Slurry wall containment performance: monitoring and modeling of unsaturated and saturated flow

Environ Monit Assess (2012) 184:607–624DOI 10.1007/s10661-011-1990-1

Slurry wall containment performance: monitoringand modeling of unsaturated and saturated flow

Daniele Pedretti · Marco Masetti ·Tomaso Marangoni · Giovanni Pietro Beretta

Received: 8 April 2010 / Accepted: 23 February 2011 / Published online: 5 April 2011© Springer Science+Business Media B.V. 2011

Abstract A specific 2-year program to monitorand test both the vadose zone and the saturatedzone, coupled with a numerical analysis, was per-formed to evaluate the overall performance ofslurry wall systems for containment of contami-nated areas. Despite local physical confinement(slurry walls keyed into an average 2-m-thickaquitard), for at least two decades, high concen-trations of chlorinated solvents (up to 110 mg l−1)have been observed in aquifers that supply drink-ing water close to the city of Milan (Italy). Re-sults of monitoring and in situ tests have beenused to perform an unsaturated-saturated numer-ical model. These results yielded the necessaryquantitative information to be used both for thedetermination of the hydraulic properties of thedifferent media in the area and for the calibrationand validation of the numerical model. Backfillmaterial in the shallower part of the investigatedaquifer dramatically affects the natural recharge

D. Pedretti (B)GHS, Dept. Geotechnical Engineeringand Geosciences, Universitat Politecnicade Catalunya—BarcelonaTech,C/Jordi-Girona 1-3, 08034, Barcelona, Spaine-mail: [email protected]

M. Masetti · T. Marangoni · G. P. BerettaDipartimento di Scienze della Terra ‘A. Desio’,Università degli Studi di Milano, via Mangiagalli 34,20133, Milan, Italy

of the encapsulated area. A transient simulationfrom wet to drought periods highlights a changein the ratio between leakages from lateral barri-ers that support a specific scenario of water lossthrough the containment system. The combina-tion of monitoring and modelling allows a reliableestimate of the overall performance of the phys-ical confinement to be made without using anyinvasive techniques on slurry wall.

Keywords Slurry wall · Groundwatermonitoring · Numerical modeling · Leakage ·Milan (Italy)

Introduction

Vertical barriers have been used since the late1970s for containment and environmental pollu-tion control at hazardous sites (Paul et al. 1992;Ryan 1987, 1994; Manassero et al. 1995; Filz andMitchell 1995) and have been extensively studiedin terms of design, performance, permanence, andother relevant issues (Paul et al. 1992; Nash 1974;D’Appolonia 1980; Hajnal et al. 1984; Evans 1993;Xanthakos 1994; Beretta et al. 2003). In Italy, thetechnique has been widely applied for groundwa-ter remediation strategies in the major contami-nated sites (Rolle et al. 2009). Given the extremelyhigh cost to clean up the sites completely in ashort period of time, vertical barriers are useful to

608 Environ Monit Assess (2012) 184:607–624

confine the source for subsequent removal whileprotecting downgradient groundwater, or simplyto contain exposure to the source. Under certainflow conditions, and when combined with pump-and-treat remediation systems, physical barriershave a dramatic effect on the contaminant plumemanagement strategy (Bayer et al. 2004). In anycase, barriers must be designed to be effective fora long-term period, which is a function of manyvariables, such as the extent and amount of conta-mination, the hydrogeological condition, and thepresence of potential risk in the surroundings.The lifetimes of these systems, therefore, rangefrom about 50 to 100 years (Inyang 2004). Themost widely used technique for containment hasbeen the soil-bentonite slurry wall. It is typicallythe most economical, utilises low-permeabilitybackfill, and usually allows reuse of all or most ofthe material excavated during trenching.

A study (US EPA 1998) on a number of sitesin the USA has showed that maintaining theefficiency of containment is strongly related todesign, construction quality assurance, construc-tion quality control, and especially monitoring ofthe systems. Monitoring the water regime of thecontainments is particularly necessary to assesstheir proper operation (Brandelik and Huebner2003). In fact, one of the most crucial points isthe long-term performance, which depends on anumber of factors, such as keying between wallsections, keying into an aquitard (Ryan and Day2003), formation of flaws, lack of continuity ofthe aquitard, and the contaminants properties(Candelaria and Matsumoto 2000); all of these canalter the performance of the system. Some authorstried to extrapolate short-time observations tolarger scales (Inyang and Tomassoni 1992), but lo-cal conditions strongly affect single performancesloss; therefore, each case can only be controlled bya reliable monitoring scheme tailored to the localfeatures of the area. Many techniques can be used,both “non-invasive” (those that do not alter thewall structure) and “invasive” (those that can alterthe wall structure). The use of the latter is usuallyrestricted to avoid problems that enhance contam-ination spreading. Each technique has advantagesand disadvantages; for instance, geophysics cangive the general state condition of the wall but isnot always reliable for detecting small flaws and

discontinuities, whereas piezocone is very sensi-tive (Manassero 1994) but many tests could berequired in an extended area and problems ofverticality can arise especially in thin and/or deepwalls. A modified slug test approach has also beenused to determine the hydraulic conductivity of avertical cut-off wall (Choi and Daniel 2006), evenif it has the main limitations of the standard tests,related to the limited extension of the volume ofwall investigated during the test. Pumping testswithin the containment area in order to measuredrawdown outside the area are expensive and aregenerally extremely sensitive to the position ofthe largest defects, so such tests are effective onlyif piezometers are placed close to them (Benson2002).

The performance of pumping tests by creatinga test cell is very time consuming and expensiveand cannot be applied if the containment areas arenot wide enough to provide an adequate amountof groundwater in order to perform the test onthe correct time scale (Ryan 1987). Moreover, itwas showed that there is a clear dependence ofthe cut-off walls’ hydraulic conductivity on thesample volume (Britton et al. 2004), so any tech-nique should also considered the actual amount ofvolume that is investigated by the test. In any case,whatever the approach, the general aim should al-ways be to obtain a reliable assessment of the bulktransmissivity, that is, to quantify the effectivenessof the system considering its overall performance.

In this study, a specific 2-year program of mon-itoring and testing both the vadose zone and thesaturated zone, coupled with a numerical analy-sis, was performed to evaluate the overall per-formance of slurry wall systems for contaminatedarea containment. The activities were focused onfinding out whether deficiencies in the integrity ofthe vertical barrier systems or the local geologicalconditions could be more likely to determine therelease of contaminants into the aquifers. Due torestrictions of the local Environmental Agency,all the activities have to be done using only non-invasive techniques on the slurry wall systems.The evaluation of containment effectiveness andthe dynamics of contaminant release from it arestrategic to perform risk assessment of ground-water contamination, that should always be usedas an important support for planning remediation

Environ Monit Assess (2012) 184:607–624 609

system. Given that risk assessment in these con-texts should always be evaluated within a proba-bilistic framework (Tartakovsky and Winter 2008;Winter and Tartakovsky 2008; de Barros et al.2009; Bolster et al. 2009), a proper understand-ing of the impact of source behaviour is essen-tial to reduce uncertainty. So, the aim was to:(a) assess failures in the containment systems,including deficient cut-off walls or inappropriatedesign of the geological scheme; (b) estimate thewater budget in an equilibrium state and thedischarge/recharge behaviour, (c) define reliablehydraulic parameters to support successive reme-diation action; and (d) obtain a quantitative evalu-ation of groundwater losses through the confinedarea, in order to support future analysis of thefate of chlorinated compounds released into theaquifers. All these aspects are necessary to builda reliable conceptual model and to refine ana-lytical estimate of the probability of failure ofa remediation effort for future probabilistic riskassessment (Bolster et al. 2009).

The study area

The industrial growth in northern Italy that oc-curred over the last 50–70 years, together with thelack of specific laws on waste management, hascontributed to the gradual depletion of vulnerablegroundwater resources. Chemical facilities had agreat role in this as well. The former “Bianchi”chemical facility at the Rho district, located fewkm northwest to the city of Milan (Fig. 1), is anexample of how hazardous wastes produced byindustrial activities have been stored for decadeswithout proper knowledge of environmental mat-ters and have caused a potential risk situation thatcurrently is a matter of special concern. In thelate 1970s, the first evidence of intense contami-nation with chlorinated compounds arose and thesupposed main source was immediately confinedin a specific area (Fig. 2), later indicated as zone1, through the construction of vertical slurry cut-off walls keyed into a low hydraulic conductiv-ity layer (aquitard). This containment system wasnever coupled with other remediation techniques,causing a progressive spread of the contaminationduring the following years into the present. A

second confinement solution, that of a hydraulicbarrier was set only much later, in 2006, to stop themigration of a well-defined dense non-aqueousphase liquids (DNAPL) plume in the downgra-dient areas (zone 2). Again, in 2007, however,concentrations of chlorinated solvents were mea-sured up to 110 mg l−1 near the source at thearea of the former “Bianchi” chemical facility, inthe southern side of the municipality. Pollutionhas been detected within a depth of 45 m. Thecity of Milan is located immediately downstreamand, consequently, public and private pumpingwater wells for human direct consumption andfarming are directly exposed to the chlorinatedplume threat (Fig. 1).

Hydrogeological setting

The geological structure of the area may be clas-sified by adopting the traditional regional subdivi-sions for the northern Po Plain (e.g. ENI-AGIP2002). The vulnerable aquifer is the uppermost,multilayered aquifer (A), which is commonly de-tected within a depth of 60 m in the whole westernpart of the Milan province. However, several finelayers with low hydraulic conductivity divide thislayer into two semi-confined aquifers (A1 andA2), even if, in some zones, the subunits mergebecause of a thickness reduction or local lackof such low conductivity horizons. On the topof the A1 aquifer, an unconfined (or sometimesperched) aquifer (A0) occasionally appears. ThisA0 aquifer has a variable depth (7–9 m from theground surface), and overlays a semi-permeable,discontinuous aquitard (Fig. 2). At the formerchemical domain scale, the discontinuous geom-etry of the aquitard, and the consequent depen-dence on DNAPL migration to small-scale anom-alies, are already shown in Pedretti et al. (2009)(Fig. 3).

The recharge of aquifers is mainly due to pre-cipitation and irrigation water (with a period go-ing from April to mid-October), while the contri-bution of the small river crossing the area is negli-gible (Alberti et al. 2007). Within the containmentarea, recharge is given by precipitation only, evenif the nature of backfilling dramatically affects therecharge rate and dynamics, as described later.


Fig. 1 Location of the area and main hydrogeological features (modified from www.provincia.milano.it). The metropolitanarea of Milan is located a few km downstream the polluted site, along the main direction of the regional piezometric trend

The characteristics of the containment sys-tem are only partially known. The geometry iscommonly referred to as a “sarcophagus-shape”polygon (Fig. 4), keyed into the aquitard (Fig. 2).The composition is composed of typical cut-offslurry backfilled walls.

Focusing on the containment system, some as-pects become strategic in the conceptual modeldescribing the general water balance. The aquiferconnections and the recharge/discharge relation-ship through the vadose zone affect DNAPLs fate

in the groundwater, involving various aspects ofthe migration of contaminants (acting physicaland chemical properties of the compounds). Largeuncertainties are thus associated with the verticalbarrier system, especially in term of its hydraulicconductivity, that have never been tested in thepast and that are not possible to test currently dueto the risk of enhancing contaminant migration inthe subsoil. Only a few historical monitoring datawere available; moreover, these data were ex-tremely discontinuous and basically were not very

http://www.provincia.milano.it


Fig. 2 Alternative conceptual models describing the hy-drogeological division of A aquifer in the Rho area. TheDNAPLs leak into aquifer A1 depending on the aquitardefficiency below the encapsulated area. Case (1): Theaquitard is more efficient to contaminant leakage than

lateral barriers. Spreading into A1 mainly occurs throughdiscontinuities along the aquitard. Case (2): Aquitard andlateral barrier efficiencies are equivalent; spreading intoA1 occurs also at the bottom of the encapsulated area

useful. The only reliable and available historicaldata were the chlorinated compounds concentra-tions in wells and in piezometers downgradientfrom the containment area. At the Rho site, thehigh and continuous contaminant concentrationsin groundwater indicated that the containmentsystem seems to have lost its effectiveness since itwas constructed in 1981. Two possible conceptualmodels are most likely to explain this, as shownin Fig. 2. In the first case (Fig. 2a up), consis-tent flow occurs outward from the encapsulatedarea due to lateral spreading and the fact thatthe aquitard is almost impermeable below sourcearea; the A1 aquifer is polluted by deepening ofDNAPLs through discontinuities in the aquitard,

and from long-term diffusion processes. In thesecond case (Fig. 4b below), flow comes fromlateral spreading directly into the A0 aquifer, andinto the A1 aquifer through the bottom aquitard.The integrity and efficiency of the lateral barrieris contrasted by poor hydrodynamic efficiency ofthe geological barriers.

Site characterization: in situ monitoringand laboratory tests

As a method to cope with problems related to thecontainment area, a 2-year monitoring and testingprogram was developed, due to the need to obtain


Fig. 3 3D mapping of theA0/A1 aquitard atRho-Bianchi formerchemical facility, afterPedretti et al. (2009). Theshadow area is wheregradients are maximumtowards the southernportion of the domain,thus acting as a potentialpreferential migrationzone of DNAPLs

the necessary information in order to increase theknowledge of the contaminant migration and tosupport the ultimate remediation design. It shouldbe considered that a thorough hydrogeologicalcharacterization can give great benefits in uncer-tainty reduction in human health risk in cases ofsmall contaminant source, as the one analysed inthis study (de Barros et al. 2009).

Because the major uncertainty came from theunknown geometry and permeability of the bar-rier, there was a need to deeply investigate thegeological features of both the vadose zone andthe saturated domain in the source area. To avoid

the risk of further spreading the contaminationby direct surveys in the slurry wall, a monitoringsystem of both the vadose zone and the aquiferswas designed in order to support the definition ofa reliable conceptual model (Hudak and Loaiciga1999). The system was also used for groundwatermodelling in the unsaturated and saturated condi-tions (Chen et al. 1999). A great number of sur-veys and measurements have been performed tocharacterise the study area, using both in situ andlaboratory tests, in order to collect geologic andhydraulic information of the area and to developits geologic conceptual model. The location of the


Fig. 4 Synthetic sketch of the local set-up, depicting piezo-metric isolines (m) of the A0 aquifer (taken from regionalpiezometry); the SEEP model A’–A trace; the relative

position of observation wells, small S-boreholes and EVSdevices. The encapsulated zone is Zone 1; the external areais Zone 2

surveys is shown in Fig. 4. Standard monitoring ofgroundwater level and contaminant concentrationof the A0 and A1 aquifers has been done bothwithin the encapsulated area (herein, Zone 1)and outside it (herein, Zone2). A specific detailedprogram of activities was planned in zone 1 andincluded the following.

– Drilling six small boreholes (S-points in Fig. 4)with the direct push technique, to the maxi-mum depth of 3.6 m, for a detailed reconstruc-tion of the stratigraphy of the vadose zone,including the partial recovery of drilled soils.

– Performing double-ring infiltrometery tests todetermine the vertical saturated conductivityof superficial soils, and pumping and slug testson aquifers and aquitard A0–A1.

– Installing two continuous electrical data log-gers for the automatic monitoring of soil mois-ture content to a maximum depth of 150 cm.based on the frequency domain reflectometry(FDR) technique. Each device consists of fourprobes, located at different depths.

– Installing three tensiometers at differentdepth for the measure of pore-water pressurein the vadose zone (manual reading) to a max-imum depth of 120 cm.

– A recording station for temperature andrainfall.

– Standard laboratory tests on sampled soil havebeen done, including grain-size analysis, or-ganic content, specific weight, and permeabil-ity tests.

Drilling

The analysis of the stratigraphy of the six S-points(Fig. 4) showed that the deposits in zone 1 arevery heterogeneous (Fig. 5—SG stands for SoilGas). This is probably due to the different na-ture of backfill used and to the presence of theremains of ancient anthropogenic structures in thesubsoil. Human impact is remarkable and affectsthe hydrodynamic properties of the soils. In thepast, the pilot area was dug out and then backfilledby waste materials from former structures of thechemical facility. At a depth of approximately 1–2 m, an undistinguished, fine, reddish horizon canbe found locally; it is mainly constituted by bricksand other waste material (Fig. 5). This horizon isnot continuous and has also been found randomlyin the containment area during some surveys per-formed in the past.


Fig. 5 Stratigraphy and depth of soil samples for laboratory test. The soil is extremely heterogeneous both in the horizontaland in the vertical direction. In some case, anthropogenic materials were observed at shallow depths

Hydraulic conductivity tests

Saturated hydraulic parameters were obtainedfrom multi-scale sampling tests. Depending onthe investigated representative volume, differentorders of magnitude should be obtained, due toan increase in soil heterogeneities for differentand controversial reasons. However, even if large-scale tests (i.e., pumping tests) are assumednormally to deliver better global effective trans-missivity values, here the interest was also tocharacterize small-scale hydraulic properties. Infact, small features (e.g., higher conductivity path-ways) are crucial in the solute migration fate intolower conductivity media (Zheng and Bennett1997).

Falling head permeability laboratory tests werecarried out on aquitard A0–A1 samples and anaverage 1.2·10−7 m s−1 was obtained. As shown bydouble-ring infiltrometery test results, the surficiallayer showed different degrees of compaction.This significantly affects infiltration at a localscale, even if the actual influence of this fac-tor on the overall infiltration in zone 1 needsto be investigated through a numerical analy-sis. According to these field tests, vertical satu-rated hydraulic conductivity was measured witha total variation of three orders of magnitude(from 10−4 to 10−7 m s−1), with a consequentialhigh sensitivity to the compaction of the surficiallayer. Traditional pumping tests and a slug testwere also performed to investigate the hydraulic


Table 1 Experimental hydraulic conductivities at differenthorizons of the Rho aquifer using different testing devices

Layer Saturated hydraulic Methodconductivity (Ks) m/s

A0 3.50E-04 Pumping testA1 2.00E-03 Pumping testAquitard A0–A1 8.30E-08 Slug test

Mean 1.20E-07 LaboratorySt. Dev. 1.35E-07 Laboratory

For laboratory experiments, mean and standard deviationsresulting from repeated measurements are also reported

conductivity of the two aquifers, A0 and A1, andof aquitard A0–A1, respectively.

Overall results are presented in Table 1.

Vadose zone monitoring

Monitoring of the vadose zone was a key point inthe program, because it allowed the quantificationof the recharge rate in the area. Figure 6 rep-resents the observation of soil moisture contentusing the two continuous electrical data loggers.EVS-1 records are shown on the left side ofFig. 6, while EVS-2 are plotted on the right. Thedataset series we considered are limited for bothdevices to a close time window within the mon-itoring program. Actually, this monitoring lastedtwelve months because of instrument failure dur-ing an exceptionally warm summer period. Duringthe considered time, at least two major rainfall

D−06 J−07 F−07

20

30

40

10

50

Moi

stur

e C

onte

nt

0

10

20

30

40EVS1

10

20

30

40

50

0

10

20

30

40

Rai

nfal

l (m

m)

EVS2

rainfall 90cm

80cm

60cm

120cm

30cm

D−06 J−07 F−07

20cm

60cm

40cm

Fig. 6 Experimental soil moisture values at differentdepths and rainfall heights. It can be seen that the soil isheterogeneously stratified by layer of different moisture

content. For the same depths, the device recorded in mostcases different values, which reflect the horizontal hetero-geneous spatial distribution of hydraulic properties


events (gray line) occurred, at a rate of more than5 mm h−1. The texts within the plots indicate thedepth of the lecture probes, and thus at whichdepth the corresponding dataset refers to. Thefunctionality of probes was manually tested withoccasional surveys.

Some critical aspects concerning the heteroge-neous distribution of the soil hydraulic propertiesmay be observed and commented upon. Be-cause of the absence of rainfall events, all probes(excepting the shallower probe of EVS-2) main-tain a constant value of moisture content. Inspecific, probes at 60-, 90- and 120-cm depths forEVS show values above 30% with no remarkableresponse to rainfall events. This behaviour is typ-ical of fine-grained deposits, which are sparselydistributed at different depths in the study area(see Fig. 5); indeed, EVS-2 was located in a sectorwith the anthropogenic fine and strongly consoli-dated deposits. A different question concerns the30-cm depth probe in EVS-2 which showed anunrealistically fast response to precipitation, prob-ably due to preferential flow (poor adhesion ofthe sensors to the soil at low depth). On the otherhand, EVS-1 show lower mean values of moisturecontent, and a faster response to rainfall events.We can assume that at this point the depositscan be coarser than the ones at the same depthsaround EVS-2 locations; especially, at depth 80cm the moisture content is extremely low andwe still observe a minimum response to rainfallevents. It is probable than a coarse grained for-mation is encountered at this depth and that ahigher hydraulic connections takes place duringrecharge events. In general, the results show ahigh degree of spatial variability of the hydraulicproperty of the soils, both in the vertical directionand, referring also to Fig. 5, in the horizontal plan.

Laboratory tests

Falling head hydraulic conductivity tests were pre-viously described as performed on fine-grainedsediments composing aquitard A0–A1. Other lab-oratory tests were also carried out. Grain-sizeanalysis and organic content tests were executedon samples collected from drilled core samplesat different depths. The purpose was to bettercharacterise the stratigraphy of the upper 3.6 m

and to obtain data to indirectly derive the hy-draulic function of sediments in the vadose zone.This aspect is extremely important, because theselayers influence the infiltration process withinthe containment area the most. In this way, thedetermination of reliable physical functions is akey point to obtaining a correct infiltration rateduring rainfall events, and to determine the localrecharge of the area.

Groundwater heads monitoring

Groundwater levels were monitored twice a yearusing about twenty piezometers outside the en-capsulate area (zone 2), while three points havebeen chosen to be measured almost daily. Thesepoints were located according to the followingscheme (Fig. 4): one in zone 1 within the A0aquifer (A0-int) and the other two in zone 2,just outside the walls, in A0 and A1 aquifers(A0-ext, A1-ext), respectively. The aim was todetect different hydraulic behaviors of the innerand outer zones of the isolated A0 aquifer, incomparison with the A1 aquifer. Measurementsin a long, dry period allowed us to observe thatthe progressive decrease of groundwater head atpoints A0-int, A0-ext and A1-ext occurred withthree different rates (Fig. 9). As they were notinfluenced by external factors such as precipita-tion or well pumping, they were considered opti-mal for model calibration. It is also important tohighlight that evapotranspiration processes fromthe saturated layer were not active either, mainlybecause piezometric levels were at least 4 m belowthe topographic surface, and also because of thepresence of low conductivity horizons of backfillat low depth, which inhibits such processes.

Numerical modelling

Model settings

The conceptual model used in the numericalanalyses covers the entire containment (about25 m width, see Fig. 2), up to an average depth of30 m. The conceptual model included the follow-ing assumptions, and according to the hydrostrati-graphic scheme in Fig. 2, the model is composed


Fig. 7 Hydrogeologicalconceptual model fornumerical analysis. Atshallow depth the gridhas finer refinement toreproduce theexperimental observationin the first centimeters ofthe encapsulated area

of (Fig. 7): (1) a shallow aquifer (A0), from theground surface to a depth of 9 m; (2) a fine-grainedaquitard (A0–A1), which divides A0 from thelower semi-confined aquifer and has been locatedfrom a depth of 9–11 m; (3) the second mainaquifer (A1), from a depth of 11–30 m; and (4)the lowest fine-grained aquitard (A1–A2), up to adepth of 33 m. For the top layer, a finer verticaldiscretisation was chosen to take into account theupper, thin heterogeneities and to increase theability of the model to effectively evaluate theirinfluence on the recharge dynamics. For the firstattempt, the layer was divided according to the po-sition of the device sensors for the water contentand tensiometers.

The vertical barrier is represented by meansof fine-grained materials located from the topsurface to a depth of 9.30 m (30 cm inside theaquitard, to represent the keying). The barriersystem is 17-m wide, and occupies one third of thetotal depth of aquifer A. The profile is orientedalong the local groundwater flux in the A0 aquifer(Fig. 4). An imposed-valued function (Dirichletconditions) was adopted as boundary conditionson the lateral sides. SEEP/W allows the use of“infinite elements” to simulate far conditions, lim-iting border effects on the model (GEO-SLOPEInternational Ltd. 2002 2006). The bottom sidehas an imposed-derivative function (Neuman con-dition), for which no flux was assigned forthis problem, since leakages through the secondaquitard are considered negligible.

Model calibration

Calibration and verification tasks were performedto identify unique series of proper parameters,adopting current techniques from the groundwa-ter model literature (e.g. Anderson and Woessner1992). Trial and error procedures were followed,and no automatic method was taken into account.The model was calibrated for both unsaturatedand saturated properties. Three different sets ofdata, derived from the monitoring activities, havebeen chosen for optimisation: (1) agreement withhead values in observation boreholes close to thestudied domain; (2) agreement with pore-waterpressure obtained by tensiometers monitoring;and (3) agreement with soil moisture content mea-sured by means of electric probes. As a first at-tempt, retention curves and hydraulic conductivityfunctions for materials in the vadose zone weredetermined using the Gupta and Larrson (1979)technique starting from the grain-size curve, withspecific weight and organic content determinedthrough laboratory tests. Hydraulic conductiv-ity functions were calculated from the reten-tion curves with the Green and Corey tech-nique (Green and Corey 1971) using the experi-mental saturated hydraulic conductivity.

First step: calibration of retention curves

Two pre-rainfall reference temporal momentswere chosen to reproduce initial steady-state


conditions, with different saturation profiles.Transient analysis was performed to simulate theeffect of the registered rainfall events, and appliedas a boundary condition to the topographical sur-face. During the calibration phase, soil moisturecontent and pore-water pressures have been used,and the model was considered calibrated whenthere was a good agreement (residual lower than5%) between experimental and calculated data.Soil retention curves were changed to reach thisresult, where points with the pair of pore-waterpressure and soil moisture content measured atthe same depth in the same time were consideredas fixed. Examples of results of the calibrationphase are shown in Fig. 8.

When a randomly distributed heterogeneouslayer, at a shallow depth into the soil, was as-sumed, the simulated soil moisture content fitswell with observed values at different depths (onprobe sensors) in the pilot area. Saturated prop-erties of this horizon will be discussed later. Thebackfill layer had a significant impact and wasconfigured with high moisture content and abrupttransitions to upper and lower horizons. Thislayer, in particular, was calibrated by assigningthe first shallow layer of the model a saturatedhydraulic conductivity (Ksat) = 5 · 10−6 m s−1; un-saturated hydraulic parameters were then derived

from typical low-permeability materials, for exam-ple, adopting retention curves that showed a softerdecrease of hydraulic conductivity with suction.In this case, however, volumetric soil moisturecontent holds saturation close to 100%, and theretention curve is not influential.

Second step: calibration of saturated conductivity

Following calibration of the retention curves,transient simulations were set-up for a period ofsix months, and divided into 600 time steps. Along, dry period was chosen in order to recreate agroundwater head drawdown trend of the system(drainage conditions) and to validate the model.In this way, no recharge flux was applied to the up-per model layer, since no rainfall event occurredduring the observed time. Two different functionsfor the imposed head boundary conditions of themodels were used in order to detect the actualdynamics through the borders of the encapsulatedarea. At the upper A0 aquifer, constant headvalues were applied to infinite elements for thewhole transient time in order to leave piezometriclevels free to change within the model as timestepswere solved. At the lower (A1) aquifer, time-variable values were applied to the boundaries.This function was chosen based upon observation

Fig. 8 Results of model calibration for unsaturated flow. The numerical simulation fits the experimental data if a layer withlower permeability and high moisture content is set at a shallow depth


Fig. 9 Observed andsimulated drawdown dataat two observation points.Value 0 (zero) ondrawdown axis representsthe reference datameasured in the aquifersat the beginning of thedrawdown phase

at the A1 piezometer, located immediately closeto the downstream wall border (Fig. 4). The samefunction was applied to the upper limit of the A1aquifer, but modified in absolute values accordingto the measured hydraulic gradient. The outputsof the transient state simulation were verifiedby comparing the observed exhaustion curve inaquifer A0 piezometers (Fig. 4) and the computedhead.

Initial parameter values were assigned accord-ing to laboratory measurements and the in situ testresults described in the previous paragraph. Also,for the vertical barriers, since no hydrodynamicinformation was available, an initial range of low-permeability values from 5·10−7 m s−1 (i.e., similarto the aquitard vertical conductivity evaluated bylaboratory test) to around 5·10−9 m s−1 (typicalvalues of well-consolidated slurry cut-off walls)was considered.

Figure 9 shows the good agreement betweenthe observed and computed values. It must benoted that real values tend to oscillate after reach-ing the bottom of the A0 shallow aquifer whilecalculated the curve flattened out. This behaviour

can be related to the formation of small waterpools on the small-scale irregularities at the topof the fine grain aquitard that were not analysedin the model. A randomly distributed, heteroge-neous layer at a shallow depth into the soil wasmaintained, as observed by the “stratified” soilmoisture contents (Fig. 6). Table 2 summarises thefinal hydraulic properties of both the aquifer andthe aquitards after the calibration process.

Model results

Water release incidents occurring from the iso-lated zone were assessed by evaluating flux

Table 2 Hydraulic conductivities adopted in the SEEPmodel for different geological units and anthropic features

Layer Model Ks m/s

A0 3.50E-04A0 - Brick layer 5.00E-06Aquitard A0–A1 2.00E-03A1 1.24E-7Aquitard A1–A2 1.00E-08Slurry walls 5.00E-08


sections on the lateral and bottom surfaces of theisolated area. An average value (the integral) ofthe total leakage from one boundary was thencalculated. Figure 10 shows a synthetic plot ofcalculated water leakage. While values at the bot-tom surface seem to maintain a generally stableamount at 0.5 m2 d−1 during the simulation, lateralvalues seems to decrease from an initial 0.75 to0.65 m2 d−1 in the simulation time (180 days), de-livering a general reduction of about 20% in termsof normalised flux (lateral vs. bottom spreadingratio). In any case, the results showed that outflowis given by both the lateral boundaries throughthe slurry wall, and the bottom boundary throughthe aquitard. This result matches the second hy-pothesis (Case 2) we formulated as a reliableconceptual model for the study area (Fig. 2—down). It is interesting to note that losses aregenerally higher through the slurry walls, withvalues reaching about 18 m3 d−1, while the dis-charge through the bottom is about 13 m3 d−1.This gives a value for total losses of more than30 m3 d−1, which represent a significant amount ofcontaminated water released by the containmentsystem.

Evidently, a direct correlation between the sat-urated volume of the aquifer within the contain-ment system and the lateral spreading exists, whilethe leakage from the bottom is only slightly de-pendant on the saturated volume. That means thatincreasing head within the system has only a smalleffect on the amount of total leakage; the leakagemainly increases or decreases according to thechanges in the filtration surface, which is givenonly by changes of the lateral filtration surface.

Scenario simulations in recharging conditions

According to the importance of the hydraulichead values within the containment area, as de-scribed in the previous paragraph, and in orderto describe qualitatively and quantitatively thesystems reaction to recharge, a scenario analy-sis based on two simple cases (A and B) wasset-up, based on the discharge model but withincreased total simulation time. We focused onthe infiltration contribution as a critical point. Inscenario A, the hydrogeological background wasmaintained with conditions identical to the cal-ibration processes. Measured rainfall rates wereavailable at a meteorological station near the Rhosite, and were applied as the inward flux (Neumanboundary conditions) at the nodes of the top layer.For the sake of the problem, we assumed com-plete effective infiltration of water, and also dis-carded any topographic gradient, and thus no wa-tershed may deliver secondary recharge from thetop. In scenario B, we applied the same Neumanboundary conditions, but a homogenous aquiferwas considered inside the containment system, asif no buried waste materials existed. In this way,by comparing results of the two scenarios, theactual impact of the backfill on the dynamics ofthe infiltration process was calculated. The mainlimitation was the lack of experimental data todirectly “verify” scenario outputs, which deliversuncertainty to the analysis even if the numericalmodels have been thoroughly calibrated. In ad-dition, due to high CPU calculation times (up to1 week for each simulation of full drainage condi-tion), both scenario simulations have been limited

Fig. 10 Change ofleakage during thetransient simulationthrough the containmentsystem. The values areexpressed in terms ofabsolute values of thelateral and bottom fluxper unit of wall length(left axis) and of therelative proportion oflateral flux with thebottom (right axis)


Fig. 11 Evaluation of the change of leakage in time through the containment system under recharge for different scenarios

to a short single rainfall event (5 days) after thedrought period (in which complete drainage wassimulated). The results revealed the importanceof the human impact in the dynamics of aquifer,affecting the vertical recharge profile, and thus thecalculated leakages, both through the lateral andbottom sides.

Figure 11 shows that leakages from lateralconditions are much higher in the case ofincidental heterogeneities in the soil (case A, darkcontinuous line) than in the case of homogeneousconditions (case B, grey continuous line). On theother hand, the homogeneous aquifer (case B)reacts promptly to instantaneous recharge;

Fig. 12 Qualitative effects of waste backfill and slurry wall on the seepage vectors occurring after a recharge event. Thefigure refers to the area highlighted in Fig. 7


bottom leakages react quickly, but then reach anasymptotic value close to 0 m2 d−1. Only lateralflux exists, but lateral leakages become stableat least until the end of the simulation. For theheterogeneous case (A), leakages react later butasymptotic values with non-negligible values areobtained.

This behaviour contrasts with the assumed hy-pothesis that in homogeneous conditions (B),infiltration could have been more effective, reach-ing the bottom of the aquifer, and generatingbottom flux. On the other hand, higher values areobtained from lateral spreading in heterogeneouscase (A). We can give a physical explanation as-suming that the heterogeneous layer, made upby impermeable but saturated media, acts as a“barrier” which makes preferential paths divergein lateral directions rather than vertically down-wards. Since it is located at shallower depths,water avoids infiltrating into larger unsaturatedvolumes, which would retain more water and takelonger for enough soil moisture content to accu-mulate in order to permit gravitic seepage. There-fore, since fewer volumes are affected, seepageoccurs rapidly and diverges. Figure 12 offers asnapshot of seepage direction (indicated by darkarrows) close to the upstream vertical walls duringthe initial phase of the infiltration process in theheterogeneous (A) case.

Discussion

The results showed that the poor global perfor-mance of the containment system stems from alack of knowledge about the geologic structureof the area. In fact, a remarkable exchange be-tween the inner and the outer part of the confinedsource area occurs through the aquitard horizon.The slurry walls seem to have lost most of theircontainment properties and presently they havealmost the same amount of losses as the aquitard.However, it is probable that in the past, whenwalls were close to the top of the performance,they may have played a role as a funnel andallowed the plume to head even more directlytowards the lower layers. The lack of any otherremediation system coupled with the main systemcontributed to the development of the present

groundwater contamination in an extended area.This indicates, firstly, the need for a revision of theexisting remedial action, which can be effectivelydesigned only with the addition of reliable knowl-edge of processes analysed and quantified, bothin the vadose zone and in the aquifer, through acomplete and calibrated numerical model. Other,further enhancements may include a better knowl-edge of the spatial distribution of heterogeneitiesin the soils. This is remarkably important, es-pecially for anomalous features, to better assessthe local influence of infiltration to the generalbehaviour of the source area. Further solutions byindirect analysis (via numerical method) allowedus to estimate the approximate assets of the sub-surface engineering structures. The system seemsparticularly sensitive to:

– the state of the slurry walls, in terms of hy-draulic retention capacity;

– the geometry and the extension of theaquitard on a detailed spatial scale; and

– abrupt changes in the composition of thesoil, due to the presence of waste backfilledmaterials.

Slurry walls are only partly responsible for thelack in performance of the containment system.A non-negligible flux takes place through theaquitard and inevitably contributes to the spreadof the plume. Waste backfill could intervene inthe alteration of the natural infiltration process ofthe encapsulated aquifer. The low transmissivityof this fine-grained horizon, even if it is located inheterogeneous and isolated spots, tends to reduceand delay the water infiltration into the shallowconfined aquifer.

Conclusions

The present study aimed to achieve quantita-tive results about the overall effectiveness of acontainment system, without using invasive tech-nique, and to demonstrate the feasibility of thisapproach for future uses. In order to assessglobal water release from a containment system,an unsaturated–saturated flow model has beencalibrated through an extensive monitoring sys-tem developed for the specific case study. Both


unsaturated and saturated hydraulic parameterswere used, whose reliability came from multi-scaletests. Laboratory and in situ-derived unsaturated(soil moisture content and hydraulic conductivitycurve) and saturated hydrodynamic parameterswere optimised through model calibration in bothsteady-state and transitory-state conditions. Thisallowed us to hydraulically characterise all materi-als, including waste backfilling, even if their actualexact spatial distribution remains unknown due tothe lack of information. However, the influenceof these heterogeneities on the infiltration processhas been generically established and, for a generalview, considered as an impediment to normal hy-drological cycle and recharge processes. The mostimportant thing to note is that, presently, lossesthrough the containment system come both fromlateral spreading directly into the A0 aquifer, andinto the A1 aquifer through the bottom aquitard,according to case 2 proposed in “Hydrogeologicalsetting”. Thus, the hot spot in the containmentsystem can be considered responsible for contam-ination found downgradient in both the aquifers.The results of this research will be also a reli-able base for future development of the study,concerning development of transport, multi-phasemodelling. They also furnish the initial input togenerate a quantitative analysis of the risk offurther spreading of contaminant from the pol-luted site. Nonetheless, in spite of the fact thatstochastic methods are a suitable way to deal withsoil heterogeneity, further consideration of theefficiency of the barrier may be obtained only withnew direct investigation in the site. In any case,the combination of monitoring and modelling al-lowed us to make a reliable estimate of the overallperformance of the physical confinement and itspossible influence on groundwater contamination,without using any invasive techniques on slurrywall containment systems, as is often required be-cause of the potential risk for an increased releaseof contaminants.

References

Alberti, L., Brogioli, G., Formentin, G., Marangoni, T.,& Masetti, M. (2007). Experimental studies andnumerical modeling of surface water–groundwater

interaction in a semi-disconnected system. In XXXVIAH congress, groundwater and ecosystems, Lisbon,Portugal.

Anderson, M. P., & Woessner, W. W. (1992). Appliedgroundwater modeling: Simulation of f low and advec-tive transport. San Diego: Academic.

Bayer, P., Finkel, M., & Teutsch, G. (2004). Hydraulicperformance of a combination of pump-and-treat andphysical barrier systems for contaminant plume man-agement. Ground Water, 42(6), 856–867.

Benson, C. H. (2002). Containment systems: Lessonslearned from north american failures. In Environmen-tal geotechnics (4th ICEG), Swets and Zeitlinger, Lisse,1095–1112.

Beretta, G. P., Bianchi, M., & Pellegrini, R. (2003).Linee guida per la verif ica ed il collaudo delle barriereimpermeabili per la messa in sicurezza di siti conta-minati. Technical report, Provincia di Milano, Milan(in Italian).

Bolster, D., Barahona-Palomo, M., Dentz, M., FernàndezGarcia, D., Sanchez-Vila, X., Trinchero, P., et al.(2009). Probabilistic risk assessment applied to conta-mination scenarios in porous media. Water ResourcesResearch, 45, W06413. doi:10.1029/2008WR007551.

Brandelik, A., & Huebner, C. (2003). Moisture monitor-ing in waste disposal surface barriers. EnvironmentalMonitoring and Assessment, 84, 61–70.

Britton, J. P., Filz, G. M., & Herring, W. E. (2004).Measuring the hydraulic conductivity of soil-bentonitebackfill. Journal of Geotechnical and Geoenvironmen-tal Engineering (ASCE), 130(12), 1250–1258.

Candelaria, L. M., & Matsumoto, M. R. (2000). Effectsof NAPL contaminants on the permeability of a soil-bentonite slurry wall material. Transport in PorousMedia, 38, 43–56.

Chen, D. W., Moeti, L., Carsel, R. F., & Vona, B.(1999). Assessment and prediction of contaminanttransport and migration at a Florida superfund site.Environmental Monitoring and Assessment, 57, 291–299.

Choi, H., & Daniel, D. E. (2006). Slug test analysis invertical cutoff walls. I: Analysis methods. Journalof Geotechnical and Geoenvironmental Engineering(ASCE), 132(4), 429–438.

D’Appolonia, D. J. (1980). Soil-bentonite slurry trenchcutoffs. Journal of Geotechnical and Geoenvironmen-tal Engineering (ASCE), 106(4), 399–417.

de Barros, F. P. J., Rubin, Y., & Maxwell, R. M. (2009). Theconcept of comparative information yield curves andtheir application to risk-based site characterization.Water Resources Research, 45, W06401.

ENI-AGIP (2002). Geologia degli acquiferi padani dellaRegione Lombardia. Regione-Lombardia—ENI Divi-sione AGIP.

Evans, J. C. (1993). Vertical cutof f walls. Geotechnicalpractice for waste disposal (Chapter 17). New York:Chapman & Hall.

Filz, G. M., & Mitchell, J. K. (1995). Design, construction,and performance of soil- and cement-based verticalbarriers. In R. R. Rumer, & J. K. Mitchell (Eds.), In-ternational containment technology conference (p. 63).

http://dx.doi.org/10.1029/2008WR007551


Baltimore: US DoE, US EPA, and DupontCompany.

GEO-SLOPE International Ltd. 2002 (2006). SEEP/W forf inite elements seepage analysis version 5 user’s guide,Calgary.

Green, R. E., & Corey, J. C. (1971). Calculation of hy-draulic conductivity: A further evaluation of somepredictive methods1. Soil Science Society of AmericaJournal, 35(5), 3–8.

Gupta, S. C., & Larson, W. E. (1979). Estimating soil wa-ter retention characteristics from particle size distrib-ution, organic matter percent and bulk density. WaterResources Research, 15, 1633–1635.

Hajnal, I., Marton, J., & Regele, Z. (1984). Construction ofdiaphragm walls. New York: Wiley.

Hudak, P. F., & Loaiciga, H. A. (1999). Conjunctive vadoseand saturated zone monitoring for subsurface conta-mination. Environmental Monitoring and Assessment,59, 15–29.

Inyang, H. (2004). Modeling the long-term performance ofwaste containment systems. Environmental Science &Technology, 38(17), 328–334.

Inyang, H. I., & Tomassoni, G. (1992). Indexing of long-term ef fectiveness of waste containment systems for aregulatory impact analysis (p. 29). Technical report, atechnical guidance document, Office of Solid Waste,US EPA, Washington.

Manassero, M. (1994). Hydraulic conductivity assessmentof slurry wall using piezocone test. Journal of Geotech-nical and Geoenvironmental Engineering (ASCE),120(10), 1725–1746.

Manassero, M., Fratalocchi, E., Pasqualini, E., Spanna, C.,& Verga, F. (1995). Containment with vertical cutoffwalls. In Y. B. Acar, & D. E. Daniel (Eds.), GeoEn-vironment 2000, GSP no. 46 (pp. 1142–1172). NewOrleans: ASCE.

Nash, K. L. (1974). Diaphragm wall construction tech-niques. Journal of the Construction Division (ASCE),100(4), 605–620.

Paul, D. B., Davidson, R. R., & Cavalli, N. J. (1992).Slurry walls: Design, construction and quality control.

In ASTM STO 1129. Philadephia: American Societyfor Testing and Material

Pedretti, D., Masetti, M., & Francioli, A. (2009). Geo-statistical techniques for DNAPL contamination as-sessment in polluted aquifers. The case of the for-mer Chimica Bianchi facility in the Milan-Rho district.Rendiconti online Soc. Geol. It., 2(1–3).

Rolle, E., Beretta, G. P., Majone, M., Pedretti, D.,Petrangeli-Papini, M., & Raffaelli, L. (2009). Anal-isi delle alternative tecnologiche per il conteni-mento della contaminazione di acque sotterranee.In Seiminario sulla reindustrializzazione di siti in-quinati e tecnologie di intervento sulle acque sotterra-nee sui sedimenti. Rome: Italian Ministry od EconomicDevelopment.

Ryan, C. R. (1987). Vertical barrier in soil for pollutioncontainment. Geotechnical practice for waste disposal(pp. 182–204). New York: ASCE.

Ryan, C. R. (1994). Slurry cutoff walls: Applications in thecontrol of hazardous wastes, hydraulic barriers in soiland rock. In A. I. Johnson, R. K. Frobel, N. J. Cavalli,& C. B. Pettersson (Eds.), STP 874 (pp. 9–23). Denver:ASTM.

Ryan, C. R., & Day, S. R. (2003). Soil-bentonite slurry wallspecifications. In Pan American conference on soilsmechanics & geotechnical engineering. Cambridge:Geo-Institute and MIT.

Tartakovsky, D. M., & Winter, C. L. (2008). Uncertainfuture of hydrogeology. ASCE Journal of HydrologicEngineering, 13(1), 37–39.

United States Environmental Protection Agency (EPA)(1998). Evaluation of subsurface engineered barriers atwaste sites. EPA 542-R-98-005, 148 pp.

Winter, C. L., & Tartakovsky, D. M. (2008). A reducedcomplexity model for probabilistic risk assessmentof groundwater contamination. Water Resources Re-search, 44(1), W06501.

Xanthakos, P. P. (1994). Slurry walls as structural systems(2nd ed.). New York: McGraw-Hill.

Zheng, C., & Bennett, G. D. (1997). Applied contaminanttransport modeling (2nd ed.). New York: Wiley.

Documents

Slurry wall containment performance: monitoring and modeling of unsaturated and saturated flow