Application of chemometric methods for assessment and modelling of microbiological quality data concerning coastal bathing water in Greece

Abstract

Background

Worldwide, the aim of managing water is to safeguard human health whilst maintaining sustainable aquatic and associated terrestrial, ecosystems. Because human enteric viruses are the most likely pathogens responsible for waterborne diseases from recreational water use, but detection methods are complex and costly for routine monitoring, it is of great interest to determine the quality of coastal bathing water with a minimum cost and maximum safety.

Design and methods

This study handles the assessment and modelling of the microbiological quality data of 2149 seawater bathing areas in Greece over 10-year period (1997-2006) by chemometric methods.

Results

Cluster analysis results indicated that the studied bathing beaches are classified in accordance with the seasonality in three groups. Factor analysis was applied to investigate possible determining factors in the groups resulted from the cluster analysis, and also two new parameters were created in each group; VF1 includes E. coli, faecal coliforms and total coliforms and VF2 includes faecal streptococci/enterococci. By applying the cluster analysis in each seasonal group, three new groups of coasts were generated, group A (ultraclean), group B (clean) and group C (contaminated).

Conclusions

The above analysis is confirmed by the application of discriminant analysis, and proves that chemometric methods are useful tools for assessment and modeling microbiological quality data of coastal bathing water on a large scale, and thus could attribute to effective and economical monitoring of the quality of coastal bathing water in a country with a big number of bathing coasts, like Greece.

Significance for public health

The microbiological protection of coastal bathing water quality is of great interest for the public health authorities as well as for the economy. The present study proves that this protection can be achieved by monitoring only two microbiological parameters, E. coli and faecal streptococci/enterococci instead four microbiological parameters (the two mentioned above plus Total coliforms and Faecal coliforms) that are usually monitored today. As a consequence, countries, especially those with large quantities of coastal bathing sites, can perform microbiological monitoring of their bathing waters by checking only the mentioned two parameters, thus ensuring economies of scale. Thus, funds can be used in other actions to preserve the quality of coastal water and human health. This in turn, would aid in the assessment of the quality of coastal bathing waters and provide a more timely indication of bathing water quality, hence contributing to the immediate health protection of bathers.

Introduction

In 2010, EU Member States reported 21.063 bathing waters, of which 70% are coastal bathing waters. The majority of coastal bathing waters are located at Mediterranean Sea coasts (about 9900), representing almost two thirds of all reported coastal bathing waters in Europe. A total of 24 countries reported coastal bathing waters. Italy (4896), Greece (2149), France (2012), Spain (1930), Denmark (1054), Croatia (913) and the United Kingdom (596) have the highest number of coastal bathing waters.

The quality of bathing waters is of great importance for several reasons. Contaminated (unclean) water is a major hazard to bathers (causing gastric and skin problems). For the tourist industry, clean and safe water is also a major factor in attracting visitors to an area. Based on risk assessments from the World Health Organization (WHO) and academic research sources, studies suggest that millions of gastrointestinal and severe respiratory diseases are caused by swimming and bathing in wastewater-polluted coastal waters.1-11

No indicator has proven perfect for controlling coastal water quality. Indicator bacteria including total coliforms, faecal coliforms, Escherichia coli and streptococci/enterococci have been used over time for the assessment of water quality and risk assessment in the prediction of water microbial pollution. Research supports use of E. coli and enterococci rather than the broader group of faecal coliforms as indicators of microbiological pollution. Besides their limitations, these indicator bacteria have been used successfully in many countries as a monitoring tool for microbiological contamination of water and prediction of the presence of pathogens.12-23

The aim of the present study is to analyse a large number of numerical data that concern measurements of four microbiological quality indicators of seawater (total coliforms, faecal coliforms, E. coli and faecal streptococci/enterococci) over a 10-years period using chemometric methods such as cluster analysis, factor analysis and discriminant analysis for the assessment and modelling of these data. The extraction of successful models is of great importance for effective monitoring of coastal bathing water, allowing economies of scale without compromising the health of swimmers.

Design and methods

Study areas and sampling programs

Beaches were sampled on a regular basis with an average of 13 samples collected from each beach per year, from predetermined points specified by the competent department of the Hellenic Ministry for the Environment, Physical Planning and Public Works.

Sample collection and testing

Water samples from regularly monitored beaches were taken from areas of beaches which are most frequently used by bathers. The beaches are mainly visited by bathers during June to September with the highest counts of visitors noted during July and August. During the other months there is minimal to zero visiting of the seawater bathing areas for bathing purposes. Consequently, between the months of October and April only a few water samples were collected from the coastal bathing areas. Water samples were therefore collected from May through to November. The time of sampling was almost the same for each particular beach every time. The majority of samples were taken between 10.30 (a.m.) and 17.30 (p.m.) as this was considered to be the time at which the majority of people engaged in water activities.

A volume of 450 mL of water was collected in sterile bottles of 500 mL capacity. Samples were taken 20-30 cm below the water surface level at locations with a sea depth of 0.8-1.3 meters. Samples were transferred to the laboratory on the same day of collection in a closed Esky cooler, thereby avoiding any disinfecting effect of sunlight and changes to microbial presence. All samples were processed within 24 hours after collection. The microbiological variables of the regular monitored bathing areas can be seen in Table 1.

The majority of the water samples were collected and analysed by a contracted main private laboratory. Due to the vast number of samples, over 40 public and private authorities were involved in the sampling operation and 11 public and private laboratories including the main contracted laboratory were involved in the testing of the samples. All laboratories processed samples for microbiological analysis in accordance with standard ISO methods for the detection and enumeration of E. coli (ECOL), faecal coliforms (FCOL), total coliforms (TCOL) and faecal streptococci/enterococci (STREPT).

Data collection and validation

Data included in the study were gathered from the archives of the Hellenic Ministry for the Environment, Physical Planning and Public Works and comprised microbiological test results and relevant information recorded during sampling of the regularly monitored coastal bathing areas. All data entries were subjected to data validation and any inaccuracies found in the database, due to data entry errors, were cross checked with result transcripts and corrected.

Statistical analysis

This study handles the assessment and modelling of the microbiological quality data of 2.149 seawater bathing areas in Greece over 10-year period (1997-2006) by chemometric methods.

Data consisting of the microbiological test results (four microbiological indicators per water sample: TCOL, FCOL, STREPT and E. coli) collected from the coastal bathing areas were built into a database.

Parameters distribution characteristics and data treatment

Most methods such as cluster analysis (CA) and factor analysis (FA) require variables to be at least column cantering and some of them as discriminant analysis (DA) require variables to conform to a normal distribution.

More specific, for temporal and parameter CA as also for FA, column standardization was performed. All parameters were also z-scale standardized (mean=0; variance=1) to minimize the effects of differences in measurement units and variance and render the data dimensionless. Consequently, each column had a mean of 0 and variance of 1. For spatial CA and DA in each temporal cluster, log transformation and column standardization were performed, thus each column had a mean of 0 and variance of 1. All the calculations and plots in the following sections were done with the SPSS 15.0.

Cluster analysis

Hierarchical CA, being the most common approach of CA, starts with each case in a separate cluster and joins the clusters together step by step until only one cluster remains.25-27 In this study, hierarchical CA was performed on the standardized data using Ward’s method with squared Euclidean distances as a measure of similarity.27-34

Discriminant analysis

DA constructs a discriminant function for each group of two or more naturally occurring groups as follows:24,33 where, i is the number of groups (G); ki is a constant inherent to each group; n is the number of parameters used to classify a set of data into a given group; and wij is the weight coefficient assigned by discriminant analysis to a given parameter (pij).

In this study DA was performed on standardized log-transformed data using the standard, forward stepwise and backward stepwise modes to evaluate both the temporal and spatial variations in water quality. The best discriminant functions for each mode were constructed considering the quality of the classification matrix and the number of parameters. The monitoring sites and periods were the grouping variables and the measured parameters were the independent variables.

Factor analysis

Although not commonly used in water quality analysis, several studies have used FA to identify primary sources of contamination. FA is also used to find associations between parameters so that the number of measured parameters can be reduced. Known associations are then used to predict unmeasured water quality parameters.35-38

In this study, the factor extraction was performed using the method of principal components through varimax rotation. Screeplot criterion was used for determining how many factors to use and how many to ignore, and in our study retains only those factors with eigenvalues more than 0.75. Factor rotation (varimax rotation method) was used to facilitate interpretation by providing simpler factor structure.

Results and discussion

Temporal similarity and period grouping (temporal cluster analysis)

Hierarchical CA (single linkage method of linkage, Euclidean distances as similarity measure, standardization of the input data) was used to study the temporal relationships. Temporal CA generated a dendrogram grouping the 6 months into three clusters at (Dlink/Dmax) ×100 <35 and the difference between the clusters was significant (Figure 1). As can be seen in Figure 1, the studied period is separated into three clusters as follows: Cluster 1: May; Cluster 2: Jun, Jul and Aug; Cluster 3: Sept and Oct.

The careful consideration of the content of the clusters (Figure 1) offers some interesting conclusions about the data classification. The first cluster (1st period) included May, cluster 2 (2nd period) comprised June-August and cluster 3 (3rd period) consisted of the two remaining months (September-October).

Therefore, specific patterns of the classified parameters could be offered: 1st period or Spring period pattern (May); 2nd period or Summer period pattern (June-August); 3rd period or Autumn period pattern (September and October).

Hence, the temporal variation in coastal bathing water quality was absolutely determined by local climate (in spring, summer and autumn) or hydrological conditions (dry and wet seasons) because the costal sea water quality was also related to pollution characteristics (such as discharge frequency and type).

Spatial similarity and site grouping (parameter cluster analysis)

Hierarchical CA on standardized data was also applied to reveal natural groupings (clusters) within the data set of the four microbiological parameters and to examine relationships between them. Parameter CA (Ward’s method of linkage, Squared Euclidean distance as similarity measure, standardization of the input data), was conducted for each temporal cluster (first, second and third period) and for the All samples data set, with the same results.

Parameter CA generated a dendrogram for each temporal cluster and for the All samples data set, grouping the four parameters into two clusters at (Dlink/Dmax) ×100 <35 and the difference between the clusters was significant. In Figure 2 the hierarchical dendrogram for the clustering of the determined microbiological parameters for all the studied costal sea areas of the 1st period is plotted (the other three dendrograms for the temporal clusters are similar).

More specific, from the dendrogram of Figure 2 it could be concluded that the parameters at each temporal cluster and the All samples group are separated into two similar clusters as follows: Cluster 1 (three parameters are included): ECOL, FCOL, TCOL; Cluster 2 (one parameter is included): STREPT.

Therefore, specific patterns of the classified microbiological parameters could be offered: Coliforms pattern (including ECOL, FCOL and TCOL); Streptococci pattern (including STREPT).

Factor analysis

Usually, the typical classification approach of clustering is accompanied by FA, which is a typical projection and modelling approach. FA was applied to standardized datasets (4 microbiological parameters) to examine differences between the three studied periods and the All samples group and moreover, to identify the latent factors.

Before conducting the FA, the Kaiser-Meyer-Olkin (KMO) and Bartlett’s sphericity tests were performed on the parameter correlation matrix to examine the validity of the FA. The KMO results for the three temporal clusters and All samples were 0.694; 0.703; 0.719 and 0.710 respectively, and those for Bartlett’s sphericity were 9.036; 1.2263; 8.175 and 27.765 (P<0.05), respectively, indicating that FA may be useful in providing significant reductions in dimensionality.

Based on the screeplot criterion, only the VFs with eigenvalues over 0.75 were considered significant.

FA yielded two VariFactors (VFs) for each of the above data sets (Spring, Summer, and Autumn period and All samples group), explaining 92.14%; 97.40%; 94.50% and 94.90% of the total variance, respectively.

Table 1 summarized the FA results comprising the loadings, eigenvalues, percentage of total variance (the loadings which absolute value was more than 0.7 of the total variance was significant).

In general, FA confirms the results obtained by CA. From Table 1 it is seen, that the pollution structure of the three periods was similar to that of the All samples group and almost the same in the degree of pollution.

The linkage between the microbiological parameters in the four groups is shaped as follows (Table 1).

First period:

VF1, which explained 72.73% of the total variance (TV), shows how high [0.8≤ Loading (L)] ECOL, FCOL and TCOL coincide with costal bathing areas having low STREPT (L=0.21) (this VF could be called Coliforms factor). Additionally, VF2 (19.41% of the TV) shows that high STREPT (L=0.98) is met in costal sea having low ECOL (L=0.15), FCOL (L=0.18) and TCOL (L=0.28) (Streptococci factor).

Second period:

VF1, which explained 75.85% of the TV, shows how high ECOL, FCOL and TCOL coincide with costal seas having low STREPT (L=0.16) (Coliforms factor) and VF2 (21.55% of the TV) shows that high STREPT (L=0.99) is met in costal bathing areas having low ECOL (L=0.10), FCOL (L=0.13) and TCOL (L=0.29) (Streptococci factor).

Third period:

VF1, which explained 74.17% of the TV, shows how high ECOL, FCOL and TCOL coincide with costal seas having low STREPT (L=0.19) (Coliforms factor) and VF2 (20.33% of the TV) shows that high STREPT (L=0.98) is met in costal bathing areas having low ECOL (L=0.12), FCOL (L=0.15) and TCOL (L=0.34) (Streptococci factor).

All samples:

VF1, which explained 74.15% of the TV, shows how high ECOL, FCOL and TCOL coincide with costal seas having low STREPT (L=0.18) (Coliforms factor) and VF2 (20.75% of the TV) shows that high STREPT (L=0.98) is met in costal bathing areas having low ECOL (L=0.12), FCOL (L=0.15) and TCOL (L=0.39) (Streptococci factor).

It is easily seen that the major groups of microbiological parameters interpreted by parameter CA for the studied groups (Figure 2) are also involved in the VFs loadings presented in Table 1. Thus, the classification scheme obtained by parameter CA is confirmed by FA. This confirmation is an important hint that the microbiological parameters tested are indeed related and form groups of similar indicative properties.

The degree of pollution in the monitoring sites for 1st, 2nd and 3rd period, according to the sp-cluster A, B and C defined by the 1st period, was as follows:

First period:

The degree of pollution (as the average value of VFs) in the monitoring sites for 1st period differed significantly among the three sp-clusters, and sites received more pollution from VF1 (ECOL, TCOL and FCOL) and VF2 (STREPT) in the sp-cluster C1 than in the spcluster A1 and B1, and more in the sp-cluster B1 than in the A1. Cases distributed in the region of larger values of VF1 were almost all collected from sp-cluster B1 and C1.

Second period:

The degree of pollution (as the average value of VFs) in the monitoring sites for 2nd period differed significantly among the three sp-clusters, and sites received more pollution from VF1 in the sp-cluster C1 than in the sp-cluster A1 and B1, and more in the sp-cluster B1 than in the A1. Moreover, sites received more pollution from VF2 in the sp-cluster C1 than in the sp-cluster A1 and B1, and more in the spcluster A1 than in the B1. Cases distributed in the region of larger values of VF1 were almost all collected from sp-cluster B1 and C1.

Third period:

The degree of pollution (as the average value of VFs) in the monitoring sites for 3rd period differed significantly among the three sp-clusters, and sites received more pollution from VF1 and VF2 in the sp-cluster C than in the sp-cluster A1 and B1, and more in the spcluster B1 than in the A1. Cases distributed in the region of larger values of VF1 were almost all collected from sp-cluster B1 and C1.

The degree of pollution in the monitoring sites for sp-clusters A, B and C defined by the 1st period, according to the three periods (1st, 2nd and 3rd), was as follows:

Sp-cluster A defined by the 1st period [or Group A (A1)]: The degree of pollution (as the average value of VFs) in the monitoring sites for group A differed significantly among the three periods, and sites received more pollution from VF1 (ECOL, FCOL and TCOL) and VF2 (STREPT) during the second (June-August) and the third (September-October) periods than in the first period (May). The factor scores for the three periods were not significantly regular or distinct. Cases distributed in the region of larger values of VF1 were almost all collected from second and third period.

Sp-cluster B defined by the 1st period [or Group B (B1)]: The degree of pollution (as the average value of VFs) in the monitoring sites for group B differed significantly among the three periods, and sites received more pollution from VF1 during the second and the third periods than in the first period. Moreover, sites received more pollution from VF2 during the first period than in the second and in the third periods, and more in the third period than in the second. The factor scores for the three periods were not significantly regular or distinct. Cases distributed in the region of larger values of VF1 were almost all collected from second and third period.

Sp-cluster C defined by the 1st period [or Group C (C1)]: The degree of pollution (as the mean average value of VFs) in the monitoring sites for group C differed significantly among the three periods, and sites received more pollution from VF1 during the first period than in the second and the third periods. Moreover, sites received more pollution from VF2 during the first period than in the second period and in the third period received not significantly different pollution as the other two. Cases distributed in the region of larger values of VF1 were collected from all three periods but mostly from 1st and 2nd period.

Spatial cluster analysis (on temporal clusters)

Spatial CA was conducted on standardized log-transformed data for each temporal cluster (1st, 2nd and 3rd period). Before carrying out Spatial CA on the data sets of the three temporal clusters, the following data subsets were created: the costal sea sites, of which the value of at least one of the four microbiological parameters was greater than the permitted limit (according the EU Directive 2006/7/EC) was subtracted from each temporal cluster.

Spatial CA produced a dendrogram with two spatial clusters for each temporal cluster (or two spatial clusters were created for each temporal cluster). The above subtracted costal sea sites of each temporal cluster created the third cluster in each period.

Table 2 shows the statistical descriptive of the four microbiological parameters per studied temporal cluster (period) and per spatial cluster of each period.

Next, Table 3 shows the statistical descriptive of the four microbiological parameters per studied temporal cluster (period) and per spatial cluster of the 1st period.

Consequently, spatial CA identified similar monitoring sites considering the effects of temporal differences in spatial CA.

Identification of the pollution pattern in costal bathing areas in the three different periods

As we see above, temporal CA generated a dendrogram grouping the six months into three clusters at (Dlink/Dmax) ×100 <35, and the difference between the clusters was significant (Figure 2). Cluster 1 (the first period) comprised May, cluster 2 (the second period) included June-August and cluster 3 (the third period) consisted of the two remaining months (September-October). Spatial CA identified similar monitoring sites considering the effects of temporal differences in spatial CA.

Spatial similarity analysis was conducted for each temporal cluster (first, second, and third period) with different results. Therefore, in continuance the three spatial clusters of the first time period were used for the clustering of the second and third period. In this way the All samples group was achieved to be clustered according the clusters of the 1st period (A, B and C). Group A comprised 276, group B contained 1546 and group C included 96 monitoring sites (costal bathing areas).

FA (PCA method) was carried out for the source identification in the monitoring sites. Before conducting the FA, the Kaiser-Meyer-Olkin (KMO) and Bartlett’s sphericity tests were performed on the parameter correlation matrix to examine the validity of the PCA. The KMO results for groups A, B, C and All samples were 0.598; 0,700; 0.664 and 0,710 respectively, and those for Bartlett’s sphericity were 2.561,70; 20.183,60; 1.612,22 and 27.764,82 (P<0.05), indicating that PCA may be useful in providing significant reductions in dimensionality.

FA was applied to standardized data sets (4 parameters) to examine differences between groups A, B and C and identify the latent factors. Based on the screeplot for the FA and the eigenvalues-0.75 criterion, only the VFs with eigenvalues over 0.75 were considered essential.

Table 4 summarizes the FA results comprising the loadings, eigenvalues and percentages of total variance and the loadings of which the absolute value is more than 0.7 was highlighted.

FA of the four data sets yielded two VFs for the groups A, B, C and All samples, explaining 87.02%, 94.10, 96.03% and 94.90% of the total variance in the respective costal bathing areas water quality data sets.

According the descriptive of the microbiological parameters and FA results (pollution sources) the 276 monitoring sites of group A corresponded to cleaner costal bathing areas, the 1546 sites of group B corresponded to relatively cleaner costal bathing areas and the 96 sites in group C corresponded to relatively polluted costal bathing areas.

Considering the last results of FA, the degree of pollution in the monitoring sites for groups A, B and C was as follows:

Group A:

The degree of pollution (as the mean average value of VFs) in the monitoring sites for group A differed significantly among the three periods, and sites received more pollution from VF1 (ECOL and FCOL) and VF2 (TCOL and STREPT) during the second (June–August) and the third (September-October) periods than in the first period (May).

Group B:

The degree of pollution (as the mean average value of VFs) in the monitoring sites for group B differed significantly among the three periods, and sites received more pollution from VF1 (ECOL, TCOL and FCOL) during the second and the third periods than in the first period. Moreover, sites received more pollution from VF2 (STREPT) during the first period than in the second and in the third periods, and more in the third period than in the second.

Group C:

The degree of pollution (as the mean average value of VFs) in the monitoring sites for group C differed significantly among the three periods, and sites received more pollution from VF1 (ECOL, FCOL and TCOL) during the first period than in the third period and during the second period received not significant different pollution than each of the other two separately. Moreover, sites received more pollution from VF2 (STREPT) during the first period than in the second and third periods.

Discriminant analysis: spatial variations in costal sea-water quality

based on spatial CA. The objectives of the DA were to test the significance of discriminant functions and determine the most significant parameters associated with the differences among clusters.

The spatial DA was performed using the standardized log-transformed data of the four parameters after classification into the three major clusters (first, second and third period) obtained from the spatial CA. Clusters formed the dependent categorical, and the measured parameters were the independent variables. Wilks’ lambda and the Chisquare for the discriminant functions (DFs), obtained from the standard and stepwise modes of DA for the three periods (first, second and third period), ranged from 0.337 to 0.972, and from 55.108 to 2078.428, the two methods, respectively, at P<0.0001, suggesting that the spatial DA was credible and effective.

Using the stepwise mode, the same results were received, which means that all four microbiological parameters are significant for the determination of the differences among the clusters. The standard and stepwise modes of DA constructed DFs containing all parameters. The discriminant functions (DFs), using the four discriminant variables (microbiological parameters) yielded classification matrices (CMs) correctly assigning 93.60%, 90.80% and 88.20% of all the cases for the 1st, 2nd and 3rd period respectively.

Conclusions

This research concerns the study of the seawater quality at bathing coasts, based on four microbiological parameters. Multivariate statistical analysis methods were applied to group the bathing beaches in order to accomplish the assessment and modelling of the microbiological quality data of them.

CA was applied to investigate the effect of seasonality on water quality of the bathing beaches surveyed. CA results indicated that the studied bathing beaches are classified in accordance with the seasonality in three groups. The first group consists of May, the second comprises June to August and the third group September and October (Figure 2). Therefore, it appears that the water quality of coastal bathing is completely dependent on seasonality and varies from spring to summer and then autumn.

By implementing CA in each seasonal group of bathing coasts and all coasts together, a new grouping arises clearly clustering the determined parameters in two new groups. The first group includes parameters ECOL, FCOL and TCOL, while the second includes the parameter STREPT (Figure 2). Furthermore, FA was applied to investigate possible determining factors in each of above groups that resulted from the CA. As shown in Table 1, in each group two new parameters were created (VF1 and VF2), where VF1 includes ECOL, FCOL and TCOL and VF2 includes STREPT.

By applying the CA in each seasonal group, three new groups of coasts were generated, group A (ultraclean), group B (clean) and group C (infected). The above analysis is confirmed by the application of DA.

CA and DA, as well as the FA give identical results, grouping the studied parameters in two groups or two new parameters, respectively. Based on the results it can be concluded that the study of water of the bathing beaches as to its microbiological quality does not require the identification of all four parameters as it is sufficient to identify only two: ECOL and STREPT. Taking into account all the studied beaches, ECOL interprets approximately 74% and STREPT about 21% of the total variance, while both together interpret nearly 95% of the total variance in the quality of the coastline (Table 1).

Considering the above, it is understood that the results of this research can contribute to economies of scale in determining the water quality of coastal bathing waters.

Specifically, in Greece the bathing coasts are audited at least 13 times a year for the above four microbiological parameters in accordance with the law (the Greek legislation is in line with the new Directive 2006/7/EU regarding the management of bathing water quality and repealing, Directive 76/160/EEC). Given that today in Greece four microbiological parameters are identified and the results of this research show that only two are needed, cost saving that can be achieved is approximately 50%. Therefore, from an approximate initial cost of 1.700.000 Euro (based on 2.149 swimming locations tested 13 times a year (28.037 samples, ~60 euro per sample) about 850.000 Euro can be economized without risking the health of bathers, since the determination of the parameters ECOL and STREPT interprets nearly 95% of the variation in quality throughout the studied coasts. Moreover, the parameter ECOL by itself seems to be a reliable quality marker, since it corresponds to 74% of the total variance, according FA results. In this case the cost reduction would be about 75% or around 1.250.000 Euro. These funds can be used in other actions to preserve the quality of coastal water and human health. This in turn, would aid in the assessment of the quality of coastal bathing waters and provide a more timely indication of bathing water quality, hence contributing to the immediate health protection of bathers.

The results of this study show that observing the quality of coastal bathing water can be accomplished by specifying and monitoring the parameters ECOL and STREPT. As a consequence, countries, especially those with large quantities of coastal bathing sites, can perform microbiological monitoring of their bathing waters by checking only the mentioned two parameters, thus ensuring economies of scale.

References

1.

M Bouvy, E Briand, MM Boup. Effects of sewage discharges on microbial components in tropical coastal waters (Senegal, West Africa). Mar Freshwater Res 2008;59:614-26.

2.

LJ Beversdorf, SM Bornstein-Forst, SL McLellan. The potential for beach sand to serve as a reservoir for Escherichia coli and the physical influences on cell die-off. J Appl Microbiol 2007;102:1372-81.

3.

V Cabelli. Health effects criteria for marine recreational waters. U.S. Environmental Protection Agency; 1983. EPA-600/1-80-031.

4.

D Kay, JM Fleischer, RL Salomon. Predicting likelihood of gastroenteritis from sea bathing: results from randomised exposure. Lancet 1994;344:905-9.

5.

EK Lipp, SA Farrah, JB Rose. Assessment and impact of microbial fecal pollution and human enteric pathogens in a coastal community. Mar Pollut Bull 2001;42:286-93.

6.

EK Lipp, R Kurz, R Vincent. The effects of seasonal variability and weather on microbial fecal pollution and enteric pathogens in a subtropical estuary. Estuaries 2001;24:266-76.

7.

KC Schiff, SB Weisberg, JH Dorsey. Microbiological monitoring of marine recreational waters in southern California. Environ Manage 2001;27:149-57.

8.

H Shuval. Estimating the global burden of thalassogenic diseases: human infectious diseases caused by wastewater pollution of the marine environment. J. Water Health 2003;1:53-64.

9.

LM Smith, JM Macauley, LC Harwell, CA Chancy. Water quality in the near coastal waters of the Gulf of Mexico affected by Hurricane Katrina: before and after the storm. Environ Manage 2009;44:149-62.

10.

TJ Wade, N Pai, JN Eisenberg, JM Colford Jr. Do US Environmental Protection Agency water quality guidelines for recreational waters prevent gastrointestinal illness? (A systematic review and meta-analysis). Environ Health Perspect 2003;111:1102-9.

11.

M Ostoich, E Aimo, D Fassina. Biologic impact on the coastal belt of the province of Venice (Italy, Northern Adriatic Sea): preliminary analysis for the characterization of the bathing water profile. Environ Sci Pollut Res 2011;18:247-59.

12.

A Basset. Aquatic science and the water framework directive: a still open challenge towards ecogovernance of aquatic ecosystems. Aquatic Conserv: Mar Freshw Ecosyst 2010;20:245-9.

13.

A Chandran, AAM Hatha, S Varghese. Increased prevalence of indicator and pathogenic bacteria in Vembanadu Lake: a function of salt water regulator, along south west coast of India. J Water Health 2008;6:539-46.

14.

PJ Cinotto. Occurrence of fecal-indicator bacteria and protocols for identification of fecal-contamination sources in selected reaches of the West Branch Brandywine Creek, Chester County, Pennsylvania: US Geological Survey Scientific Investigations Report 2005-5039. p 91.

15.

EU Directive 2006/7/EC of the European Parliament and of the Council. Directive of 15 February concerning the management of bathing water quality and repealing Directive 76/160/EEC. Official J European Communities 2006;L64:37-51.

16.

U.S. Environmental Protection Agency (USEPA). Ambient water quality criteria for bacteria. Washington, DC; 1986. EPA 440/5-84-002.

17.

U.S. Environmental Protection Agency (USEPA). Quality criteria for water Washington, DC; 1986. EPA 440/5-86-001.

18.

WHO. Monitoring bathing waters: a practical guide to the design and implementation of assessments and monitoring programmes. Geneva: WHO; 2000.

19.

WHO. Water quality: guidelines, standards and health. London: IWA; 2001.

20.

WHO. Guidelines for safe recreational-water environments: Vol. 1. Coastal and fresh-waters. Geneva: WHO; 2003.

21.

WHO. Addendum to the WHO guidelines for safe recreational water environments. Vol. 1. Coastal and fresh waters. WHO, Geneva; 2009.

22.

P Papastergiou, V Mouchtouri, M Karanika. Analysis of seawater microbiological quality data in Greece from 1997 to 2006: association of risk factors with bacterial indicators. J Water Health 2009;7:514-26.

23.

C Almeida, SO Gonzalez, M Mallea, P Gonzalez. A recreational water quality index using chemical, physical and microbiological parameters. Environ Sci Pollut Res 2012;19:3400-11.

24.

J Lattin, D Carroll, P Green. Analyzing multivariate data. New York: Duxbury; 2003.

25.

J McKenna. An enhanced cluster analysis program with bootstrap significance testing for ecological community analysis. Environ Modell Softw 2003;18:205-20.

26.

M Otto. Multivariate methods. In: R Kellner, JM Mermet, M Otto, HM Widmer, Eds. Analytical Chemistry. Weinheim: Wiley-VCH; 1998.

27.

A Astel, S Tsakovski, V Simeonov. Multivariate classification and modeling in surface water pollution estimation. Anal Bioanal Chem 2008;390:1283-92.

28.

T Kowalkowski, R Zbytniewski, J Szpejna, B Buszewski. Application of chemometrics in river water classification. Water Res 2006;40:744-52.

29.

S Shrestha, F Kazama. Assessment of surface water quality using multivariate statistical techniques: a case study of the Fuji River Basin, Japan. Environ Modell Softw 2007;22:464-75.

30.

V Simeonov, JA Stratis, C Samara. Assessment of the surface water quality in Northern Greece. Water Res 2003;37:4119-24.

31.

KP Singh, A Malik, D Mohan, S Sinha. Multivariate statistical techniques for the evaluation of spatial and temporal variations in water quality of Gomti River (India): a case study. Water Res 2004;38:3980-92.

32.

T Venugopal, L Giridharan, M Jayaprakash. Application of chemometric analysis for identifying pollution sources: a case study on the River Adyar, India. Marine Freshwater Res 2009;60:1254-64.

33.

DA Wunderlin, MP Diaz, MV Ame. Pattern recognition techniques for the evaluation of spatial and temporal variations in water quality. A case study: Suquia river basin (Cordoba-Argentina). Water Res 2001;35:2881-94.

34.

F Zhou, HC Guo, Y Liu, YM Jiang. Chemometrics data analysis of marine water quality and source identification in Southern Hong Kong. Mar Pollut Bull 2007;54:745-56.

35.

I Gupta, S Dhage, R Kumar. Study of variations in water quality of Mumbai Coast through multivariate analysis techniques. Indian J Mar Sci 2009;38:170-7.

36.

L Jagadeesan, M Manju, P Perumal, P Anantharaman. Temporal variations of water quality characteristics and their principal sources in tropical Vellar Estuary, south east coast of India. Res J Environ Sci 2011;5:703-13.

37.

A Papaioannou, E Dovriki, N Rigas. Assessment and modeling of groundwater quality data by environmentric methods in the context of public health. Water Resour Manag 2010;24:3257-78.

38.

A Papaioannou, A Mavridou, C Hadjichristodoulou. Application of multivariate statistical methods for groundwater physicochemical and biological quality assessment in the context of public health. Environ Monit Assess 2010;170:87-97.

Figure 1.

Dendrogram showing temporal clustering of monitoring periods (JUN = June, JUL = July, AUG = August, SEPT = September and OCT = October).

Figure 2.

Dendrogram for the microbiological parameters in all the studied costal bathing areas of the 1st period.

Table 1.

Loadings, eigenvalues, and percentage of total variance (TV) of the measured parameters on significant VFs of 1st, 2nd and 3rd period, and all samples.

Parameter First period Second period Third period All samples
VF1 VF2 VF1 VF2 VF1 VF2 VF1 VF2
ECOL 0.962 0.147 0.985 0.099 0.968 0.124 0.971 0.120
FCOL 0.964 0.180 0.984 0.134 0.972 0.152 0.976 0.146
TCOL 0.837 0.283 0.922 0.289 0.867 0.340 0.886 0.289
STREPT 0.205 0.976 0.163 0.985 0.186 0.979 0.178 0.982
Eigenvalue 2.909 0.776 3.034 0.862 2.967 0.813 2.966 0.830
% TV 72.728 19.412 75.851 21.550 74.172 20.329 74.152 20.753
[i] VF, varifactors; ECOL, E. coli; FCOL, faecal coliforms; TCOL, total coliforms; STREPT, faecal streptococci/enterococci.
Table 2.

Statistical descriptive [mean, standard error (SE), standard deviation (SD) and CI-95% for mean] of the four microbiological parameters per studied temporal and spatial cluster.

Parameter, spatial cluster N. Mean SD SE 95% CI for mean
First period
ECOL
  A1 276 8.53 3.60 0.22 8.11 8.96
  B1 1546 33.73 12.36 0.31 33.12 34.35
  C1 96 175.32 184.83 18.86 137.87 212.77
  All samples 1918 37.19 53.86 1.23 34.78 39.61
FCOL
  A1 276 9.82 3.98 0.24 9.35 10.29
  B1 1546 36.63 14.31 0.36 35.92 37.35
  C1 96 190.36 188.38 19.23 152.19 228.53
  All samples 1918 40.47 56.55 1.29 37.94 43.00
TCOL
  A1 276 27.98 19.84 1.19 25.63 30.33
  B1 1546 50.24 44.89 1.14 48.00 52.48
  C1 96 272.37 237.65 24.26 224.22 320.52
  All samples 1918 58.16 83.42 1.90 54.42 61.89
STREPT
  A1 276 6.85 3.36 0.20 6.45 7.24
  B1 1546 13.82 5.23 0.13 13.56 14.08
  C1 96 22.81 9.61 0.98 20.87 24.76
  All samples 1918 13.27 6.25 0.14 12.99 13.55
Second period
ECOL
  A2 1598 29.26 9.95 0.25 28.78 29.75
  B2 237 56.32 13.15 0.85 54.64 58.00
  C2 83 187.99 275.40 30.23 127.86 248.13
  All samples 1918 39.48 66.52 1.52 36.50 42.46
FCOL
  A2 1598 31.46 11.00 0.28 30.92 32.00
  B2 237 71.00 14.23 0.92 69.18 72.82
  C2 83 226.79 306.48 33.64 159.86 293.71
  All samples 1918 44.80 76.23 1.74 41.39 48.21
TCOL
  A2 1598 46.80 24.81 0.62 45.58 48.02
  B2 237 103.76 42.36 2.75 98.34 109.18
  C2 83 320.76 387.69 42.55 236.11 405.42
  All samples 1918 65.70 102.26 2.33 61.12 70.27
STREPT
  A2 1598 10.07 3.75 0.09 9.88 10.25
  B2 237 16.91 4.98 0.32 16.27 17.55
  C2 83 21.78 14.20 1.56 18.67 24.88
  All samples 1918 11.42 5.77 0.13 11.16 11.68
Third period
ECOL
  A3 1121 26.04 9.00 0.27 25.51 26.57
  B3 699 46.23 16.20 0.61 45.02 47.43
  C3 98 138.29 197.54 19.95 98.69 177.89
  All samples 1918 39.13 52.33 1.19 36.79 41.47
FCOL
  A3 1121 28.80 9.41 0.28 28.25 29.36
  B3 699 56.68 18.99 0.72 55.27 58.09
  C3 98 185.57 237.41 23.98 137.97 233.17
  All samples 1918 46.97 65.15 1.49 44.06 49.89
TCOL
  A3 1121 44.94 19.53 0.58 43.79 46.08
  B3 699 91.70 45.97 1.74 88.28 95.11
  C3 98 283.34 310.61 31.38 221.07 345.61
  All samples 1918 74.16 93.40 2.13 69.98 78.34
STREPT
  A3 1121 9.53 2.36 0.07 9.40 9.67
  B3 699 15.67 4.62 0.17 15.33 16.01
  C3 98 20.08 8.25 0.83 18.43 21.74
  All samples 1918 12.31 5.11 0.12 12.08 12.54

Table 3.

Statistical descriptive [mean, standard error (SE), standard deviation (SD) and CI-95% for mean] of the four microbiological parameters per studied temporal cluster (period) and per spatial cluster of the 1st period.

Parameter, spatial cluster N. Mean SD SE 95% CI for mean
First period
ECOL
  A1 276 8.53 3.60 0.22 8.11 276
  B1 1546 33.73 12.36 0.31 33.12 1546
  C1 96 175.32 184.83 18.86 137.87 96
  All samples 1918 37.19 53.86 1.23 34.78 1918
FCOL
  A1 276 9.82 3.98 0.24 9.35 276
  B1 1546 36.63 14.31 0.36 35.92 1546
  C1 96 190.36 188.38 19.23 152.19 96
  All samples 1918 40.47 56.55 1.29 37.94 1918
TCOL
  A1 276 27.98 19.84 1.19 25.63 276
  B1 1546 50.24 44.89 1.14 48.00 1546
  C1 96 272.37 237.65 24.26 224.22 96
  All samples 1918 58.16 83.42 1.90 54.42 1918
STREPT
  A1 276 6.85 3.36 0.20 6.45 276
  B1 1546 13.82 5.23 0.13 13.56 1546
  C1 96 22.81 9.61 0.98 20.87 96
  All samples 1918 13.27 6.25 0.14 12.99 1918
Second period
ECOL
  A2 276 16.77 9.96 0.60 15.59 17.95
  B2 1546 40.04 53.79 1.37 37.35 42.72
  C2 96 95.74 193.02 19.70 56.64 134.85
  All samples 1918 39.48 66.52 1.52 36.50 42.46
FCOL
  A2 276 18.62 11.63 0.70 17.25 20.00
  B2 1546 45.27 62.77 1.60 42.13 48.40
  C2 96 112.53 214.82 21.93 69.00 156.06
  All samples 1918 44.80 76.23 1.74 41.39 48.21
TCOL
  A2 276 60.41 56.98 3.43 53.66 67.16
  B2 1546 59.98 72.29 1.84 56.38 63.59
  C2 96 172.87 323.11 32.98 107.40 238.34
  All samples 1918 65.70 102.26 2.33 61.12 70.27
STREPT
  A2 276 11.41 7.98 0.48 10.47 12.36
  B2 1546 11.05 4.90 0.12 10.81 11.30
  C2 96 17.31 7.71 0.79 15.75 18.87
  All samples 1918 11.42 5.77 0.13 11.16 11.68
Third period
ECOL
  A3 276 16.35 10.85 0.65 15.07 17.64
  B3 1546 40.99 51.76 1.32 38.41 43.58
  C3 96 74.64 92.25 9.42 55.95 93.33
  All samples 1918 39.13 52.33 1.19 36.79 41.47
FCOL
  A3 276 19.25 12.72 0.77 17.75 20.76
  B3 1546 49.06 65.05 1.65 45.81 52.30
  C3 96 93.12 109.07 11.13 71.02 115.22
  All samples 1918 46.97 65.15 1.49 44.06 49.89
TCOL
  A3 276 55.87 39.82 2.40 51.16 60.59
  B3 1546 71.73 85.07 2.16 67.49 75.98
  C3 96 165.78 210.21 21.45 123.19 208.37
  All samples 1918 74.16 93.40 2.13 69.98 78.34
STREPT
  A3 276 10.79 3.78 0.23 10.34 11.24
  B3 1546 12.22 4.78 0.12 11.98 12.45
  C3 96 18.18 8.54 0.87 16.45 19.91