), JDS19—river km 1794 (/Moson DanubeArm-end), JDS47—river km 1,252 (DownstreamNovisad) and JDS87–JDS96, in the lower Danube be-tween river km 295 (Upstream Cernavoda) and riverkm 0 (Sf.Gheorghe arm), for which TSS and OSS variedin the intervals 15.73–87.60 mg L−1 and 5.40–10.00 mg L−1, respectively. Factor 3 (Fig. 2c) identifiedthe locations JDS22—river km 1,761 (Iza/\Szony),JDS31—river km 1,642 (Rackeve-Soroksar DanubeArm—start), JDS34—river km 1,586 (Rackeve-SoroksarDanube Arm—end), JDS50—river km 1,200(Downstream Tisa/Upstream Sava) and the stretchJDS54–JDS68—river km 1,132 (Grocka)—riverkm 795 (Calafat), sites for which DO % absolute devia-tion from equilibrium ranged between 12.2 and 60.7 %.Factor 4 (Fig. 2d) highlighted the sites JDS46–JDS50,

Fig. 2 Factor scores versus JDS2 monitoring sites from the main course of the Danube River: a “Nitrates” factor; b “Turbidity” factor;c “Feedback effect” factor; d “Algal response” factor; e “Phosphorous” factor


between river km 1,262 (Upstream Novisad) and riverkm 1,200 (Downstream Tisa/Upstream Sava) andJDS62–JDS75, between river km 954 (Iron Gate reser-voir) and river km 602 (Downstream Olt), stretches inwhich high biomass and N–NO2 values were measured(6.05–6.96 mg L−1 and 0.047–0.072 mg L−1 N–NO2).The influence of factor 5 (Fig. 2e) is visible in monitoring

sites JDS82—river km 488 (Downstream Ruse/Giurgiu)and JDS83—river km 434 (Upstream Arges), in whichhigh total phosphorous concentrations were recorded(0.910 and 0.410 mg L−1 P, respectively).

Based on the significant factors obtained byPCA/FA—variables linked to both causes (nutrientloading) and effects (biomass, turbidity, pH, alkalinity,

Fig. 3 Dendrogram showing the cluster analysis results based on factors identified by PCA/FA for the 78 JDS2 monitoring siteslocated on the main course of the Danube River

Table 4 JDS2 monitoring sites (on the Danube River) grouped according to cluster analysis results and proposal for sites which couldbe excluded from the JDS3 sampling network (in brackets and bold character—total number of sites in each cluster)

Cluster JDS2 monitoring site number in clustera JDS2 monitoring site proposed to be excluded

C1 (25) 7–18, 20, 23–30, 32–33, 35–36, 52–53, 83 (25) 9, 11, 23, 27, 28, 30, 33 (7)

C2 (12) 68–80, 85–86 (12) 68, 70, 72, 73, 76, 80 (6)

C3 (15) 22, 31, 34, 54–67 (15) 55, 58, 61, 64 (4)

C4 (8) 87–96 (8) 87, 88, 89, 94 (4)

C5 (11) 19, 38–48 (11) 40, 44 (2)

C6 (5) 1–5 (5) 4, 5 (2)

C7 (2) 50, 82 (2) –

Total number of sites 78 25

a Numbers in brackets and bold character—total number of sites in each cluster


dissolved oxygen deviation from equilibrium)—clusteranalysis (CA) was further applied in order to explore thespatial similarities among the sampling sites located onthe Danube River. The resulting dendrogram ispresented in Fig. 3, in which the 78 monitoring siteslocated on the main course of the Danube River weregrouped in seven clusters (C1–C7). Starting from thestatistically significant clusters at (Dlink/Dmax) x100,several specific river parts or highlighted “hot spots”are presented in Table 4 as JDS2 monitoring site num-ber. Based on sequence in the association of sites in eachcluster, sampling network of the next Joint DanubeSurvey can be designed in a pragmatic manner, whileensuring the data comparability. Thus, starting from thecontinuous stretches (underlined in Table 4), sites fromeach cluster can be chosen to be excluded from theJDS3, keeping the distinctive (singular) siteshighlighted both by CA and factor scores of FA,as it follows: C1 (25 sites) comprised 8 sites fromthe upper Danube (JDS7–JDS14), 16 sites from themiddle Danube (JDS18, 20, 23–30, 32–33, 35–36, 52–53) and 1 (one) site from the lower Danube (JDS83), thelast site (Upstream Arges) being also visible in Fig. 2e;within this cluster, 7 sites are proposed to be excluded(see column 3 from Table 4) as sites being locatedupstream a dam (JDS9—Upstream dam Abwinden-Asten, JDS11—Upstream dam Greifenstein), upstreamend of an island or side arm (JDS27—Upstream end ofSzentendre Island, JDS28—Upstream end of SzentendreIsland—arm, JDS30—end of Szentendre Island) and sitesfor which the measured variables are highly comparable(JDS23—Sturovo/Esztergom and JDS33—Adony/Lórév).C2 included 12 sites from the lower Danube (JDS68–JDS80, JDS85 and JDS86). From this cluster, six sitesare proposed to be taken out, as being located up-stream a tributary or a point source or characterised byhighly homogenous similarity within the cluster(JDS68—Calafat, JDS70—Upstream Iskar, JDS72—Downstream Iskar, similar to JDS73—Upstream Olt andJDS76—DownstreamTurnu-Magurele/Nikopol, similar toJDS80—Upstream Ruse). C3 (15 sites) grouped the siteslocated in the Iron Gate reservoir area (JDS54–JDS67), aswell as distinctive sites JDS22 (Iza/\Szony), JDS31(/Rackeve-Soroksar Danube Arm—start) and JDS34(/Rackeve-Soroksar Danube Arm—end), also distin-guished in Fig. 2c. Four sites are proposed to be excludedfrom this cluster based on the following: located upstreama tributary (JDS55—Upstream Velika Morava), locate tooclose to another site (JDS58—Starapalanka Ram is only 2

km near to JDS59—Banatska Palanka/Baziaswhich is aninternational border), highly similarity within the cluster(JDS61—Donji Milanovac and JDS64—Iron Gate II,similar to JDS59 and JDS65, respectively). C4 comprisedthe last eight sites of the lower Danube (JDS87–96),from which four sites are proposed to be taken out:located upstream a point source and located on a sidecanal (JDS87—Upstream Cernavoda, JDS88—Giurgeni,JDS89—Braila and JDS94—Bystroe canal). C5 (11 sites)included JDS19 (/Moson Danube Arm-end)—identifiedalso in Fig. 2b and a river stretch (JDS38—JDS48) locatedin themiddle Danube, stretch fromwhich JDS40—Batinaand JDS44—Dalj (similar to JDS43—DownstreamDrava and JDS39—Hercegszanto, respectively) couldbe excluded without loss of information. C6 included thefirst five sites located in the upper Danube (JDS1–JDS5),highlighted also in Fig. 2a. Based on the homogeneitywithin the cluster, JDS4—Daggendorf and JDS5—Niederalteich (similar to JDS3—Geisling power plantand JDS2—Kelheim, respectively) are proposed to beexcluded. The maximum linkage distance was reachedby cluster C7, comprising JDS50—Downstream Tisa/Upstream Sava and JDS82 –Downstream Ruse/Giurgiu.Given the specificity of these two sampling sites(highlighted also by Fig. 2c–e), no exclusion is proposedfrom this cluster. As it can be seen, the number of exclud-ed sites gradually decreased with increasing of the linkagedistance, Dlink/Dmax Based on the results obtained byapplication of CA, the total number of sites that could beskipped for the general physico-chemical measurementsduring the JDS3–2013 is 25, representing a reduction of32 %, percentage that can lead to a corresponding costsreduction.

Conclusions

In this study, water quality data for 14 selected sup-portive physico-chemical and biological indicatorswere investigated by means of multivariate statisticaltechniques in order to design the JDS3-2013 samplinglist in which the general physico-chemical variablesare going to be measured. The dataset used in thispaper comprised results from 78 sites located alongthe main course of the Danube River. The correlationmatrix between the biological and physico-chemicalindicators provided statistically significant correla-tions (p<0.05) between causes (nutrients), responseindicator (phytoplankton biomass) and secondary


effects (dissolved oxygen absolute deviation, pH andturbidity as direct result of suspended solids content).Results from PCA/FA indicated that most of thedataset variance (above 75 %) was gathered by fivevarifactors described by water quality indicators main-ly loaded with nutrient inputs, feedback effects ofbiological activity and algal response. Therefore, theproposed set for physico-chemical variables to bemeasured during the JDS3-2013 consists of: pH, alka-linity, dissolved oxygen, transparency by Secchidepth, total suspended solids, N–nitrates and totalphosphorous.

Starting from the five varifactors obtained from FAand the representation of the factor scores versus mon-itoring sites, the most relevant locations related to theinfluence of each factor could be identified and furtherconfirmed by CA. Based on the detected similaritygroups among the sampling sites as resulted in theseven clusters showed by CA, the spatial structure ofthe next survey could be established in such a way thatonly representative sites from each cluster as well asdistinctive hot spots can be included. Thus, from thetotal number of 78 sites, 25 were proposed to beeliminated from JDS3 sites in which supportivephysico-chemical variables are going to be measured.The approach of target oriented sampling strategybased on the previous data offers a good possibilityto significantly adjust of the number of analysis andthe corresponding costs.

The multivariate statistics techniques applied tophysico-chemical and selected biological qualityelements from Joint Danube Survey 2 provideduseful information regarding the opportunity toalternatively investigate the monitoring datasets inorder to design the next Joint Danube Survey(JDS3-2013). However, one has to be aware ofthat the outcomes of this paper are valid for the JDS-type investigation only and cannot replace the currentlong-term monitoring strategy at the Danube basin-widelevel.

Acknowledgment This work would not have been possiblewithout numerous people. The authors of this paper wish toexpress their gratitude to all of them, but especially to the JDS2Core Member Teams as well as to the National Teams through-out the Danube Basin for their strong support and work teamspirit. Special appreciation is given to Professor Martin T.Dokulil (Institute for Limnology, Austrian Academy of Sci-ences) and Christina M. Kaiblinger, members of the JDS2 CoreTeam, who kindly provided the biological data.

References

Adams, M. J. (1988). The principles of multivariate data anal-ysis. In P. R. Ashurst & M. J. Dennis (Eds.), Analyticalmethods of food authentication (p. 350). London: BlackieAcademic & Professional.

Astel, A., Biziuk, M., Przyjazny, A., & Namieśnik, J. (2006).Chemometrics in monitoring spatial and temporal varia-tions in drinking water quality. Water Research, 40, 1706–1716.

Battegazzore, M., & Renoldi, M. (1995). Integrated chemicaland biological evaluation of the quality of the riverLambro (Italy). Water, Air, and Soil Pollution, 83, 375–390.

Bengraine, K., & Marhaba, T. F. (2003). Using principal com-ponent analysis to monitor spatial and temporal changes inwater quality. Journal of Hazardous Materials, B, 100,179–195.

Birk, S., & Hering, D. (2009). A new procedure for comparingclass boundaries of biological assessment methods: a casestudy from the Danube Basin. Ecological Indicators, 9(3),528–539.

Bouza-Deaño, R., Ternero-Rodríguez, M., & Fernández-Espinosa, A. J. (2008). Trend study and assessment ofsurface water quality in the Ebro River (Spain). Journalof Hydrology, 361, 227–239.

Bu, H., Tan, X., Li, S., & Zhang, Q. (2010). Temporal andspatial variations of water quality in the Jinshui River ofthe South Qinling Mts., China. Ecotoxicology and Envi-ronmental Safety, 73, 907–913.

Chapman, D. (1996). Water quality assessment: a guide to useof biota, sediments and water in environmental monitoring.In: D. Chapman, D. (Ed.) on behalf of UNESCO, WHOand UNEP, Chapman and Hall, London, 585 pp.

Dokulil, T., M. and Kaiblinger, M., C. (2008). Phytoplankton.In: Liška I., Wagner, F., Slobodnik, J. (Ed.) Joint DanubeSurvey 2, Final Scientific Report. ICPDR—InternationalCommission for the Protection of the Danube River, Vien-na (Austria), pp. 242.

EuropeanCommission. (2000). Directive 2000/60/EC of the Eu-ropean Parliament and of the Council of 23 October 2000establishing a framework for Community action in the fieldof water policy. Official Journal of European Community.

Gangopadhyay, S., Gupta, A. D., & Nachabe, M. H. (2001).Evaluation of ground water monitoring network by princi-pal component analysis. Ground Water, 39, 181–191.

Greenberg, A.E., Clesceri, L.S., Eaton, A.D. ( 2005). Standardmethods for the examination of water and wastewater, 21stedn. American Water Works Association, pp, 1368 ISBN:0875530478.

Hamchevici, C. and Craciun, M. (2008). Full report on generalphysico-chemical quality elements (nutrients). In: Liška I.,Wagner, F., Slobodnik, J. (Ed.) (2008). Joint Danube Sur-vey 2, Final Scientific Report. ICPDR—InternationalCommission for the Protection of the Danube River, Vien-na (Austria), pp. 242.

Hamchevici, C. and Udrea, I. (2013). In: I. Liška (ed) TheDanube River Basin, Pollution by Nutrients in the DanubeRiver Basin. Series: The Handbook of EnvironmentalChemistry, Springer (in press).


Helena, B., Pardo, R., Vega, M., Barrado, E., Fernandez, J. M.,& Fernandez, L. (2000). Temporal evolution of groundwa-ter composition in an alluvial aquifer (Pisuerga RiverSpain) by principal component analysis. Water Research,34, 807–816.

Huang, F., Wang, X., Lou, L., Zhou, Z., &Wu, J. (2010). Spatialvariation and source apportionment of water pollution inQiantang River (China) using statistical techniques. WaterResearch, 44, 1562–1572.

Janauer, G. A., Schmidt-Mumm, U., & Schmidt, B. (2010).Aquatic macrophytes and water current velocity in theDanube River. Ecological Engineering, 36(9), 1138–1145.

Johnson, R.A., Wichern, D.W. (2002). Applied MultivariateStatistical Analysis, 5th edn. ISBN 0-471-41889-7, NewYork, 708+22 pp.

Kazi, T. G., Arain, M. B., Jamali, M. K., Jalbani, N., Afridi, H.I., Sarfraz, R. A., et al. (2009). Assessment of water qualityof polluted lake using multivariate statistical techniques: acase study. Ecotoxicology and Environmental Safety, 72(2),301–309.

Kovalkovski, T., Zbytniewski, R., Szpejna, J., & Buszewski, B.(2006). Application of chemometrics in river water classi-fication. Water Research, 40, 744–752.

Li, S., & Zhang, Q. (2009). Geochemistry of the upper HanRiver basin, China 2: seasonal variations in major ioncompositions and contribution of precipitation chemistryto the dissolved load. Journal of Hazardous Materials,170, 605–611.

Liška I., Wagner, F., Slobodnik, J. (Ed.) (2008). Joint DanubeSurvey 2, Final Scientific Report. ICPDR—InternationalCommission for the Protection of the Danube River, Vien-na (Austria), pp. 242.

Literáthy, P., & László, F. (1999). Micropollutants in the DanubeRiver basin. Water Science and Technology, 40(10),17–26.

Literáthy, P., Koller-Kreimel, V. & Liska, I. (Ed.) (2002). JointDanube Survey—Technical Report of the ICPDR—International Commission of the Danube River, pp. 261.

Loos, R., Locoro, G., & Contini, S. (2010). Occurence of polarorganic contaminants in the dissolved water phase of theDanube River and its major tributaries using SPE-LC-MSanalysis. Water Research, 44, 2325–2335.

McKenna, J. E., Jr. (2003). An enhanced cluster analysis pro-gram with bootstrap significance testing for ecologicalcommunity analysis. Environmental Modelling and Soft-ware, 18, 205–220.

Mendiguchia, C., Moreno, C., Galindo-Riaño, M. D., & Garcia-Vargas, M. (2004). Using chemometric tools to assessanthropogenic effects in river water, a case study: Guadal-quivir River (Spain). Analytica Chimica Acta, 515, 143–149.

Otto, M. (1998). Multivariate methods. In R. Kellner, J. M.Mermet, M. Otto, & H. M. Widmer (Eds.), Analyticalchemistry. Weinheim: Wiley. 916 pp.

Ouyang, Y. (2005). Application of principal component andfactor analysis to evaluate surface water quality monitoringnetwork. Water Research, 39, 2621–2635.

Ouyang, Y., Nkedi-Kizza, P., Wu, C. T., Shinde, D., & Huang,C. H. (2006). Assessment of seasonal variations in surfacewater quality. Water Research, 40, 3800–3810.

Perkins, R. G., & Underwood, G. J. C. (2000). Gradientsof chlorophyll a and water chemistry along an eutro-phic reservoir with determination of the limiting nu-trient by in situ nutrient addition. Water Research, 34,713–724.

Razmkhah, H., Abrishamchi, A., & Torkian, A. (2010). Evalu-ation of spatial and temporal variation in water quality bypattern recognition techniques: a case study on JajroodRiver (Tehran, Iran). Journal of Environmental Manage-ment, 91, 852–860.

Shine, J. P., Ika, R. V., & Ford, T. E. (1995). Multivariatestatistical examination of spatial and temporal patterns ofheavy metals contamination in New Bedford Harbor ma-rine sediments. Environmental Science and Technology,29, 1781–1788.

Shrestha, S., & Kazama, F. (2007). Assessment of surface waterquality using multivariate statistical techniques: a casestudy of the Fuji river basin, Japan. Environmental Model-ing and Software, 22(4), 464–475.

Simeonov, V., Stratis, J. A., Samara, C., Zachariadis, G., Voutsa,D., Anthemidis, A., et al. (2003). Assessment of the surfacewater quality in Northern Greece. Water Research, 37,4119–4124.

Singh, K. P., Malik, A., Mohan, D., & Sinha, S. (2004).Multivariate statistical techniques for the evaluation ofspatial and temporal variations in water quality ofGomti River (India): a case study. Water Research, 38,3980–3992.

Singh, K. P., Malik, A., & Sinha, S. (2005). Water quality assess-ment and apportionment of pollution sources of Gomti river(India) using multivariate statistical techniques: a case study.Analytica Chimica Acta, 538, 355–374.

Slobodník, J., Hamchevichi, C., Liška, I., Shearman, A., Csányi,B., Makovinská, J. et al. (2005). Final report on sam-pling, chemical analysis and ecotoxicological studies.Project no. 505428 (GOCE), AquaTerra—IntegratedModelling of the river-sediment-soil-groundwater system;advanced tools for the management of catchment areasand river basins in the context of global change, Integrat-ed Project, Thematic Priority: Sustainable development,global change and ecosystems, Deliverable No.: BASIN5.11.

Sommerwerk, N., Baumgartner, C., Bloesch, J., Hein, T.,Ostojić, A., Paunović, M.et al. (2009) In: Rivers of Europe,The Danube River Basin, 59–112. ISBN: 978-0-12-369449-2.

Tauler, R., Barcelo, D., & Thurman, E. M. (2000). Multivariatecorrelation between concentrations of selected herbicidesand derivatives in outflows from selected US Midwesternreservoirs. Environmental Science and Technology, 34,3307–3314.

Vega, M., Pardo, R., Barrado, E., & Deban, L. (1998). Assess-ment of seasonal and polluting effects on the quality ofriver by exploratory data analysis. Water Research, 32,3581–3592.

VesnaMicić, V., & Hofmann, T. (2009). Occurrence and behav-iour of selected hydrophobic alkylphenolic compounds inthe Danube River. Environmental Pollution, 157(10),2759–2768.

Voutsa, D., Manoli, E., Samara, C., Sofoniou, M., & Stratis, I.(2001). A study of surface water quality in Macedonia,


Greece: speciation of nitrogen and phosphorus. Water, Air,and Soil Pollution, 129, 13–32.

Woitke, P., Wellmitz, J., Helm, D., Kube, P., Lepom, P., &Literáthy, P. (2003). Analysis and assessment of heavymetal pollution in suspended solids and sediments of theriver Danube. Chemosphere, 51(8), 633–642.

Wunderlin, D. A., Diaz, M., Ame, M. V., Pesce, S. F., Hued, A.C., & Bistoni, M. (2001). Pattern recognition techniques

for the evaluation of the spatial and temporal variations inwater quality. A case study: Suquia River Basin (Cordoba-Argentina). Water Research, 35, 2881–2894.

Yu, J. C., Quinn, J. T., Dufournaud, C. M., Harrington, J. J.,Roger, P. P., & Lohani, B. N. (1998). Effective dimension-ality of environmental indicators: a principal componentanalysis with bootstrap confidence intervals. Journal ofEnvironmental Management, 53, 101–119.


Documents

art%3A10.1007%2Fs10661-013-3268-2