Combined use of heat and saline tracer to estimate aquifer properties in a forced gradient test

Combined use of heat and saline tracer to estimate

aquifer properties in a forced gradient test

N. Colombani1, B.M.S. Giambastiani

2#, M. Mastrocicco

3

1 Department of Earth Sciences, “Sapienza” University, P.le A. Moro 5, 00185 Rome, Italy 2 Interdepartmental Research Centre for Environmental Sciences (CIRSA), University of Bologna, Via S. Alberto 163,

48123 Ravenna, Italy 3 Department of Physics and Earth Sciences, University of Ferrara, Via Saragat 1, 44122 Ferrara, Italy # corresponding author: [email protected]

Abstract

Usually electrolytic tracers are employed for subsurface characterization, but the interpretation of

tracer test data collected by low cost techniques, such as electrical conductivity logging, can be

biased by cation exchange reactions. To characterize the aquifer transport properties a saline and

heat forced gradient test was employed. The field site, located near Ferrara (Northern Italy), is a

well characterized site, which covers an area of 200 m2 and is equipped with a grid of 13

monitoring wells. A two wells (injection and pumping) system was employed to perform the forced

gradient test and a straddle packer was installed in the injection well to avoid in-well artificial

mixing. The contemporary continuous monitor of hydraulic head, electrical conductivity and

temperature within the wells permitted to obtain a robust dataset, which was then used to accurately

simulate injection conditions, calibrate a 3D transient flow and transport model and to obtain

aquifer properties at small scale. The transient groundwater flow and solute-heat transport model

was built using SEAWAT. The result significance was further investigated by comparing the results

with already published column experiments and a natural gradient tracer test performed in the same

field. The test procedure shown here can provide a fast and low cost technique to characterize

coarse grain aquifer properties, although some limitations can be highlighted, such as the small

value of the dispersion coefficient compared to values obtained by natural gradient tracer test, or the

fast depletion of heat signal due to high thermal diffusivity.

Keywords saline tracer; heat tracer; numerical modelling; aquifer properties; dispersivity.

INTRODUCTION

Tracer tests are regularly employed methods to characterize some of the key hydrogeological

parameters that control groundwater flow and transport (Ptak et al., 2004). Unfortunately,

comprehensive tracer tests require a substantial effort in terms of chemical analysis of a large

number of samples, especially where grids of multilevel sampling wells are involved (Sudicky,

1986). The high cost for sampling and chemical analysis may be efficiently reduced by employing

in situ instruments that provide continuous measurements, such as electrical conductivity (EC)

probes equipped with data loggers, which are able to identify the variables governing nonreactive

solute transport both in laboratory and in field (Cirpka et al., 2007; Vogt et al., 2010). Also

fluorescent dyes are often used for the same purpose (Kung et al. 2000; Richardson et al. 2004).

Numerous techniques have been developed to limit the duration of experiments and therefore to

minimize sampling and analytical costs. Forced gradient tracer tests, for example, allow for faster

breakthrough times by increasing natural groundwater velocities (Chen et al., 1999; Sutton et al.,

2000). However, sampling requirements in many cases might not be dictated by the absolute length

of the experimental period, but rather by the temporal resolution required to resolve breakthrough

signals at sufficient detail.

Saline solutions are the most commonly used hydrological tracers because of their availability,

relative low costs, simple handling, easy monitoring by geoelectrical measurements (Cassiani et al.,

2006; Perri et al., 2011) and the potential for continuous measurement. Unfortunately, these low-

cost techniques only provide information on the total concentration of ions in solution and they

cannot solve the ionic composition of the aqueous solution. This limitation can introduce a bias in

the estimation of aquifer parameters where ion exchange and sorption phenomena between saline

tracers and sediments become relevant (Mastrocicco et al., 2011a). Salt tracers are mainly useful in

small-scale experiments, such as soil column tests or investigations in small surface water bodies. In

general, only selected anions, such as chloride (Cl-) and bromide (Br

-), are recognized to be

transported unretarded and they are referred to as conservative tracers or mobile anions (Leibundgut

and Seibert, 2013). Br- is often used for tracer experiments (Parsons et al., 2004; Sambale et al.,

2000) because of its lower background concentration in natural waters (about 300 times lower than

Cl-), low toxicity, and low sorption to soil particles.

However, cations of salt tracers within the saline tracer may interact with the soil matrix through a

range of processes such as ion exchange, surface complexation and via physical mass-transfer

phenomena. Heterogeneous reactions with minerals or mineral surfaces may not be negligible

where aquifers are composed of fine alluvial sediments.

Also heat is an almost free and readily available tracer that provides easy and cost-effective

observation to parameterize groundwater systems, despite the fact that it could be usually used for

small travel distances. The heat is a tracer widely used for detecting areas of groundwater recharge

and discharge (Conant, 2004; Ferguson et al., 2003; Rau et al., 2014; Silliman and Booth, 1993), to

estimate infiltration rates and near-river groundwater velocities (Molina-Giraldo et al., 2011) and to

determine the spatial and temporal infiltration dynamics during managed aquifer recharge (Racz et

al., 2011). Moreover, many thermal injection and recovery experiments have been simulated, both

in the field (Hermans et al., 2015; Palmer et al., 1992; Read et al. 2013; Vandenbohede et al., 2011)

and in laboratory by physical models (Giambastiani et al., 2013; Rau et al., 2012a, 2012b; Saeid et

al., 2014).

In general, heat can be used to define the physical parameters of the aquifer (Anderson, 2005;

Vandenbohede et al., 2011), but inaccurate results or misleading values can be obtained using only

heat as a tracer in fine-grained alluvial systems characterized by low groundwater velocities and

dominated by conduction (Giambastiani et al., 2013). In this case, for instance, it would not be

possible to distinguish between diffusivity and dispersivity if only a heat tracer test is performed,

without any solute tracer test. The non-conservative nature and retardation of heat limit the distance

that the use of heat as an applied tracer will be appropriate under a natural gradient (Irvine et al.

2013). In addition, the use of heat as a sole tracer could lead to unrealistic dispersivity values also in

dipole tracer tests in heterogeneous alluvial aquifers (Wildemeersch et al., 2014) and highly damped

temperature signals within a short distance from the injection well (Doro et al. 2014). Dispersion is

typically based on assumed values of longitudinal dispersivity, which is a notorious “adjustable

factor” in calibration of contaminant transport models, and this practice can introduce an

unacceptable level of uncertainty to risk or remedial considerations. So in many cases, to overcome

these issues and accurately estimate aquifer parameters, coupling heat and chemical tracer

experiments with numerical simulations is unavoidable (Ma et al. 2012; Rau et al., 2012a;

Vandenbohede et al., 2009; Wildemeersch et al., 2014). This coupling is performed for taking

advantage of the similarities between heat transfer and solute transport in porous media in order to

facilitate the separation of heat transfer processes and to identify related parameters: (i) effective

porosity simultaneously governs heat transfer by convection and solute transport by advection, but

it is estimated by fitting the chemical tracer breakthrough curve (Wildemeersch et al., 2014); (ii)

longitudinal solute dispersivity and thermal diffusivity could be inferred accurately from chloride

and temperature data sampled from the injection/extraction well during a push and pull test

(Vandenbohede et al., 2009), respectively.

In this context, the current study presents an in situ experiment designed for estimating aquifer

physical parameters in a small scale shallow aquifer near Ferrara (Po River plain, Italy). This

experiment consisted in performing a saline and heat forced gradient test and monitoring the

evolution of groundwater temperature and concentration in the recovery well. Hydraulic head and

concentration data were then used to calibrate a 3D transient groundwater flow, solute and heat

transport model using SEAWAT. The heat data was used in the validation step to adjust the saline

tracer retardation due to cation exchange that could eventually affect the test. Model results were

compared with results of previous studies conducted in the same area: a natural-gradient test

(Mastrocicco et al., 2011b) and a series of column displacement experiments (Mastrocicco et al.,

2011a).

The differences between the results obtained with the abovementioned methods and the ones

obtained by the combined use of heat and saline tracer in a forced gradient test are discussed in

order to highlight advantages and limitations of the latter method in estimating aquifer properties.

MATERIALS AND METHODS

Site description

The field site is a shallow unconfined aquifer located in Ferrara (Northern Italy, Figure 1). It is a

200 m2 field, equipped by a monitoring network of 13 monitoring wells, 5 m spaced each other. All

wells are 5-10 cm in inner diameter and are screened from 1 to 8 m below ground level. The

monitoring wells are installed in the unconfined aquifer of the paleochannel, which is composed of

Holocenic fluvial sandy deposits with small clay and silt lenses.

Figure 1 – Field site location.

The water table usually is found at 1.5 m below ground level. Underlying is a confining unit

(aquiclude), consisting of impermeable clay and silt sediments, 10 m in thickness, rich in organic

matter and peat. The location of the coarse sand lens boundaries were initially estimated by joint

multi-level slug tests and electrical resistivity tomography (Mastrocicco et al. 2010). Then, to locate

the coarse lens boundary, 9 stratigraphic cores were drilled (5 m spaced) manually with an

Ejielkamp Agrisearch auger forming a regular grid around the two wells (Figure 2). The hydraulic

conductivity values were obtained by multi-level slug tests data (Mastrocicco et al. 2010) and they

were used as first guess in the model setup as described in the following paragraph.

Hydrogeological parameters of the aquifer are reported in Table 1.

The hydraulic gradient of the shallow unconfined aquifer (1-3‰) is manly controlled by canals and

drainage ditches and, generally, it is northward. The flow velocity is slow (on average 10 m/y),

while the thermal conditions are typical of shallow unconfined aquifers with a mean temperature of

about 14 °C and an annual thermal excursion of about 5 °C. The groundwater has nearly constant

EC around 1±0.1 mS/cm. More details about the site stratigraphy, hydrogeology and groundwater

quality have been reported in previous studies (Mastrocicco et al., 2010; 2011b).

Analytical and field method

A two wells system, consisting of an injection and a pumping well located 2.5 m away (Figure 1),

was used to carry out a coupled heat and solute tracer test. The forced gradient tracer test was

conducted by injecting 10 g/l NaBr solution (10 litres) at 20°C (groundwater temperature = 12.7°C;

ΔT 7°C) into the injection well at a rate of 0.6 l/min and pumping water from the pumping well

using a centrifuge pump at 2 l/min. Once 10 litres of solution was completely injected, the pumping

was kept constant during the whole test duration in order to force the hydraulic gradient.

A straddle packer system was installed in the injection well in order to inject the solution at the

desired depth and avoid in-well artificial mixing. The wells do not have a gravel pack around the

case to avoid artificial mixing within the gravel pack. Hydraulic head, EC and temperature were

continuously monitored in both wells during the test by Solinst leveloggers, to detect breakthrough

curves. The accuracy of the ceramic pressure transducer was 1 cm, while for the temperature sensor

was 0.1 °C and for the EC sensor was 0.01 mS/cm, as reported on the Solinst manual.

The measured EC values were converted to dissolved saline tracer concentrations using a linear

relationship obtained by Mastrocicco et al. (2011a), where 0.86 was found to be the conversion

coefficient between EC and NaBr concentration with a regression coefficient of 0.995.

13 groundwater samples were collected in the pumping well and analyzed in laboratory for Br- by

using an isocratic dual pump ion chromatography ICS-1000 Dionex, to compared measured and

converted Br- concentrations. The monitoring dataset was then used to calibrate a 3D transient flow

and transport model to obtain aquifer properties at small scale.

Model setup

The 3D transient groundwater flow, solute and heat transport model was built using SEAWAT

(Langevin et al., 2008), which couples MODFLOW-2005 (Harbaugh, 2005) and MT3DMS (Zheng

and Wang, 1999) and allows to represent the simultaneous transport of heat and solutes, which

might lead to non-uniform solution densities and a subsequent effect on flow.

To simulate heat transport within the context of the SEAWAT framework, an MT3DMS species is

used to represent temperature. Simulation of heat transport with MT3DMS is based on the analogy

between solute and heat transport equations and the effectiveness of using MT3DMS to simulate

heat transport has been verified by several studies (Giambastiani et al., 2013; Hecht-Méndez et al.,

2010; Sethi and Molfetta, 2007). Heat transport is determined by convection and conduction, which

are analogous to advection and diffusion in the solute transport equation. Based on these analogies,

the solute transport equation is reinterpreted to simulate heat transport as widely documented by

Langevin et al. (2008), Thorne et al. (2006) and Vandenbohede et al. (2011).

Transient state groundwater flow and the transport simulations are referred to a portion of the field

site, which covers an area of 10 x 10 m, horizontally discretized in cells of 0.05 x 0.05 m with

vertical discretization of 0.2 m (Figure 2) for a total of 10 layers. The dense discretization was

found to be fundamental in order to avoid numerical oscillation and obtain model calibration

relatively to Br- and temperature. Constant head and constant concentration boundary conditions

were applied to the cells at the border of the model domain and numerical tests were done to

exclude boundary effects on the simulated results (e.g. doubling the size of the model domain did

not produce different model results in the monitored wells). Initial heads were extracted from the

piezometric contour map shown in Figure 1 and initial concentrations were set equal to 1 mS/cm

throughout the model domain.

In addition to the standard transient advection-dispersion equation (ADE), MT3DMS also

implements a dual-domain formulation (DD) that accounts for physical non-equilibrium processes.

In this formulation, the pore space is conceptually divided into two distinct domains: mobile and

immobile. In the mobile domain, transport is assumed to be governed by advection and

hydrodynamic dispersion. Within the immobile domain, no advective transport is assumed to occur.

Mass exchange between both domains is assumed to occur through a rate-limited mass-transfer

process. In the model, the total porosity (θ) is separated into a fraction representing the mobile

domain (θm) and a fraction representing the immobile domain (θim) (Table 2). The mass

conservation equation for the DD approach is (Zheng and Bennett, 2002):

x

Cvθ

x

CDθ

t

Cθ

t

Cθ m

mmm

mmim

imm

m

2

2

(1)

where Cm (ML−3

) and Cim (ML−3

) denote the mobile and immobile solute concentrations,

respectively, as functions of distance x (L) and time t (T). Under steady state flow conditions, the

hydrodynamic dispersion coefficient for the mobile region Dm (L2T

-1), the mean mobile pore-water

velocity vm (LT-1

) and the volumes θ (L3L

−3), θm, (L

3L

−3) and θim (L

3L

−3) are assumed to be

constant. Dm is defined as Dm = αLvm, where αL (L) is the longitudinal dispersivity. When θm = θ,

Eq. 1 reduces to the single-domain ADE. The solute-mass transfer between mobile and immobile

regions is defined as:

immim

im CCt

Cθ

(2)

where ω (L3L

−3T

−1) is a rate coefficient (or mass transfer coefficient). Eq. 1 is combined with Eq. 2

to give:

imm

m

mm

m

mm

m

m CCx

Cv

x

CD

t

C

2

2

(3)

The main characteristics that distinguish DD from ADE breakthrough curves (BTCs) of tracers are

the so called “early breakthrough”, related to accelerated transport via preferential pathways and

“tailing”, due to diffusion driven processes into stagnant zones. Thus, with increasing importance of

preferential pathways and stagnant zones the ADE approach becomes increasingly inappropriate

while solute transport behaviour can still be successfully described by the DD approach.

The hybrid method of characteristics (HMOC), which combines the method of characteristics and

the modified method of characteristics schemes by using an automatic adaptation of the solution

process to the nature of the temperature field, is used to solve the convection term, while the

generalized conjugate gradient (GCG) solver is used for the non-convection terms (Zheng and

Wang, 1999). A linear sorption isotherm is used in the chemical reaction package in which the

thermal retardation (R) is given by:

d

e

b Kn

R

1 (4)

where ne [-] is the effective porosity of the porous medium; ρb (Kg/m3) is the bulk density of the

porous medium, and Kd (m3/kg) is the thermal distribution coefficient defined as:

ff

sd

C

CK

(5)

where ρfCf (J/m3

°C) is the volumetric heat capacity of the water, and Cs (J/kg°C) is the specific heat

capacity of the soil.

The bulk thermal diffusivity (Dt) or thermal diffusivity (m2/d) is defined as:

ffe

bT

CnD

(6)

with

sefeb nn )1( (7)

where λb is the bulk thermal conductivity defined by λf and λs (W/m°C), which are the thermal

conductivities of fluid and soil, respectively.

Parameters values for the flow and transport model are listed in Tables 1 and 2.

The calibration procedure followed a stepwise approach, firstly calibrating the flow component of

the model versus the observed heads adjusting the K values. The calibrated K values, both for

coarse and fine sediments, are consistent with the ones obtained in the field via slug tests and

reported in Mastrocicco et al. (2010).

The transport component of the model was calibrated by using Br- concentrations in the pumping

well and adjusting the αL, θim, θm and ω values. The ratio between αL/αTH and αL/αTV were assumed

to be constant and derived from a natural gradient tracer test (Mastrocicco et al., 2011b). Finally,

the temperature was used to validate the calculated values, adjusting only the retardation factor.

Figure 2 - Model domain, boundary conditions and grid discretization. Shown are also injection

and pumping wells (black colour), and the auger cores (green colour) drilled to characterize the

field site stratigraphy.

Table 1 - Hydrogeological parameters used in the flow model; the calibrated parameters values are

denoted with the symbol *. K values were initially obtained by multilevel slug tests (Mastrocicco et

al., 2010) and then adjusted during the calibration procedure; Ss and Sy were obtained by previous

model (Mastrocicco et al., 2011b).

Sediment Type Parameter Unit Value

Coarse Sand

Hydraulic conductivity (K) m/s 4.0e-4

*

Specific storage (Ss) 1/m 1e-5

Specific yield (Sy) - 0.2

Silty Sand

Hydraulic conductivity (K) m/s 9.3e-6

*

Specific storage (Ss) 1/m 1e-5

Specific yield (Sy) - 0.2

Table 2 – Parameters used in the SEAWAT model; the calibrated parameters values are denoted the

symbol *.

Parameter Unit Value

Mobile porosity (θm) - 0.10*

Immobile porosity (θim) - 0.15*

Mass transfer coefficient (ω) l/d 0.346*

Bulk density of the porous medium (ρb) (kg/m3) 1600

Density of the solid matrix (ρs) (kg/m3) 2650

Specific heat capacity of the soil (Cs) (J/(kg °C)) 700

Volumetric heat capacity of the water (ρfCf) (J/(m3 °C)) 4.19e

6

Thermal distribution coefficient (Kd) (L/mg) 6.0e-8*

Retardation factor (R) - 1*

Bulk thermal diffusivity (DT) (m2/d) 8.64e

-5

Longitudinal dispersivity (αL) (m) 7.4e-2*

Horizontal transverse dispersivity (αTH) (m) 7.4e-3

Vertical transverse dispersivity (αTV) (m) 7.4e-4

Model performance was evaluated by calculating modelling efficiency (EF) (Nash and Sutcliffe,

1970) and Mean Absolute Error (MAE) values (Reusser et al., 2009; Zheng and Bennett, 2002)

based on equations 4 and 5, respectively:

(8)

where is the simulated value corresponding to the observed , and is the mean value of

observed data; and

simobs xxn

MAE1

(9)

where xobs is the observed parameter (hydraulic head, temperature and solute concentration) and xsim

the corresponding simulated value.

To further investigate model response to parameter changes, a sensitivity analysis was performed by

carrying out several simulations, in each of which one parameter is perturbed by a certain

percentage from its base value (i.e. that identified during the model calibration). After each

simulation the normalized sensitivity coefficient (NSC) is calculated to quantify the impact on the

simulated responses of the aquifer due to variations in the model parameters, thus identifying which

parameters are most sensitive. Being normalized, the sensitivity coefficients calculated after each

simulation can be ranked in order of importance. This coefficient is calculated from (Zheng and

Bennett, 2002):

(10)

where is the kth

parameter value; is the perturbation of the parameter value; is the

simulated output of the model selected for sensitivity analysis (e.g. hydraulic head for flow models

or contaminant concentration for transport models); is the change of the output due to the

perturbation of parameter .

A sensitivity analysis on the calculated heads was performed by changing K, Ss and Sy values using

a parameter perturbation of 50%. Then, αL, θim, θm and ω were changed from the base case

(calibrated model) to test the sensitivity of the model with a parameter perturbation of 50% for both

Br- and temperature.

Model performance, as well as sensitivities analysis, were evaluated by considering observed and

simulated hydraulic head, Br- concentration and temperature in the injection and pumping wells for

the whole duration of the test.

RESULT AND DISCUSSION

Figure 3 and 4 and indicate that a good calibration is achieved between modelled and observed

values, as also confirmed by EF and MAE values for hydraulic head, temperature and solute

concentration (Table 3).

Table 3 – Modelling efficiency (EF) and mean absolute error (MAE) values of the calibrated

model. The higher EF values and the lower MAE values, the better model calibration.

Hydraulic head (m) Temperature (°C) NaBr (mg/l) MAE 0.03 0.09 42.63

EF 0.93 0.85 0.96

Figure 3 shows an initial rise of hydraulic heads in both wells induced by the injection cycle. Then,

after less than one hour, the heads stabilize in both wells with lowest values in the pumping well,

indicating quasi-stationary conditions. A quite steep head gradient develops with an abstraction rate

of just 2 l/min. This is due to the low hydraulic conductivity of the silty sand unit constituting the

major portion of the aquifer (Table 1 and Figure 2) and the fully screened wells captured this

behaviour. The sensitivity analysis on the hydraulic parameters highlights a strong influence of K

on the calculated hydraulic heads, with changes of even 1 m respect to the calibrated model. This is

not surprisingly, since the flow models are mainly influenced by K values and distribution, while Ss

and Sy were much less effective in influencing the computed heads in both the injection and

pumping wells (Table 4). The similar sensitivity values of Sy and Ss is due to the fine vertical

discretization of the model grid, since SEAWAT treats as confined the fully saturated cells, thus

only few model cells are considered unconfined.

The transport parameters that most influence the computed concentrations are both αL and θm, since

they shift and modify the observed breakthrough curves of Br- and temperature, while ω and θim

were less effective in changing the computed concentrations.

Table 4 – Normalized sensitivity coefficients (NSC) calculated for selected parameters of the

calibrated model.

K field Ss Sy αL ω θim θm

Heads NSC 2.35 0.01 0.01 - - - - Br

- NSC - - - 1.83 0.36 0.15 1.39

Temperature NSC - - - 1.72 - - 1.22

The saline tracer breakthrough curve are well simulated by the model, while modelled temperature

profiles show a small retardation in the injection well and slightly smaller values in the pumping

well compared to observed values (Figure 4). This is probably due to heat dissipation within the

injection well, although a straddle packer system was used to minimize in-well artificial mixing

processes, often reported as cause of bias in groundwater quality monitoring (Elci et al., 2003;

McMillan et al., 2014).

Figure 3 – Modelled (red lines) and observed (open circles) hydraulic head in the injection and

pumping wells. Drawn are also error bars indicating the instrument accuracy.

Figure 4 – DD modelled (red lines) and observed (open circles) concentration and temperature data

in the injection and pumping wells. The best fit ADE model (grey lines) are also plotted for a direct

comparison with the DD, note that for the injection well the ADE and DD coincide. For the

pumping well the Br- concentrations measured via IC (blue crosses) are also plotted after

conversion into NaBr concentrations. Drawn are also error bars indicating the instrument accuracy.

Bias between modelled and observed temperature values were reduced by increasing temporal

discretization; in fact, increasing the stress periods from 4 to 8 to better discretize the heat injection

(ΔT 7°C) resulted in an improved match between modelled and observed values. The temperature

monitoring in the injection well was fundamental to reproduce the effective temperature decrease

experienced within the aquifer, due to heat loss along the well casing and pumping equipment. In

fact, without the injection monitoring of the boundary condition applied would be a constant

temperature of ΔT 7°C for the 17 minutes of injection, which would for sure impair the model

results in the recovery well.

The temperature plume, created by the forced gradient test, rapidly mixes with the ambient

groundwater, producing a maximum temperature variation of just 0.3 °C after 3.3 hours in the

pumping well. The saline tracer plume develops only within the highly permeable lens (Figure 5)

creating a self-sharpening plume with a very limited vertical and horizontal spreading.

Figure 5 - Results of the 3D model after two hours from the injection. To be noted: a self-

sharpening Br- plume developing within the high permeability lens, and the drawdown induced by

the pumping.

The correct position of the saline tracer breakthrough peak was obtained via trial and error, by

fitting θm and θim, αL and ω (Table 2). All the attempts to calibrate the breakthrough curve of Br- in

the pumping well with the ADE approach failed. The obtained EF for Br- was very poor (0.16)

giving a complete mismatch between observed and calculated concentrations. Instead, using the DD

approach permitted to decrease the simulated peak time and replicate the early breakthrough, related

to accelerated transport via preferential pathways, and the curve tail, related to diffusion driven

processes into stagnant zones. It has to be noted that, even using a θm value of 0.15, that is quite

small for sandy lenses, the peak concentration was reached after 3.8 hours from the injection, while

the observed peak occurred at 2.8 hours. Even try an unrealistic θm value of 0.1 with the ADE, did

not produced reliable results (Figure 4).

For this forced gradient test, a αL of 7.4 cm has been obtained (Table 5); this value is higher than the

αL value of 0.59 cm obtained through column displacement experiments carried out by Mastrocicco

et al. (2011a). In this latter study, column displacement experiments were performed with repacked

alluvial aquifer materials collected from this test site. Although the sediments come from the same

alluvial deposit, their repackaging during the column assemblage and the increased flow path from

1 m of the column to 2.5 m of this field test, both contributed to increase the field αL value (Bromly

et al. 2007). The study allowed quantifying the bias between the aquifer parameters estimated via

model-based interpretation of EC measurements of six selected saline tracers (LiCl, KCl, NaCl,

LiBr, KBr and NaBr) versus their respective anions (Cl- and Br

-).

Table 5 – Transport model parameters obtained in this study compared with previous studies in the

same area.

αL

(cm) ω

(1/d) θim

(-) θm

(-)

Forced gradient field test (This study) 7.4 0.346 0.10 0.15

Column experiment (Mastrocicco et al. 2011a) 0.59 0.297 0.04 0.36

Natural gradient field test (Mastrocicco et al. 2011b) 53 - - 0.31

On the contrary, αL value obtained during the forced gradient test is lower than the αL value of 53

cm achieved through a natural gradient test conducted in the same test site (Mastrocicco et al.,

2011b). In Mastrocicco et al. (2011b) a natural gradient test was performed in the same field site

and a transient groundwater flow and contaminant transport model was built; Cl- was used as

environmental tracer to quantify groundwater velocity, while NO3- was treated as reactive species to

identify the mechanism of NO3- attenuation. The small αL value obtain through this forced gradient

test compared to the natural gradient test is due to the high flow field, with an average velocity of

20 m/d, which is more similar to the column experiment rather than the natural gradient test. In fact,

the ambient groundwater flow velocities are in the order of 0.1 m/d, while the column experiment

was run at an average velocity of 4.7 m/d. In the natural gradient test there was no need to use a

dual domain approach, while in the forced gradient test the best results were obtained using an ω

value of 0.346 1/d, which is quite similar to the ω value of 0.297 1/d obtained by Mastrocicco et al.

(2011a). Nevertheless, results of the sensitivity analysis on ω show that for a parameter perturbation

of 100%, quite different breakthrough curves can be obtained. The ω value found in this experiment

is quite different from 1.37 10-2

1/d obtained by Bianchi et al. (2011) for a dipole tracer test

conducted in an extremely heterogeneous aquifer. An even lower ω value of 3 10-3

1/d was recently

found in a large-scale tracer experiment in a sub-irrigated buffer zone (Mastrocicco et al., 2014).

This highlight the need of site-specific estimation of ω values for different hydrological conditions,

like for example different flow velocities as pointed out by (Jørgensen et al. 2004). The heat

breakthrough was not well reproduced using the same ω value of the saline tracer, while a ω value

approaching to zero was necessary to fit the data. This implies that the heat has travelled only

through the active pore space, while in the stagnant zones the heat exchange is negligible. It must be

pointed out that the heat travels through both fluid filled pores and the sediment matrix, and the

storage and release from the aquifer matrix is the reason for the retardation of heat transport with

respect to solute. Heat resulted to be a useful tracer only in the presence of high groundwater

velocity (in this case the forced gradient test was performed in a sandy aquifer) and small cation

exchange capacity. In fact, it has been demonstrated that in fine-grained sediments characterized by

low groundwater velocities and thermal diffusion dominated, it is difficult to precisely distinguish

and quantify diffusivity and dispersivity components without also considering solute tracer tests

(Rau et al., 2012a, b). In fine graded alluvial sediments, the use of heat as groundwater tracer to

define aquifer properties can be problematic, since heat transport is relatively insensitive to the

longitudinal dispersivity, which is a relevant parameter for solute transport modelling (Giambastiani

et al., 2013).

While the only use of EC as a tracer can lead to an erroneous parameterization in fine-grained

sediment (Mastrocicco et al., 2011a), NaBr tracer turned to be reliable as a saline tracer because of

the interactions between dissolved Na+ and the soil matrix are limited due to the high groundwater

flow velocity imposed. In case of slow velocity flow fields, cation exchange reactions can occur

changing the EC of the tracer solution and causing tailing effects in the breakthrough curves. Thus,

the use of just one of the two employed tracers could lead to non-unique solutions or to biased

parameterization of the aquifer properties.

Another important consideration has to be discussed about the simulation code used for the

simulations. With the limited temperature gradient recorded during the test (7°C in the injection

well and 0.3°C in the pumping well), the effects of density and viscosity (explicitly considered in

SEAWAT) may be neglected for greater computational efficiency without any significant loss of

accuracy. As demonstrated by Ma and Zheng (2010), MT3DMS can provide accurate

approximation of heat transport under the assumption of constant fluid density and viscosity, when

the maximum temperature difference across the simulation domain is below 15°C, thus being a

reasonable compromise between accuracy and efficiency. In this case, the simulations by SEAWAT

code take 90 minutes to run, while simulations with identical model setting but performed by

MT3DMS are significantly more efficient, taking only 15 minutes, decreasing the computation time

of 84%. For the modelled tracer test, using MT3DMS instead of SEAWAT produced the same

MAE and EF values for heads, Br- and temperature reported in Table 3. Despite of this, for each

case it is recommendable to test the validity of the above assumption by running a full density

driven simulation with the site specific boundary conditions and stresses before switching to

simplified numerical models.

CONCLUSIONS

The combined use of continuous heads and solute monitoring in both the injection and recovery

wells to model a forced gradient test is a fast and low-cost technique to characterize coarse grain

aquifer properties. The use of heat in the validation phase of the numerical model was proven to be

robust and affordable. The test procedure described in this paper can give a fast and inexpensive

framework able to identify transport parameters responsible for preferential pathways in

sedimentary aquifers. Although some limitation can be highlighted, such as the different values of

the mass transfer coefficient between mobile and immobile porosity gained by heat and saline

tracer, or cation exchange reactions between the saline tracer and the aquifer matrix. This procedure

can be efficiently used in sandy aquifer, with injection and pumping wells not far away from each

other and using limited temperature gradient that permits to neglect the effect of density and

viscosity on groundwater flow. The numerical model could be reasonably fitted only using a dual

domain approach, while the advection dispersion equation approach was unsuitable to describe the

breakthrough of Br- in the pumping well. The comparison of the obtained transport parameters

gained for the same aquifer materials or similar experimental settings suggest the site specificity of

these parameters, which should be estimated in the field.

Nevertheless, this framework could be adopted to characterize the preferential pathways near

pumping wells fields where flow velocities are artificially increased, e.g. in pump and treat

remediation systems or in riverbank filtration facilities for drinking water supply.

REFERENCES Anderson, M.P., 2005. Heat as a ground water tracer. Ground Water 43(6), 951-968.

Bianchi, M., Zheng, C., Tick, G. R., Gorelick, S.M., 2011. Investigation of small‐scale preferential flow with a

forced‐gradient tracer test. Ground Water 49(4), 503-514.

Bromly, M., Hinz, C., Aylmore, L.A.G., 2007. Relation of dispersivity to properties of homogeneous saturated repacked

soil columns. Eur. J. Soil Sci. 58(1), 293-30.

Cassiani, G., Bruno, V., Villa, A., Fusi, N., Binley, A.M., 2006. A saline trace test monitored via time-lapse surface

electrical resistivity tomography. J. Appl. Geophys. 59, 244-259. Chen, J.S., Chen, Ch.S., Gau, H.S., Liu, Ch.W., 1999. A two well method to evaluate transverse dispersivity for tracer

tests in a radially convergent flow field. J. Hydrol. 223, 175-197.

Cirpka, O.A., Fienen, M.N., Hofer, M., Hoehn, E., Tessarini, A., Kipfer, R., Kitanidis, P.K., 2007. Analyzing bank

filtration by deconvoluting time series of electric conductivity. Ground Water 45(3), 318-328.

Conant, B.J., 2004. Delineating and quantifying ground water discharge zones using streambed temperature. Ground

Water 42(2), 243-257.

Doro, K.O., Cirpka, O.A., Leven, C., 2014. Tracer Tomography: Design Concepts and Field Experiments Using Heat as

a Tracer. Groundwater 53, 139-148.

Elci, A., Flach, G.P., Molz, F.J., 2003. Detrimental effects of natural vertical head gradients on chemical and water

level measurements in observation wells: identification and control. J. Hydrol. 28, 70-81.

Ferguson, G., Woodbury, A.D., Matile, G.L.D., 2003. Estimating deep recharge rates beneath an interlobate moraine

using temperature logs. Ground Water 41(5), 640-646.

Giambastiani, B.M.S., Colombani, N., Mastrocicco, M., 2013. Limitation of using heat as a groundwater tracer to define aquifer properties: experiment in a large-tank model. Environ. Earth Sci. 70(2), 719-728.

Harbaugh, A.W., 2005. MODFLOW-2005, the U.S. Geological Survey Modular Ground-water Model - The Ground-

water Flow Process: U.S.G.S. Techniques and Methods 6-A16.

Hecht-Méndez, J., Molina-Giraldo, N., Blum, P., Bayer, P., 2010. Evaluating MT3DMS for heat transport simulation

of closed geothermal systems. Ground Water 48(5), 741-756.

Hermans, T., Wildemeersch, S., Jamin, P., Orban, P., Brouyere, S., Dassargues, A., Nguyen, F., 2015. Quantitative

temperature monitoring of a heat tracing experiment using cross-borehole ERT. Geothermics 53, 14-26.

Irvine, D.J., Simmons, C.T., Werner, A.D. and Graf, T., 2013. Heat and solute tracers: how do They compare in

heterogeneous aquifers? Ground Water doi: 10.1111/gwat.12146

Jørgensen, P. R., Helstrup, T., Urup, J., & Seifert, D. (2004). Modeling of non-reactive solute transport in fractured

clayey till during variable flow rate and time. J. Cont. Hydrol. 68(3), 193-216. Konikow, L.F. and Bredehoeft, J.D., 1978. Computer model of two-dimensional solute transport and dispersion in

ground water. U. S. Geological Survey. Water Resources Investigation. Book 7, Chapter C2, 90 pp.

Kung, K.J., Steenhuis, T.S., Kladivko, E.J., Gish, T.J., Bubenzer, G., Helling, C.S., 2000. Impact of preferential flow on

the transport of adsorbing and non-adsorbing tracers. Soil Sci. Soc. Am. J. 64(4), 1290-1296.

Langevin, C.D., Thorne, D.Jr., Dausman, A.M., Sukop, M.C., Guo, W., 2008. SEAWAT Version 4: A computer

program for simulation of multi-species solute and heat transport. Techniques and Methods Book 6, Chap A22,

USGS.

Leibundgut, C., Seibert, J., 2013. 2.09 – Tracer hydrology. In: Earth Systems and Environmental Sciences. Treatise on

Water Science, 215-236.

Ma, R., Zheng, C., 2010. Effects of density and viscosity in modelling heat as a groundwater tracer. Ground Water

48(3), 380-389.

Ma, R., Zheng, C., Zachara, J.M. and Tonkin, M., 2012. Utility of bromide and heat tracers for aquifer characterization affected by highly transient flow conditions. Water Resour. Res. 48, W08523, doi: 10.1029/2011WR011281

Mastrocicco, M., Vignoli, G., Colombani, N., Abu Zeid, N., 2010. Surface electrical resistivity tomography and

hydrogeological characterization to constrain groundwater flow modelling in an agricultural field site near Ferrara

(Italy). Environ. Earth Sci. 61(2), 311-322.

Mastrocicco, M., Prommer, H., Pasti, L., Palpacelli, S., Colombani, N. 2011a. Evaluation of saline tracer performance

during in situ electrical conductivity monitoring. J. Contam. Hydrol. 123(3-4), 157-166.

Mastrocicco, M., Colombani, N., Castaldelli, G., Jovanovic, N., 2011b. Monitoring and modeling nitrate persistence in

a shallow aquifer. Water Air Soil Poll. 217(1-4), 83-93.

Mastrocicco, M., Boz, B., Colombani, N., Carrer, G. M., Bonato, M., Gumiero, B., 2014. Modelling groundwater

residence time in a sub-irrigated buffer zone. Ecohydrol. 7, 1054-1063.

McMillan, L.A., Rivett M.O., Tellam J.H., Dumble P., Sharp H., 2014. Influence of vertical flows in wells on groundwater sampling. J. Contam. Hydrol. DOI: 10.1016/j.jconhyd.2014.05.005.

Molina-Giraldo, N., Bayer, P., Blum, P., Cirpka, O.A., 2011. Propagation of seasonal temperature signals into an

aquifer upon bank infiltration. Ground Water 49(4), 491-502.

Nash, J.E., Sutcliffe J.V., 1970. River flow forecasting through conceptual models part I — A discussion of principles.

J. Hydrol. 10(3), 282-290.

Palmer, D.P., Blowes, D.W., Frind, E.O., Molson, J.W., 1992. Thermal energy storage in an unconfined aquifer.1. Field

injection experiment. Water Resour. Res. 28(10), 2845-2856.

Parsons, D.F., Hayashi, M., van der Kamp, G. 2004. Infiltration and solute transport under a seasonal wetland: bromide

tracer experiments in Saskatoon, Canada. Hydrol. Process. 18, 2011-2027.

Perri, M.T., Cassiani, G., Gervasio, I., Deiana, R., Binley, A., 2011. A saline tracer test monitored via both surface and

cross-borehole electrical resistivity tomography: Comparison of time-lapse results. J. Appl. Geophys. 79, 6-16.

Ptak, T., Piepenbrink,M., Martac, E., 2004. Tracer tests for the investigation of heterogeneous porous media and stochastic modelling of flow and transport — a review of some recent developments. J. Hydrol. 294, 122-163.

Racz, A.J., Fisher, A.T., Schmidt, C.M., Lockwood, B.S., Los Huertos, M., 2011. Spatial and temporal infiltration

dynamics during managed aquifer recharge. Ground Water 50(4), 562-570.

Rau, G.C., Andersen, M.S., Acworth, R.I., 2012a. Experimental investigation of the thermal dispersivity term and its

significance in the heat transport equation for flow in sediments. Water Resour. Res. 48(3),

DOI:10.1029/2011WR011038.

Rau, G.C., Andersen, M.S., Acworth, R.I., 2012b. Experimental investigation of the thermal time-series method for

surface water-groundwater interactions. Water Resour. Res. 48(3), DOI:10.1029/2011WR011560.

Rau, G.C., Andersen, M.S., McCallum, A.M., Roshan, H., Acworth, R.I., 2014. Heat as a tracer to quantify water flow

in near-surface sediments. Earth-Sci. Rev. 129, 40-58.

Read, T., Bour, O., Bense, V., Le Borgne, T., Goderniaux, P., Klepikova, M.V., Hochreutener, R., Lavenant, N.,

Boschero, V., 2013. Characterizing groundwater flow and heat transport in fractured rock using fiber‐optic

distributed temperature sensing. Geophys. Res. Lett. 40(10), 2055-2059. Reusser, D.E., Blume, T., Schaefli, B., Zehe, E., 2009. Analysing the temporal dynamics of model performance for

hydrological models. Hydrol. Earth Syst. Sci. 13, 999-1018.

Richardson, S.D., Willson, C.S., Rusch, K.A., 2004. Use of rhodamine water tracer in the marshland upwelling system.

Ground Water 42(5), 678-688.

Sambale, C.H., Peschke, G., Uhlenbrook, S., Leibundgut, C.H., Markert, B., 2000. Simulation of vertical water flow

and bromide transport under temporarily very dry conditions. IAHS Publ. 262, 339-345.

Sethi, R., Molfetta, A.D., 2007. Heat transport modelling in an aquifer downgradient a municipal solid waste landfill in

Italy. Am. J. Env. Sci. 3, 106-110.

Silliman, S.E., Booth, D.F., 1993. Analysis of time-series measurements of sediment temperature for identification of

gaining vs. losing portion of Juday Creek, Indiana. J. Hydrol. 146, 131-148.

Sudicky, E.A., 1986. A natural gradient experiment on solute transport in a sand aquifer: spatial variability of hydraulic

conductivity and its role in the dispersion process. Water Resour. Res. 22, 2069-2082. Sutton, D.J., Kabala, Z.J., Schaad, D.E., Ruud, N.C., 2000. The dipole-flow test with a tracer: a new single-borehole

tracer test for aquifer characterization. J. Contam. Hydrol. 44, 71-101.

Thorne, D., Langevin, C.D., Sukop, M.C., 2006. Addition of simultaneous heat and solute transport and variable fluid

viscosity to SEAWAT. Comput. Geosci. 32, 1758-1768.

Vandenbohede, A., Hermans, T., Nguyen, F., Lebbe, L., 2011. Shallow heat injection and storage experiment: heat

transport simulation and sensitivity analysis. J. Hydrol. 409, 262-272.

Vandenbohede, A., Louwyck, A., Lebbe, L., 2009. Conservative solute versus heat transport in porous media during

push-pull tests. Transport Porous Med. 76, 265-287.

Vogt, T., Hoehn, E., Schneider, P., Freund, A., Schirmer, M., Cirpka, O.A., 2010. Fluctuations of Electrical

Conductivity as a Natural Tracer for Bank Filtration in a Losing Stream. Adv. Water Res. 33(11), 1296-1308.

Wildemeersch, S., Jamin, P., Orban, P., Hermans, T., Klepikova, M., Nguyen, F., Brouyère, S., Dassargues, A., 2014. Coupling heat and chemical tracer experiments for estimating heat transfer parameters in shallow alluvial aquifers.

J. Contam. Hydrol. DOI: 10.1016/j.jconhyd.2014.08.001.

Zheng, C., Bennett, G.D., 2002. Applied Contaminant Transport Modeling, 2nd Edition. John Wiley & Sons, New

York. 621 pp.

Zheng, C., Wang, P.P., 1999. MT3DMS, A modular three-dimensional multispecies model for simulation of advection,

dispersion and chemical reactions of contaminants in groundwater systems; Documentation and User’s guide. US

Army Engineer Research and Development Center Contract Report SERDP-99-1. USAERDC, Vicksburg, MS.