1 23
Arabian Journal of Geosciences ISSN 1866-7511 Arab J GeosciDOI 10.1007/s12517-012-0814-0
Assessment of the effect of a plannedmarina on groundwater quality in Enfidaplain (Tunisia)
Mounira Zammouri, Faten Jarraya-Horriche, Bio Oumaro Odo & SihemBenabdallah
1 23
Your article is protected by copyright and all
rights are held exclusively by Saudi Society
for Geosciences. This e-offprint is for personal
use only and shall not be self-archived in
electronic repositories. If you wish to self-
archive your work, please use the accepted
author’s version for posting to your own
website or your institution’s repository. You
may further deposit the accepted author’s
version on a funder’s repository at a funder’s
request, provided it is not made publicly
available until 12 months after publication.
ORIGINAL PAPER
Assessment of the effect of a planned marina on groundwaterquality in Enfida plain (Tunisia)
Mounira Zammouri & Faten Jarraya-Horriche &
Bio Oumaro Odo & Sihem Benabdallah
Received: 20 August 2012 /Accepted: 21 December 2012# Saudi Society for Geosciences 2013
Abstract The Enfida aquifer system is of importance to theeconomic activity of the eastern center of Tunisia. Theplanning of a marina is likely to present a significant riskon groundwater. Classical physically based modeling isused to better understand salty water intrusion in the aqui-fers. The transport model (MT3DMS) is coupled with thegroundwater model (Modflow). Model calibration was car-ried out over the period 1972–2005. Four scenarios werethen simulated for a 50-year period, to assess the effects ofboth planned marina and future abstraction regime. Wepredict a rise in the groundwater salinity generated by theplanned pumping infrastructure. The impact of the plannedproject will be observed only near the marina. However,limited measurements of transmissivity may affect the mod-el’s results. Thus, the second part of the paper is aimed toassess the models output error due to the uncertainty intransmissivity, using a stochastic approach. Hundred real-izations of a log-normal random transmissivity field hadbeen performed. According to the most pessimistic realiza-tions, the uncertainty may reach 49 % in the sector of an
important pumping field. Accordingly, the calculated con-centration may reach 6.5 g/l in 2055 instead of 3.2 g/l.
Keywords Hydrogeology . Groundwater . Flowmodeling .
Solute transport modeling . Kriging . Uncertainty .
Stochastic simulation . Enfida plain
Introduction
The impact of urbanization may affect water resources in termsof quality and quantity (Appleyard 1995; McAllister et al.1996; Changming et al. 2001; Lerner 2002; Schiedek et al.2007; Carlson et al. 2011 etc.). In such situation, modeling is anefficient tool to assist managers and decision makers in assess-ing their impact before implementation. The planning of amarina over an area of 30 km2 in the coastal zone of Enfidaplain, in Tunisia, is likely to deteriorate the plain ecosystem andpresent significant risk on groundwater. In fact, the planned 20-m depth basin to be dragged on theMediterranean Sea threatenfuture water supply in the Enfida plain where groundwater isthe most important freshwater resource. The Enfida plain ischaracterized by a Mediterranean environment where precipi-tation is erratic and moderate. It contains shallow and deepaquifers with variable spatial quality. It is also one of majoraquifer in eastern center of Tunisia with regard to its extent andits relevant yield. Renewable resources, estimated to 16×106 m3/year, are exploited for irrigation and drinking watersupply. The current exploitation, rising to 13.5×106 m3/year,indicates a strong demand for groundwater, a key resource toeconomic development in the region. The possibility of sea-water intrusion and consequently of an increase in salinitywould lead to direct water supply difficulties.
In this paper, a groundwater model, coupled to a transportmodel, is built in order to study the impact of the planned
M. ZammouriFaculté des Sciences de Tunis,Campus Universitaire, Tunis El Manar 2092, Tunisiae-mail: [email protected]
F. Jarraya-Horriche (*) : S. BenabdallahCentre de Recherches et des Technologies des Eaux,BP 273, Soliman 8020, Tunisiae-mail: [email protected]
S. Benabdallahe-mail: [email protected]
B. O. OdoEcole Supérieure des Ingénieurs de l’Equipement Rural de Medjezel bab, Route du Kef, km 5, 9070,Jendouba, Tunisiae-mail: [email protected]
Arab J GeosciDOI 10.1007/s12517-012-0814-0
Author's personal copy
marina on the Enfida groundwater quality. The adopted meth-odology combines a deterministic model with geostatisticaltool (stochastic simulations). Deterministic models are com-monly used to simulate groundwater flow and solute transportin aquifers (Bear and Verruijt 1987; Anderson and Woessner1992; Bear and Cheng 2009; Konikow 2011). Limited knowl-edge combined with natural spatial variability may causeuncertainty in values of the predictive variables of the models.Several authors reported the inaccuracy of deterministic mod-els and the uncertainty in predicting contamination in suchsituations (Fried 1975; Anderson 1979; Matheron andMarsily1980; Simmons 1982 etc.). Thus, coupling stochastic anddeterministic modeling approaches provides a way of evalu-ating uncertainties on model predictions and can be used as atool to help decision makers (Naji et al. 1998; Cinnirella et al.2005; Al-Bitar and Ababou 2005; Abarca 2006; Renard 2008etc.). The purpose of this paper is to assess the impact of aplanned marina on the groundwater of Enfida plain. This isdone by integrating two approaches. Firstly, a geologic, hy-drologic, and hydrogeologic database is elaborated and ana-lyzed using kriging techniques for the aquifer systemcharacterization and the conceptual model development. Infact, classical deterministic modeling approaches are usuallybased on local measurements that generally do not cover thewhole study area. Kriging techniques (Matheron 1970;Delhomme 1976; Journel and Huijbregts 1978; Isaak andSrivastava 1989; Chiles and Delfiner 1999; Kitanidis 2000;Clark and Harper 2000) provide appropriate tools to analyzedata and assess their distributions needed for model elabora-tion, especially when setting model parameters, referencepiezometry for the steady state calibration and initial condi-tions for the transport model. Secondly, a groundwater flowmodel and a transport model using Modflow and MT3DMSare developed to determine more insights into the groundwa-ter flow in the vicinity of the marina. Stochastic simulation’stechnique is applied to transmissivity to analyze the predictivesensitivity and uncertainty of model’s results.
Materials and methods
Study area description
The Enfida plain is located at the eastern center of Tunisia, andextends over an area of about 1,200 km2 (Fig. 1). It is encom-passed completely by mountains and hills from the north,west, and southeast boundaries. It is bounded by salty marshes(Kalbia sabkha) from the southwest and the Hammamet gulffrom the east. The plain is located in a vast syncline oriented tothe southwest–northeast direction. It is limited by the DraaSouatir monocline in the west and by the Kalaa Kebira anti-cline in the south. The bedrock is formed by the middleMiocene marls. Two different subsidence basins are
identified: Sidi Abich-Chegarnia in the north and Kondar-Sidi Bou Ali in the south. The basins primarily contain sandand clay formations of the Mio-Pliocene age and wadis allu-via. The sediments are extremely heterogeneous, comprisingconglomerates, clays, coarse sandstone, sands, and clayeysands. The Enfida plain center is characterized by the presenceof several sabkhas bordering the coast. The sedimentation ofthe sabkhas has been evaporitic and lacustrine since the latePleistocene. They are characterized by clays and salty materi-als (gypsum) for a depth of about 60 m. The annual averagerainfall is 340 mmwith high seasonal variability, a mean dailytemperature of 20 °C, and a yearly potential evapotranspira-tion reaching 1,730 mm. According to Martonne (1942), thisbasin is located in a semiarid climate with an aridity index of11.5 mm/°C. The plain contains an alluvial water table whichis important for the economic activity of the region. Enfidagroundwater is primarily used not only for agriculture anddrinking water supply but also for industry to a lesser degree.The water table is mainly recharged from ephemeral streams(wadis), with the most important stream, El Khairat wadi.Runoff from wadis, feeding groundwater, flows towards thesea without reaching it, with the exception of strong floods, asthey disappear into terrain depressions, so-called sabkhas. In1999, El Khairat wadi was dammed upstream of the plain inorder to protect the Enfida town against flooding. The damreservoir is also used for groundwater artificial recharge andfor cereal irrigation. Many artificial recharge operations werecarried out to re-establish the natural recharge affected by thegroundwater abstraction (Zammouri et al. 2006).
Hydrogeologic characterization of Enfida plain
The hydrogeology literature is scarce for this region (El Batti1974; Manaa 1989; Manaa et al. 1996). Most information wasgathered from Tunisian Water Authorities at the national andlocal levels (DGRE and CRDA of Sousse). It related to hydro-geology, rainfall, runoff, groundwater levels, groundwaterabstraction, surface water, and groundwater quality. A geo-logical database was elaborated in order to realize lithostrati-graphical cross-sections, and to map the elevation of theaquifers’ base and confining beds and to determine theirrelative thickness in order to identify the hydrostratigraphicunits (Horriche and Zammouri 2008). Drilling logs (DGREand CRDA of Sousse) and geophysical survey results (DGP2006; SCET 2007) were analyzed. For the steady state con-dition in 1972, we elaborated a piezometric map for thephreatic aquifer using 110 measurement values available froma local survey (El Batti 1974). The values are the average ofobservations during two field campaigns, representative of thehigh and low waters. We used universal kriging interpolationwhich is one of the basic tools used for water level analysisand largely described and discussed in the literature(Matheron 1970; Delhomme 1976; Journel and Huijbregts
Arab J Geosci
Author's personal copy
Fig. 1 Location and overview of the Enfida plain
Arab J Geosci
Author's personal copy
1978; Isaak and Srivastava 1989; Chiles and Delfiner 1999;Kitanidis 2000; Clark and Harper 2000; Theodossiou andLatinopoulos 2006). For the deep aquifer, we elaborated apiezometric map based on the universal kriging interpolationand using 47 measurement values. The kriged map is repre-sentative of the year 1972. Hydrographs of groundwater leveland discharge rates were analyzed to study the historicalevolution over the period 1973–2005. Groundwater levelmonitoring network comprises about 60 observation wells(Fig. 1) with 17 for the deep aquifer. Biannual piezometricsurveys started in the 1970s for the phreatic aquifer and since1980 onwards for the deep aquifer.
Groundwater modeling approach
In order to assess the future impact of the planned marina onthe Enfida aquifers over the next 50 years, a quasi three-dimensional finite-difference flow model was developedusing MODFLOW-2000, a modular finite differencegroundwater model code (Harbaugh et al. 2000). The gov-erning equation of groundwater flow (Bear and Verruijt1987; Anderson and Woessner 1992) is:
@
@xTx;i
@hi@x
� �þ @
@yTy;i
@hi@y
� �þ liþ1 hiþ1 � hið Þ
þ li�1 hi�1 � hið Þ ¼ Si@hi@t
þ qi
ð1Þ
where hi, hi+1, hi−1 are hydraulic heads (L) of layers i, i+1and i−1, respectively; Tx,i and Ty,i are components of trans-missivity (LT−1) parallel to the x and y axes within a layer i,respectively; 1i+1 and 1i−1 are leakance (T
−1) between layersi and i+1, and i and i−1, respectively; Si is the storagecoefficient (1) of layer i; qi is source/sink of water (LT−1)in layer i; x, y are Cartesian coordinates (L); and t (T) is thetime.
For model pre- and post-processing, processingMODFLOW was used (Chiang and Kinzelbach 2001). Indeterministic modeling approach, setting initial conditionsof the model and validating its results are important issues.The reference period for the steady state calibration is 1972,during which the aquifers were considered to be in a state ofequilibrium on a basin wide level. Steady state calibrationconsisted of reproducing the general piezometric maps forthe phreatic and deep aquifers. The time period from 1973 to2005 was taken as the reference period for transient calibra-tion. The reproduction of the drawdown trend was taken ascalibration criteria. Horizontal transmissivities and storagecoefficients for both phreatic and deep aquifers, specificyield for phreatic aquifer, leakage between phreatic anddeep aquifers and hydraulic resistances between phreatic
aquifer and soil surface at the level of the wadis are thecalibration parameters. The calibration process was manual.
Hydrodynamic model requires aquifer parameterization inaddition to the aquifer geometry and boundary conditions.Local measurements are unequally distributed over the aquiferand may not be representative of the large areas. Kriging wasused to estimate distributed parameters introduced as initialguesses for the model calibration. Data about vertical trans-missivity, specific yield, porosity, and phreatic aquifer trans-missivities were mainly estimated from the geological data.Local measurements of these parameters are scarce or unavail-able. Data related to deep aquifer transmissivities reflect greatheterogeneity. Thus, this parameter was interpreted fromdrawdown data for short- and long-term pumping tests. Dataon specific capacity are also used. In fact, they are generallymore numerous as they are simpler to measure. The logarith-mic empirical relationship between transmissivity (T) andspecific capacity (Qsp), a well-recognized relationship in theliterature (Delhomme 1976; Aboufirassi and Marino 1984;Ahmed and Marsily 1987; Razack and Huntley 1991;Razack and Lasm 2006), was computed based on 40 mea-sured dataset within the samewells. Measured transmissivitiesand those estimated using specific capacity count 54.
After the flow model calibration, a transport model wasbuilt by using the modular three-dimensional multi-speciestransport model MT3DMS (Zheng and Wang 1999). Thetransport module simulates the advection and dispersion ofconservative and dissolved substances. The governing equa-tion for solute transport (Bear and Verruijt 1987; Andersonand Woessner 1992) is:
@
@xiDij
@C
@xj
� �� @
@xiCvið Þ � C0q
we¼ @C
@tð2Þ
where C is the solute concentration (ML−3), Dij are thedispersion coefficients (L2T−1), vi are the components ofthe velocity vector (LT−1), ωe the effective porosity (1), C 0
is a known source concentration (ML−3), q is a source/sinkterm (T−1), xi are Cartesian coordinates (L); and t (T) is time.
Thus, the layers of the flow model were further discretizedvertically in order to avoid instantaneous vertical mixing. Eachaquifer was divided into three layers. The vertical and hori-zontal permeabilities, geometry, and boundary conditions ofthe six-layer model were correspondingly adopted to conformto the existing two layer flow model. Due to the lack of tracertest data, dispersivities were calibrated. Resistivity values pro-vided by electrical diagraphy were used to estimate the effec-tive porosity. The initial salinity distribution of the aquifers in1972 was based on available geochemical data. Water sampleswere taken from shallow wells and analyzed with regard togroundwater salinity, during two field campaigns that tookplace in May and September 1972 as part of a research studycarried out by El Batti in 1974. Representative salinity is given
Arab J Geosci
Author's personal copy
in total dissolved solids (TDS). Measurements of each cam-paign amount to 123 values of TDS are distributed over largeparts of the basin except the northwest and southwest limits.They are representative of the two seasons, high and lowwaters. Kriging estimates based on the average of 123 TDSvalues were mapped to reproduce the spatial TDS distributionand the standard deviation error. For the deep aquifer, salinityand TDS distribution were derived from surveys related todrilling wells within the region. The collected informationrelates to 91 deep wells are used. Hydrographs of groundwatersalinity are analyzed to study the groundwater quality evolu-tion and to calibrate the transport model. Annualsystematic measurements started in 1995 with the im-plementation of the groundwater quality monitoring net-work in 1995. It comprises 17 control points (Fig. 1)with nine in the phreatic aquifer and the others in thedeep one.
Uncertainty assessment of model prediction
The prediction of saline water migration using deterministicmodels is usually faced with insufficient information relatedto transmissivity, specific yield, and porosity. This lack ofinformation affects the estimation of groundwater level intime and space. This topic was subject of intensive researchin the literature (Fried 1975; Anderson 1979; Matheron andMarsily 1980; Fried 1981; Carrera 1993; Ackerer et al. 1994etc.). Deterministic and stochastic modeling approachesmay be coupled to assess the effects of the model parameteruncertainties on model predictions (Zhou and Van Geer1992; Naji et al. 1998; Rentier 2003; Cornaton 2004;Cinnirella et al. 2005; Al-Bitar and Ababou 2005; Abarca2006; Renard 2008; Kerrou 2008 etc.). The transmissivity isthe most important parameter that controls the groundwaterflow. It influences the transport velocity and thus the transfertime of the contaminant. Its spatial variability can have asignificant effect on the field-length dispersion of contami-nant plumes. The effective porosity affects also contaminanttransport. However, the uncertainty effect of the effectiveporosity is much less significant than that of the transmis-sivity. In fact, the effective porosity, whose interval variationlies between 1 and 30 %, does not show a degree of spatialvariability as high as the transmissivity which can varyseveral orders of magnitude on a few meters (Freeze 1975).
For the case of Enfida plain, local measurements of thephreatic aquifer transmissivities are insufficient to use thestochastic approach. Accordingly, we consider herein onlyuncertain transmissivity of the deep aquifer. A stochasticmethod, based on Mejia’s algorithm (Mejia and Rodriguez-Iturbe 1974; Rodriguez-Iturbe et al. 1998; Ferraris et al.2002) was used to generate these uncertain values.Hundred realizations of transmissivity field were producedusing the random log-normal field generator developed by
Frenzel (1995) and available within the software PMWIN(Chiang and Kinzelbach 2001). These generated valueswere introduced to the deterministic model, under equiprob-able realizations as a possible representative of the real fieldvalues, to simulate groundwater flow. This method was usedin a wide range of hydrologic researches and engineeringproblems involving groundwater pollution modeling(Brovelli et al. 2010; Robinson et al. 2009; Dreybrodt etal. 2009; Cornaton 2004).
Results and discussion
Enfida model setting
The geological database analysis confirms a multilayerstructure of the aquifer: two aquifers (Fig. 2). The superficialand deep aquifers are separated by semipervious layers ofclays, sandy clays, and clayey sands. Their thickness variesbetween few meters and 60 m. The thickness of superficialand deep aquifers is important (50 to 200 m) in the SidiAbich-Chegarnia basin and in the valley of the El Khairatwadi. The semipervious layer is thicker in the north than inthe valley of El Khairat wadi. Both aquifers and semiper-vious layers are thinned towards the littoral. In the centerand the south, the thickness of the first and second aquiferranges from 50 to 100 m. The semipervious thickness isalmost constant (30 m). The superficial aquifer is thinned inthe southwest and disappears in the northwest.
The phreatic aquifer is contained in the superficial onecorresponding to alluvial fillings of wadis, sand dunes, andcontinental or marine sediments of the old Quaternary. Thedeep aquifer is contained in the Mio-Plio-Quaternary sedi-ments overlying the middle Miocene marls. It is confinedover the whole area except in the west where the Mio-Plio-Quaternary formations outcrop. The shallow wells tap thephreatic aquifer, whereas the deeper ones tap the deepaquifer. Aquifers get recharged by infiltration throughfloods descending from the mountain ranges located to thenorthwest, and west of the Enfida plain and by infiltrationthrough surface runoff descending from southern hills. Thephreatic aquifer gets also recharged by direct infiltration inthe north and along the littoral which is characterized bysands, fine sands, and gravels. The water table piezometrydistribution was obtained by universal kriging interpolation.The experimental variogram was computed on the basis ofthe 110 available data. A variogram model is fitted with thefollowing basic structures: a nugget effect and an exponen-tial model (Fig. 3a). Validity of the selected structural modelwas checked by cross-validation (Isaak and Srivastava1989) giving a satisfactory result, with the average of theactual errors close to zero and with their variance minimum(Table 1). The scatter diagram (Fig. 3b), indicating a
Arab J Geosci
Author's personal copy
correlation coefficient of 0.85, shows a great magnitude inthe differences between measured and estimated values forisolated points. These points indicate that either the mea-surement is incorrect or the measurements density of thearea where the observation well is located is low. Phreaticgroundwater is in general from the west and southwesttowards the east and the northeast. A part gets removed byevaporation and the remainder flows to the MediterraneanSea. For the deep aquifer, the experimental variogram wascomputed, adjusted, and tested by cross-validation(Table 1). The final result was a spherical variogram. Asshown in the piezometric map, obtained by universal
kriging interpolation (Fig. 4), groundwater flows from thewest towards the east in the north and in the plain center. Atthe level of Kalbia sabkha, flow occurs mainly from thesouthwest towards the northeast. Mediterranean Sea andsabkhas are also the discharge area for groundwater fromthe deep aquifer.
Hydrochemical investigations started in the early 1970s.However, systematic observations were only available since1995 onwards. An average experimental variogram is firstcomputed for the two field campaigns May and September1972 (El Batti 1974). A nugget effect and a spherical modelare fitted to this experimental variogram (Table 1). Kriged
Fig. 2 Lithostratigraphic cross-section through the Enfida plain (southwest–northeast)
0
10
20
30
40
0 10 20 30 40
Est
imat
ed v
alue
s (m
)
Measured values (m)
0 2000 4000 6000 8000 10000 12000 14000
Lag Distance (m)
0
5
10
15
20
25
Var
iogr
am
(a) (b)
Empirical and theoretical variogram Correlation between measured and estimated hydraulic head
Fig. 3 a, b Experimental variogram and fitted model for phreatic aquifer piezometry (piezometry in 1972, in meters)
Arab J Geosci
Author's personal copy
TDS map, elaborated for the field campaign of 1972 (ElBatti 1974), shows values ranging from 1.5 to more than20 g/l in the phreatic aquifer (Fig. 5). The low valuescorrespond to the recharge zones mainly the wadis and inSidi Abich basin. The high values are located in the center,at the level of the sabkhas and the bordering regions whichare the discharge area. For the deep aquifer, after testingseveral types of theoretical variograms, the best fit to theexperimental variogram was the spherical type (Table 1).Groundwater salinity distribution in the deep aquifer (Fig. 6)shows a low salinity at the level of Sidi Abich basin, El
Khairat wadi valley, and the sectors of Draa Souatir andKalaa Kebira that correspond to the recharge areas. In thebasin center, groundwater quality is as poor as that of thephreatic aquifer.
The economic expansion of the Enfida plain from theearly 1980s onwards was mainly accomplished by the large-scale allocation of groundwater resources involving a grad-ual transition from traditional groundwater abstraction fromhand-dug wells to high-volume abstraction by motor pumpsand from deep wells. Water quantities extracted from thedeep aquifer increased from 0.08 m3/s in 1972 to 0.17 m3/sin 2005. Withdrawals from shallow wells reached 0.22 m3/sin 2005. The total abstraction increased more than threefoldfrom 0.11 m3/s in 1972 to 0.39 m3/s in 2005. As a conse-quence, regional decline of groundwater level is observed.Over the last few decades, groundwater levels in the deepaquifer had fallen at an average rate of 0.2 m/year in the SidiAbich region, 0.4 m/year in the El Khairat wadi valley and0.3 m/year in the south (see Fig. 8). In the phreatic aquifer,observation wells show different behavior with a loweredgroundwater level at a rate ranging from 0.1 to 0.3 m/year insome wells (see Fig. 8), and with a quasi-stationary ground-water level in others. The observation wells for water qualitydo not show any deterioration of groundwater.
Flow and transport model’s calibration
The study area was horizontally discretized into square cellsof 500 m side length. The chosen mesh size is coherent withthe average data density and allows representing the systemaccurately. The model boundaries follow the natural Enfidabasin limits except in the south. The southern limit wasartificially chosen with null flux by tracing an orthogonallimit to the contour lines of piezometric maps developed for1972 data. Vertically, the upper model layer corresponds tothe phreatic aquifer. Accordingly, the lower model layercorresponds to the deep aquifer. The two aquifers wereconnected by a leakage term. Boundary conditions withfixed head defined the Mediterranean Sea. Rainfall infiltra-tion through permeable zones was represented by flux con-ditions. Imposed inflow conditions were also used to define
Table 1 Results of cross-validation test performed for piezometry, TDS, and transmissivity data
Variable Count (N) Variogram model Nugget effect Sill Range (km) ME MSE
Water table piezometry (m) 110 Spherical 0 20.1 8 0.02 11.57
Deep aquifer piezometry (m) 31 Spherical 0 448.3 15 0.57 344.1
Phreatic aquifer TDS (g/l) 122 Spherical 3.4 11.8 3.2 0.05 14.5
Deep aquifer TDS (g/l) 91 Spherical 0 5.3 15 0.16 4.13
Log10(T) (T in m2/s) 54 Spherical 0.05 0.2 6.5 0.016 0.29
T transmissivity of deep aquifer, ME mean error: Σ(h*−h)/N where h indicates the measured variable in question and h* the kriged value, MSEmean squared error: Σ(h*−h)2 /N
Scale
0 10 20 Km
Hergla
Enfida
Sidi Bouali
Kalaa Kebira
Fig. 4 Kriged piezometric map for the deep aquifer in 1972 (piezo-metric contour lines in meters)
Arab J Geosci
Author's personal copy
the recharge by infiltration from floods from the northwest-ern and western mountains, Kalaa Kebira hills and the DraaSouatir monocline. Cauchy boundary conditions representthe drainage of the groundwater by wadis and sabkhas. Inthis context, a digital elevation model was interpolated byordinary kriging data determined by the digitalized topo-graphic maps. General head boundary conditions were cho-sen to represent the lateral inflow from southwestern modelboundary. Direct recharge from precipitation, drainage andevapotranspiration occur from the upper layer only.
Based on a previous work carried out in the Kairouanplain located in central Tunisia (Nazoumou 2002), the aqui-fer recharge through the beds of wadis was estimated to30 % of the annual runoff. In fact, the Enfida plain presentsclimatic conditions similar to those in the North of Kairouanbasin which is an alluvial plain situated in a semi-aridclimate, with mean annual rainfall of 400 mm. In theEnfida plain, the mean annual runoff vary between 0.3×106 m3/year for small catchments to 9×106 m3/year for ElKhairat wadi, draining an area of 159 km2. Artificial re-charge by releases from El Khairat dam constructed in 2002
was also modeled. Released water volume during rechargeoperations varied between 1.6×106 and 2×106 m3. Theresults of a previous modeling work indicated an infiltrationyield ranging from 60 to 90 % of the released water(Zammouri et al. 2006).
According to the results of previous studies (El Batti1974; Besbes 1978; Kerrou 2008), recharge from precipita-tion may be estimated to about 5 to 15 % of yearly rainfall.It was set to 7 % of average annual rainfall along the coast,which corresponds to a rate of 8×10−10 m/s. Elsewhere, itvaries between 0.2×10−10 and 4×10−10 m/s according to thesoil permeability.
To assess the phreatic aquifer transmissivities, the modeledarea was divided into several zones with uniform value on thebasis of geological setting. The major part of the north and thecenter of the plain was characterized by transmissivities vary-ing between 6×10−3 and 8×10−3m2/s. The southern part wascharacterized by lower transmissivities ranging between 7.5×10−4 and 1.5×10−3 m2/s. With regard to the deep aquifertransmissivities, initial distribution was initiated using krigingtechniques. The empirical relationship between log T and log
0
2
4
6
8
10
15
20
Kalaa Kebira
Hergla
Enfida
Sidi Bouali
TDS (g/l)
0 10 20 Km
Scale
Fig. 5 Total dissolved solidscontent in phreatic aquifer in1972 (2 g/l intervals from 0–10,5 g/l intervals from 10–25)
Arab J Geosci
Author's personal copy
Qsp, was fitted with an acceptable correlation coefficient (0.8).The empirical relationship is written as:
T ¼ 0:936� Qsp0:97 ð3Þ
where T and Qsp are expressed in square meters per second.This relationship remains rather significant since the values
of T and Qsp extend over several orders of magnitude. Thetransmissivity values range between 1.9×10−4 and 0.05 m2/sand the specific capacity values vary between 2.6×10−4 and5×10−3m2/s. They reveal a great heterogeneity within theaquifer. The experimental variogram of log10 Twas computedon the basis of the 54 available data. Indeed, it was previouslypublished that the variogram of the transmissivity logarithm ismuch more structured than those of the transmissivity(Delhomme 1976; Ahmed et al. 1988). A spherical variogramwith nugget was obtained (Fig. 7). The plausibility of fit waschecked by cross-validation (Table 1).Thus, the kriged mapwas used as a starting point for the calibration process. Few
modifications were brought to this map during calibration.The initial horizontal transmissivities for the phreatic aquiferwere modified several times during the calibration process inorder to minimize calibration errors in most of the observationwells. The hydraulic resistances, between phreatic aquifer andsoil surface along the wadis, were adjusted to reproduce aclose value to the drainage flow from the wadis. The initialvertical hydraulic conductivities were slightly modified. Inareas with thin semi-previous layers, vertical hydraulic con-ductivities were increased to 6×10−6 m/s. Lower verticalhydraulic conductivities of less than 10−10 m/s were set tothe area with thick semi-pervious layers.
The comparison of the available observed and calculatedpiezometric values in 1972 indicates a satisfactory modelcalibration with a correlation coefficient (R2) of 0.85 and0.63, respectively for the phreatic and deep aquifers and aroot mean squared error (σ) of 3.5 and 4.9 m, respectively.The distribution of the calculated evaporation rate indicatesthat the modeled evaporation area overlays the actual
0
2
4
6
8
10
12
14
16
0 10 20 Km
Scale
Enfida
Hergla
Sidi Bouali
Kalaa Kebira
TDS (g/)
Fig. 6 Total dissolved solidscontent in deep aquifer (2 g/l intervals from 0–18) in 1972
Arab J Geosci
Author's personal copy
location of the sebkhas. Thus, the calibration provided betterresults for the phreatic aquifer. Improving its quality for thedeep aquifer is not essential because of the low precision onthe piezometric data. In addition, it is illusory of undertakingto reduce the error calibration lower than the standard devi-ation of estimation error derived from the application of thekriging techniques, in the areas where piezometric observa-tions are not available. As depicted in Fig. 8, the calculateddrawdown over the period 1973–2005 compares well to theobserved ones for most piezometers. The increase of pump-ing over the period 1973–2005 results in the reduction ofnatural outflows. As for the year 2005, the outflow into theMediterranean Sea decreased more than threefold from 156to 44 l/s (see Table 2). Losses by evaporation are reduced by
half from 76 to 28 l/s. Drainage by wadis had practicallyvanished by 2003. The reservoir depletion averaged over theperiod 1972–2005 represents 34 % of the mean recharge. Itconfirms the ongoing aquifer mining in some zones such asthe El Khairat wadi valley. The model results show negativepiezometric levels ranging from 1.5 to 1.8 m below sea levelin the deep aquifer near Enfida and Sidi Bou Ali regions,due to the numerous pumping fields.
With regard to transport modeling, a uniform porosityvalue of 0.2 was firstly chosen. After calibration, this param-eter was finally set varying from 0.01 to 0.25. The initialsalinity distribution of the phreatic aquifer was that of 1972(see Fig. 5). For the deep aquifer, it was initiated from thekriged map assumed to be representative of the 1970s (seeFig. 6). In the Mediterranean Sea, a uniform salinity of 38 g/l was adopted corresponding to averaged observation values.This boundary condition was assigned as fixed concentration.The concentration in the sabkhas was uniformly imposed to20 g/l, according to actually measured concentration values(El Batti 1974). Calculated longitudinal dispersivity variesbetween 100 m on the coast to 1,200 m in the center.
Based on these assumptions, a more consistent initialconcentration was calculated by flushing the aquifer in theundisturbed state over 100,000 years to arrive to an approx-imation of the state in 1972. Figure 9 shows the resultsobtained in the selected control points over the period1973–2005. Generally, the transport model reproduces thenatural salinization processes in the Enfida plain since thetime-scales of the observed and modeled phenomena matchwell for the whole basin. Despite local imperfections, thetransport model can be used to assess the threat of a plannedmarina on the groundwater quality in the plain.
Assessment of the impact of the planned marina
The transport model was used to assess the impact of the long-term application of the plannedmarina project on groundwaterquality. For this purpose, groundwater simulation model was
0 5000 10000 15000 20000 25000 30000
Distance (m)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Var
iogr
am
Fig. 7 Experimental variogram for transmissivity logarithms (deepaquifer). The available data comprises measured values and thoseestimated using specific capacity; transmissivity is in square metersper second
-2
2
6
10
14
18
1970 1975 1980 1985 1990 1995 2000 2005
Dra
wdo
wn
(m)
Phreatic aquifer : Valley of the Khairat wadi
0
2
4
6
8
1970 1975 1980 1985 1990 1995 2000 2005
Dra
wdo
wn
(m)
Deep aquifer : Sidi Bou Ali region
Fig. 8 Temporal reproduction of drawdown (1973–2005) in two selected observation wells. Straight line corresponds to calculated values andmeasured values are marked by connected filled symbols
Arab J Geosci
Author's personal copy
extended to the year 2055 with various management alterna-tives. In the first scenario (S1), present groundwater with-drawals were maintained over the whole basin, without anyadditional project. In the second scenario (S2), presentgroundwater extraction was maintained with implementingthe planned marina from 2006 onwards. The third scenario
(S3) considers an increase in groundwater pumping over thewhole plain according to the Water Authority projections ex-cluding the construction of the plannedmarina. Planned ground-water withdrawals for the phreatic aquifer will reach 1.25 timesthe pumped volume of 2005. The abstraction from deep wellswill increase twofold by 2055. In the fourth scenario (S4), thegroundwater extraction increase according to S3, with the im-plementation of the planned marina. The boundary conditionsrepresenting the marina location (see Fig. 1) were assigned anull head in the groundwater model as a fixed head and a TDSof 38 g/l in transport model as an imposed concentration.
In the S1, the maintenance of the withdrawals as those of2005, over the next 50 years, will result in an additionaldrawdown reaching 1.5 m. The loss of water by evaporationwill increase by 45%. The outflow to the seawill be reduced by43 %. The general groundwater flow direction will remain stilltowards the Mediterranean Sea. The transport model results donot show any significant difference in the calculated concen-trations compared to those of 2005. In fact, the piezometriclevels above the sea level near the coast prevent from marineintrusion. According to S2 which considers the marina con-struction starting 2006, an additional drawdown reaching 0.5 mis obtained compared to piezometric results of S1. Flow stilloccurs towards the sea. The comparison of the transport modelresults for the two scenarios revealed a TDS increase reaching10 g/l. The planned marina impact is located within a sector of2 km in radius in the north and northwest direction, 1.5 km inthe east direction, and 5 km in the southwest direction.
The S3 is carried out to study the impact due to the increasein groundwater withdrawals. The exploitation of the deepaquifer will reach 350 l/s by 2055 compared to 170 l/s in theyear 2005. An increase of 54 l/s of abstraction is projected inthe phreatic aquifer. The exploitation of this aquifer will reach270 l/s by 2055. The calculated piezometric levels for bothaquifers in 2055 show negative values in the valley of ElKhairat wadi and the Chegarnia-Sidi Abich basin. A gradientinversion will occur and an important reversed leakage fromthe TDS-enriched Mediterranean Sea to the aquifer systemwill be found for the year 2055. In fact, the calculated waterbalance indicates that in 1972, groundwater flows towards thesea but with the augmentation of withdrawals, the exchangewith the sea will occur in the two directions. The inflow fromthe sea is low in 2005 (6.6 l/s). It will increase slightly in thecase of the scenarios 1 and 2 to reach 16 l/s in 2055. For S3, itwill become significant (85 l/s in 2055), due to the abstractioneffect. For S4, the situation is the worst with the plannedmarina. An increase in saline water amounts to 34 l/s, com-pared to the S3, will involve an even more pronounced TDSdevelopment. Mediterranean salty water migration will occurmainly in the phreatic aquifer with an inflow of 105 l/s. Thequantity flowing to the deep aquifer is solely 14 l/s.
Compared to S1, the transport model results shows a TDSincrease from 2 to 17 g/l in S3. The high values concern
Table 2 The water balances calculated for 1972 and 2005 according tothe regional flow model (liters per second)
1972 2005
Enfida aquifer system
Inflow (l/s)
Reservoir depletion – 80
Recharge 421 377
Lateral inflow from southwestern model boundary 16 18
Total 437 475
Outflow (l/s)
Losses by evaporation 76 38
Pumping 194 393
Drainage by wadis 11 0
Leakage to Mediterranean sea 156 44
Total 437 475
Phreatic aquifer
Inflow (l/s)
Reservoir depletion – 63
Exchange with deep aquifer 319 87
Recharge 85 116
Lateral inflow from southwestern model boundary 16 18
Total 420 284
Outflow (l/s)
Exchange with deep aquifer 117 0
Losses by evaporation 76 38
Pumping 110 218
Drainage by wadis 11 0
Leakage to Mediterranean sea 106 28
Total 420 284
Deep aquifer
Inflow (l/s)
Reservoir depletion – 17
Exchange with phreatic aquifer 117 0
Recharge 336 260
Total 453 277
Outflow (l/s)
Exchange with phreatic aquifer 319 87
Pumping 84 174
Leakage to Mediterranean sea 50 16
Total 453 277
Arab J Geosci
Author's personal copy
primarily the northern part of the Enfida plain. In the areabordering the location of the planned marina, the TDS willincreases from 2 to 15 g/l. This increase is only due to theover-abstraction before the construction of the marina. In S4,the TDS increase will be pronounced in the vicinity of themarina. It will reach respectively 23 and 26 g/l in phreatic anddeep aquifers, compared to S2. The impact of the projectedmarina will cover a sector of 6 km in radius to the west, 4 kmto the northwest, and 3 km to the southwest.
In order to better assess the marina impact on the ground-water salinity, a fictional network of 16 wells (see Fig. 1) wasused to compare the water quality in the aquifer given by thedifferent scenarios. The nearest one is well no. 1 located
500 m west of the marina and the farthest one is well no. 14situated at 16 km to the northwest. Comparing S1 and S2,Fig. 10 shows the higher TDS increase for S2 of 4.6 g/l. Thedifference between scenarios vanishes as we go farther fromthe marina. Thus, well no. 4, located at 4.8 km southwest ofthe marina, shows an increase of only 0.2 g/l while there is noimpact on well no. 2 located at 2 km. The results of S3 and S4,compared to those of S1 show a different behavior. In thezones bordering the planned marina, the exploitation effect isadded to that of the marina. However, the marina effect ismore pronounce compared to that of the exploitation. Thisbehavior is revealed in some control points such as wells 1, 2,and 8 (Fig. 10). The calculated TDS increase in S4 compared
5000
10000
15000
20000
25000
30000
2005 2015 2025 2035 2045 2055
TD
S (
mg/
l)
W1_S 1 W1_S 2 W1_S 3 W1_S 4
W2_S 1 W2_S 2 W2_S 3 W2_S 4
2500
4000
5500
7000
8500
10000
11500
13000
14500
16000
2005 2015 2025 2035 2045 2055
TD
S (
mg/
l)
W8_S 1 W8_S 2 W8_S 3 W8_S 4
2000
4000
6000
8000
10000
12000
14000
16000
2005 2015 2025 2035 2045 2055
TD
S (
mg/
l)
W16_S1 W16_S 2 W16_S 3 W16_S 4
W7_S 1 W7_S 2 W7_S 3 W7_S 4
W4_S 1 W4_S 2 W4_S 3 W4_S 4
2500
4000
5500
7000
8500
10000
2005 2015 2025 2035 2045 2055
TD
S (
mg/
l)
W12_S 1 W12_S 2 W12_S 3 W12_S 4
Fig. 10 TDS development according to the different scenarios. Wm_Sn corresponds to well m and scenario n; calculated total dissolved solids(TDS) content of the scenarios S1 to S4 are compared for selected wells (1, 2, 4, 7, 8, 12, and 16)
0
1
2
3
4
5
6
1973 1978 1983 1989 1994 2000 2005
TD
S (
g/l)
TD
S (
g/l)
Phreatic aquifer: Valley of the Khairat wadi
0
1
2
3
4
5
6
1973 1978 1983 1989 1994 2000 2005
Deep aquifer : Chegarnia region
Fig. 9 Temporal reproduction of total dissolved solids (TDS) content (1973–2005). Straight line corresponds to calculated values and the filledsymbols are observations of TDS
Arab J Geosci
Author's personal copy
to S3 ranges from 2 to 10.8 g/l. While moving away from thefuture site of the planned marina towards the zones exploitedby shallow and deep wells, the TDS rise becomes primarilydue to the over-abstraction. In fact, S3 and S4 do not show anysignificant difference in the calculated concentration values.This result is illustrated in wells 4 and 12 (Fig. 10). Wells 7and 16 show similar results in all simulations (Fig. 10). In fact,these wells are located in the area where the impact of theplanned marina is weak and without additional withdrawals.Note that the pollution propagation varies in the differentdirections. Wells located northwest of the marina show aneven more pronounced TDS development than those situatedat the east and in the southwest of the marina. Comparing S1and S2, wells 3 and 8, located respectively at 1.5 km northwestand southwest of the marina, show a respective TDS rise of0.4 and 1.6 g/l. Preferential pathways of the influx from themarina are mapped using the PMPATH code. The results of S4indicate a severe situation since they indicate that a part ofsaline water will reach El Khairat pumping field.
Effect of transmissivity uncertainties on model’s predictions
In order to assess the effect of transmissivity uncertainties ofdeep aquifer on model’s predictions, the stochastic approachwas used to generate hundred realizations of transmissivityfields. The statistics of the 100 fields of the transmissivity
are shown in Fig. 11. Individually, the statistics for eachsimulated field of transmissivity are comparable with thoseof the observations since stochastic simulations are condi-tioned with these same observations. However, the meanfield corresponding to the average of the 100 fields showsless variability (Table 3). Afterwards, a hundred of equi-probable distributions of concentration are generated (Odo2008). The 100 simulations results are analyzed in order todetermine the possible variation range for the concentra-tions. For this purpose, the analysis of concentration, piez-ometry, and water balance results is carried out to eliminatethe realizations that strongly deviate from reality. The com-parison of the concentration calculated over the period1973–2005 at the various observation wells to the observedvalues is lengthy and hard. Accordingly, the comparison isfocused on the phreatic aquifer. The majority of the 100fields of transmissivity shows higher values. More than athird of realizations was eliminated. The comparison of thecalculated piezometry in steady state to the observed one, atthe shallow and deep observation wells, is based on thegraphic visualization in a scatter diagram and on the corre-lation coefficient and the standard deviation. The most ofthe scatter diagrams shows a more or less acceptable disper-sion around the diagonal. However, some realizations showa shift of calculated values below the diagonal indicating askewed piezometry. The shift is most likely caused by hightransmissivities at the downstream and thus corresponds tounrealistic distribution. For the phreatic aquifer, the correla-tion coefficient varies from 0.65 to 0.85 while the standarddeviation ranges between 3.5 and 5 m (Fig. 12). Theyindicate acceptable results but with less matching than thoseobtained by the calibrated transmissivity field. For the deepaquifer, the correlation coefficient and the standard devia-tion vary between 0.3 and 0.75 and between 4.15 and 6.5 m,respectively (Fig. 12). Finally, the analysis of piezometryled to the elimination of about 30 realizations.
The analysis of the various terms of the calculated waterbalance can also help to examine the realizations plausibil-ity. A great number of simulations shows a higher leakage toMediterranean Sea than the calibrated value, and in contrast,lower losses by evaporation (Table 4). Measurement forthese groundwater outflows is not available. The order of
-5
-4
-3
-2
-1
0
0 20 40 60 80 100
logT
(m
2 /s)
Mean MinimumMaximum Mean+standard deviationMean-Standard deviation
Fig. 11 Statistics of the 100 fields of transmissivity
Table 3 Basic statistics of thefield transmissivity Observations Average of 100 fields Mean field
Mean (T) 1.44×10−3 3.78×10−3 3.78×10−3
Minimum (T) 9.88×10−5 1.30×10−4 1.35×10−3
Maximum (T) 1.40×10−2 1.25×10−1 1.11×10−2
Mean (logT) −2.84 −2.42 −2.42
Minimum (logT) −4.01 −3.88 −2.87
Maximum (logT) −1.85 −0.90 −1.95
Standard deviation (logT) 0.54 0.46 0.16
Arab J Geosci
Author's personal copy
magnitude of the computed values may be then consideredacceptable. However, the drainage by wadis is measurableand can be assessed with an acceptable precision. By refer-ring to the measurements, the simulations leading to drain-age by wadis exceeding two orders of magnitude ofobserved values are considered unrealistic and cannot betaken into account. Finally, about 20 realizations were up-held. They allow surrounding the possible fluctuations inthe saline water migration. For these upheld realizations, thegroundwater and transport simulation models were extendedto the year 2055 according to the S4. Calculated concen-trations over the period 2006–2055 are compared to theresults obtained with the calibrated transmissivity field.The comparison was based on the relative deviation.
Results of three informative simulations in eight selectedobservation wells are reported in Fig. 13. In general, therelative deviation increases in time. However, for few realiza-tions, it is high in some observation wells at the beginning ofsimulation but it decreases to lower value in 2055. It is thecase of the realization 8 in the observation well 9 as shown inFig. 13. Accordingly, the analysis is based on the relativedeviation averaged over the period 2006–2055. The highestaverage relative deviation is located to the northwestern re-gion of the marina (observation wells 8 and 9) where it reaches57% in the observation well 8, in the realization 1. In the close
vicinity of the marina, all the realizations show a low relativedeviation. For example, it is 3 % in the observation well 1. Theaverage relative deviation increases when moving towards thesouthwestern of the marina, but it remains low in the majorityof the realizations. In the observation wells 2, 3, and 4, itranges from 3 to 13 %. Occasionally, it is even negative as inthe case for realization 8 (wells 2 and 3) and realization 1 (well4) where the calculated concentration is lower by 6 % than thepredicted values given by the calibrated model. These results,indicating a low uncertainty varying between 3 and 13 %,confirms the predictive quality of the calibrated model in theregions close to the marina and to the southwestern areabordering the marina. In the valley of the El Khairat wadi,where important groundwater withdrawals fields are located,the average relative deviation is variable. In observation well10, it ranges from 3 to 13 % in the various realizations. Itindicates that the forecasted results by the calibrated modelcan be considered credible. In contrast, in the observation well11, the average relative deviation that varies between 26 and49% is high. According to the most pessimistic realizations (1and 93), the calculated concentration could reach 6.5 g/l in2055 in the observation well 11, instead of 3.2 g/l. This resultis alarming since a bad knowledge of the transmissivity maycause a major quality deterioration of the shallow and deepgroundwater over the course of time.
3
4
5
6
0 50 100
Sta
ndar
d de
viat
ion
(m)
Fields
Phreatic aquifer
0 50 100
Sta
ndar
d de
viat
ion
(m)
Fields
Deep aquifer
0.6
0.7
0.8
0.9
0 50 100
)m(tneiciffe
ocn
oitalerro
C
Fields
Phreatic aquifer
4
5
6
7
0.2
0.4
0.6
0.8
0 50 100
Cor
rela
tion
coef
ficie
nt (
m)
Fields
Deep aquifer
Fig. 12 Comparison ofpiezometric results of the 100fields of transmissivity toobservations; the dashed linecorresponds to steady statemodel calibration
Table 4 Natural groundwateroutflow results of the 100 fieldsof the transmissivity
Discharge (l/s) Minimum Maximum Mean Calibrated value
Drainage by wadis 0 63 18 11
Leakage to Mediterranean sea 52 225 176 156
Losses by evaporation 32 97 57 76
Arab J Geosci
Author's personal copy
Conclusions
The exploitation in the Enfida plain had increased more thantwofold between 1972 and 2005. Simulation results over thisperiod show a decrease of natural outflows and decline ingeneral piezometric levels. In the coast, the piezometryremains above the seawater level preventing from marineintrusion. In Enfida and Sidi Bou Ali regions where pumpingfield is concentrated, the model shows negative piezometriclevels in the deep aquifer, indicating the ongoing aquifermining. The scenario, investigating the conservation of thepresent pumping scheme over the next 50 years, shows areduction of natural outflow, a low additional drawdown andno significant rise in the calculated concentrations. The im-plementation of the planned marina will generate a TDSincrease reaching 10 g/l in the bordering area within a sectornot exceeding 5 km in ray. The TDS rise will be located inmarshy regions naturally characterized by low permeabilityand salty groundwater and will not generate any deteriorationof groundwater quality in the pumping fields. The plannedextraction from both phreatic and deep aquifers based onincreasing the present withdrawals by 1.25 times from shallowwells and by two times the current abstraction from deep wellsare not sustainable. ATDS increase in 2055 varying between 2to 17 g/l will occur in the coastal area. In the case of the
implementation of the planned marina, the TDS increase willbe pronounced in the vicinity of the marina.
Stochastic modeling of the uncertainty present in the trans-missivity field allowed the quantification of these uncertaintieson models predictions. In the regions bordering the marina,low uncertainty varying between 3 and 13 % does not call inquestion the predictive quality of the deterministic model. Inthe valley of the El Khairat wadi that is characterized by freshresource and consequently by important withdrawals fields,stochastic simulations indicate high uncertainty reaching lo-cally 49 %. If this proves the weak representativeness of thedeterministic model because of the inaccuracy of the calibratedtransmissivity in this area, the situation is likely to worsen inthe future with further degradation of groundwater quality.According to the most pessimistic realizations, the calculatedconcentration might reach locally 6.5 g/l, in 2055, instead of3.2 g/l as it is predicted by the deterministic model.
Control measures to avoid Enfida groundwater saliniza-tion would include the following. First, planned extractionprojects have to be abandoned. The model shows a futureTDS increase. Since overall pumping in the basin particu-larly in El Khairat valley is responsible for this occurring,the construction of the marina is only of limited effect.Second, stochastic modeling reveals, in some area, highuncertainty in the transmissivity field. Model results could
4
8
12
16
20
24
28
2006 2013 2020 2027 2034 2041 2048 2055
TD
S (
g/l)
R1_W1 R8_W1 R93_W1 S4_W1
R1_W2 R8_W2 R93_W2 S4_W2
3
10
17
24
2006 2013 2020 2027 2034 2041 2048 2055
TD
S (
g/l)
R1_W3 R8_W3 R93_W3 S4_W3
R1_W8 R8_W8 R93_W8 S4_W8
0
5
10
15
20
25
30
2006 2013 2020 2027 2034 2041 2048 2055
TD
S (
g/l)
R1_W9 R8_W9 R93_W9 S4_W9
R1_W10 R8_W10 R93_W10 S4_W10
0
4
8
12
16
20
2006 2013 2020 2027 2034 2041 2048 2055
TD
S (
g/l)
R1_W4 R8_W4 R93_W4 S4_W4
R1_W11 R8_W11 R93_W11 S4_W11
Fig. 13 TDS development according to stochastic simulations. Rm_Wn corresponds to realization m and well n; calculated total dissolved solids(TDS) content of selected realizations (1, 8, and 93) are compared to those of the scenario S4 in chosen wells (1, 2, 4, 8, 9, 10, and 11)
Arab J Geosci
Author's personal copy
certainly be improved by better knowledge of the distribu-tion of this parameter and also of the other model parame-ters. The model predictive quality may be affected by thedensity effects. However, the relevance of this problem wasnot treated in the case of Enfida plain. Third and finally, theimplementation of quality control points in the regionsbordering the planned marina would give important infor-mation about the groundwater salinity development.
Acknowledgments Financial support for this research was providedby “Agence de Protection et d’Aménagement du Territoire” (APAL).The authors wish to acknowledge Dr. H. Ben Moussa (APAL) and Dr.F. Maalal (CRDA Monastir) for their valuable help in the datagathering.
References
Abarca E (2006) Seawater intrusion in complex geological environ-ments. PhD Thesis, Technical University of Catalonia, Spain.154p
Aboufirassi M, Marino MA (1984) Cokriging of aquifer transmissivityfrom field measurements of transmissivity and specific capacity.Math Geol 16(1):19–35
Ackerer P, Magnico P, More R (1994) Réflexions sur la modélisationde la propagation de polluants dans les hydrosystèmes souter-rains. Revue des sciences de l'eau/J Water Sci 7(2):201–212
Ahmed S, de Marsily G (1987) Comparison of geostatistical methodsfor estimating transmissivity using data on transmissivity andspecific capacity. Water Resour Res 23(9):1717–1737
Ahmed S, de Marsily G, Talbot A (1988) Combined use of hydraulicand electrical properties of an aquifer in a geostatistical estimationof transmissivity. Ground Water 26(1):78–86
Al-Bitar A, Ababou R (2005) Random field approach to seawaterintrusion in heterogeneous coastal aquifers: unconditional simu-lations and statistical analysis. Geostat Environ Appl 233–248.doi:10.1007/3-540-26535-X_20
Anderson MP (1979) Using models to simulate the movement ofcontaminants through groundwater flow systems. Crit RevEnviron Controls 9(2):97–156
Anderson MP, Woessner WW (1992) Applied groundwater modeling.Academic, San Diego, 381p
Appleyard S (1995) The impact of urban development on recharge andgroundwater quality in a coastal aquifer near Perth, WesternAustralia. Hydrogeol J 3(2):65–75. doi:10.1007/s100400050072
Bear J, Verruijt A (1987) Modeling groundwater flow and pollution. D.Reidel Publishing Company, New York, 414p
Bear J, Cheng AH-D (2009) Modeling groundwater flow and contam-inant transport. Springer, Dordrecht, 834 p
Besbes M (1978) L'estimation des apports aux nappes souterraines. Unmodèle d'infiltration efficace. Thèse Doc. es. Sc. Univ. Paris VI. 270p
Brovelli A, Carranza-Diaz O, Rossi L, Barry DA (2010) Designmethodology accounting for the effects of porous medium het-erogeneity on hydraulic residence time and biodegradation inhorizontal subsurface flow constructed wetlands. Ecol Eng.doi:10.1016/j.ecoleng.2010.04.031
Carlson M, Lohse K, McIntosh J, McClain J (2011) Impacts of urbani-zation on groundwater quality and recharge in a semi-arid alluvialbasin. J Hydrol 409:196–211. doi:10.1016/j.jhydrol.2011.08.020
Carrera J (1993) An overview of uncertainties in modeling groundwa-ter solute transport. J Contam Hydrol 13:23–48
Changming L, Jingjie Y, Kendy E (2001) Groundwater exploitationand its impact on the environment in the North China plain. WaterInt 26(2):265–272
Chiang WH, Kinzelbach W (2001) 3D-groundwater modeling withPMWIN a simulation system for modeling groundwater flowand pollution. Springer, Berlin, 346p
Chiles JP, Delfiner P (1999) Geostatistics: modeling spatial uncertain-ty. Wiley, New York, 695p
Cinnirella S, Buttafuoco G, Pirrone N (2005) Stochastic analysis toassess the spatial distribution of groundwater nitrate concentra-tions in the Po catchment. Environ Pollut 133:569–580
Clark I, Harper WV (2000) Practical geostatistics 2000. Ecosse NorthAmerica, Columbus, 342p
Cornaton F (2004) Deterministic models of groundwater age, lifeexpectancy and transit time distributions in advective-dispersivesystems. PhD thesis, Fac. Sc. Univ. Neuchâtel. 147p
Delhomme JP (1976) Application de la théorie des variablesrégionalisées dans les sciences de l’eau. Docteur-Ingénieur thesis,Ecole des Mines de Paris, Univ. Paris VI. 130p
DGP « Direction Générale de la Privatisation » (2006) Mission d’étudeet d’assistance au projet de réalisation, de financement et d’ex-ploitation d’un port en eaux profondes sous forme de concession.Ref. EuropeAid/119858/D/SV/TN. Rapport de la mission 3.Rapport sur les études d’impact sur l’environnement. PremierMinistère. Tunisie. 17p
Dreybrodt W, Romanov D, Kaufmann G (2009) Evolution of isolatedcaves in porous limestone by mixing of phreatic water and surfacewater at the water table of unconfined aquifers: a model approach.J Hydrol 376:200–208. doi:10.1016/j.jhydrol.2009.07.027
El Batti (1974) Hydrogéologie de la plaine d’Enfida. Doc. 3ème cyclethesis. Univ. Scientifique et médicale de Grenoble. 122p
Ferraris L, Gabellani S, Rebora N, Provenzale A (2002) A comparisonof stochastic models for spatial rainfall downscaling. WaterResour Res 39:1368–1384
Freeze RA (1975) A stochastic-conceptual analysis of one-dimensionalgroundwater flow in non uniform homogeneous media. WaterResour Res 11(5):725–741. doi:10.1029/WR011i005p00725
Frenzel H (1995) A field generator based on Mejia’s algorithm. Institutfür fuer Umwelt-physik. University of Heidelberg, Germany
Fried JJ (1975) Groundwater pollution. Elsevier, Amsterdam, 330pFried JJ (1981) Groundwater pollution mathematical modeling: im-
provement or stagnation? Sci Total Environ 21:283–298Harbaugh AW, Banta ER, Hill MC, McDonald MG (2000) Modflow-
2000, the U.S. geological survey modular ground-water model:user guide to modularization concepts and the groundwater flowProcess. US Geological Survey, Reston, 121p
Horriche F, Zammouri M (2008) Etude sur modèles mathématiques dela plaine d’Enfida: construction, calage et exploitation desmodèles d’écoulement et de transport de soluté. Mission d’exper-tise : Etude hydrogéologique de la zone d’implantation du port eneaux profondes dans la région d’Enfida. A.P.A.L, Ministère Env.Dévelop. Durable, Tunise. 117p
Isaak E, Srivastava R (1989) An introduction to applied geostatistics.Oxford University Press, New York, 592p
Journel A, Huijbregts CJ (1978) Mining geostatistics. Academic, NewYork, 600p
Kerrou J (2008) Deterministic and probabilistic numerical modelingtowards sustainable groundwater management. Application toseawater intrusion in the Korba aquifer (Tunisia). PhD thesis.Univ. Neuchâtel. Switzerland. 181p
Kitanidis PK (2000) Introduction to geostatistics. Application inhydrogeology. Cambridge University Press, Cambridge, 249p
Konikow LF (2011) The secret to successful solute-transport modeling.Ground Water 49(2):144–159
Lerner DN (2002) Identifying and quantifying urban recharge: a review.Hydrogeol J 10(1):143–152. doi:10.1007/s10040-001-0177-1
Arab J Geosci
Author's personal copy
Manaa M (1989) Etude hydrogéologique préliminaire du Sahel deSousse. Direction des Ressources en Eau. Arrondissement deSousse. Ministère de l’Agriculture, Tunisie
Manaa M, Chaieb H, Amri R (1996) Etude par modèle hydrogéologiquede la nappe alluviale de l’oued El Khairat. DGRE. Ministère del’Agriculture, Tunisie
de Martonne E (1942) Nouvelle carte mondiale de l’indice d’aridité.Annale de géographie 288:241–250
Matheron G (1970) La théorie des variables régionalisées et ses appli-cations., Les Cahiers du centre de Géostatistique 5. EcoleNationale Supérieure des Mines de Paris. 212p
Matheron G, de Marsily G (1980) Is transport in porous media alwaysdiffusive? A counter-example. Water Resour Res 16(5):901–917.doi:10.1029/WR016i005p00901
McAllister TL, Overton MF, Brill ED (1996) Cumulative impact ofmarinas on estuarine water quality. Environ Manag 20(3):385–396. doi:10.1007/BF01203846
Mejia JM, Rodriguez-Iturbe I (1974) On the synthesis of random fieldsampling from the spectrum: an application to the generation ofhydrologic spatial processes. Water Resour Res 10(4):705–711.doi:10.1029/WR010i004p00705
Naji A, Cheng AHD, Ouazar D (1998) Analytical stochastic solutionsof saltwater/freshwater interface in coastal aquifers. Stoch HydrolHydraul 12:413–430
Nazoumou Y (2002) Impact des barrages sur l’alimentation des nappessouterraines de Kairouan. Doc. thesis ENIT, Univ. El Manar,Tunis. 221p
Odo BO (2008) Etude de la sensibilité des modèles de l’écoulement etdu transport de soluté dans la plaine d’Enfida. Mém. Fin d’études.ESIER, Medjez el Bab. Tunisie. 115p
Razack M, Huntley D (1991) Assessing transmissivity from specificcapacity in a large and heterogeneous alluvial aquifer. GroundWater 29(6):856–861
Razack M, Lasm T (2006) Geostatistical estimation of the trans-missivity in a highly fractured metamorphic and crystallineaquifer (Man-Danane Region, Western Ivory Coast). J Hydrol325:164–178
Renard P (2008) La fin des certitudes? Bulletin d’Hydrogéologie 22,23–42. Centre d’Hydrogéologie, Univ. Neuchâtel. Ed. Peter Lang
Rentier C (2003) Méthode stochastique de délimitation des zones deprotection autour des captages d’eau—Conditionnement par desmesures de conductivité hydraulique K, de hauteur piézométriqueh et de résistivité électrique ρ, Doc. Thesis, Univ. Liège. 215p
Robinson C, Brovelli A, Barry DA, Li L (2009) Tidal influence onBTEX biodegradation in sandy coastal aquifers. Adv WaterResour 32:16–28. doi:10.1016/j.advwatres.2008.09.008
Rodriguez-Iturbe I, Marani M, D’Odorico P, Rinaldo A (1998) Onspace–time scaling of cumulated rainfall fields. Water Resour Res34(12):3461–3469
Simmons CS (1982) A stochastic-convective transport representationin dispersion in one-dimensional porous media systems. WaterResour Res 18(4):1193–1214
SCET (2007) Etude de gestion des eaux de ruissellement dans la régionCentre-Est (Salloum-Hergla). Phase II : Avant projet sommaire,O.N.A.S., Ministère Env. Dévelop. Durable, Tunisie
Schiedek T, Beier M, Ebhardt G (2007) An integrative method toquantify contaminant fluxes in the groundwater of urban areas. JSoils Sediments 7(4):261–269. doi:10.1065/jss2007.04.221
Theodossiou N, Latinopoulos P (2006) Evaluation and optimisation ofgroundwater observation networks using the Kriging methodolo-gy. Environ Model Softw 21:991–1000
Zammouri M, Ounaies S, Jarraya-Horriche F (2006) Modélisation de larecharge artificielle par lâchers dans le lit d’un cours d’eau. Cas dela plaine d’Enfida (Tunisie). Las aguas subterraineas en los paisemediterraneos. Instituto Geologico y Minero de Espana. Serie :Hidrogeologica y Aguas Subterraneas, N°17, 407–412
Zheng C, Wang PP (1999) A modular three-dimensional multispeciestransport model for simulation of advection, dispersion and chemi-cal reactions of contaminants in groundwater systems; documenta-tion and user’s guide. In: SERDP-99-1. Contract report, US ArmyEngineer Research and Development Center, Vicksburg, MS. 169p
Zhou Y, Van Geer FC (1992) KALMOD, a stochastic-deterministicmodel for simulating groundwater flow with Kalman filtering.Hydrol Sci J 37(4):375–389
Arab J Geosci
Author's personal copy
