A soil non-aqueous phase liquid ( NAPL ) flushing laboratory 1 experiment based on time domain reflectometry ( TDR ) and 2 modeling 3

The term non-aqueous phase liquid (NAPL) refers to a group of organic compounds with scarce solubility in 9 water. They are the products of various human activities and may be accidentally introduced into the soil system. Given 10 their toxicity level and high mobility, NAPLs constitute a serious geo-environmental problem. Contaminant distribution 11 in the soil and groundwater entails fundamental information for the remediation of polluted soil sites. The present research 12 explored the possible employment of time domain reflectometry (TDR) to estimate pollutant removal in a silt-loam soil 13 that was primarily contaminated with a light hydrocarbon and then flushed with diverse washing solutions. Known 14 mixtures of soil and NAPL were prepared in the laboratory to achieve soil specimens with diverse pollution levels. The 15 prepared soil samples were repacked into plastic cylinders and then placed in testing cells. Washing solutions were then 16 injected upward into the contaminated sample, and both the quantity of remediated oil and the bulk dielectric permittivity 17 of the soil sample were determined. The above data was also used to develop a dielectric model (the  mixing model) 18 which permits the volumetric NAPL content (θNAPL) within the contaminated sample to be determined and quantified 19 during the different decontamination stages. Our results demonstrate that during a decontamination process, the TDR 20 device is NAPL-sensitive: the dielectric permittivity of the medium increases as the NAPL volume decreases. Moreover, 21 decontamination progression can be monitored using a simple (one-parameter) mixing model. 22


Introduction
Soil and groundwater contamination with NAPL from point or nonpoint sources is a severe problem of considerable complexity (Fitts, 2002;Fetter, 1993).The repercussions concern not only the deterioration of the soil's physical, mechanical and chemical properties, but also account for a potentially severe hazard to the well-being of humans and other living species (Freeze, 2000).
Soil flushing is the technical procedure used for treating polluted soils with water, surfactants and co-solvents (such as methanol, ethanol and propanols).Surfactant-enhanced flushing was developed from the conventional pump-and-treat Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.method.The success of this approach is related to the capacity of such chemical compounds to greatly enhance the aqueous solubility of oils (Pennell et al., 1994;Parnian and Ayatollahi, 2008).
There is high interfacial tension between NAPL and water molecules that makes water a non-efficient cleaning material in removing NAPL from the soil.Instead, surfactants and co-solvent agents can promote the enhanced removal of NAPL from the subsurface through mobilization and solubilization (Martel et al., 1998;Rinaldi and Francisca, 2006;Parnian and Ayatollahi, 2008).
Primary remediation refers to the removal of the NAPL free phase by pumping.This extraction mechanism returns appreciable effects if there is a region of high NAPL saturation.After primary pumping, a considerable portion of NAPL remains constrained within the soil as capillary forces overcome viscous and buoyancy forces.This discontinuous NAPL phase is referred to as trapped residual NAPL (or NAPL residual saturation), and its remediation is referred to as secondary remediation (Parnian and Ayatollahi, 2008).Residual NAPL is a long-term source of soil and groundwater pollution (Mercer and Cohen, 1990;Troung Hong and Bettahar, 2000).
To develop powerful decontamination procedures, the characterization of polluted soils is required.Practices usually employed to characterize polluted soil sites are coring, soil sampling and the installation of monitoring wells for the collection of water samples from aquifers (Mercer and Cohen, 1990).Since the aforementioned procedures are costly, the time domain reflectometry (TDR) technique is suggested as a valid technical alternative since it exhibits adequate sensitivity for the characterization of NAPL saturation (Persson and Berndtsson, 2002;Mohamed and Said, 2005;Moroizumi and Sasaki, 2006;Francisca and Montoro, 2012;Comegna et al., 2013a;Comegna et al., 2016).
The purpose of this study was to the following: i) investigate a possible extension of TDR technology to assess the effects of NAPL removal in soils, and ii) build, on the basis of the acquired data and the experimental results, a dielectric model to predict the volumetric amounts of NAPL (θNAPL) within the contaminated soil during the decontamination process.

Theoretical concepts of TDR
TDR is a geophysical technique employed to determine the dielectric permittivity of liquids and solids.In general, the bulk dielectric permittivity is a complex term ( * ), which may be expressed as follows (Robinson et al., 2003): where r '  is the real part of dielectric permittivity, which gives the energy stored in the dielectrics at a certain frequency and temperature, and ) of an electromagnetic wave along the wave guide across the investigated medium by the following expression: where c (= 8 10 3 m s -1 ) is the velocity of an electromagnetic wave in vacuum (Topp et al., 1980) and t is travel time, i.e. the time required by the generated signal to go back and forth through the TDR probe of length L (m).This can be calculated as the following: The direct dependence of the signal's travel time t upon soil dielectric permittivity is expressed by equation 3.

Estimating volumetric NAPL content during a decontamination process in soils
Dielectric mixing models, in their classical application, have been proposed to estimate the bulk dielectric permittivity of a multi-phase medium, that is, a combination of three or four dielectric phases, and to couple the dielectric permittivity of the medium to the dielectric permittivity of each single phase (Hilhorst, 1998).Recently, after analyzing the effects of organic contaminants on soil dielectric properties, the above models were further developed to estimate the dielectric properties of NAPL-polluted soils (Persson and Berndtsson, 2002;Francisca andMontoro, 2012, Comegna et al., 2013a;Comegna et al., 2016;Comegna et al., 2017).
Based on such models, in the present study, we analyze the possibility of predicting the correlations between the volumetric contents of NAPL (θNAPL) and the dielectric response (εb) of contaminated soil during the progression of a steady-state remediation process.
In the present research, we chose the so-called  model (Birchack et al., 1974;Roth et al., 1990): where i V is the volume and εi is the permittivity of each component of the complex medium; the exponent  is a fitting parameter ( varies between -1 and 1), which may be related to the internal structure of the investigated medium (Hilhorst, 1998;Coppola et al., 2013;Coppola et al., 2015).Under the following hypothesis: i) the soil is homogeneous from a textural point of view, and ii) the soil porosity () is constant, equation 4 was reformulated for our purposes.For mixtures of soil (s) saturated with a certain amount of washing solution (ws), in rearranging the model formulation of Rinaldi and Francisca (2006), the  model yields the following: where εs-ws is the soil-washing solution permittivity, and εs and εws are the permittivities of soil particles and washing solutions, respectively.By the same token, for soil organic (s-NAPL) compounds at saturation, the  model can be expressed as the following: where εs-NAPL is the permittivity of the soil-NAPL mixture, and εNAPL is the oil permittivity.
A medium consisting of soil particles, washing solution and NAPL (s-ws-NAPL) can be viewed as a mix of soil-washing solution (equation 5) and soil-NAPL (equation 6): where  is the relative volume of NAPL contained in the whole fluid phase: where θf is the volumetric fluid content (cm 3 /cm 3 ), sum of the volumetric washing solution content ( ws  ) and volumetric NAPL content ( NAPL
To estimate NAPL  , equation 7 is first reformulated in terms of : Substituting equation 8 into equation 9, and considering that for a saturated medium, the volumetric fluid content is equal to soil porosity (i.e.θf =), θNAPL can be calculated as the following: A silt-loam Anthrosol (IUSS Working Group WRB, 2006) from the region of Puglia (Italy) was used for this study.The soil texture was measured by means of the hydrometer method (Day, 1965), while the Walkley-Black procedure (Allison, 1965) was used to determine soil organic C content.The method developed by Miller and Curtis (2007) was used to measure soil electrical conductivity (ECw), while soil pH was determined on the basis of a 1:1 soil/water ratio (Eckert, 1988).In textural terms, the soil comprised 15.7% sand, 11.6% clay and 72.4% silt.Soil porosity was 0.57%, organic content 1.84%, ECw 0.17 dS/m and soil pH 8.40.

Experimental setup
As illustrated in Figure 1, the experimental layout consisted of the following: i) a Techtronix (model 1502C) cable tester; ii) a three-wire TDR probe 14.5 cm long, introduced vertically into the soil samples; iii) a testing cell 15 cm high and 8 cm in diameter; iv) a peristaltic pump used for upward movement of the washing solution.

Sample preparation and testing procedures
After oven-drying at 105°C and sieving at 2 mm, the soil was mixed with oil in known quantities and then placed for 24 hours in plastic bags to prevent evaporation and ensure a complete distribution of oil in the soil.The samples were then allocated to cylindrical boxes.With a view to achieve different degrees of soil contamination, volumetric NAPL content (NAPL) was varied from 0.05 to 0.40 (in steps of 0.05).In all, each washing solution comprised eight oil-contaminated soil samples.
For all experiments, the soil samples were placed in the vessels in various steps at a bulk density of 1.13 g/cm 3 .During TDR measurements, the soil samples were conserved at a temperature of 25°C by using a thermostat box.Remediation was performed using an upward flux of diverse pore volumes T of three washing solutions (wd, wda#1 and wda#2) supplied at the rate of 90 cm 3 /h, corresponding to a Darcian velocity of 1.8 cm/h.After collection of the outflow from the The obtained data series were employed to calibrate the proposed dielectric model of equation 10.

Numerical indices for model performance evaluation
The goodness of equation 10 was evaluated using two different criteria: i) the mean bias error (MBE), and ii) the model efficiency (EF), computed according to the following relations (Legates and McCabe Jr, 1999): where i E and i O are respectively the expected and the observed value, O is the mean of the observed data, and N is the number of observations.
MBE measures the differences between model-simulated data and measured values (positive MBE values are used to indicate average overprediction, while negative values indicate underprediction).The model's ability to forecast θNAPL is described by parameter EF, according to which EF=1 indicates perfect accord between predicted and measured data.) for the soil specimens initially polluted with oil.As the washing solution started to remove oil, the dielectric permittivity rose due to the larger dielectric permittivity of the flushing mixture.As the remediation solution continued to move upward, the rising rate of the dielectric permittivity decreased and asymptotically approached a constant value.This steady value was smaller than that observed when the soil specimens were completely saturated by only the flushing solution (i.e.wd, wda#1 or wda#2), which in our tests corresponds to the condition of a completely decontaminated soil.This difference in values is undoubtedly due to oil confined in soil pores (i.e.NAPL residual saturation).For the same reason, residual saturation may explain why insignificant oil remediation was observed for NAPL  values less than 0.15.This aspect may be explained by the fact that for low volumetric NAPL contents, the non-wetting fluid (oil) is disconnectedly distributed (i.e.immobile) in the soil samples, which means that NAPL  is close to the limiting residual value, and thus NAPL loses its ability to move in the soil in response to a hydraulic gradient [i.e.capillary retention forces are greater than gravitational forces, which tend to immobilize the NAPL (Brost and DeVaull, 2000)].

Influence of NAPL removal on bulk dielectric permittivity
Figure 3 showed, for different washing solutions, the percentages of NAPL volumes recovered (VNAPL-Rem) with respect to the initial volume of NAPL present in the soil sample (V0).For all the three cleaning solutions adopted, the experiments ultimately demonstrate (for a fixed θNAPL) the same results in terms of soil decontamination, and they show that NAPL removal increases with increasing θNAPL.In some cases (i.e.θNAPL=0.15,0.20 and 0.30), contaminated samples flushed with the wda#1 solution yield slightly higher removal efficiency values compared to the samples flushed with wd and wda#2.Martel et al. (1998) suggest the need to investigate the best water-surfactant-alcohol combination in order to enhance NAPL solubilization in soil.

Model calibration and validation
For the model (equation 10) calibration methodology, with reference to the three washing solutions (wd, wda#1 and wda#2), we analyze the effect of the measured dielectric permittivity on volumetric NAPL content (θNAPL) in order to estimate the  parameter of the model.The complete data set of estimated  parameters is reported in Table 1.
A permittivity value of 3.70 was adopted for the solid phase.This value was determined using the "immersion method" which is commonly employed for estimating the εs of soils (Robinson et al., 2003;Kameyama and Miyamoto, 2008;Comegna et al., 2013a;Coppola et al., 2013).
For the sake of brevity, a selection of the experimental εs-ws-NAPL-θNAPL relationships (validation dataset) is reported in figures 4a, b, c, d, e and f.The data in figures 4 (except for figures 4e, f) show that some of the model-simulated values tend to overestimate the measured data.This behavior is mostly restricted to the beginning of the remediation process, when a rapid change in dielectric permittivity may be observed.This behavior was also verified in other tests (not shown here) and may be explained by invoking both NAPL properties such as liquid density, surface tension and viscosity, and soil properties including moisture content, relative permeability, soil heterogeneity and porosity (Brost and DeVaull, 2000;Wang et al., 2013).Mercer and Cohen (1990) referred to the existence, in NAPL-contaminated soils, of a "double fluid domain," defined as the composition of the following: i) mobile pools, which are NAPL-connected phases that move in the soil and ii) immobile residuals (i.e.low permeability regions), which depend on small disconnected blobs or ganglia within the contaminated soil (see also section 5.1 above).As long as the flushing continues, mobile pools are reduced and the oil tends increasingly to be trapped in the immobile areas.This means that, during soil cleaning, the capacity of non-wetting fluids to respond to gravitational forces gradually diminishes (Luckner et al., 1989).From a dielectric point of view, this mechanism may appear as a rapid dielectric permittivity increase (identified in figures 4 as fast oil mobility region) within a few pore volumes.When this fast mobility mechanism is dominant, the predictions of equation 10 fail.However, since the phenomenon is mostly limited to the initial part of the washing process, overall model effectiveness is not compromised, as also shown in Table 2, which summarizes the goodness-of-fit statistical indices.
Overall, both graphical and quantitative evaluations in terms of MBE and EF reveal the suitability of the dielectric model adopted to estimate the volumetric NAPL content in the NAPL range 0.15-0.40.

Conclusions
This paper presented an extensive dataset of remediation experiments that were conducted at a laboratory scale using corn oil as a soil contaminant, and three different solutions for soil cleaning.The results of these tests were employed to investigate the potential of the TDR technique in monitoring the development of a steady-state decontamination process.
Dielectric data analysis showed that, during soil flushing, dielectric permittivity behavior is highly dependent on the initial volumetric content and intrinsic permittivity of the specific NAPL: removal of NAPL produces an increase in bulk     Tables Table 1.Estimated  parameter of equation 10 for all three washing solutions (wd, wda#1 and wda#2) and volumetric NAPL content (θNAPL) tested.


is the imaginary part due to relaxations.The zero frequency conductivity , the angle frequency ω, the imaginary number 1   j and the permittivity ε0 in free space contribute to define  * .Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.When the frequency of a TDR cable tester ranges between 200 MHz to 1.5 GHz, dielectric losses can be considered minimal and the bulk dielectric permittivity εb (  the real part of permittivity) of a probe of length L is determined from the propagation velocity v(= t L 2 Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.

)4
Equation 10 correlates the dependence of volumetric NAPL content with soil porosity; θNAPL can be estimated (within the contaminated soil) during the progression of a remediation process once the dielectric permittivity of the soil-Materials and Methods4.1 Soil and fluid propertiesHydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.
Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.soil columns, the oil was separated from the washing solution and the quantity of oil remediated from the soil was determined.

Figures
Figures 2a, b, c, d, e and f, with reference to the most representative experimental results, reveal the influence of pore volumes T on evaluated bulk dielectric permittivity ( NAPL ws s    Hydrol.Earth Syst.Sci.Discuss., https://doi.org/10.5194/hess-2019-149Manuscript under review for journal Hydrol.Earth Syst.Sci. Discussion started: 29 April 2019 c Author(s) 2019.CC BY 4.0 License.
dielectric permittivity, due to the low value of oil permittivity.The experiments conducted also allowed us to calibrate and validate a dielectric mixing model (equation 10).The model outcomes are encouraging; the calculated statistical indices confirmed a high accuracy in NAPL predictions of the -model at different stages during soil cleaning, with the only exception of the very initial cleaning stage (confined to the low values of T) where the eventual presence of a fast flow region may limit its applicability.