This action might not be possible to undo. Are you sure you want to continue?
Khmiri ISSN: 1735-1472 © IRSEN, CEERS, IAU
Evaluation of ground water quality using multiple linear regression and structural equation modeling
*I. Chenini; 2S. Khemiri
Minerals Resources and Environment Laboratory, Department of Geology, Faculty of Sciences of Tunis, 2092 Tunis El Manar, Tunisia
U. R. Structural and applied geology, Department of Geology, Faculty of Sciences of Tunis, 2092 Tunis El Manar, Tunisia
Received 2 January 2009; revised 18 April 2009; accepted 5 May 2009; available online 1 June 2009
ABSTRACT: A methodology for characterizing ground water quality of watersheds using hydrochemical data that mingle multiple linear regression and structural equation modeling is presented. The aim of this work is to analyze hydrochemical data in order to explore the compositional of phreatic aquifer groundwater samples and the origin of water mineralization, using mathematical method and modeling, in Maknassy Basin, central Tunisia). Principal component analysis is used to determine the sources of variation between parameters. These components show that the variations within the dataset are related to variation in sulfuric acid and bicarbonate, sodium and cloride, calcium and magnesium which are derived from water-rock interaction. Thus, an equation is explored for the sampled ground water. Using Amos software, the structural equation modeling allows, to test in simultaneous analysis the entire system of variables (sodium, magnesium, sulfat, bicarbonate, cloride, calcium), in order to determine the extent to which it is consistent with the data. For this purpose, it should investigate simultaneously the interactions between the different components of ground water and their relationship with total dissolved solids. The integrated result provides a method to characterize ground water quality using statistical analyses and modeling of hydrochemical data in Maknassy basin to explain the ground water chemistry origin. Keywords: Ground water quality; Hydrochemical; Mathematical method; Water-rock interaction
INTRODUCTION Water quality concept has been evaluated in the last years owing to greater understanding of water mineralization process and greater concern about its origin (Shane and Jerzy, 2003). Water quality shows water-rock interaction and indicates residence time and recharge zone confirmation (Cronin et al., 2005). Thus, water quality indicators must reflect mineralization process, integrate reservoir properties and be sensitive to ground water recharge rate and flow direction (Andre et al., 2005). The concept of ground water quality seems to be clear, but the way of how to study and evaluate it still remains tricky. The chemical composition of ground water is controlled by many factors that include the composition of precipitation, mineralogy of the watershed and aquifers, climate and topography. These factors can combine to create diverse water types that change in composition spatially and temporally. The use of major ions as natural tracers has become a common method to delineate flow paths in aquifers
*Corresponding Author Email: firstname.lastname@example.org Tel.: +216 97 282 335; Fax: +216 70 860 325
(Back, 1966). Generally, the approach is to divide the samples into hydrochemical facies, that is groups of samples with similar chemical characteristics that can then be correlated with location. The observed spatial variability can provide insight into aquifer heterogeneity and connectivity, as well as the physical and chemical processes controlling water chemistry. In recent decades, multivariate statistical methods have been employed to extract significant information from hydrochemical datasets in compound systems. These techniques can help resolve hydrological factors such as aquifer boundaries, ground water flow paths, or hydrochemical components (Liedholz and Schafmeister, 1998; Locsey and Cox, 2003; Seyhan et al., 1985; Suk and Lee, 1999; Usunoff and Guzman-Guzman, 1989) identify geochemical controls on composition (Adams et al., 2001; Alberto et al., 2001; Lopez-Chicano et al., 2001; Reeve et al., 1996). Different techniques have been used in attempt to evaluate water quality, essentially based on chemical ions correlation (Piper and sholler diagramm) and some ions rapports to predict
all the variables would be included simultaneously into single model in order to test the potential interactions between the independents variables using the structural equation modeling (SEM). Since the data obtained in this study had multivariate nature and several of the variables were correlated. the deep aquifer made up by two fractured limestone levels extending to depth between 180 and 250 m and the upper level of deep aquifer forms the main source of water supply in the studied area. 2002. potassium. PCA reduces a large number of variables to a small number of variables. S. [Ca]) to the independent variable Vi using the following form: . The shallow aquifer extending to varying depths of 30150 m. similarities and dissimilarities among chemical properties of sampled water were carried out according to PCA method (Jolliffe. The ground water samples were analyzed for parameters which include pH. Then. TDS were determined grametrically. In this geographic area. which reasonably represent the ground water quality in the study region. Temperature was measured on the site using mercury thermometer. MLR met hod provides equation li nkin g a dependant variable Vd ([Na]. 2000. bicarbonates (HCO3-).0 statistical program. Hidalgo and Cruz-Sanjulian. without sacrificing too much of the information (Qian et al.. were immediately transported to the laboratory under low temperature conditions in ice-box and stored in the laboratory at 510 4 °C until processed analyses. Chenini. 1996). All the samples. This approach has been used to extract related variables and infer the processes that control water chemistry (Helena et al. Muller et al. in October 2006. The study basin has a three tiers aquifer system.. Among the major cations. Multiple linear regression (MLR). Chemometric analyses were used to differentiate the water samples on the basis of their composition and origin (Singh et al. [SO4]. A total of 28 water samples were collected in the months of October-November 2005 (Fig. total dissolved solids (TDS). sulfate (SO42+).. The sampling network and strategy were collected in equilibrium state of aquifer. 2003). All analyses were completed within a week time in laboratory. sodium (Na+).. [Cl]. [Mg]. As a multivariate data analytic technique. Major anions were analyzed using modular ion chromatograph. PCA combines two or more correlated variables into one variable. climate is arid Mediterranean type. [HCO3]. calcium (Ca2+) and magnesium (Mg2+ ). pH was measured using a pH-meter. More concisely.. Before establishing the statistical models. The ground water ionic strength is dominated by major cations and anions. electrical conductivity (EC). Electrical conductivity was measured at 25 °C with a conductivity meter. potassium (K+). collected in tight capped high quality polyethylene bottles. chloride (Cl -).. The present study attempts to establish a mathematical water quality model. 1994). Sampling and ground water analysis were done in Minerals Resources and Environement Laboratory. 2001). sodium. Khmiri the origin of the mineralization (Bennetts et al. The analytical data quality was ensured through careful standardization. since it generates a minimum data set of indicators (Doran and Parkin. the detection of outliers and the elucidation of trends. Hydrochemistry of the phreatic aquifer of Maknassy Basin was summarized through the statistical analysis and mathematical modeling of ground water properties. MATERIALS AND METHODS The study site covers an area of about 1250 km² located in southern central Atlas of Tunisia. All other parameters were determined in laboratory following the standard protocols (Apha. based on chemical ground water properties in phreatic aquifer level in Maknassy basin.I. 2005). 2007). The general purpose of multiple linear regressions is to quantify the relationship between several independent or predictor variables and a dependant variable. Principal component analysis (PCA) analysis methods were used for the interpretation of the data. the ionic charge balance of each sample was within ± 5 %. 2008). Therefore. 2006. Pulido-Leboeuf et al. 2008). This method is successfully used by different authors to establish statistical model (Ghasemi and Saaidpour. 1985). 1). calcium and magnesium were analyzed by flame photometer. PCA method was performed using STATISTICA 6. The measured variables included the characteristic water quality parameters.. the results of ground water quality data analyses in a sequential fashion have been fully integrated to better constrain the interpretation and include the statistical modeling in the process. is given as an accurate tool to evaluate ground water quality. The altitude of water table in the region varies from 180 to 320 m and movement of ground water is from east to west and from borders to center basin (Chenini et al.
(n) is the number of water samples.Int. Vi descriptors are used to describe water quality and cation dependence. Chenini. The best equation is selected 511 while being based on the highest multiple correlation coefficients (R). 1996). S. Structural equation modeling is a statistical method that takes a confirmatory fashion to the analysis of a structural theory bearing on some phenomena (Kline. Tech. Typically this theory represents “causal” .. 1: The study area and sampling wells locations Groundwater sampling Analyses: Water quality parameters PCA analysis MLR regression analyses: Investigate the relationships between TDS and water properties SEM: Potential interaction between parameters Fig. lowest standard deviation (SD) and F-ratio value. The MLR modeling method was performed by the SPSS statistical program. The reduction in the number of descriptors (variables) is included in the study to minimize the information overlap in variables. Relationships between variables were established using the forward stepwise regression method (Bernstein. J. Summer 2009 I. Khmiri 520000 3850000 28 18 17 6 7 5 E 4 16 8 s Sm 19 ad 49 Q 52 El Guallel 15 1 u d a 14 24 Leben y 50 22 13 12 29 G71 30 43 a 21 G74 r G73 25 23 48 jou 47 Ma ay G72 46 ud ur 3 a dho 40 32 31 ar Q r 2 Na ma m A h 33 La 53 t Ou lad 28 Maknassy ar 10 Ks 27 Ko udia 38 37 36 Menzel 34 41 Bouzaien 35 44 20 Es Sened Q ud ay EIM echacha 28 530000 540000 550000 560000 3850000 Africa 3840000 Quady 3840000 Alegria 3830000 Tunisia 3820000 Mediterranean Sea 3810000 42 3830000 45 11 3820000 0 5 10 km 3810000 520000 Pluviometric station Hydrometric station 530000 540000 Hydrographical station 550000 560000 Sampled well and boring Hydrological basin limit Fig. Sci. 2: Flowchart of the adopted approach Vd = β0 + β 1Vi1 + … + β nVin (1) Wh en th e i ntercept β 0 an d th e regr essi on coefficients of descriptors (βi) are determined by least square method (Green and Carroll. 2005). 1988). Environ. 509-519. 6 (3).
[K] and T contribute most strongly to the third component (PC3) that explains 10. The [Na]. 3 shows the projection of the first two PC scores in a scatter plot. 1988). approach to the data analysis. The SEM conveys that the causal processes under study are represented by a series of structural equations and that these relations can be modeled. After obtaining various equations with ground water samples from the phreatic qauifer levels. SEM provides explicit estimates of these error variance parameters (Byrne. 2001). which is associated with the variables [Na]. The Variance Inflation Factor (VIF) value obtained was close to one and thus. the three significant PCs are cumulatively accounted for 70 % of the total data variance. [SO4]. Second. According to PCA. Furthermore. although traditional multivariate procedures are incapable of either assessing of correcting measurement error. so that hypothesis testing is difficult.4 %). Regression analysis (SPSS 14. [SO4] and TDS. 99 % for the total variance for the estimation of TDS is explained by the MLR model. The data were input to Amos 7. the water properties measures are statistically significant in estimating TDS (P < 0. Among all candidate equations. of discharge wastewater effluent An assessmentI. the equation where this ratio was closer to 1 were selected. PC2 explains 12. First. The application of PCA reveals that the classification of ground water sampled was achieved according to their chemical and physical properties. PC1 contains hydrochemical variable originating from weathering process and reservoir formation dissolution.0 (Arbuckle. According to R square statistic. but TDS is grouped with pH. 7 analysts were randomly selected to constitute the training set for the construction of statistical models.072 Mg + 0. [Mg].141 HCO3 + 0. Fig. An examination of tvalues also reveals an identical descending order of the factors that contribute to the estimation of TDS in Maknassy basin phreatic aquifer water (Bring.2 % of the total variance. Th e second step of t his st udy i ncl udes simultaneously all the variables in a conceptual model 512 . [Cl]. An analysis of residuals was developed and R² values were studied.285 Cl + 0. [SO4].6 % of the variance and is mainly related to [Mg] and [HCO3]. The descriptors and the regression coefficient of this model are presented in Table 2. must other multivariate procedures are essentially descriptive by nature. Several aspects of SEM set are apart from the older generation of multivariate procedure (Fronell. S. [Ca]). RESULTS AND DISCUSSION PCA was applied to the combined ground water data set of the shallow aquifer (Table 1) to examine relations between water properties analyzed and to identify the factors that influence the concentration of each one. [Cl]. It confirms the PCA classification. [HCO3]. [Ca]. [Ca] were considered as independent variables and TDS as a dependent variable. The model was also checked to see if it was prone to any multicollinearity effect. The multiple R coefficient indicates that the correlation between water properties and TDS is moderate (the multiple R > 0. In the other hand. it can squabble that the [SO4] makes the largest contribution across the model. As can be seen in the case of all the MLR regression analyses.0) was conducted to investigate the relationships between TDS and water properties. [K] and T. 2 summarizes the steps of the study. 1994).062 Ca + 0. SEM lends itself well to the analysis of data for inferential purpose.of the impact Khmiri processes that generate observations on multiple variables (Bentler.047 + 0. [SO4] is important in water samples that’s why it can be classified as an own class of water properties. The selected equation for shallow aquifer ground water in Maknassy basin is: TDS = 0. The positive sign of the beta coefficients and t-values pertaining to these variables indicates that there is a positive relationship between TDS and elements of ground water properties ([Na]. The model can then be tested statistically in a simultaneous analysis of the entire system of variables to determine to which it is consistent with the data. [K].Chenini. The variables pH. 2006) for SEM analysis.247 Na + (2) 0.99). there was no evidence of multicollinearity (Hair and Anderson. The analysts (chemical properties) were classified and presented in a dendogram (Fig. 1982). it takes a confirmatory. Most of the variance is contained in PC1 (48.00). rather than an exploratory. 4). ε is the error of estimation in the statistical regression model. The distribution suggests a more continuous variation of properties of the samples. The flowchart of Fig. whereas PC2 and PC3 are related to bicarbonates and physical properties.451 SO4 + ε Where. 1998). The best model was derived by the application of MLR method. [Cl]. In terms of the relative importance of the estimation of a dependent variable. [Mg]. [HCO3].
46 7.34 650.2 15.86 270 180. 6 S.34 232.1 19 21.86 13.25 253.5 112.34 181.73 7.52 188.63 9.3 110.989 2.5 253 345.5 818.67 13.6 180.35 595.37 263.7 23 22 22.8 430 847 885.4 195.44 267.3 14.6 7.Chenini.56 186.2 7.145 4.446 2.67 684.75 7.45 25.5 174.4 230.377 2.14 336.59 162.74 7.34 282.08 720 320. (3).7 823.83 124.67 134.9 906 365.78 1454.78 212.4 345. 509-519.76 185.78 281.4 720.62 515.Khmiri Shallow aquifer 513 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 TDS (g/L) 2.7 21.5 20.67 242.85 1041.37 198.61 7.67 245.6 7.65 190.75 123.44 20.78 353.34 669.56 362.581 1.59 K (mg/L) 10.33 340.1 296.05 21 18.5 140.4 195.9 7.45 1100. Sci.24 275 210.75 745.5 7.577 7.3 125 102.67 145.23 165.46 316.3 21.985 3.5 294.71 471.83 1161.65 293.319 3.429 1.83 7. Summer 2009 Int.24 490.64 307.56 121.56 630 243.65 182 178.34 752.9 201.326 1.818 1.82 7.45 14.5 7.58 17. I.93 110.7 128.7 Ca (mg/L) 235.2 815.435 2.62 240.87 355.1 1767.1 647.7 215.73 7.45 20.61 221.34 125.916 1.3 202.84 120.83 480.3 19.71 7.5 268..69 7.945 2. J.2 19.72 678.32 713.66 568. Environ.77 123.45 HCO3 (mg/L) 247.36 16.67 7.73 270.7 9.57 100.5 19 21.496 1.65 250.688 1.8 273.8 162.65 7.815 2.47 320.75 298.2 1151.4 137.52 218.3 18.3 22 21.24 25.62 7.21 226.18 232.447 4.5 826.78 1650.032 1.78 20.3 724.54 1002.84 745.569 1.15 180.701 1.48 130.78 367.54 7.7 20 21.3 7.45 17.2 18.95 821.21 214.39 13.71 124.23 283.95 Mg (mg/L) 210.56 264.8 23.6 23.4 198.144 1.2 125 114.64 220.45 303.7 Cl (mg/L) 543.42 15.903 2.8 T (°C) 20.64 225.75 115.54 215.377 1.45 13.86 150.23 1085 810.843 2.68 24.Table 1: Physico-chemical properties of water samples for the Maknassy Basin shallow aquifer pH No.43 115.42 7.2 388.84 320.779 3.78 391.35 265.78 10.67 152.28 271.9 220.12 19.65 231.8 148.8 18.84 1.5 720.42 17.67 15.71 20.5 19.2 208.65 250.1 125.8 368.3 7.898 2.59 16.19 213.87 486.3 21.67 780.53 122.17 24.47 1065.45 292.31 230.67 873.45 118.73 . Tech.78 1892.2 145.38 Na (mg/L) 315.53 7.76 274.24 335.45 7.34 150.74 7.53 7.42 145.45 18.7 18 19.83 292.3 262.34 SO4 (mg/L) 1252.
2 PC 2 0.413 11.994 0. Ca) and their contribution on the dependant variable (TDS).062 0.8 -1.6 -0.483 0.000 Sig.2 0.4 0.047 0.2 -1.4 0.0 -0.4 -0. e3 and e4 are added to the statistical model in order to reduce the error estimation of path value (relation) between variables.6 Fig. The structural equation modeling was used to achieve the goal. Cl.058 5959.000 in order to test the potential interaction between them.451 t-value 5.0 -0.141 0.247 0. Khmiri 0.248 3. the sulphated facies ground water could be remaining to the dissolution and/or leaching of the abundant 514 . SO4.418 23.6 -0.000 0. This model hypothesizes the potential interactions between the independents variables (Na.689 18. Thus. 3: Plots of the PCA results showing the distribution of analysts 5000 4000 Linkage distance 3000 2000 1000 0 SO4 Cl Na HCO3 Mg Ca K T pH TDS Fig. [HCO3] and [Cl] result of water-rock interaction.285 0. e2. 4: Dendogram generated from hydrochemical data showing relations between variables Table 2: Summary of the linear regressions predicting TDS Ca Mg Na HCO3 Cl SO4 Constant Multiple R R square Adjusted R square Standard error F test statistics Significance β 0.19 0. Chenini.000 0.000 0.000 0. S.0 0. Mg.4 PC 1 -0.t 0.6 0. 5. e1.994 0. A generalized conceptual model was developed which is illustrated in Fig.2 -0.072 0.000 0.2 0.549 8.8 -0. HCO3.977 0.I. TDS in ground water sampled is constrained by [SO4].
number of estimated parameters) of the hypothesized model in the assessment of overall model fit. (1999) further cautioned that GFI and AGFI values can be overly influenced by sample size. The GFI and AGFI can be classified as absolute index of fit (Bentler.001 TDS e4 Cl TDS e4 0. The PGFI takes in to account the complexity (i. 1990).998 TLI 0.I. For Joreskog and Sorbom (1993) this two indices is impossible for them to be negative. The adjusted goodness of fit index (AGFI) differs from the goodness of fit index (GFI) only in the fact that it adjusts for the number of degree of freedom in the specified model. [SO4] and [Cl] are all endogenous variables. Environ. in a well-fitting model this value will be smaller than 0. it can be seen that the RMR value for this model is 0.001 e1 Na 0. with values close to 1 being indicative of good fit. [Mg] and [HCO3] are the exogenous variables in the structural model.934.001 Cl Fig. both indices range from 0 to 1. An exogenous variable’s causes lie outside the model. In the current model (Fig. was introduced to address the issue of parsimony in SEM. 1989). All factors loadings that were tested had t-values greater than 1.984 RMSEA 0.988 and 0. are represented by a simple index (PGFI).176 would seen to be consistent with fit statistics.996 CFI 0. concluded that this model fits the sample data fairly well. Parsimony goodness fit index (PGFI). Khmiri Int. Typically. Tech.e.96 and all of the path coefficients were significant.992 RFI 0. Summer 2009 Table 3: Fit indices for the structural model RMR 0. respectively). to logically interdependent pieces of information. It is concluded that the model fit the data well. J. In contrast to exogenous variables. Turning to Table 3. Although.13 0. 5 depicts the causal relationship among exogenous and endogenous variables.471 0.001 Ca 0. [Ca]. The goodness of the fit of the model as measured by the GFI and the parsimony of the model. Chenini.934 GFI 0. The root mean square (RMR) residual represents the average value across all standardized residuals and ranges from 0 to 1.05. They address the issue of parsimony by incorporating a penalty for the inclusion of 515 additional parameters. it can be.000.189 -0.. 6: A structural model of factors influencing TDS gypsum in geological formations and soil in Maknassy basin.621 SO4 -0. The goodness of fit indices for the structural model that are shown in Table 3 indicated the model has a good fit of the data. TDS. 6 (3).063 e3 e3 Mg Ca Na Mg SO4 HCO3 0. as such.. 509-519. the postulated causes of endogenous variables are included in the model. Normed fit index (NFI) has been the practical criterion .512 2. Fan et al. 5: A conceptual model of factors influencing TDS Fig.001 0. 6). [Na].55 0. S. Thus these findings of a PGFI value of 0.. thereby providing a more realistic of evaluation of the model (Mulaik et al. Fig.255 e1 1.001 e2 0.988 PGFI 0.998 IFI 0. In these causes. Based on GFI and AGFI values reported in Table 3 (0. parsimony based indices have lower than the threshold level generally perceived as acceptable for other normed indices of fit.176 NFI 0. Sci.000 AGFI 0. once again.361 e2 HCO3 4.
984 indicating superior fit (Hu and Bentler. first PCA is applied to identify the most important variables that control chemical variability. are the influence of a variable on another that is not mediated by any other variable. [HCO3] and [Ca] also indirectly influenced TDS. These factors can combine to create diverse water types that change in composition spatially and temporally. [SO4]. S.95 (Bentler. If an independent variable has more than one route of indirect influence on a dependent variable.90 was originally considered representative of a well fiting model. [HCO3]. it is clear that the RMSEA value for this model is 0. However. As such its computation is basically the same as the NFI. Direct effects. 1993). 1989). MLR is generated on the basis of chemical variables previously identified. The RFI coefficient values range from 0 to 1 with values close to 0. As shown in Table 3. 1986). thus it is concluded that the model fits the data well. The relative fit index (RFI) represents a derivative of the NFI and the CFI (Bollen.05 indicates good fit (Browne and Cudeck. Turning to Table 3. a dendogram show possible association between variables. [Cl] and [Ca].998 is consistent with that the CFI in reflecting a well fitting model. [Cl] and [Ca]). but it is the total effect that is more relevant. 1999). 1992). Then. 1999). [SO4]. [SO4]. topography and climate. addressing evidence that the NFI has shown a tendency to under estimate fit in small samples. Values less than 0. according to Bollen (1986). the Tucker Lewis index (TLI) (Tucker and Lewis. [Mg]. except that degree of freedom are taken into account. Bollen (1989) noted that the direct and indirect effects can help to answer important questions regarding the influence of one variable on another.063. 1992) has recently been advised (Hu and Bentler. Values for both the NFI and CFI range from 0 to 1. 2006).I.001). 1973). [Na]. Indirect effects are calculated by multiplying all the path coefficients for each route of indirect influence. Finally. some ions rapports and chemometric analysis were used to explore water samples quality in the bases of their composition. using AMOS. Bentler (1999) revised the NFI to take sample size into account and proposed the comparative fit index (CFI). Once it is verified that the model fits the data well. yield values ranging from 0 to 1 (Bollen. [Cl] and [Ca] had a direct effect on TDS (0. . The root mean square error of approximation (RMSEA) is one of the most criteria in covariance structure modeling. [HCO3].996) and CFI (0. [Na]. 1989). TDS is used as a factor defining general ground water salinity (Grassi and Cortecci. as evidenced in large part by the current classic of status of its original paper (Bentler. These studies often use parameters variations or samples variations. A revised cut off value close to 0. in order to determine the extent to which it is consistent with the data. He explained that the direct effect could be misleading when the indirect effect has an opposite sign. [SO4]. The incremental fit index was developed by Bollen (1986) to address the issue of parsimony and sample size which were known to be associated with the NFI. Thus. consistent with the other index noted here. both the NFI (0.. it is not surprising that finding of IFI = 0. [Na]. impact of discharge An assessment of the Chenini. the SEM allows to test in a simultaneous analysis the entire system of variables (TDS. then the indirect effects for each path are summed to calculate the overall indirect effects of the independent variable on the dependent variable (Bollen. Chemical ions correlations (Bennetts et al. It takes into account the error of approximation. the relationship between water properties are explored (variables). for in such cases the total effect may not be as strong as the direct effect shows. Finally. 2005). Each provides a measure of complete covariation in the data. Khmiri wastewater effluent of choice.998) were consistent in suggesting that the model represented an adequate fit of the data. In addition. [Mg]. The chemical composition of ground water is contr oll ed by man y fact ors th at include t he composition of precipitation. Fig. [HCO3]. [Mg]. but to date none has fully integrated the results of analyzed parameters in a mathematical model to better constrain the interpretation and include the possible interactions between the parameters. 516 Significant interpretations have been extracted from hydrochemical data using multivariate statistical methods to evaluate ground water quality. mineralogy of the watershed and aquifer. In this studies. Indirect effects are ones that are mediated by at least one other variable and the total effects are the sum of direct and indirect effects. it constitutes an accurate tool to evaluate relationship between TDS and other chemical properties of the ground water. 6 indicates that [Na]. The general purpose of this study was to investigate simultaneously the interaction between the different chemical components of the ground water of the phreatic aquifer of Maknassy basin. although a value > 0.
Hydrol. Atteia. P. Results of the st udy suppor t t h e pr evi ous h ydr och em ica l interpretation using ions correlations and ions rapport to evaluate the possible origin of the ground water mineralization. Environ. H. PCA were used to differentiate the water samples on the basis of their chemical compositions. u si ng hydrochemica l a nd isotopic evidence.. 4 0-6 2 (23 pag e s). Pouchan. Stone. MLR was used to evaluate relationship between TDS and other chemical water properties. Applied mu ltiva ria te a na lysis. (2001). WEF. J. J. O. Although. AWWA. New York.. Ame rica n Wa ter Work s Associa t ion. Hydrol. 1 1 2 (3). A case study: Suquia River basin (Cordoba-Argentina). 28812 89 4 (14 pag e s)... 2 38– 246 (9 pages). Valeria.. Tredoux. the illustrated approach exhibits the limitation that results of the study only verify that the proposed relationships among variables in the conceptual model were supported by the sample data collection (number of the analystes).. Compu ter progra m. Del Pilar.471). 2 nd . D. Alberto. Arbu ck le. (2006).. Ca u sa l mode ling via st ru ctu ra l equation system. south-west of France. the proposed methodology can be useful in many typical aquifers where hydrochemical data are available. (1 9 8 8 ).. Bernstein.S. [Cl] and [Mg] result of water-rock interaction (carbonate and halite dissolution). Sta nda rd methods for the ex a mi na ti on of wa t er a nd wa stewa ter (1 8 th. D. Ed. S. Andre. B. A. The model offers statistical integration of hydrochemical variables producing a more robust interpretation. 3 05 . Moreover. 35.Int... 1 7 8– 1 9 2 (1 5 pag es).. 2 41 . P.. D. (1 9 8 8 ). Therefore. Hydrochemica l characteristics of a qu ifers near Sutherland in the western Karoo. M. Water Res. Then. It is confirmed by a largely direct path influencing [Na] to [Cl]. Pietersen. Titus R. S. Moreover. Fabiana. P. Psycho l. M. Understanding the salinisation process for ground wa ter in a n a r ea of sou th-e a st ern Au str a li a . Com pa r a ti ve fit in dex es in structural models. W. Amos (Version 7 . G. On the fit of models to covariances a nd methodolo gy to the bu lletin. 31 7– 33 5. REFERENCES Adams. . M. J. Chenini. Although. P. (2005). ( 1 9 9 0 ). Handbook of multivariate experimental psychology.. J. L. (2 00 1).. Hydrol. De los Angles. 3 7 3 -4 1 2 .. M. W. Springer. [SO4] is directly influenced by [Ca] and [Na] while. 509-519. 1 07 (2).. (196 6). (19 8 5 ).. 517 . Geological su rvey professiona l pa per 4 98 A. It appears that the methodology illustrated in this paper allows to incorporate hydrochemical information (variables) into a statistical model that takes in consideration all possible interactions between the variables. [Ca] largely influenced [SO4] (4. A..0 ). It is possible to identify the major process controlling hydrochemical variation in studied aquifer. [HCO3] and [Cl]. (2 0 0 6 ). A. 4 2.. . Back. Franceschi. A.. Bentler. I.. M. P. Using geochemica l da ta a nd modeling to enha nce the understanding of grou nd water flow in a regiona l deep aquifer. Ben tler. Cecilia. D. Khmiri Summer 2009 Tech. SEM infers that the ground water quality in the phreatic aquifer of Maknassy basin is dominated by [SO4]. C. M.. H. Bennetts. SEM pr ovi des a n a dequa te expla n a t ion of t h e simultaneously interaction between the variable i n cl uded i n th e con ceptua l model . (1992). M. CONCLUSION Ground water datasets of the phreatic aquifer of Maknassy basin in central Tunisia was investigated for their chemical composition differences with a dominance of [SO4]. PCA rendered considerable variable reduction and clearly distinguish between variables group. Ed . The procedure is aimed to explore ground water quality... [Ca] to [SO4] ratio is lower than 1 since [Ca] participate to the process of basic exchange with clay during ground water flow. South Africa. Harris. Chicago. J.. U. Bu ll. L. The correlation between [Na] and [Cl] and the [Na] to [Cl] ratio in close proximity to 1 may be two indicators of halite dissolution during ground water transit. SPSS.. I. Sci.. 6 (3). [HCO3]. New York . J. Webb. APHA. Plenum..) . [Mg] indirectly influenced [SO4]. Be ntl er. 3 2 3 (1-4 ). S. M.. J. M. Wa shin gton DC. Hydrochemical facies and ground water flow patterns in the northern part of the Atlantic Coastal Pla in.. Hill. Pattern recognition techniques for the evaluation of spatial and temporal variations in water quality. K. It is a complementary tool to MLR result testing the potential interaction between variables. D. 91 -10 3 (13 pag e s). Goodness of fit of this model can be worked by multiplying water sample and analyzed parameters. It is clear that TDS is dominated by [Ca] and [Mg] for cations and [SO4] for anions. Psychol. 4 0 0 -4 0 4 (5 pag es). Bu ll. The integration of PCA.. MLR analysis and structural equation modeling to evaluate ground water quality of the phreatic aquifer of Maknassy basin showed that water chemistry is related to water-rock interaction and/or dissolution of evaporitic formation from where dominance of [SO4]. Aquitaine Basin.
(1999). Environ. (1989). J. Multidisci.. SO4 2"/Cl ” a nd Mg 2+ /Ca 2 + ra tios a s tra cers for the evolu tion of seawater into coastal aquifers: The example of Castell de Ferro aquifer (SE Spain). A. Liedholz.. B.. J.. G. Ac. New York. Hydrol.. Ghasemi.. 16 (7-8). (2 0 0 5 ). Jerzy. A. Structural equa tion modeling with AMOS. Byrne. (20 05 ). Hida lgo.. Bull.. hydrochemical evolution a nd mass transfer in a regional detrital aquifer (Baza Basin... Anal. J. New York. southern Spain). Pulido. Methods for a ssessing soil qu a lity. 211 (1-2 ). R. Wiley. Black. Chim. 2 98-30 3. G. Lind.. R.. Cronin. 11 79-1192 (1 4 pag es).. 18 1 (1 -4).. Jolliffe. A. A. (2 0 0 2 ). northea stern Au stra lia .. Glaser. Onta rio. Wust. Sta tistica l a nd hydrochemical methods to compa re basalt and basement rock-hosted grou nd wa ters: Atherton Ta blelands.. M. L.. T. 6 (1). 3 4 (3). R. Ca lva che. Principa l components selection by the criterion of the minimu m mean difference of complexity. A. P. North India. 385-399 (15 pages). Ben Mammou. P. S. S. Pardo. Multivariate data analysis with reading. LISREL8: Structural equation modeling with the SIMPLIS command language: Scientific software internationa l.. J. Special publication. A... M. Fernandez. Appl... USA. J. Helena. Geochemical controls on peatland pore wa ter from the Hudson Ba y Lowland: A multivariate statistical approach. (2000). 99-106 (8 pages). K. Singh. M. J. I. 103910 48 (1 0 page s).. (1989).. R. Res. Bentler. Student Ed.. Ga bor. E.. In Dora n. Central New South Wales. J. Modl. Joreskog.. 56-83 (28 pages)... 8 2-91 (10 page s). Springer-Verlag. Wa ter Res. Eds. 48 (3). P. Carroll. Siegel. Sociologic. Martinez-Cortizas. Potenza . R. Tatham. How to standardize regression coefficients.. A. J. K... Sample size and Bentler and Bonett’s non normed fit index. (2003). Saaidpour. A. Fa ctors which determine the hydrogeochemical behaviour of k arstic springs: A case study from the Betic Cordilleras. R. (19 98). 49.. C. M.. Geochem. Parkin. van-de-Griend. Chem.. J. Geochim. Kline.... J. Eq.. 4 49 463 (1 5 pages).. Green. D. (200 3). V. E. Cosmochim. Appl. Fronell.. Recha rge velocity a nd geochemica l evolu tion for the Permo-Tria ssic Sherwood sa ndstone. A. New York.. J. G. 550 (1-2). M... M. (1986). (1 9 82 )... Ground water resources in Maknassy Basin (central Tunisia): Hydrological data analysis and water budgeting. 111-1 34 (24 page s). Principles and pra ctice of structu ral equation modeling. Hydrogeology a nd geochemistry of the multilayered confined aquifer of the Pisa plain (Tuscany – central Italy)... Kylandern... Madison. Lopez-Chicano. L.. Thompson. Multidisci. 25-37. S. Barrado. Locsey.. A... E. B. Stilwell. 4 1-54 (1 4 pag es). G. Psychol. Hu.I.. Seyhan. (2007). 4 3 . L... M. Muller. Weiss. M. 5 5-7 5 (21 pages).. Northern Irela nd. Ca na da : Interna tiona l Associa tion for Ma thema tical Geology. M. D. Kingston. (19 85). M. Structural equation with latent variables. K. D. estima tion methods a nd model specifica tion on stru ctu ra l equ a tion modeling fit indexes. Cudeck. 1-5 5 (55 page s).. (2 00 1). S. C. Ma pping of hydrochemica l grou nd wa ter regimes by mea ns of mu ltiva ria testa tistica l a na lyses. 209-213 (5 pages). Guilford Press. T. Cox. Bennett. J. Cortecci. Mathematical tools for applied multivariate analysis. Cutoff criteria for indexes in covaria nce stru ctu re analysis: Conventional criteria versus new alternatives. K. J. Geosci. W. (2005 ). Elliot. (1994). 21(2 ).. Cruz-Sanjulian. Spain.. L... X. R. Chemometric analysis of ground wa ter qu ality da ta of alluvial a quifer of Ga ngetic plain.. B. Barth. (2005).. (2008). Sinha. 285 -304 (2 0 pages). T.. 3 7 5 -3 7 7 (3 pa g e s).. Geochem. G.. Pu lido-Bosch. Ground water composition.. C.. Shane. Ischia.. T he u se of principle component a na lyses in characterising trace and major elemental distribution in a 55 kyr peat deposit in tropical Australia: Implications to paleoclimate.. Engelen. J. J. 43 0-44 5 (1 6 pag es). Stat. In Proceedings of the fourth annual conference of the International Association for Mathematical Geology. Turki. pu blishers. L. 5 1 (3 ). Quantitative indicators of soil qu a lity: a minimu m data set. Effects of sample size. Appl. 335 (14). A. S. Anal. 230258 (2 9 pages). J.. Na rdi a nd R.... 520. S.. M. 315 (1-4). Geol. Vega. T... WI: SSSA. Multivariate Anal. J. Anderson. Wang. C. E.. J. Hair.. M.. P.. (2001). 21 (7). A. C. S.. Geosc. Bring. Chenini. 6 9 8 -7 1 3 (16 page s). (2 0 0 8 ). J.. K. Gu pta. 1.. 20 (1 ). Reeve. 1010-1024 (15 pages). Evaluation of goodness-of-fit indices for structural equation models. Doran. Blake. Mahwah. A.. 518 . Fernandez. M. Methods. 6 (1). L. R. Fan. P. Vanaltine.. 7 2 (2).. Strontium. New Jersey. F.. Pu lido-Leboeu f. M. James. Bouamama Vallejos.. N. W. A. C. A second generation of mu ltivariate analysis. Psychometrik a .. USA. Gra ssi. Andreu. 80 7-81 6 (1 0 pag es). Vallejos.. D.. Austra lia. B. 16 (9-10). northern Italy. Mercier. B. New York. Hydrochemistry a nd isotopic composition of Na –HCO3-rich ground wa ters from the Ballimore Region. (1996). (1 993 ). 7 45-75 8 (1 4 pages). (1996). 49 (1 ). I. R. Mulaik. Scha fmeister. (2 0 0 3 ). M. (1999). E. W. B. (2 0 0 1 ). G. L. K. Garcia-Sanchez. Chica go. Praeger Vol.. Acta. Bollen. Bu ccia nti. Struct. I. K. Coles. Chenini... 5 th Ed.. B.. Ed. Quantitative structure– property rela tionship study of n-octanol–water pa rtition coefficients of some of diverse drugs using multiple linear regression. Mohan. S. Singh... Malik. H. D.. A. Upper Saddle River (NJ: Prentice-Hall). Spain) by principal component analysis. Khmiri Bollen. (1996). 308-3 24 (17 pages). 604 (2). Jones. Acta. Qia n. A. Res. R. Alternative ways of assessing model fit. P. Chim.. Geol. (Eds) Lawrence Erlba um associates. B.. A. W. Geochem. E.. R. Itala y. M. (1993). Hydrol. Principa l component a na lysis. (1 9 9 8 ). A. 12 (4).. Kalin. T. J. M.. S. October 5 –9. Browne. (1 9 9 4 ). A. Temporal evolution of ground water composition in an alluvial aquifer (Pisuerga River. Water Resour. Academic Press. T. Sorbom. Mu ltiva ria te a na lysis a nd interpreta tion of the hydrochemistry of a dolomitic reef aquifer. Meth. 105 (3). J. Am.
. Psychometrika. Faculty of Sciences of Tunis. S. Summer 2009 Tech. Multivariate a nalysis in hydrochemistry: An exa mple of the u se of factor and correspondence analysis.. Khemiri. Department of Geology. Tucker. E. Lee. Environ.. 2092 Tunis El Manar. 27 (1).. K. 519 . Sci.R. I.... (1973). Department of Geology. L. Lewis. Grou nd Wa ter. S.fr This article should be referenced as follows: Chenini. (1989 ). Sci.. 1–10 (9 pages). 6 (3). Structural and applied geology. Email: fantsam01@yahoo. 358-366 (9 pag es).. J. 38 (1). (1999). A.fr Khemiri. Senior lecturer and Ph. AUTHOR (S) BIOSKETCHES Chenini. H. 27-34 (8 page s). 509-519. K. Characterization of a ground wa ter hydrochemica l system throu gh mu ltiva ria te a na lysis: Clu stering into grou nd wa ter zones. 2092 Tunis El Manar. K.. Usunoff. Khmiri Suk. Minerals resources and Environement Laboratory. Chenini. Tunisia. J.. C. Faculty of Sciences of Tunis. Associate professor in hydrogeology. Tech.. I. (2009).509-519. 6 (3). A reliability coefficient for maximum likelihood factor a nalysis. J. research candidate in U. Environ. Email: chenini_ismail@yahoo.D. Int.Evaluation of ground water quality using multiple linear regression and structural equation modeling.Int. Tunisia. 37 (3 ). Guzman-Guzma n. Ground Water.I. S.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.