Journal of Contaminant Hydrology 161 (2014) 24–34

Contents lists available at ScienceDirect

Journal of Contaminant Hydrology journal homepage: www.elsevier.com/locate/jconhyd

The transport behaviour of elemental mercury DNAPL in saturated porous media: Analysis of field observations and two-phase flow modelling Thomas Sweijen a,b,⁎, Niels Hartog b,c, Annemieke Marsman a, Thomas J.S. Keijzer d a b c d

Deltares, Princetonlaan 6, 3584 CB Utrecht, The Netherlands Utrecht University, Department of Earth Sciences, Environmental Hydrogeology Group, Budapestlaan 4, 3584 CD Utrecht, The Netherlands KWR Watercycle Research Institute, Nieuwegein, The Netherlands Philips Innovation Services, Eindhoven, The Netherlands

a r t i c l e

i n f o

Article history: Received 15 November 2013 Received in revised form 23 March 2014 Accepted 24 March 2014 Available online 30 March 2014 Keywords: Mercury PCE DNAPL Multi-phase flow Camera push-probe STOMP

a b s t r a c t Mercury is a contaminant of global concern. The use of elemental mercury in various (former) industrial processes, such as chlorine production at chlor-alkali plants, is known to have resulted in soil and groundwater contaminations worldwide. However, the subsurface transport behaviour of elemental mercury as an immiscible dense non-aqueous phase liquid (DNAPL) in porous media has received minimal attention to date. Even though, such insight would aid in the remediation effort of mercury contaminated sites. Therefore, in this study a detailed field characterization of elemental mercury DNAPL distribution with depth was performed together with two-phase flow modelling, using STOMP. This is to evaluate the dynamics of mercury DNAPL migration and the controls on its distribution in saturated porous media. Using a CPT-probe mounted with a digital camera, in-situ mercury DNAPL depth distribution was obtained at a former chlor-alkali-plant, down to 9 m below ground surface. Images revealing the presence of silvery mercury DNAPL droplets were used to quantify its distribution, characteristics and saturation, using an image analysis method. These field-observations with depth were compared with results from a one-dimensional two-phase flow model simulation for the same transect. Considering the limitations of this approach, simulations reasonably reflected the variability and range of the mercury DNAPL distribution. To further explore the impact of mercury's physical properties in comparison with more common DNAPLs, the migration of mercury and PCE DNAPL in several typical hydrological scenarios was simulated. Comparison of the simulations suggest that mercury's higher density is the overall controlling factor in controlling its penetration in saturated porous media, despite its higher resistance to flow due to its higher viscosity. Based on these results the hazard of spilled mercury DNAPL to cause deep contamination of groundwater systems seems larger than for any other DNAPL. © 2014 Elsevier B.V. All rights reserved.

1. Introduction The historical use of elemental and ionic forms of mercury in various industrial proce'sses has caused mercury to become ⁎ Corresponding author at: Utrecht University, Department of Earth Sciences, Environmental Hydrogeology Group, Budapestlaan 4, 3584 CD Utrecht, The Netherlands. Tel.: +31 302 532 497. E-mail address: [email protected] (T. Sweijen).

http://dx.doi.org/10.1016/j.jconhyd.2014.03.001 0169-7722/© 2014 Elsevier B.V. All rights reserved.

a contaminant of global concern. This is due to its long-range transport in the atmosphere, its persistence in the environment, its ability to bioaccumulate in ecosystems and its substantial negative effect on human health and the environment (Horvat et al., 2003; Ullrich et al., 2001; Walvoord et al., 2008; World Chlorine Council, 2011). Contaminated industrial sites include former wood-preservation and mining sites as well as (former) chlor-alkali plants (e.g. Arbestain et al., 2009; Bernaus et al.,

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

2006; Biester et al., 2002a, 2002b; Brooks and Southworth, 2011; Miller et al., 2013). To date, the scientific research on mercury contamination in wastewater, soil, and groundwater systems has focussed largely on the behaviour and ecotoxicological effects of ionic and methylated Hg forms (Gabriel and Williamson, 2004; Horvat et al., 2003; Jackson, 1998; Johannesson and Neumann, 2012; Lopes et al., 2013; Schuster, 1991). Mercury species have contaminated groundwater due to their use in industrial processes. For example through the use of HgCl2 in wood-preservation (Bollen et al., 2008). Other mercury species may result from the redox transformation of elemental mercury to mercury(II) compounds by oxidation or to methylmercury by subsequent methylation under reducing conditions. The extent to which mercury undergoes these redox transformations strongly depends on the soil and groundwater redox chemistry (Hu et al., 2013; Schuster, 1991). Under sufficiently reducing groundwater conditions, reduction of ionic mercury species may result in the formation of elemental mercury phase (Bollen et al., 2008). Nowadays, chlor-alkali plants are still responsible for the largest industrial use of elemental mercury, although its global use is rapidly decreasing (World Chlorine Council, 2011) and the European chlorine producers are phasing out its usage by 2020 (EuroChlor, 2010). At numerous (former) industrial sites, liquid elemental mercury in the subsurface has been observed, particularly at (former) chlor-alkali plants. For example, liquid mercury was found at a former chlor-alkali plant near Onodaga Lake, New York, at depths down to 17 m below surface (Deeb et al., 2011; ITRC, 2012). Also, liquid mercury was observed in soil samples, and wells in Lavaca Bay (Texas), surrounding a former chlor-alkali plant (Scanlon et al., 2005). Moreover, at the Oak Ridge Y-12 National Security Complex, 11 million kg mercury was used during 1950 to 1963 from which an estimated 193,000 kg was lost to the soil (Brooks and Southworth, 2011). At such sites exposure to mercury DNAPL during handling is a key difficulty (Deeb et al., 2011). Being the heaviest known liquid, with a density of 13.5 kg L− 1 at standard conditions and immiscible with water, liquid elemental mercury can be considered as a dense non-aqueous phase liquid (DNAPL). Analogue to better known DNAPLs, sites contaminated with mercury DNAPL are likely to act as long-lasting sources for more mobile and toxic mercury compounds, either as mercury vapour or ionic mercury species in groundwater. However, in contrast with more common DNAPL types (e.g. creosote, carbon tetrachloride, trichloroethylene (TCE) and perchloroethylene: PCE), scientific studies on the infiltration behaviour of mercury DNAPL into soils and aquifers are still very limited (Devasena and Nambi, 2010). Contrasting behaviour for mercury DNAPL is however expected as liquid mercury is not only an order of magnitude denser than more commonly studied DNAPLs, but also its viscosity and particularly its surface tension deviate (Table 1). To date however, it is unknown how these different properties affect the transport and distribution behaviour of elemental mercury infiltrating the subsurface. Improved insight in the infiltration behaviour of mercury DNAPL, and how it differs from more common DNAPLs, will aid the characterisation, riskassessments and evaluation of remediation options of mercury DNAPL contaminated sites. Only recently, the saturation behaviour of mercury DNAPL in water-saturated sands has been studied by using short

25

column experiments (Devasena and Nambi, 2010). As known for other DNAPLs, their results indicated that mercury DNAPL flow is governed by gravitational and capillary forces, hence its high density and surface tension, respectively. Moreover, they showed that for the water–sand system, mercury DNAPL can be considered as a non-wetting liquid because an entry-pressure was required for mercury to infiltrate the water-saturated sands. Following Cohen and Mercer (1990), this implies that water or air preferentially wets the surface of sand grains rather than mercury. In this respect, mercury acts as a non-wetting DNAPL similar to PCE (e.g. Schwille, 1988). However as shown in the study of Devasena and Nambi (2010), the distinctively different fluid properties may result in much lower residual saturations for mercury (0.04) than for TCE (0.14) and PCE (0.17). The recent pioneering study by Devasena and Nambi (2010) provides valuable insight in the characteristics of the two-phase mercury DNAPL–water system. However the dynamic aspects of mercury infiltration and distribution remain unclear, particularly at the field scale. Therefore, in this study we combined field characterisation data of mercury DNAPL contaminated soil at a former chlor-alkali site in The Netherlands with a multi-phase flow model. Goals of this study were 1) to assess the field scale characteristics of mercury DNAPL infiltration and distribution, 2) to assess the ability of the multi-phase flow model for mercury to reproduce field observations using literature-derived input values and 3) to assess the differences and similarities in the behaviour of liquid elemental mercury and a common DNAPL (PCE). 2. Materials and methods 2.1. Field-site The field study was performed at a former chlor-alkali plant in the Netherlands. During a previous study at this site, detailed geochemical analysis of shallow (b 30 cm depth) soil samples confirmed contamination by various mercury species in the unsaturated zone (Bernaus et al., 2006). The site is located close to a river in an alluvial plain. Consequently the aquifer consists of river sediments, such as sandy and clayey materials. The groundwater level at the site is approximately 1 m below ground level. In this study, a cone penetration test (CPT) was performed at the site. 2.1.1. CPT-probe characterisation Cone penetration tests are frequently used for describing the lithology of the subsurface. Measured parameters include the cone-pressure, which is the pressure at the tip of the cone, and the resistance of pushing the probe into the soil. The latter parameter was normalized to the cone-pressure, to obtain the friction number. The CPT-cone used was equipped (in-house) with a camera to acquire in-situ images to investigate the depth distribution of the mercury DNAPL contamination. The CPT-cone had a diameter of 36 mm, the overlying tube had a diameter of 56 mm and the length of the CPT-cone with camera was 1.2 m. In the CPT-cone a sapphire glass window (12 × 18 mm) was present, located 76 cm above the CPT-cone tip. Behind the glass window a mirror construction enabled a 2MP camera (1200 × 1600 pixels) equipped with LED-lights to visualize the subsurface, using a resolution of 10 μm by 11 μm per pixel. The camera visualized the soil profile including

26

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

Table 1 Summary of fluid properties of mercury DNAPL, PCE and water at 20° to 25 °Ca. Parameter

Unit

Mercury

HgO

PCE

Water

Density Viscosity Surface tension Interfacial tension with water Vapour-pressure Solubility

kg·m−3 10−3 Pa·s Dynes·cm−1 Dynes·cm−1 Pa μg·L−1

13,500a 1.55c 485c 375c 0.07 a,d 27.4e

– – – – – 5.3 · 104f

1630b 0.89b 32.6b 47.8 1867b 2.00 · 105b

1000 1.00 72 – – –

a b c d e f

CRC (2014). Schwille (1988). U.S. DOE (2001) and references therein. At 10 °C. At 10 °C, Sanemasa (1975). Wallschläger (1996) as referred in Schroeder and Munthe (1997).

the vertical mercury distribution at a 1 cm resolution over a 9 m vertical transect. The presence of elemental mercury was visually apparent in the appearance of bright silvery blobs. A description of the lithological characteristics was made using the visual observations, as well as the conventional cone-pressure and friction data from the CPT probe. Following e.g. Nohl and De Boer (2008), low friction numbers and high cone pressures indicate sandy lithologies, whereas high friction numbers and low cone pressures indicate clay lithologies. 2.1.2. Image analysis method Image analysis was performed to determine the vertical mercury DNAPL distribution, as observed through the CPTmounted camera. No smearing of DNAPL on the sapphire glass-window was observed, indicating that the observed DNAPL presence was relatively undisturbed by the probing. The image analysis was performed similar to the methods used previously in laboratory studies, to visualize DNAPL distribution and to determine saturations (Darnault et al., 1996; Kechavarzi et al., 2000; Luciano et al., 2010). Here, we assumed the percentage of pixels representing mercury in the in-situ images to represent the mercury saturation. First step in the image processing was to manually colour the observed mercury droplets. Mercury DNAPL was recognized by its silvery reflection and rounded droplet shapes. Mercury recognition could not be done automatically, because image-processing software was not able to recognize mercury droplets, due to wide colour range of the background and

subtle colour difference in the mercury droplets depending on light orientation. Images were then opened in the open source software R, including package: EBimage (Pau et al., 2010). Photos were written in a grey-scaled matrix and filtered, such that pixels containing mercury were given a value of 1 and the other pixels a value of 0. Where mercury DNAPL was observed, it filled up the space between the sediment grains and the sapphire-glass window, therefore the ratio of pixels containing mercury was taken as a measure of the mercury saturation. The representativeness of the derived values for pore saturation was verified using the concept of Representative Elementary Volume (REV). The REV was visualized by changing the area of interest on a photo from the maximum size, 1200 × 1600 pixels (i.e. one whole photo) to a minimum size, 12 by 16 pixels over 100 steps. As a consequence the mercury pore saturation was a function of the area of interest. With increasing areas, the saturation stabilized, indicating that the determined mercury saturation was a representative average value for the scale of the image (12 × 18 mm). The spatial correlation of the image-derived mercury saturations was analysed using one-dimensional variogram analysis, using a spherical model. These results were compared to the variogram analysis of the cone-pressure and friction number. 2.2. Comparison of mercury and PCE DNAPL properties Mercury DNAPL has a high potential to infiltrate to substantial depths due to its exceptionally high density

Table 2 Soil properties used in this study. Parameter Porosity Hydraulic conductivity van Genuchten (α) van Genuchten (n) Irreducible water content Residual Hg0 saturation Residual PCE saturation Scaling factor (β) mercury Scaling factor (β) PCE a b c d e

Carsel and Parrish (1988). Devasena and Nambi (2010). Ippisch et al. (2006). Busby et al. (1995). Scaled values.

Unit

Sand a

cm·h−1 cm−1

0.43 29.70a 0.145a 2.68a 0.045a 0.10 – 0.192 –

Silty sand

Silty clay

a

a

0.46 0.25a 0.016a 1.37a 0.034a 0.10 – 0.192 –

0.36 0.02a 0.005a 1.09a 0.070a 0.10 – 0.192 –

Medium sand b

0.33 120 0.32b 4.3b 0.1/0.288b 0.08b 0.275b 0.192 1.6

Fine sand a

0.43 80 0.024c 4.4c 0.045a 0.049e 0.17 0.192 1.6

Crozier loam 0.41a 0.21 0.009d 1.38d 0.095a 0.014e 0.13 0.26 2.2

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34 Fig. 1. Analysis of in-situ images, from left to right: in-situ image, in-situ image with black coloured mercury, and the volumetric mercury DNAPL content as function of the area of interest: A) 4.00 m depth and B) 3.53 m depth. 27

28

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

Fig. 2. Soil profiles at a former chlor-alkali plant. A) Observed friction number. B) Observed cone-pressure. C) Interpretation of cone-pressure and friction number in terms of soil classification. D) Mercury DNAPL saturation profiles for field-observations by the CPT-probe (solid line) and model results for a simplified case model (dotted line). (1) Indicates mercury DNAPL accounted for by the model, (2) indicates mercury DNAPL accumulations which were not accounted for by the model.

(13.5 kg·L− 1) and low maximum residual saturation in sands (Devasena and Nambi, 2010). For any DNAPL to infiltrate a lithological sequence, an entry-pressure has to be overcome. Both PCE and mercury DNAPL are non-wetting, thus water has to be pushed out of the pores before infiltration occurs. Assuming Leverett scaling to apply (Devasena and Nambi, 2010) the relation between required entry-head pressures for PCE and mercury DNAPL can be estimated as follows (e.g. Lenhard, 1994): Hg

PCE

Pe ¼ Pe



σ Hg  σ PCE



ð1Þ

where Pe denotes the entry-pressure induced by capillary pressure between two fluids within a pore throat [Pa], σ denotes the interfacial tension with water [N·m−1]. Next, assume no flow and a hydrostatic fluid pressure, such that Pe = gρh. Where g is the gravitational constant [m·s−2], ρ the density [kg m−3] and h the entry head [m]. Rewrite Eq. (1): Hg

PCE

he ¼ he



ρPCE 

 ρHg

σ Hg 

σ PCE

 :

ð2Þ

Using the parameters from Table 1, Eq. (2) yields that the required entry-heads for mercury and PCE are similar PCE (hHg e = 1.03he ). The analysis thus indicate that the effect of the higher density and surface tension on the entry-head of mercury DNAPL balance each other, such that the entry-head is similar to that of PCE. Consequently, the ability to infiltrate Table 3 Fitting results for semi-variograms.

Sill Range [m] Nugget effect

Friction number

Cone-pressure

Mercury DNAPL fraction

0.9166 0.1003 3.39·10-11

20.80 1.097 7.36·10-10

0.00125 0.07423 2.66E·10-14

low permeable layers is expected to be similar for mercury DNAPL and PCE under hydrostatic conditions. 2.3. Numerical multi-phase flow model To model the two-phase flow processes of mercury DNAPL– water and PCE–water in the saturated zone, the module water– oil in STOMP (Subsurface Transport Over Multiple Phases) was used (White and Oostrom, 2006). This model numerically solves differential equations for multi-phase flow by using Newton–Raphson iteration method, including a convergence criterion of 10−6, maximum number of iteration of 20 and an acceleration factor of 1.25. The model is based on the extended Darcy's equation for multi-phase flow. It, therefore, assumes a continuous flow on the Darcy scale, which implies process descriptions to be at the macro-scale and not at the pore-scale. The relative permeability, which is a function of saturation, was described by using the van Genuchten and Mualem's model (Lenhard and Parker, 1987a,b; Parker and Lenhard, 1987; van Genuchten, 1980). Following Devasena and Nambi (2010) the van Genuchten function was used for simulating the capillary-saturation behaviour. The capillary saturation function was scaled for different fluid–fluid systems, or different interfacial tensions, by using Leverett-scaling (Leverett, 1941). For common NAPLs, Leverett-scaling is proven to work well in permeable layers and reasonable well for impermeable layers (Busby et al., 1995, Lenhard and Parker, 1987b). Following the applicability of Leverett-scaling on common NAPLs, mercury capillary saturation functions were obtained by using Leverettscaling, which seemed to apply on mercury DNAPL in permeable saturated porous media, based on limited experimental literature (Devasena and Nambi, 2010). 2.3.1. Model scenario of field-investigation Using literature-derived input values for the two-phase flow of mercury DNAPL, and the soil properties such as hydraulic conductivity (Table 2), a one-dimensional two-phase flow

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

29

Fig. 3. Model results for a saturated homogeneous medium sand after a 1.35 m3 spill and 70 h: A) mercury DNAPL and B) PCE.

model scenario was designed based on the lithological information from the CPT log measurements. The model included the first 10 m of the subsurface, discretized by 1000 cells, and the lithology according to the cone-pressure and friction number (Fig. 2). For the focus of this study the whole model domain was initially assumed fully water-saturated. Attempts for a twodimensional simulation did not yield numerical convergence due to size of the model and the high resolution used to describe the lithological heterogeneities. The hydraulic conductivity, porosity, pressure-saturation curve in terms of van Genuchten parameters, and the irreducible water content, the fraction of irreplaceable water within the pores, were derived from typical parameters (Carsel and Parrish, 1988), for the sand, silty sand (silt) and silty clay, identified in the CPT-cone results. The residual mercury saturation found at the former chlor-alkaliplant was known, as described before. At the site, the location, volumes and rates of elemental mercury spills were unknown. Therefore, we assumed one mercury spill right above the characterised soil section, by using Neumann's boundary condition. We assumed, that the infiltration area was limited to 1 m2 of surface area, with a volume of 0.24 m3 spilled over 8 h. The initial and boundary conditions maintained a fullysaturated aquifer with a hydrostatic groundwater distribution. The modelled mercury DNAPL infiltration was considered for the depths below the approximate groundwater level of 1 m, below ground surface. The distribution obtained by the multiphase flow modelling was compared to the distribution observed in the field. 2.3.2. Model comparison of PCE and mercury DNAPL In addition to the one-dimensional case scenario for the field site, we simulated various two-dimensional multi-phase flow scenarios to compare the transport and distribution for the infiltration of mercury and PCE DNAPL, in homogeneous (sand) and heterogeneous (with Crozier loam) aquifers. The model dimensions used were X = 25 m, Y = 1 m and Z = 50 m with a discretization of 25 × 1 × 50 cells, respectively. For the initial and boundary conditions a hydrostatic

groundwater distribution in a saturated aquifer was maintained, by using a constant hydraulic pressure at Z = 0 (water pressure was assumed to be 0.59 MPa) and hydraulic pressure distributions at the west and east side of the model domain. A Dirichlet boundary condition simulated a pulse injection of 1.5 m3 DNAPL of either PCE or mercury DNAPL, over a 2 m2 injection area and assuming an 0.21 MPa infiltration pressure, equivalent to 90 cm mercury head. This volume approximates the average release during 8 major mercury spill events at the Oak Ridge Y-12 National Security Complex (~1.8 m3 per spill), as derived from the total estimated mercury release of 193,000 kg (Brooks and Southworth, 2011). The soil- and fluid parameters used (Table 2) were derived from literature, for medium sand, fine sand and Crozier loam (Busby et al., 1995; Carsel and Parrish, 1988; Ippisch et al., 2006). To obtain a set of parameters as realistic as possible for the modelling of mercury DNAPL the residual saturation of mercury was linearly extrapolated for fine sand and Crozier loam by using known residual saturation for mercury in medium sand and using known data for PCE in medium sand, fine sand and Crozier loam (Parker et al., 1995). The following three scenarios were used to compare the main characteristics of mercury and PCE migration and final distribution in a saturated porous media: 1) mercury and PCE spill in homogeneous medium sand, 2) sensitivity analyses on the influence of density and viscosity on mercury migration and 3) mercury and PCE spill in fine sand including a 2 m thick loam layer at 5 m depth. Spatial moments were used to characterise the DNAPL bodies and to allow a quantitative comparison (Freyberg, 1986; Kueper and Frind, 1991). 3. Results and discussion In Section 3.1 results are shown of field-observations of a soil section surrounding a former chlor-alkali plant, and the according two-phase flow simulations of infiltrating mercury DNAPL. Whereas in Section 3.2 a model comparison between mercury DNAPL and PCE is made.

30

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

Fig. 4. Centre of mass for the sensitivity analyses of mercury DNAPL in saturated homogeneous medium sand.

3.1. Characterization and modelling of mercury DNAPL at a field site 3.1.1. Image-derived mercury DNAPL morphology and vertical distribution The in-situ images obtained with the CPT mounted camera typically revealed mercury DNAPL present in small and discontinuous droplets (b 2 mm in diameter), representing ganglia of interconnected DNAPL between multiple grains. These ganglia were considerably larger than the estimated average blob sizes (b10 μm) in field studies on PCE DNAPL zones (Hartog et al., 2010). In some occasions, mercury DNAPL droplets were interconnected forming larger ganglia (2–3 mm in diameter), with elongated shapes spanning N6 mm, suggesting spatially continuous mercury DNAPL accumulations, particularly in the vertical direction (Fig. 1A). However, since

the distribution of ganglia is 3-dimensional, the 2-dimensional images under estimated their interconnectedness. Therefore, elemental mercury that appears in the images as isolated droplets might also be horizontally linked ganglia. The vertical distribution of image-derived mercury pore saturations is shown in Fig. 2, along with CPT-derived conepressures and friction numbers. As the calculated pore saturations stabilized with increasing area within the image (Fig. 1), the obtained values were representative averages for the scale of the images. The vertical mercury DNAPL distribution was characterised by generally low saturations (b0.8%) and intervals of relative large saturations up to 13.2%. Theoretically, the lowest detectable mercury saturation would be that of one pixel (10 μm by 11 μm). However, the image analysis method was dependent on the ability to recognize blobs visually. Therefore, based on the smallest mercury droplet (60 by 66 μm) identified in the images, an operational detection limit of 0.002% was derived for an image with a single droplet that size. However, the determined saturations for images with observable mercury were at least 400 times larger, all in excess of 0.8%, suggesting that mercury DNAPL was not present at lower saturations. Moreover, the observed range of the image-derived saturations was in keeping with the range of the residual saturations (1–8%; Table 2) observed in experiments by Devasena and Nambi (2010). Integration of the image-derived pore-saturations over depth yielded a total amount of 13 L mercury DNAPL per m2 surface area. The average mercury DNAPL content for this 9 m investigated sediment profile, was therefore 1.4 L m−3 for mercury DNAPL, or an estimated average pore saturation of 0.35% (assuming a porosity of 0.40).

3.1.2. Lithological heterogeneity and spatial correlation with mercury DNAPL The CPT-based lithological characterisation down to 9 m below ground surface indicated a dominantly sandy stratigraphy with alternating silty clay layers of up to 0.5 m thickness (Fig. 2). These lithological variations were revealed by the

Fig. 5. Model results for saturated fine sand including a Crozier Loam layer at 5 m depth. A) 1.48 m3 mercury DNAPL after 10 h and B) 1.48 m3 PCE after 10 h.

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

in-situ images, whereas minor heterogeneities present in the images were not always included in the CPT-data. Within this investigated aquifer section, the mercury DNAPL appeared to have accumulated on top of coarse to fine stratigraphic transitions, some of which were minor as indicated by subtle changes in friction number and cone-pressure (Fig. 2). All of these intervals were thin (b 9 cm), spanning at maximum only several times the vertical image resolution (18 mm). For example the thickness of the mercury accumulation interval at 3.6 m, 4 m, and 6.7 m depth, where 2 cm, 4 cm and 8 cm, respectively. To evaluate the overall spatial correlation of mercury DNAPL with depth in relation with the lithological heterogeneity observed by the CPT-probe, variogram analysis was performed. The CPT-derived friction number and cone-pressures, both reflect variation in lithological properties. But as can be seen from Fig. 2, the friction number reflected more strongly lithological boundaries whereas the cone-pressure variation reflected various lithological types. Consequently, this resulted in a more irregular and heterogeneous pattern for the friction number, which was reflected in the fitted variogram values, using a spherical model with a nugget effect (Table 3). The higher resolution of measurements for the friction number and cone-pressure resulted in smoother variograms, whereas the variogram for mercury DNAPL was more angular. However, the variograms could be fitted with a relative low intrinsic (nugget) uncertainty (Table 3). The range fitted for the vertical mercury distribution was 7 cm, similar to the range fitted for the friction number (10 cm) and considerably lower than for the cone-pressure (110 cm), reflecting the sensitivity of the mercury DNAPL distribution to lithological transitions. 3.1.3. Modelling analysis of field-observations The simulations with the one-dimensional model that included the major lithological units, resulted in mercury DNAPL accumulations on top of the clay-layers at 5.5 m, 6.35 m and 8.37 m, in keeping with the image-derived field observations (Fig. 2D, indicated by 1). However, the model did not reflect all accumulations of mercury directly beneath clay-layers (indicated by 2). This could be related to minor lithological changes not included in the model, but which were reflected by small variations in the friction number and cone-pressure. The magnitude of simulated mercury pore saturations was in general within range of the saturations found in the field investigation. Although, overestimations of mercury DNAPL pore saturation occurred on top of silty clay layers. Overall the modelling results suggested that the input parameters used in the mercury flow model (Table 2) were reasonable. Generally, however, the model overestimated the thickness of DNAPL accumulations, also some accumulations were predicted by the model but not observed in the field. This could be due to the one-dimensional model assumption that the mercury DNAPL infiltration occurred directly above the modelled stratigraphic sequence and thus represents a maximum of mercury DNAPL retention capacity. Likely, mercury DNAPL infiltration occurred elsewhere, where the observed mercury DNAPL accumulations resulted from lateral spreading. In addition, the field DNAPL saturations may have decreased from initial values to some extent by ageing through volatilization and dissolution, as has been described for chlorinated

31

solvent DNAPL source zones (Parker et al., 2003). However, since both the vapour pressure and the solubility of elemental mercury are very low, e.g. four orders of magnitude lower than PCE (Table 1), mercury DNAPL zones seem considerably less prone to ageing. Nevertheless, the extent to which these physical as well as chemical processes affect DNAPL distribution characteristics over time should be addressed in future studies. 3.2. Mercury DNAPL and PCE: a comparison of transport behaviours The lack of other mercury DNAPL field studies precludes the evaluation of site-specific conditions on mercury DNAPL transport and source zone evolution in the saturated zone. However, in dealing with mercury DNAPL contaminated sites one major question is to what extent mercury DNAPL migration and distribution behaviour differs from more commonly studied DNAPL phases, such as perchloroethylene (PCE). To explore the differences in transport behaviour of mercury DNAPL and PCE their contrasting properties (Table 1) were tested in several simulations under homogeneous and heterogeneous hydrogeological conditions. 3.2.1. Mercury transport in saturated homogeneous porous media Mercury DNAPL and PCE infiltration was modelled in a homogeneous sandy aquifer consisting of medium sand, using equal infiltration pressure and volume for both DNAPLs. Results indicated distinctively different transport behaviours in terms of both transport rate and distribution. The simulations showed that mercury infiltrated 3 times deeper than PCE (Fig. 3), mainly due to its lower residual saturation (Table 2). Mercury DNAPL distribution stabilized after 20 h, compared to 5 h for PCE. The final vertical mercury front (saturation N 0.01%) reached a depth of 32 m below groundwater, with the centre of mass of the mercury DNAPL at 11.65 m depth. In contrast, the PCE DNAPL front only reached a depth of 11 m, with a centre of mass at 3.63 m depth. The relative large horizontal variance of PCE (2.33 m2) compared to that of mercury (1.36 m2) was due to the high infiltrationpressure (90 cm mercury head). Consequently, mercury infiltrated deeper than PCE due to the low residual saturation and limited horizontal spreading. Therefore, another PCE spill was simulated with an infiltration pressure equivalent to 90 cm PCE, which resulted in an infiltration depth of 16 m with a reduced horizontal variance of 1.37 m2 and a centre of mass at 5.6 m. Steady-state was achieved after 15 h, which was closer to the 20 h for mercury DNAPL. Clearly, the infiltration rate and final distribution of both PCE and mercury was dependent on infiltration conditions, however under all scenarios tested elemental mercury migrated substantially deeper than PCE. The observed differences in the infiltration behaviours between mercury and PCE DNAPL can be attributed to different properties of both DNAPLs. To evaluate the importance of density and viscosity on mercury DNAPL migration, a sensitivity analysis was performed (Fig. 4). First, the density of mercury was changed to that of PCE (i.e. 13,500 kg·m−3 to 1650 kg·m−3). This caused the infiltration depth to decrease substantially, while increasing the horizontal variance from 1.36 m2 to 3.55 m2. The centre of mass was shifted from the upper half of the DNAPL zone to the lower half, from 11.65 m

32

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

to 5.60 m depth. Moreover, the time to reach entrapment was increased from 20 to 25 h. These observations indicated that the density is the major driving force for mercury DNAPL distribution during infiltration. In a second model run where only the viscosity of mercury was changed to that of PCE (from 1.55 · 10−3 Pa·s to 0.89 · 10−3 Pa·s) the infiltration depth did not change substantially, although entrapment occured slightly faster (Fig. 4). The horizontal variance decreased from 1.37 m2 to 0.92 m2. The time required for reaching entrapment decreased from 20 to 15 h, which is in keeping with a decrease in viscosity, i.e. lower resistance to flow. Though all these factors impacted on mercury DNAPL migration, its final distribution was mainly controlled by its high density, while its higher penetration potential is due to its low residual saturation.

method allows for determination of the pore saturation along with morphological characteristics directly from visual observation. However, since morphology affects visibility, absolute quantification through calibration of the measurements is challenging and derived pore saturation values should be considered approximations. In this study for example, the operational detection limit was determined at 0.002% per image, based on the smallest detected mercury droplet (60 by 66 μm). Since pore saturations below 0.8% were not observed, it is unlikely that significant mercury accumulations were missed. However, other field-specific conditions, such as DNAPL type, lithological characteristics or significant ageing, might have resulted in finely dispersed DNAPL morphologies with dominant dimensions smaller than 10's of micrometres, that would have been too small for visible detection.

3.2.2. Mercury transport in saturated heterogeneous porous media To assess the influence of heterogeneities within the soil system on mercury DNAPL migration in comparison to PCE, a spill was simulated in a fine sand aquifer containing continuous 5 m thick loam layer at 5 m depth. Although both DNAPLs infiltrated into the loam layer the extent of the infiltration depth differed substantially. Mercury DNAPL permeated through the entire loam layer, whereas PCE only infiltrated the upper part of the loam layer (Fig. 5). As a result mercury reached a far greater depth (20.5 m) than PCE (6 m). Strikingly, the horizontal spreading of mercury DNAPL on top of the loam layer resulted in a variance of 6.91 m2, which was substantially larger than for PCE (3.40 m2). In contrast, the height of PCE accumulation was higher, resulting in the centre of mass located at a more shallow depth of 3.34 m, compared to 6.83 m for mercury. Both DNAPLs required 10 h to reach entrapment, even though both DNAPLs migrated differently. Overall, the two-phase flow modelling results indicated the higher penetration potential of mercury DNAPL compared to PCE, which was further emphasized in the presence of layered heterogeneities.

3.4. Considerations for the characterisation of mercury DNAPL field sites

3.3. Image-derived in-situ characterization of DNAPL distributions To the best of our knowledge, our study is the first to present the use of in-situ derived images to determine DNAPL distribution in the field. This allowed the characterisation of mercury DNAPL droplets at the mm scale (Fig. 1), which could be up-scaled using the concept of representative elementary volume (REV) to represent the variation of mercury DNAPL saturation at a cm scale along a depth of 9 m (Fig. 2). In principle, other DNAPL phases, such as creosote can be characterised by this method, as long as they present a phase that provides sufficient contrast with the sedimentary background to allow visual detection. This dependence on visibility impedes detection of colourless DNAPL types, such as PCE, unless they could be high-lighted in-situ (e.g. Sudan dye) prior to camera observations. Compared to detailed core-sampling and analysis, as conducted previously for chlorinated solvent DNAPL characterization (e.g. Hartog et al., 2010; Parker et al., 2003), the advantages of the optical in-situ DNAPL characterization method are the rate of data collection and not having to bring these heavily contaminated sediments to surface for analysis. In addition, the

With this study we aimed to provide a first-order understanding of the behaviour and distribution of elemental mercury DNAPL in the saturated zone at field conditions and scales, taking into account the experience and insight in more commonly studied DNAPL types, such as PCE. In comparison with PCE, the low residual saturation that results from the combined properties of mercury DNAPL, result in a significantly stronger vertical penetration potential under relatively homogenous conditions. Under heterogeneous conditions with horizontal layering, the vertical penetration potential of infiltrating mercury DNAPL is further increased compared to PCE by its high density. The current study confirmed the presence of mercury DNAPL to at least 9 m depth, while at another former chloralkali plant mercury DNAPL was found at 17 m depth below surface (Deeb et al., 2011; ITRC, 2012). However, it is likely that in some locations mercury DNAPL has penetrated considerably deeper, since chlorinated solvent DNAPL have been found at greater depths (N 50 m), and our comparison indicates that the penetration potential of mercury DNAPL in the saturated zone is significantly stronger than that of PCE DNAPL. In contrast to PCE, the high surface tension of elemental mercury with respect to air (Table 1) is likely to cause mercury to behave as a non-wetting phase with respect to dry soil in the unsaturated zone, which would require some pooling of mercury to overcome entry pressures (Sweijen, 2013). Whether, where and to what extent mercury DNAPL will reach the saturated zone at a particular site, will, therefore, largely depend on the ability of mercury DNAPL to permeate or bypass (e.g. via surface water or sewer system) the unsaturated zone. The behaviour of elemental mercury and the distribution of mercury DNAPL in the unsaturated zone are further complicated by mercury vapour transport and dynamic volatilization– condensation processes (e.g. Walvoord et al., 2008). To further increase the understanding of elemental mercury transport behaviour at field sites, studies focussing on the fundamental principles governing three-phase flow (water–air–mercury) in conjunction with the dynamic aspects of elemental mercury vapour transport in the unsaturated zone are critically important. In addition to the physical factors influencing the mercury DNAPL behaviour at field sites, elemental mercury in the saturated zone is affected by chemical processes. As indicated

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

earlier, dissolution and volatilization of elemental mercury may result in source zone ageing reducing overall pore saturations. Although, based on the vapour pressure and solubility of elemental mercury (Table 1), the potential of these processes for the reduction of mercury DNAPL seems limited compared to chlorinate solvent DNAPLs (Parker et al., 2003), redox transformations may result in more soluble and volatile species. For instance, elemental mercury can be oxidized to result in mercury-oxide, which has a substantially higher solubility than pure phase mercury (Table 1), or mercury can be transformed into methyl-bound mercury species under reducing conditions (e.g. Gabriel and Williamson, 2004). To which extent and under which conditions these redox transformation processes may significantly enhance the ageing of mercury DNAPL zones is a pressing research topic considering the toxicity of these transformation products and the persistence with which mercury DNAPL can act as a long-term source of contamination. 4. Conclusion Through analysis of both field data and two-phase flow modelling, we present in this paper a first assessment of mercury DNAPL transport and distribution behaviour in saturated porous media. Also, we present a novel methodology of interpreting in-situ DNAPL saturations by the processing of images obtained with a CPT-probe. Using this method, a soil section surrounding a former chlor-alkali plant was studied, revealing mercury DNAPL to be present in saturations up to 13.2%, which were located at lithological interfaces. Using the multi-phase flow model STOMP (Subsurface Transport Over Multiple Phases), the results from a onedimensional mercury infiltration model representing this characterised soil section, suggested that the input parameters used in the mercury flow model (Table 2) were reasonable. A model comparison of the transport behaviour of mercury DNAPL and PCE in the saturated zone indicated that mercury DNAPL had a higher potential to penetrate to greater depths than PCE DNAPL. In the presence of horizontally layered heterogeneity the penetration potential of mercury DNAPL was further enhanced relative to PCE DNAPL. Compared to commonly known DNAPLs (e.g. PCE) elemental mercury is an extraordinary DNAPL, with a high density, surface tension and a low residual saturation. Due to its low maximum residual saturation and its high density mercury DNAPL may migrate to substantial depths. Acknowledgements We would like to thank Prof. Dr. S. M. Hassanizadeh for valuable discussions on this research. Also, we appreciate the constructive comments and suggestions from three anonymous reviewers that helped improve the manuscript. References Arbestain, M.C., Rodriguez-Lado, L., Bao, M., Macias, F., 2009. Assessment of mercury-polluted soils adjacent to an old mercury-fulminate production plant. Appl. Environ. Soil Sci. 2009. Bernaus, A., Gaona, X., van Ree, D., Valiente, M., 2006. Determination of mercury in polluted soils surrounding a chlor-alkali plant: direct speciation by X-ray

33

absorption spectroscopy techniques and preliminary geochemical characterisation of the area. Anal. Chim. Acta. 565-1, 73–80. Biester, H., Müller, G., Schöler, H.F., 2002a. Estimating distribution and retention of mercury in three different soils contaminated by emissions from chlor-alkali plants: part I. Sci. Total Environ. 284, 177–189. Biester, H., Müller, G., Schöler, H.F., 2002b. Binding and mobility of mercury in soils contaminated by emissions from chlor-alkali plants. Sci. Total Environ. 284, 191–203. Bollen, A., Wenke, A., Biester, H., 2008. Mercury speciation analyses in HgCl2contaminated soils and groundwater — implications for risk assessment and remediation strategies. Water Res. 42, 91–100. Brooks, S.C., Southworth, G.R., 2011. History of mercury use and environmental contamination at the Oak Ridge Y-12 Plant. Environ. Pollut. 159, 219–228. Busby, R.D., Lenhard, R.J., Rolston, D.E., 1995. An investigation of saturationcapillary pressure relations in tow-and-three-fluid systems for several NAPLS in different porous media. Groundw. 33 (4) (July–August 1995). Carsel, R.F., Parrish, R.S., 1988. Developing joint probability distributions of soil water retention characteristics. Water Resour. Res. 24-1, 755–769. Cohen, M.C., Mercer, J.M., 1990. A review of immiscible fluids in the subsurface: properties, models, characterization and remediation. J. Contam. Hydrol. 6, 107–163. CRC, 2014. Handbook of Chemistry and Physics, 94th ed. (Internet Version 2014). Darnault, C.J.G., Throop, J.A., DiCarlo, D.A., Rimmer, A., Steenhuis, T.S., Parlange, J.Y., 1996. Visualisation by light transmission of oil and water contents in transient two-phase flow fields. J. Contam. Hydrol. 31 (1998), 337–348. Deeb, R., Hawley, E., Kell, L., O'Laskey, R., 2011. Assessing alternative endpoints for groundwater remediation at contaminated sites. ESTCP-Project ER-200832. Devasena, M., Nambi, I.M., 2010. Migration and entrapment of mercury in porous media. J. Contam. Hydrol. 117, 60–70. EuroChlor, 2010. http://www.eurochlor.org/chlorine-industry-issues/mercury. aspx (visited at 3-February-2013). Freyberg, D.L., 1986. A natural gradient experiment on solute transport in a sand aquifer 2. Spatial moments and the advection and dispersion of nonreactive tracers. Water Resour. Res. 22-13, 2031–2046. Gabriel, M.C., Williamson, D.G., 2004. Principal biogeochemical factors affecting the speciation and transport of mercury through the terrestrial environment. Environ. Geochem. Health 26, 421–434. Hartog, N., Cho, J., Parker, B., Annable, M.D., 2010. Characterization of a heterogeneous DNAPL source zone in the Borden aquifer using partitioning and interfacial tracers: residual morphologies and background sorption. J. Contam. Hydrol. 115, 79–89. Horvat, M., Nolde, N., Fajon, V., Jereb, V., Logar, M., Lojen, S., Jacimovic, R., Falnoga, I., Liya, Q., Faganeli, J., Drobne, D., 2003. Total mercury, methyl mercury and selenium in mercury polluted areas in the province Guizhou, China. (2002) Sci. Total Environ. 304, 231–256. Hu, Haiyan, Lin, Hui, Zheng, Wang, Tomanicek, Stephen J., Johs, Alexander, Feng, Xinbin, Elias, Dwayne A., Liang, Liyuin, Gu, Baohua, 2013. Oxidation and methylation of dissolved elemental mercury by anaerobic bacteria. Nat. Geosci. 6, 751–754. Ippisch, O., Vogel, H.J., Bastian, P., 2006. Validity limits for the van Genuchten– Mualem model and implications for parameter estimation and numerical simulation. Adv. Water Resour. 29-12, 1780–1789. ITRC, 2012. Using Remediation Risk Management to Address Groundwater Cleanup Challenges at Complex Sites. p. 31. Jackson, T.A., 1998. Mercury in aquatic ecosystems. In: Langston, W.J., Bebianno, M.J. (Eds.), Metal Metabolism in Aquatic Environments. Chapman & Hall. Johannesson, K.H., Neumann, K., 2012. Geochemical cycling of mercury in a deep, confined aquifer: insight from biogeochemical reactive transport modeling. Geochim. Cosmochim. Acta 106 (2013), 25–43. Kechavarzi, C., Soga, K., Wiart, P., 2000. Multispectral Image Analysis Method to Determine Dynamic Fluid Saturation Distribution in Two-dimensional Three-fluid Phase Flow Laboratory Experiments. Kueper, B.H., Frind, E.O., 1991. Two-phase flow in heterogeneous porous media 2. Model application. Water Resour. Res. 27 (6), 1059–1070 (June 1991). Lenhard, R.J., 1994. Scaling fluid content-pressure relations of different fluid systems in porous media. Hydrology Days Conference (April 5-9, 1994). Lenhard, R.J., Parker, J.C., 1987a. A model for hysteric constitutive relations governing multiphase flow 2. Permeability–saturation relations. Water Resour. Res. 23 (12), 2197–2206 (December 1987). Lenhard, R.J., Parker, J.C., 1987b. Measurement and prediction of saturation– pressure relationships in three-phase porous media systems. J. Contam. Hydrol. 1, 407–424. Leverett, M.C., 1941. Capillary behaviour in porous solids. American Institute of Mining and Metallurgical Engineers Transactions. Petrol. Dev. Technol. 142, 152–169. Lopes, C.B., Oliveira, J.R., Rocha, L.S., Tavares, D.S., Silva, C.M., Silva, S.P., Hartog, N., Duarte, A.C., Pereira, E., 2013. Cork stoppers as an effective

34

T. Sweijen et al. / Journal of Contaminant Hydrology 161 (2014) 24–34

sorbent for water treatment: the removal of mercury at environmentally relevant concentrations and conditions. Environ. Sci. Pollut. Res. 1-14. Luciano, A., Viotti, P., Papini, M.P., 2010. Laboratory investigation of DNAPL migration in porous media. J. Hazard. Mater. 176, 1006–1017. Miller, C.L., Watson, D.B., Lester, B.P., Lowe, K.A., Pierce, E.M., Liang, L., 2013. Characterization of soils from an industrial complex contaminated with elemental mercury. Environ. Res. 125, 20–29. Nohl, W.A., de Boer, E., 2008. Voor Nederlandse bodem: Classificatie grondsoorten op basis van sonderingen. Geotechnique 2008-7. Parker, J.C., Lenhard, R.J., 1987. A model for hysteric constitutive relations governing multiphase flow 1. Saturation–pressure relations. Water Resour. Res. 23 (12), 2187–2196 (December 1987). Parker, J.C., Waddill, D.W., Johnson, J.A., 1995. UST Corrective Action Technologies: Engineering Design of Free Product Recovery Systems, EPA/600/SR96/031. U.S. Environmental Protection Agency, Washington, DC. Parker, B.L., Cherry, J.A., Chapman, S.W., Guilbeault, M.A., 2003. Review and analysis of chlorinated solvent dense nonaqueous phase liquid distributions in five sandy aquifers. Vadose Zone J. 2 (2), 116–137. Pau, G., Fuchs, F., Skylar, O., Boutros, M., Huber, W., 2010. EBImage — an R package for image processing with applications to cellular phenotypes. Bioinformatics 26-7, 979–981. Sanemasa, I., 1975. The solubility of elemental mercury vapor in water. Bull. Chem. Soc. Jpn. 48 (6), 1795–1798. Scanlon, B.R., Tachovsky, J.A., Reedy, R., Nicot, J.P., Keese, K., Slade, R.M., Merwad, V., Howard, M.T., Wells, G.L., Mullins, G.J., Ortiz, D.M., 2005. Groundwater– surface water interactions in Texas. Bur. Econ. Geol. 129–131. Schroeder, W.H., Munthe, J., 1997. Atmospheric mercury — an overview. Atmos. Environ. 32 (1998), 809–822.

Schuster, E., 1991. The behaviour of mercury in the soil with special emphasis on complexation and adsorption processes — a review of the literature. Water Air Soil Pollut. 56, 667–680. Schwille, F., 1988. Dense Chlorinated Solvents in Porous and Fractured Media Model Experiments. Lewis Publishers, Chelsia, MI. Sweijen, T., 2013. The behaviour of elemental mercury in porous media: a modelling approach on the multi-phase-flow of elemental mercury. (MSc Thesis) Faculty of Geoscience, Utrecht Universiteit, The Netherlands. U.S. Doe, 2001. Mercury Contaminated Material Decontamination Methods: Investigation and, Assessment. DE-FG21-95EW55094. Ullrich, S.M., Tanton, T.W., Abdrashitova, S.A., 2001. Mercury in the aquatic environment: a review of factors affecting methylation. Crit. Rev. Environ. Sci. Technol. 31, 241–293. van Genuchten, M.Th., 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci. Soc. Am. J. 44 (5) (September–October 1980). Wallschläger, D., 1996. Speziesanalytische Untersuchungen zur Abschätzung des Remobilisierungspotentials von Quicksilber aus kontaminierten Elbauen. (Doctoral Dissertation) University of Bremen, Bremen Germany. Walvoord, M.A., Andraski, B.J., Krabbenhoft, D.P., Striegel, R.G., 2008. Transport of elemental mercury in the unsaturated zone from a waste disposal site in an arid region. Appl. Geochem. 23, 572–583. White, M.D., Oostrom, M., 2006. STOMP subsurface transport over multiple phases. Theory Guide. World chlorine council, 2011. Reduction of mercury emissions and use from the chlor-alkali sector partnership. Update January 2011. http://www.eurochlor. org/media/9074/3-4-1-unep_global_mercury_programme_-_the_chloralkali_sector_partnership.pdf (Visited at 11-June-2013).