Arch Environ Contam Toxicol (2009) 56:654–669 DOI 10.1007/s00244-009-9310-2
Assessment of Water Quality Using Chemometric Tools: A Case Study of River Cooum, South India L. Giridharan Æ T. Venugopal Æ M. Jayaprakash
Received: 23 June 2008 / Accepted: 1 March 2009 / Published online: 20 March 2009 Ó Springer Science+Business Media, LLC 2009
Abstract Multivariate statistical techniques were applied to identify and assess the quality of river water. Thirty samples were collected from the River Cooum, and basic chemical parameters—such as pH, effect concentration, total dissolved solids, major cations, anions, nutrients, and trace metals—were evaluated. To evaluate chemical variation and seasonal effect on the variables, analysis of variance and box-and-whisker plots were performed. Cluster analysis was applied, and pre-monsoon and postmonsoon major and minor clusters were classified. The relations among the stations were highlighted by cluster analysis, which were represented by dendograms to categorize different levels of contamination. Cluster analysis clearly grouped stations into polluted and unpolluted regions. The analysis classified the upper part of the river course into one unpolluted cluster; the middle and lower parts of the river clustered together, reflecting the presence of pollution. Factor analysis revealed that water quality is strongly affected by anthropogenic activities, rock–water interaction, and saline water intrusion. Seasonal variations in water chemistry were clearly highlighted by both cluster and factor analysis. Factor-score diagrams were used successfully to delineate the stations under study by the contributing factors, and seasonal effects on the sample stations were identified and evaluated. These statistical approaches and results yielded useful information about water quality and can lead to better water resource management.
L. Giridharan (&) T. Venugopal M. Jayaprakash Department of Applied Geology, University of Madras, Chennai, India e-mail:
[email protected]
123
The chemical composition of river water and its properties depends on several factors, such as geochemical nature of the soil, precipitation, anthropogenic activities, etc. Spatiotemporal variation of hydrochemical parameters and river water quality largely depends on these factors. The quality of surface water in a particular region is basically governed by natural processes, e.g., precipitation rate, weathering processes, soil erosion, and anthropogenic effects, such as urban, industrial, and agricultural activities (Jarvie et al. 1998). Smith (2001) presented a detailed analysis of pollution loads caused by storm water events in an urban watershed. Ferrier et al. (2001) emphasized that quantitative and qualitative characteristics of a hydrologic system reflect the geomorphologic attributes of a watershed, modified by the influences of variations in climate and anthropogenic activities. Stow et al. (2001) presented a long-term study of water quality in a watershed with mixed land use by deriving regressions for time-series analysis. In many surface aquifers, domestic sewage and industrial effluents are the chief polluting sources, and surface runoff is a seasonal phenomenon that is largely influenced by the climate prevailing in the basin (Liao et al. 2006). Characterization and interpretation of various physicochemical parameters of river water requires handling a large data set. Complexity is mainly associated with the interpretation of a large number of measured variables, with high variability arising from various factors, e.g., natural and anthropogenic (Simeonov et al. 2002). Multivariate statistical analysis offers a powerful means of identifying similarities among the variables present in the chemical composition of water (Vega et al. 1998; Bengraıne and Marhaba 2003). To identify the likely factors causing variations in hydrochemical composition, multivariate statistical methods, such as principal component factor analysis, can be useful tools. Such analysis is
Arch Environ Contam Toxicol (2009) 56:654–669
especially useful because it highlights the relative significance of the combinations of chemical variables that can be evaluated. Subsequent interpretation is simplified because these statistical tools reduce and categorize complex sets of data into groups with similar characteristics. Factor analysis attempts to explain correlations between observations in terms of underlying factors, which are not directly observable (Yu et al. 2003). In this study, data were further characterized and evaluated using cluster analysis. Hierarchical cluster analysis is an objective technique, employed to identify natural groupings in a set of data, that classifies entities having similar properties (Yeung 1999). This unsupervised pattern-recognition technique uncovers the intrinsic structure or underlying behavior of a data set without making a priori assumption about the data. It classifies objects of the system into categories or clusters based on their nearness or similarity to one another (Vega et al. 1998). In this study, classification based on the sampling site was performed with cluster analysis using Ward’s method (linkage between groups); Euclidian distance was used as a similarity measure; and the clusters were synthesized into dendograms. Statistical analysis has been successfully applied by many investigators to sort out hydrogeochemical processes from commonly collected hydrochemical data (Hitchon et al. 1971; Seyhan et al. 1985; Ruiz et al. 1990; Grande et al. 2003; Kamman et al. 2005; Causape et al. 2006; Ryu et al. 2006). Several investigators have successfully demonstrated the utility of multivariate statistical analysis in identifying and characterizing pollution sources and apportioning natural versus anthropogenic contributions (Cave and Reeder 1995; Villaescusa-Celaya et al. 2000; Facchinelli et al. 2001; Yu et al. 2003; Liao et al. 2006). In this study, the effect of certain factors on river water—such as agricultural, industrial, domestic, rock–water interaction, and saline water intrusion—were studied by applying the previously mentioned multivariate statistical techniques.
Study Area In this study, water samples from the River Cooum were collected at 30 stations to evaluate the nature and quality of the water (Fig. 1). The River Cooum originates from the Kesevaram Dam, in the village of Kesavaram, which lies approximately 48 km west of Chennai. Although the River Cooum originates from this dam, excess water from the Cooum tank (79.82° latitude and 13.02° longitude) joins this course at approximately 8 km, and this point is considered the head of the River Cooum. In the upper part of the river stretch, many agricultural activities are being carried out. The river receives a sizeable quantity of sewage
655
from its neighborhood after it reaches Vanagaram near Chennai. It flows through the Kancheepuram, Tiruvallur, and Chennai districts for a distance of approximately 68 km and, after flowing through the heart of Chennai, it enters into the Bay of Bengal. This river is almost stagnant and do not carry enough water except during the rainy season. This period runs from October to December and is referred to as the ‘‘northeast monsoon season’’ in Tamil Nadu. Chennai receives the bulk of its rainfall from this monsoon. It has been observed that in the upper part of the river, there is no settlement along the bank of the river; hence this part is not polluted by domestic effluents. However, because of intense agricultural activities, the possibility of pollution caused by fertilizers and pesticides is highly expected. Currently fertilizers play a vital role in crop growth. The bulk use of these fertilizers leaves behind unused wastes, which are driven off by rain and enter into aquifers. Several investigators have reported on the release of nitrates from agricultural activities, which contaminating river waters (Prasad 1998; Pacheco 2001). In the middle and lower stretches of the River Cooum, domestic sewage water is directed into the river and appears as a sewage water stream.
Analytical Methodology The presented data include the results from two sampling periods (September 2005 [pre-monsoon period] and February 2006 [post-monsoon period]) at 30 locations along the Cooum River basin performed to evaluate seasonal variations in chemical compositions. For collection, preservation, and analysis of the samples, standard methods (Rainwater and Thatcher 1960; Brown et al. 1970; American Water Works Association 1971; Hem 1985; American Public Health Association 1995) were followed. The effect concentration and pH of water samples were measured in the field immediately after collection of the samples using pH and conductivity meters. Before each measurement, the pH meter was calibrated with reference buffer solution (pH levels 4 and 7). Na? and K? were measured using a flame photometer (Systronics Flame Photometer 128). Silica content was determined with the molybdate blue method using an ultraviolet (UV) light–visible spectrophotometer. Total dissolved solids (TDS) were measured using the evaporation and calculation methods (Hem 1991). Ca2? and Mg2? were determined titrimetrically using standard ethylenediaminetetraacetic acid. Chloride was estimated by AgNO3 titration. Sulphate was analysed using the turbidimetric method (Clesceri et al. 1998). Nitrate, nitrite, phosphate, and fluoride were analysed using a UV light– visible spectrophotometer (Rowell 1994). Standard solutions for the previously mentioned analysis were prepared
123
656
Arch Environ Contam Toxicol (2009) 56:654–669
Fig. 1 Base map of Cooum River depicting sample locations and pollution sources
from the respective analytic reagent-grade salts. Trace metals were determined by a graphite furnace atomic absorption spectrophotometer (AAnalyst 700; Perkin– Elmer). Multielement Perkin–Elmer standard solutions were used for the estimation of trace metals. Statistical Methodology Factor analysis was applied to the data matrix to reduce the data to an easily interpretable form. Before applying factor analysis, the data were standardized according to the criteria presented by Davis (2002). Standardization of variables is performed to remove the influence of different units of measurement on the data by making them dimensionless. Normalization of data is essential in factor analysis because it involves the computation of a correlation coefficient matrix, which requires equal distribution in all variables. Factors are extracted by varimax rotation, which gives values closest to –1, 0, and ?1, suggesting the negative, zero, and positive contribution, respectively, of a variable toward a factor (Briz-Kishore and Murali 1992). Commonality attached to each row of the matrix gives an appreciation of how well each variable is explained by ‘‘m’’ factors. If many commonalities B0.8, more factors are required (Klovan 1975). By examining the factor loadings and their eigenvalues, those variables belonging to a specific process can be identified. In certain cases, some variables may load high in [1 factor. Finally, factor scores are evaluated to illustrate the station-wise variation of the factors (Klovan 1975). Standardization of the values was performed after we measured skewness in the variables. The measure of
123
skewness compares the manner in which variables are distributed in a particular series with variables having a symmetrical distribution. Analysis of kurtosis and skewness are vital because most statistical methods require variables to conform to normal distribution (Papatheodorou et al. 2006). Computation of the correlation coefficient matrix is the first step in factor analysis between standardized variables. Eigenvalues quantify the contribution of a factor to total variance. The contribution of a factor is significant when the eigenvalue is greater than unity (Kaiser 1960). Initial factors are extracted and subjected to mathematical rotation. Varimax rotation procedure is used to maximize differences between the variables, thus facilitating easy interpretation of the data. The first factor accounts for as much variance as possible in the data set. The second factor accounts for as much residual variance as possible, and so forth. Factor loading indicates the degree of closeness between the variables and the factor. The largest loading, either positive or negative, suggests the meaning of the dimension: Positive loading indicates that the contribution of variables increases with increased loading in a dimension, and negative loading indicates that the contribution of variables decreases with decreased loading (Lawrence and Upchurch 1983). The study of factor scores reveals the extent of influence of each factor on overall water chemistry at all locations of sampling stations. Extreme negative scores reflect areas essentially unaffected by that particular factor, and positive scores reflect the areas most affected. Near-zero scores indicates areas affected to an average degree. In the present article, the station-wise variation of factors is indicated by line diagrams.
Arch Environ Contam Toxicol (2009) 56:654–669
657
Cluster analysis is a powerful tool for identifying and evaluating similar groups in hydrochemical data. The object of this analysis is to look for similar groups of items or for variables that group together in clusters. Cluster analysis organizes sampling entities into discrete classes or groups such that within-group similarity is maximized and among-group similarity is minimized (McGarial et al. 2000; Zeng and Rasmussen 2005). An unsupervised pattern-recognition technique uncovers intrinsic structure or underlying behavior of a data set, without making a priori assumption about those data, such that the objects of the system can be classified into categories or clusters based on their nearness or similarity to one another (Vega et al. 1998). In this study, classification based on sampling site was performed with cluster analysis using Ward’s method (Ward 1963); Euclidian distance was used as a similarity measure; and data were synthesized into dendograms. Euclidean distance is the geometric distance in a multidimensional space. Ward’s method is known to be distinct because it uses an analysis-of-variance approach to evaluate the distances between clusters. This method minimizes the sum of squares of any two (hypothetical) clusters derived at each step of analysis.
Results and Discussion The basic statistical parameters of the data matrix for both the pre-monsoon and post-monsoon periods are listed in Table 1. During the pre-monsoon period, skewness ranges from –0.68 to 1.53; during the post-monsoon period, skewness ranges from –4.34 to 4.79. Skewness results for most variables show only narrow variability, and all premonsoon variables are found to be well within the skewness index. During the post-monsoon period, few variables are found to have high skewness values. Logarithmic transformation of the data was performed to decrease skewness in the variables. Kurtosis was also applied on the data matrix to measure peakedness of the probability distribution. A distribution with positive kurtosis has a higher probability of variables than the normal distribution around the mean and also indicates a higher probability of distributed variables with extreme values. A distribution with negative kurtosis indicates a lower probability of normally distributed variables of values near the mean as well as extreme values. During the pre-monsoon period, kurtosis ranges from –1.31 to 2.5; during the post-monsoon period, kurtosis ranges from –1.62 to 24.89. During the pre-
Table 1 Summary statistics of geochemical data Geochemical data
Pre-monsoon Mean (mg/l)
pH
6.70
EC
2978.49
TDS Ca
Post-monsoon
r
Kurtosis Skewness Confidence level
r
Kurtosis Skewness Confidence level
CV
-0.42
0.05
8.15
0.53 -1.62
-0.03
0.19
6.47
0.01
379.28
35.59 1506.88
593.35 -0.70
0.39
212.32
39.38
1906.23 152.00
678.35 -0.42 48.28 0.92
0.01 -0.47
242.74 17.28
35.59 31.76
964.40 118.73
379.74 -0.70 41.93 0.47
0.39 0.87
135.89 15.01
39.38 35.32
Mg
38.98
19.68 -1.02
0.56
7.04
50.49
44.80
36.07 -0.76
0.69
12.91
80.52
Na
377.09
228.83
2.50
1.38
81.88
60.68
168.71
77.44 -0.46
0.58
27.71
45.90
K
31.26
17.62
0.15
0.81
6.30
56.36
12.25
8.36
0.37
0.97
2.99
68.29
HCO3
321.35
114.01
2.34
1.08
40.80
35.48
18.80
9.06
1.50
0.76
3.24
48.17
SO4
268.73
36.24 -0.49
0.29
12.97
13.49
111.93
45.23 -1.54
-0.23
16.18
40.41
Cl
565.53
114.56
56.61
440.73
190.81 -0.14
0.60
68.28
43.29 132.63
1.34
1059.93
320.14
0.31
0.66
0.56
2.17
Mean (mg/l)
0.42
F
0.15 -0.35
CV
1.03
0.47
0.20
41.67
0.60
0.79 15.47
3.65
0.28
NO3
62.15
42.67 -0.94
0.58
15.27
68.66
8.36
3.80
3.01
1.27
1.36
45.45
NO2
0.57
0.34 -0.13
0.44
0.12
60.22
1.87
2.22 -0.56
0.94
0.79
118.35
PO4
8.33
5.87 -0.22
0.91
2.10
70.46
1.12
1.42 -0.10
1.23
0.51
127.11
SiO2
24.94
4.20 -0.93
-0.68
1.50
16.82
26.97
2.53 21.85
-4.34
0.90
9.38
Cu
0.07
0.01 -0.81
0.28
0.01
21.10
0.09
0.09 24.89
4.79
0.03
104.79
Co
0.06
0.03 -0.53
-0.56
0.01
40.60
0.05
0.06 21.86
4.37
0.02
114.66
Zn Fe
0.03 0.44
0.02 1.74 0.34 -1.31
1.53 0.31
0.01 0.12
78.56 77.03
0.02 0.45
0.01 2.01 0.30 -0.10
1.51 0.80
0.00 0.11
65.27 64.95
Pb
0.44
0.25 -0.67
0.57
0.09
56.83
0.27
0.19 -0.16
0.80
0.07
70.31
Cr
0.51
0.24 -1.14
-0.05
0.08
46.25
0.25
0.14 -0.66
0.43
0.05
53.52
123
658
monsoon period, almost of all variables lie near zero value, reflecting that the distribution is normal ‘‘mesokurtic.’’ During the post-monsoon period, high positive values are obtained for silicates, fluoride, and heavy metals Cu and Co, indicating peaked distribution. Peaked distribution of the previously mentioned variables indicates surface runoff of sporadic high emissions from certain point sources. Coefficient of variation (CV) was used to compare the variability of C 2 series. The series of data for which the CV is large indicates that the group is more variable. Among the variables, pH shows the minimum CV, reflecting that there is not much variation in pH throughout the river course during both seasons. The seasonal effect is apparent with respect to the CVs of many variables. In general, heavy metals show much higher CVs among the variables during both seasons. Analysis of variance (ANOVA) is a statistical tool that permits the testing of significant differences between several means by comparing variances. ANOVA tests the variation between the mean values of the given variable. In the present study, it was observed that among the major ions, Mg shows p C 0.05, and all other major ions and nutrients show p \ 0.05, indicating that seasonal effect is significant with respect to all major ions and nutrients except Mg. In the case of heavy metals, Cu, Co, and Fe show p [ 0.05, whereas the other metals show p \ 0.05, reflecting that seasonal effect is significant for Cu, Co, and Fe. Box-and-whisker plots (Fig. 2) of individual variables were constructed to evaluate chemical variation and seasonal effect on the variables. These plots were constructed to evaluate different patterns associated with spatial variations in river water quality. Most monsoon variables showed deviations from normal distribution and included outliers and extremes. Cluster Analysis Cluster analysis was applied to reveal the relation among the stations and to elucidate water chemistry. It is a useful tool in organizing a particular set of data from various points into clusters or groups and determining relations between the various points (McGarial et al. 2000). Cluster analysis is also helpful in determining the seasonal effect on each station. During the pre-monsoon period (Fig. 3a), cluster 1 (stations 1 through 9) is characterized as an unpolluted region that lies in the upper part of the river. In this region, there are no settlements and no point sources of pollution near the river stretch. Cluster 2 (anthropogenic and saline water intrusion) (stations 17, 25, 26, and 30) is classified as a highly polluted region. This region lies in the eastern part of the river where settlements along the river course are dense, and domestic effluents are directed into the river. Moreover, the influence of saline water intrusion
123
Arch Environ Contam Toxicol (2009) 56:654–669
into river water is high at station 30. In the case of station 17, both industrial as well as domestic effluents influence the chemical composition of the water, and all of these stations are characterized by high TDS. Cluster 3 (anthropogenic), comprising the remaining 18 stations, lies in the middle and lower parts of the river stretch. In this region, it has been observed that the industrial and domestic effluents are directed into river water. Industries are concentrated in the suburban (middle part of the river) and intense settlements near the river course, and the formation of slums on the river bank influences the quality of river water. During the post-monsoon season, the data are also classified into three major groups (Fig. 3b). Station 30 alone comes under one category and is mainly influenced by saline water intrusion. Cluster 1 (stations 1 through 5, 7 through 9, and 13) is located in the upper part of the river course and is apparently unpolluted. Because there is no point source of pollution and because it receives heavy precipitation, this water behaves almost like freshwater. Cluster 2 (stations 11-15 and 6-10-12-14) also lies in the upper part of the river. The TDS of the water is slightly higher than the other grouping in the same region but lower than the limit set by the World Health Organization. Stations 28 and 29 form cluster 3, which is located near the eastern end of the River Cooum and the ocean, where population density is considerably high. The quality of river water at these stations is influenced by domestic effluents as well as saline water intrusion. Cluster 4 (stations 16 through 26) is comprises suburban and urban areas, where domestic and industrial effluents are directed into the river, thus degrading water quality. Station 30 forms a separate cluster that is directly influenced by saline water because this station lies at the river’s confluence point. A distinct variation is caused by the seasonal effect, which is reflected by the grouping of stations falling into different clusters. Correlation Studies Analytic results of the geochemical data were analysed using Statistical Package for Social Sciences (SPSS version 11.5; Chicago, IL) for factor analysis as described by Nie et al. (1985). Before the data were investigated, the raw data were standardized. Standardization of the variables is performed to eliminate the influence of different units of measurement on the data by making them dimensionless. Close inspection of correlation matrix is useful because it can reveal associations between variables that can show overall coherence of the data set and indicate participation of the individual chemical parameters that influence factors, a phenomenon that commonly occurs in hydrochemistry (Helena et al. 2000). The correlation
Arch Environ Contam Toxicol (2009) 56:654–669
659
Fig. 2 Box-and-Whisker plot representing chemical and seasonal variation of variables
coefficient values exhibiting ?1 or –1 between the variables show that a strong correlation exists, and a value of zero indicates that there is no relation between them. In general, geochemical parameters showing a correlation coefficient [0.7 are considered to be strongly correlated, whereas values between 0.5 and 0.7 shows moderate correlation. In this study, the relation between various elements was studied. The correlation matrix (Tables 2, 3) shows a distributive pattern of positive and negative correlations among the variables. During both the pre- and post-monsoon periods,
the correlation results are as follows: During the premonsoon period, Ca2? and Na? have a positive correlation with HCO3, and the correlation matrix of the post-monsoon period shows that Mg2? and Na? have a significant correlation with HCO3, suggesting that rock–water interaction as well as precipitation and leaching contribute to water chemistry. During the pre-monsoon period, Ca2? has a low correlation with SO42–; however, it does not have any correlation at all with NO3, suggesting that agricultural activities may not contribute to the source of these ions.
123
660
Arch Environ Contam Toxicol (2009) 56:654–669
Fig. 2 continued
During the post-monsoon period, although Ca2? has some correlation with SO42–, as in the case of the pre-monsoon period, it does not have any correlation with NO3. If pollution is related to the influence of domestic sewage, then there will be some association of NO3, Na, and Cl ions because all of these constituents are usually enriched in sewage. The correlation matrix for the pre-monsoon period shows that Na? is strongly correlated with Cl– and also that the variation of Na? with Cl– is significantly correlated to nitrate, reflecting that the mixing of domestic sewage water into river water strongly contributes to water chemistry
123
during the post-monsoon period. Although Na and Cl have a strong correlation, the variation of Na with Cl is only moderately correlated with NO3. This correlation result reveals that although sewage water is polluting river water, precipitation dilutes the effect of sewage water on the chemical composition of the water during the post-monsoon period. During both periods, some pairs of constituents show moderate to strong correlation (r [ 0.6), e.g., the major exchangeable ions Na–Ca and Na–Mg correlate significantly in both the pre- and post-monsoon periods, respectively. Moreover, the pairs of constituents
Arch Environ Contam Toxicol (2009) 56:654–669
661
Fig. 3 a Dendogram showing the relation among pre-monsoon river water samples.b Dendogram showing the relation among post-monsoon river water samples
—i.e., Cl–SO4, Mg–SO4 Ca-SO4, and Na–SO4—show little or no correlation, suggesting that overall water chemistry is not predominated by a dissolution/precipitation reaction; rather, extraneous sources, such as pollution caused by anthropogenic activities, dominate the water’s chemical make-up. Factor Analysis Characterization and interpretation of various parameters is often a complex problem. However, factor analysis offers a powerful means of identifying the similarities among variables that represent water chemistry. To identify the likely factors causing variations in hydrochemical compositions, multivariate statistical methods of analyzing hydrochemical data, such as factor analysis, can be useful tools. Such an analysis is especially useful because it reveals the relative significance of the combinations of
chemical variables available for evaluation. Subsequent interpretation is simplified because these statistical tools reduce and categorize complex sets of data into smaller groups with similar characteristics. Factor-score diagrams of the pre- and post-monsoon periods are presented in Figs. 4 and 5, respectively. The first six factors, which account for approximately 80% of variance during the pre-monsoon period and 83% of variance during the post-monsoon period (all of which have eigenvalues [ 1), were extracted from the principal factor matrix after varimax rotation (Tables 4, 5). Pre-Monson Factors Factor 1 during the pre-monsoon period, which explains 33% of total variance, has high loadings of the ions Na, Cl, SO4, Mg, F, and NO3. The loading pattern of the previously mentioned variables indicates that their source of origin
123
123
0.02
0.27
0.11
0.01
-0.07
0.15
0.20
-0.09
0.10
0.13
NO2
PO4
SiO2
Cu
Co
Zn
Fe
Pb
Cr
-0.16
F
NO3
-0.10
K
Cl
0.09
Mg Na
0.13
0.07 -0.04
Ca
SO4
0.30
TDS
-0.01
-0.02
EC
HCO3
1.00
-0.02
pH
pH
-0.19
-0.12
-0.15
0.24
0.22
-0.08
0.61
0.26
0.30
0.58
0.59
0.97
0.71
0.60
0.41
0.55 0.93
0.39
1.00
1.00
EC
-0.19
-0.12
-0.15
0.24
0.22
-0.08
0.61
0.26
0.30
0.58
0.59
0.97
0.71
0.60
0.41
0.55 0.93
0.39
1.00
TDS
0.07
0.02
0.25
0.02
-0.03
0.10
0.14
0.36
0.32
0.04
0.39
0.26
0.48
0.47
0.30
0.02 0.16
1.00
Ca
-0.18
-0.11
-0.20
0.18
0.13
-0.20
0.30
-0.01
0.24
0.37
0.32
0.57
0.52
0.06
-0.04
1.00 0.59
Mg
-0.11
-0.25
-0.27
0.23
0.14
-0.07
0.47
0.08
0.16
0.52
0.48
0.95
0.60
0.40
0.27
1.00
Na
Table 2 Pre-monsoon correlation coefficient matrix
1.00
0.21
0.02
0.05
0.14
0.01
0.55
0.58
0.42
0.22
0.20
0.27
0.17
0.59
-0.27
K
-0.27
0.09
0.30
-0.08
0.30
0.13
0.50
0.70
0.25
-0.01
0.34
0.48
0.21
1.00
HCO3
0.02
0.06
-0.01
0.39
0.11
0.01
0.34
0.02
0.44
0.61
0.53
0.66
1.00
SO4
-0.22
-0.19
-0.22
0.17
0.18
-0.15
0.51
0.13
0.25
0.58
0.57
1.00
Cl
1.00 0.49
0.02
-0.03
-0.12
0.19
-0.20
0.08
0.32
-0.14
-0.05
F
-0.11
-0.03
-0.45
0.58
0.11
-0.14
0.65
-0.20
0.15
1.00
NO3
-0.40
0.35
0.35
0.00
0.42
-0.16
0.25
0.54
1.00
NO2
-0.39
0.36
0.50
-0.18
0.53
0.07
0.37
1.00
PO4
-0.31
0.09
-0.12
0.27
0.44
-0.13
1.00
SiO2
0.22
0.39
0.32
0.08
-0.23
1.00
Cu
-0.35
0.34
0.33
0.19
1.00
Co
0.21
0.03
-0.26
1.00
Zn
-0.23
0.45
1.00
Fe
-0.17
1.00
Pb
1.00
Cr
662 Arch Environ Contam Toxicol (2009) 56:654–669
0.31
-0.35
0.41
-0.21
Pb
Cr
-0.08
Co
Zn
-0.24
Cu
Fe
-0.29
0.11
F
SiO2
-0.69
Cl
-0.64
-0.58
SO4
PO4
-0.01
-0.58
HCO3
0.01
-0.67 -0.71
Na K
-0.42
-0.45
Mg
NO3
-0.59
Ca
NO2
0.13
-0.72
TDS
0.34
-0.40
0.43
-0.45
0.05
0.15
0.33
0.79
0.44
0.98
0.79
0.62
0.96 0.81
0.76
0.67
1.00
-0.72
EC
1.00
1.00
EC
pH
pH
0.34
-0.40
0.43
-0.45
0.05
0.15
0.33
0.79
0.44
-0.01
0.13
0.98
0.79
0.62
0.96 0.81
0.76
0.67
1.00
TDS
0.03
-0.19
0.40
-0.29
-0.11
0.07
0.34
0.38
0.08
-0.14
-0.13
0.60
0.72
0.31
0.60 0.54
0.12
1.00
Ca
0.27
-0.39
0.15
-0.39
0.25
0.25
0.19
0.83
0.70
0.14
0.27
0.79
0.44
0.52
0.68 0.65
1.00
Mg
0.38
-0.29
0.51
-0.44
-0.10
0.03
0.31
0.65
0.31
-0.05
0.15
0.93
0.78
0.59
1.00 0.68
Na
Table 3 Post-monsoon correlation coefficient matrix
1.00
0.83
0.52
0.62
0.37
-0.53
0.13
-0.17
0.14
0.20
0.21
0.83
0.51
0.03
-0.08
K
0.52
-0.40
0.03
-0.19
0.03
0.01
-0.10
0.51
0.29
0.09
0.02
0.59
0.45
1.00
HCO3
0.14
-0.26
0.44
-0.43
0.04
0.16
0.25
0.46
0.10
-0.22
0.33
0.67
1.00
SO4
0.36
-0.38
0.39
-0.42
0.03
0.13
0.33
0.82
0.50
0.03
0.06
1.00
Cl
1.00
0.01
-0.08
0.14
-0.18
0.15
0.04
0.08
0.00
-0.10
-0.27
F
0.30
-0.01
-0.27
0.27
0.22
0.09
0.24
0.15
0.44
1.00
NO3
0.20
-0.42
0.06
-0.18
0.34
0.36
0.20
0.77
1.00
NO2
0.36
-0.55
0.13
-0.24
0.29
0.32
0.22
1.00
PO4
-0.09
-0.07
0.27
-0.22
0.02
0.05
1.00
SiO2
-0.06
-0.15
0.17
-0.17
0.89
1.00
Cu
0.03
-0.23
-0.02
-0.06
1.00
Co
0.26
0.01
-0.44
1.00
Zn
0.02
-0.05
1.00
Fe
-0.52
1.00
Pb
1.00
Cr
Arch Environ Contam Toxicol (2009) 56:654–669 663
123
664
Arch Environ Contam Toxicol (2009) 56:654–669
Fig 4 Pre-monsoon factorscore line diagrams
arises anthropogenically, such as agriculture, industrial, and domestic effluents. Moreover, the percentage abundance of the variables suggests saline water intrusion in the lower part of the river course. The high loadings of Na, Cl, and NO3 indicate that pollution caused by mixing sewage water into river water affects the chemical composition of the water. The factor-score diagram shows that factor scores gain significance from station no. 14 and that high positive values are observed at the eastern part of the river, indicating that the lower part of the river is more affected than the central region by this factor. The intrusion of saline water into the river course at the eastern part of the river is reflected in the factor scores; ion loadings also support this point. The middle part of the river runs through the urban area and receives the bulk of domestic sewage and industrial effluents (Ramesh et al. 1995). The factorscore diagram and the correlation of the ions Na, Cl, and NO3 clearly indicate that sewage water is being mixed into the middle part of the river and that the chemical composition of river water is highly influenced by this factor. Factor 2 during the pre-monsoon period, which accounts for 17% of total variance, has high loadings of Co, NO2, and PO4. The factor-score diagram shows that the eastern and western parts of the river are not affected by this factor. Significant scores are observed only in the central part of
123
the river, suggesting that the composition of river water in this region is highly influenced by these constituents. Riverine nitrite and phosphate originate mainly from agricultural or domestic effluents. In the region near the middle part of the river, no agricultural activity is being carried out; hence the high loadings of nitrite and phosphate must be attributed to domestic effluents. The concentration of NO2–N indicates fresh input of organic pollution load into the water system. In this region, domestic effluents containing water softeners contribute to the higher concentration of phosphates in the water (Rajmohan and Elango 2005). The increased abundance of Co in the middle part of the river, along with the high concentration of nutrients in a region where organic matter pollution is high, suggests that this may be caused by sediment fluxes or anthropogenic sources (Windom et al. 1988). The concentration of many metallurgical and chemical industries near the course of the river adjacent to the urban area, as well as the uncontrolled direct mixing of the effluents into river water, accounts for the higher concentration of this ion in the water. Factor 3 during the pre-monsoon period, accounting for approximately 9% of total variance, has high loadings of K, HCO3, PO4, and SiO2 and may be considered as being affected by silicate-weathering and anthropogenic factors.
Arch Environ Contam Toxicol (2009) 56:654–669
665
Fig 5 Post-monsoon factorscore line diagrams
The weathering reaction of microcline yields K, HCO3, and SiO2. The factor-score diagram shows that the middle part of the river is affected with respect to this factor. It has been observed that SiO2 is positively correlated with K and HCO3, suggesting that the sources of these ions have the same origin. The fact that fertilizer surface runoff remains near the upper part of the river explains the highly significant factor scores in this region. In the case of the middle and lower parts of the river, domestic effluents containing water softeners contribute to the higher concentration of phosphate ions in river water (Rajmohan and Elango 2005). Factor 4 during the pre-monsoon period, which account for 8% of total variance, has high loadings of NO3 and Zn. The factor-score diagram shows high positive scores at the middle and lower parts of the river course. It is known that nitrate is stable in aquifers because dissolved oxygen is present (Hamilton and Helsel 1995). It has been observed that no agricultural activities are being carried out in the middle and lower parts of the river; however, at many points in this region, domestic sewage water is discharged into the river. Domestic sewage loads are the main source of organic matter in this region. The oxidation products of many organic loads in the river system lead to the increased concentration of NO3-N in river water. Previous studies of
this river water have also indicated higher nitrate concentrations in the middle and lower stretches of the river (Ramesh et al. 1995). Many industrial activities (chemical, paint, chrome plating, tanning industries) are being carried out near the river course, and effluents are discharged into the river. These industrial effluents must be the cause of the increased abundance of Zn in this region of the river. Factor 5 during the pre-monsoon period, which accounts for 6.5% of total variance, is explicitly a heavy-metal factor, with high loadings of Cu, Pb, and Fe. The factorscore diagram shows a distributive pattern of this factor, with many significant scores in the middle and lower parts of the river. In addition to industrial activities, the atmospheric depositions resulting from automobile pollution (Varrica 2003; Sharma 2003) as well as urban runoff caused by precipitation may lead to the increased concentration of Pb in aquifers. Factor 6 during the pre-monsoon period, accounting for 6% of total variance, has high loadings of Ca, with some positive values of NO2. The factor-score diagram shows a distributive pattern, with significant scores at many stations, near the middle part of the river. Anthropogenic activities and rock–water interaction in this region increase the concentration of these ions in the water.
123
666
Arch Environ Contam Toxicol (2009) 56:654–669
Table 4 Pre-monsoon rotated component matrix
Factor 1
Factor 2
Factor 3
Factor 4
Factor 5
Factor 6
pH
-0.135
0.107
0.024
0.204
-0.149
0.814
EC
0.911
0.099
0.339
0.116
-0.073
0.044
TDS
0.911
0.099
0.339
0.116
-0.073
0.044
Ca
0.353
-0.143
0.359
-0.231
0.225
0.655
Mg
0.688
0.280
-0.295
0.076
-0.180
0.033
Na
0.881
0.026
0.157
0.122
-0.179
-0.062
K
0.131
0.156
0.826
0.144
0.020
0.104
HCO3
0.383
0.138
0.781
-0.243
0.154
0.045
SO4
0.783
0.094
-0.089
0.245
0.201
0.368
Cl
0.932
0.094
0.192
0.073
-0.156
-0.066
F
0.691
-0.327
0.157
0.127
0.206
-0.054
NO3
0.541
0.052
0.045
0.747
-0.102
-0.059
NO2
0.226
0.686
0.130
-0.057
0.105
0.438
PO4
0.012
0.546
0.659
-0.279
0.186
0.171
0.391 -0.045
0.323 -0.295
0.565 0.078
0.460 0.018
-0.092 0.805
-0.128 -0.074
SIO2 Cu
Extraction method was principal component analysis. Rotation method was varimax with Kaiser normalization
Co
0.032
0.795
0.169
0.170
0.017
0.048
Zn
0.172
-0.037
-0.066
0.803
0.107
0.183
Fe
-0.120
0.408
0.072
-0.476
0.635
0.079
Pb
-0.190
0.446
0.076
0.177
0.700
0.044
Cr
-0.130
-0.661
-0.202
0.162
0.097
0.310
Eigen value
7.01
3.56
1.89
1.74
1.36
1.27
% Variance
33.37
16.96
8.99
8.29
6.48
6.02
Cumulative variance
33.37
50.33
59.33
67.62
74.09
80.11
Post-Monsoon Factors Factor 1 during the post-monsoon period, which explains 43% of total variance, has high loading of the ions Mg, K, Na, Cl, PO4, HCO3, and NO2. The factor-score diagram shows that the western part of the river is unaffected by this factor. In this region, there is no point source of pollution, and the intense rainfall during this period dilutes the water; hence river water in this region behaves almost like freshwater. The factor-score diagram shows that there is a gradual increase from the western (upstream) to the eastern part (downstream) of the river. The eastern part of the study area is adjacent to the ocean, and saline water intrusion contributes to the chemical composition of river water. The factor-score diagram shows less significant values from stations 12–22, demonstrating that the middle part of the river is only moderately affected. The seasonal effect on the chemical composition of water in this region is highly significant because the point sources of pollution in this urban area are highly diluted. Factor 2 during the post-monsoon period, accounting for 13% of total variance, has high loadings of Ca, Na, SO4, Cl, and Fe, as well as a positive loading of silicate, and thus may be described as having been influenced by a silicateweathering factor. It is apparent that the seasonal effect is
123
highly significant for this factor. The factor-score diagram shows significant scores in the middle part of the river; in this region, the flow of river water is almost stagnant. This stagnancy of river water facilitates more rock–water interaction, thus leading to augmentation of these ions in river water. It is observed from the correlation matrix that pH is negatively correlated with all of these ions, suggesting that an increase in hydrogen ion concentration in the water increases the previously mentioned ions in the chemical make-up of the water. The factor-score diagram also depicts low positive scores at the downstream area, showing that the water in this part is only moderately affected by this factor. In the downstream region, saline water intrusion increases the concentrations of these ions in the water. The higher loading of Fe is attributed to rock– water interaction as well as anthropogenic activities. The high concentration of Fe observed in the middle stretch of the river is caused by the strong association of dissolved Fe with finer particles and colloids (Sholkovitz 1976). Moreover, effluents from the industries near the river course also contribute Fe to river water. Factor 3 during the post-monsoon period, which accounts for 10% of total variance, has high loadings of Cu and Co. The factor-score diagram illustrates that except at one station (station 26), all other stations are only
Arch Environ Contam Toxicol (2009) 56:654–669 Table 5 Post-monsoon rotated component matrix
Extraction method was principal component analysis. Rotation method was varimax with Kaiser normalization
667
Factor1
Factor 2
Factor 3
Factor 4
Factor 5
Factor 6
pH
-0.541
-0.581
-0.135
-0.130
0.271
0.010
EC
0.753
0.621
-0.019
0.117
0.082
0.057
TDS
0.753
0.621
-0.019
0.117
0.082
0.057
Ca
0.193
0.852
-0.044
0.031
-0.288
0.014
Mg
0.919
0.047
0.137
0.000
0.286
0.102
Na
0.664
0.644
-0.167
0.119
0.157
0.054
K
0.737
0.389
0.102
0.281
-0.211
-0.028
HCO3
0.644
0.191
-0.089
0.406
-0.062
-0.270
SO4
0.370
0.775
0.030
0.072
0.234
-0.109
Cl
0.803
0.528
-0.043
0.102
0.034
0.097
F
0.023
0.079
0.071
0.027
0.927
-0.066
NO3
0.172
-0.338
0.088
0.261
-0.234
0.711
NO2
0.749
-0.147
0.332
0.002
-0.143
0.340
PO4
0.865
0.200
0.241
0.168
-0.073
0.105
SIO2 Cu
0.075 0.117
0.432 0.119
0.005 0.951
-0.144 -0.068
0.089 -0.020
0.775 0.005
Co
0.101
-0.084
0.945
0.080
0.108
0.066
Zn
-0.409
-0.360
-0.061
0.594
-0.227
0.125
Fe
0.034
0.684
0.080
-0.180
0.216
0.065
Pb
-0.391
-0.138
-0.244
-0.586
-0.044
0.063
Cr
0.292
0.016
-0.084
0.849
0.085
0.056 1.00
Eigen value
9.04
2.69
2.05
1.47
1.21
% Variance
43.04
12.84
9.76
6.99
5.76
4.76
Cumulative variance
43.04
55.89
65.65
72.63
78.39
83.16
moderately affected by this factor. It is apparent that the seasonal effect is highly significant for this factor. Both of these heavy metals have no significant lithologic origin; hence it must be attributed to anthropogenic activities. Factor 4 during the post-monsoon period, accounting for 7% of total variance, has high loadings of Cr and Zn. In the suburban area and near the entry point of the urban area, a number of industries and small tanneries are situated along the river course. Effluents from these industries increase the concentration of these metals in the water. Factor 5 during the post-monsoon period, accounting for 6% of total variance, is explicitly a fluoride factor. The factor-score diagram illustrates that water is moderately affected by this factor throughout the stretch of the river except at a few stations in the middle part of the river. The lithology of the study area has no significant fluoridebearing minerals; hence it is logical to assign the high fluoride level to anthropogenic activity and, to a lesser extent, to chemical weathering (Saxena et al. 2003; Subbarao 2003). Factor 6 during the post-monsoon period, accounting for approximately 5% of total variance, has high loadings of silicate and nitrate. The factor-score diagram shows that the upper part of the river has significant factor scores, reflecting that this region is affected by this factor.
Agricultural activities lead to fertilizer remaining in the soil, and the use of NPK fertilizers is high in this region. Fertilizer surface runoff contributes to the abundance of nitrate in river water. Heavy precipitation and subsequent soil–water interaction leads to increased loadings of silicates in the water.
Conclusion The application of chemometric techniques for characterization and evaluation of hydrochemical data has been successfully demonstrated. Variation within and between variables are highlighted using parametric and nonparametric statistical techniques. Box-and-whisker plots demonstrate seasonal and chemical variations of various chemical parameters of river water. The significance of the seasonal effect on variables is evaluated by ANOVA. The results show that among the major ions, Mg is nonsignificant regarding seasonal effect, whereas Cu, Co, and Fe are found to be nonsignificant. Cluster analysis also highlights seasonal variation among the stations, and results show that the upper part of the river is grouped into one unpolluted major cluster. The middle and lower parts of the river are grouped into another major cluster for which industrial and domestic
123
668
sewage and saline water intrusion influence river water chemistry. The results of factor analysis clearly illustrate the various factors responsible for water chemistry. The contribution of anthropogenic activities, such as discharge of industrial and domestic sewage effluents into river water, silicate-weathering reactions, and the influence of saline water on river water, is clearly demonstrated by factor analysis. Factor scores clearly delineate the influence of various factors on the sample stations. The seasonal effect on the chemical composition of river water and the underlying factors responsible for water quality are clearly illustrated.
References APHA (1995) Standard methods for the examination of water and wastewater, 19th edn. American Public Association, Washington, DC, pp 1467 AWWA (1971) Water quality and treatment. McGraw-Hill, New York, NY Bengrame K, Marhaba TF (2003) Using principal component analysis to monitor spatial and temporal changes in water quality. J Hazard Mater 100:179–195. doi:10.1016/S0304-3894(03) 00104-3 Briz-Kishore BH, Murali G (1992) Factor analysis for revealing hydrochemical characteristics of a watershed. Environ Geol Water Sci 19:3–9. doi:10.1007/BF01740571 Brown E, Skougslad MW, Fishman MJ (1970) Methods for collection and analysis of water samples for dissolved minerals and gases. United States Geological Survey, Techniques for water resources investigations, book 5, chapter A1 Causape J, Quilez D, Aragues R (2006) Irrigation efficiency and quality of irrigation return flows in the Ebro River basin: an overview. Environ Monit Assess 117:451–461. doi:10.1007/s10661-0060763-8 Cave M, Reeder S (1995) Reconstitution of in-situ compositions obtained by aqueous leaching of drill core: An evaluation using multivariate statistical deconvolution. Analyst 120:1341–1351. doi:10.1039/an9952001341 Clesceri LS, Greenberg AE, Eaton AD (1998) Standard methods for the examination of water and wastewater, 20th ed. American Public Health Association, American Water Works Association, Water Environment Federation, Washington, DC Davis JC (2002) Statistics and data analysis in geology. Wiley, New York, pp 526–540 Facchinelli A, Sacchi E, Mallen L (2001) Multivariate statistical and GIS-based approach to identify heavy metal sources in soils. Environ Pollut 114:313–324. doi:10.1016/S0269-7491(00) 00243-8 Ferrier RC, Edwards AC, Hirst D (2001) Water quality of Scottish rivers: spatial and temporal trends. Sci Total Environ 265(1– 3):327–342. doi:10.1016/S0048-9697(00)00674-4 Grande JA, Borrego J, de la Torre ML, Sa´inz A (2003) Application of cluster analysis to the geochemistry zonation of the estuary waters in the Tinto and Odiel Rivers (Huelva, Spain). Environ Geochem Health 25:233–246 Hamilton PA, Helsel DR (1995) Effects of agriculture on groundwater quality in five regions of the United States. Groundwater 33:217–226. doi:10.1111/j.1745-6584.1995.tb00276.x Helena B, Pardo R, Vega M, Barrado E, Fernandez J, Fernandez I (2000) Temporal evolution of groundwater composition in an
123
Arch Environ Contam Toxicol (2009) 56:654–669 alluvial aquifer (Pisuerga River, Spain) by principal component analysis. Water Res 34:807–816 Hem JD (1985) Study and interpretation of the chemical characteristics of natural water. United States Geological Survey Water Supply Paper 2254 Hem JD (1991) Study and interpretation of the chemical characteristics of natural water, 3rd edn. Scientific publication no. 2254, Jodhpur Hitchon B, Billings GK, Kolvan JE (1971) Geochemistry and origin of formation waters in the Western Canada sedimentary basin. III. Factors controlling chemical composition. Geochim Cosmochim Acta 35:567–598. doi:10.1016/0016-7037(71)90088-3 Jarvie HP, Whitton BA, Neal C (1998) Nitrogen and phosphorus in east coast British rivers: Speciation, sources, and biological significance. Sci Total Environ 10:79–109. doi:10.1016/S00489697(98)00109-0 Kaiser HF (1960) The application of electronic computers to factor analysis. Educ Psychol Meas 20:141–151. doi:10.1177/001316 446002000116 Kamman NC, Chalmers A, Clair TA, Major A, Moore RB, Norton SA et al (2005) Factors influencing mercury in freshwater surface sediments of northeastern North America. Ecotoxicology 14:101–111 Klovan JE (1975) R and Q mode factor analysis. In: McCammon O (ed) Concepts in geostatistics. Springer, New York Lawrence FW, Upchurch SB (1983) Identification of recharge areas using geochemical factor analysis. Ground Water 20:680–687. doi:10.1111/j.1745-6584.1982.tb01387.x Liao S-w, Lai W-l, Chen J-j, Sheu J-y, Lee C-g (2006) Water quality during development and apportionment of pollution from rivers in Tapeng Lagoon, Taiwan. Environ Monit Assess 122:81–100 McGarial K, Cushman S, Stafford S (2000) Multivariate statistics for wildlife and ecology research. Springer, New York Nie NH, Hull QH, Jenkins JC, Stembrenner K, Bernnt DH (1985) SPSS PC, Statistical Package for Social Sciences. McGraw Hill, New York Pacheco J (2001) Nitrate temporal and spatial patterns in 12 watersupply wells, Yucatan, Mexico. Environ Geol 40:708–715. doi: 10.1007/s002540000180 Papatheodorou G, Demopoulou G, Lambrakis N (2006) A long-term study of temporal hydrochemical data in a shallow lake using multivariate statistical techniques. Ecol Modell 193:759–776. doi:10.1016/j.ecolmodel.2005.09.004 Prasad R (1998) Fertilizer urea, food security, health and the environments. Curr Sci 75:667–683 Rainwater FH, Thatcher LL (1960) Methods for collection and analysis of water samples. United States Geological Survey Water Supply Paper 1454 Rajmohan N, Elango L (2005) Nutrient chemistry of groundwater in an intensively irrigated region of southern India. Environ Geol 47:820–830. doi:10.1007/s00254-004-1212-z Ramesh R, Shivkumar K, Eswaramoorthi S, Purvaja GR (1995) Migration and contamination of major and trace elements in groundwater of Madras City, India. Environ Geol 25:126–136. doi:10.1007/BF00767869 Rowell DJ (1994) Soil science: methods and applications. Longman Scientific and Technical Ruiz F, Gomis V, Blasco P (1990) Application of factor analysis to the hydrogeochemical study of a coastal aquifer. J Hydrol 119:169–177. doi:10.1016/0022-1694(90)90041-U Ryu JS, Lee KS, Kim JH, Ahn KH, Chang HW (2006) Geostatistical analysis for hydrogeochemical characterization of the Han River, Korea: Identification of major factors governing water chemistry. Environ Contam Toxicol 76:1–7. doi:10.1007/s00128-0050882-x
Arch Environ Contam Toxicol (2009) 56:654–669 Saxena VK, Shakeel A (2003) Inferring the chemical parameters for the dissolution of fluoride in groundwater. Environ Geol 43:731–736 Seyhan EV, Van de Caried AA, Engelen GB (1985) Multivariate analysis and interpretation of the hydrochemistry of a dolomite reef aquifer, Northern Italy. Water Resour Res 21:1010–1024. doi:10.1029/WR021i007p01010 Sharma R, Pervez S (2003) Enrichment and exposure of particulate lead in a traffic environment in India. Environ Geochem Health 25:297–306 Sholkovitz ER (1976) Flocculation of dissolved organic and inorganic matter during the mixing of river water and seawater. Geochim Cosmochim Acta 40:831–845. doi:10.1016/0016-7037(76)90035-1 Simeonov V, Einax JW, Stanimirova I, Kraft J (2002) Environmetric modeling and interpretation of river water monitoring data. Anal Bioanal Chem 374:898–905. doi:10.1007/s00216-002-1559-5 Smith E (2001) Pollutant concentrations of stormwater and captured sediment in flood control sumps draining an urban watershed. Water Res 35:3117–3126. doi:10.1016/S0043-1354(01)00008-2 Stow CA, Borsuk ME, Stanley DW (2001) Long-term changes in watershed nutrient inputs and riverine exports in the Neuse river, North Carolina. Water Res 35:1489–1499. doi:10.1016/S00431354(00)00402-4 Subba Rao N, John Devadas D (2003) Fluoride incidence in groundwater in an area of Peninsular India. Environ Geol 45:243–251. doi:10.1007/s00254-003-0873-3 Varrica D, Dongarra G, Sabatino G, Monna F (2003) Inorganic geochemistry of roadway dust from the metropolitan area of Palermo, Italy. Environ Geol 44:222–230
669 Vega M, Pardo R, Deban L (1998) Assessment of seasonal and polluting effects on the quality of river water by exploratory data analysis. Water Res 32:3581–3592. doi:10.1016/S0043-1354 (98)00138-9 Villaescusa-Celaya JA, Gutierrez-Galendo EA, Flores-Monoz G (2000) Heavy metals in the fine fraction of coastal sediments from Baja California (Mexico) and California (USA). Environ Pollut 108:453–462. doi:10.1016/S0269-7491(99)00222-5 Ward JH (1963) Hierarchical grouping to optimize an objective function. J Am Stat Assoc 58:236–244. doi:10.2307/2282967 Windom HL, Smith R, Rawlinson C, Hungspreugs M, Dharmvanij S, Wattayakorn G (1988) Trace metal transport in a tropical estuary. Mar Chem 24:293–305. doi:10.1016/0304-4203(88) 90037-0 Yeung IMH (1999) Multivariate analysis of the Hong Kong Victoria Harbour water quality data. Environ Monit Assess 59:331–342. doi:10.1023/A:1006177824327 Yu S, Shang J, Zhao J, Guo H (2003) Factor analysis and dynamics of water quality of the Songhua River, northeast China. Water Air Soil Pollut 144:159–169. doi:10.1023/A:1022960300693 Zeng X, Rasmussen TC (2005) Multivariate statistical characterization of water quality in Lake Lanier, Georgia, USA. J Environ Qual 34:1980–1991. doi:10.2134/jeq2004.0337
123