Simplified behaviors from increased heterogeneity: II. 3-D Uranium transport at the decimeter scale and intertank comparisons

Journal of Contaminant Hydrology 148 (2013) 51–66

Contents lists available at SciVerse ScienceDirect

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

Simplified behaviors from increased heterogeneity: II. 3-D Uranium transport at the decimeter scale and intertank comparisons Andrew W. Miller a, b,⁎, Derrick R. Rodriguez a, Bruce D. Honeyman a a b

Colorado School of Mines, Golden, CO, 80401, United States Sandia National Laboratories, Albuquerque, NM, 87123, United States

a r t i c l e

i n f o

Article history: Received 17 October 2011 Received in revised form 13 December 2012 Accepted 19 December 2012 Available online 23 January 2013 Keywords: Metals Radionuclides Mineral weathering Upscaling

a b s t r a c t Upscaling from bench scale systems to field scale systems incorporates physical and chemical heterogeneities from atomistic up to field scales. Heterogeneities of intermediate scale (~10−1 m) are impossible to incorporate in a bench scale experiment. To transcend these scale discrepancies, this second in a pair of papers presents results from an intermediate scale, 3-D tank experiment completed using five different particle sizes of uranium contaminated sediment from a former uranium mill field site. The external dimensions of the tank were 2.44 m×0.61 m×0.61 m (L×H×W). The five particle sizes were packed in a heterogeneous manner using roughly 11 cm cubes. Small groundwater wells were installed for spatial characterization of chemical gradients and flow parameters. An approximately six month long bromide tracer test was used for flow field characterization. Within the flow domain, local uranium breakthrough curves exhibited a wide range of behaviors. However, the global effluent breakthrough curve was smooth, and not unlike breakthrough curves observed in column scale experiments. This paper concludes with an inter-tank comparison of all three experimental systems presented in this pair of papers. Although there is a wide range of chemical and physical variability between the three tanks, major chemical constituent behaviors are often quite similar or even identical. © 2013 Elsevier B.V. All rights reserved.

1. Introduction In the first of this pair of papers (Miller et al., 2013) alterations to particle size distributions led to local variability in pH and calcite dissolution reactions. This in turn has localized effects on uranium desorption and transport and a previously determined surface complexation model (SCM) did not describe the data particularly well. However, at the macroscopic scale, local chemical equilibrium appeared to dominate uranium behavior: uranium concentrations did not rebound after stop flow events, and changes to the influent chemistry led to expected changes in effluent uranium concentrations. The first two tanks start to define a continuum of heterogeneity, with tank #1 being physically homogenous and tank #2 being

⁎ Corresponding author at: Sandia National Laboratories, Albuquerque, NM, 87123, United States. Tel.: + 1 505 844 2910; fax: + 1 505 844 2348. E-mail address: [email protected] (A.W. Miller). 0169-7722/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jconhyd.2012.12.011

physically heterogeneous as well as having a temporal shift in influent water chemistry. The continuum is completed in this second paper with results from a 3-D experimental tank. The third tank is considerably more heterogeneous, as three more particle sizes are introduced. The addition of the extra particle sizes leads to a more heterogeneous permeability distribution and a non-uniform initial distribution of uranium relative to the first two tanks. Also, while there are no temporal shifts in influent water chemistry, dissolved silicon was added to the influent to test whether the pH was being controlled by clay weathering reactions. This paper also includes an inter-tank comparison. In upscaling small scale variability in chemical and physical heterogeneities to transport scenarios, there are two potential extreme outcomes in observed behavior at a larger scale: small scale heterogeneities will be directly observable in the breakthrough curve, or the behavior will be smoothed as local heterogeneities counterbalance to give an averaged response to desorption and transport. The first of these would

52

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

exhibit itself as a breakthrough curve punctuated by sharp peaks and valleys as the effects of local heterogeneities elute from the system. The second would exhibit itself as a smooth breakthrough curve, where concentration does not change dramatically between sampling points. The second of these results indicates possible simplifying assumptions to describe transport, the first result indicates the need for better site characterization and explicit inclusion of heterogeneities in model domains. The results of the inter-tank comparison give evidence for the smoothing of transport behaviors. Despite the differences between the tanks in the continuum, global scale behaviors are similar. 2. Methods 2.1. Tank construction The outer walls and bottom of the tank were constructed from 2.5 cm thick Plexiglas, and were supported on all sides by a welded/bolted steel frame. The walls and floor of the tank were epoxied in place using the external frame for support. Once the walls were together, a bead of caulk was run along the interior of all the seams. The external dimensions were 2.44 m × 0.61 m × 0.61 m (L× H × W, see Fig. 1). A hole was drilled in both endplates 8.3 cm from the bottom to allow for water flow in and out of the tank. On the inside of the tank, two endfilters were constructed to hold the sediment in place. These endfilters were constructed of 7.6 cm plastic slats which were screwed together to create a frame equal to the internal dimensions of the tank. A perforated aluminum plate was screwed to the plastic frame. Aluminum #750 mesh screening was spot-epoxied to the aluminum plate. These endfilters were placed in the up and down gradient ends adjacent to the endplates. Once packed (described below), a Plexiglas lid was placed on the top of the tank. Pre-drilled holes allowed access to sampling wells. A bead of caulk was run around each well hole, as well as along the entire wall/lid boundary to ensure an air-tight seal. Overflow constant head boundary tanks were used to control the flow. Small ground water wells were constructed for porewater withdrawal throughout the 3-D spatial domain. These wells were constructed from 0.16 cm OD, 0.08 cm ID aluminum tubing. Each hole was screened over a depth of about 1 cm with the #750 mesh described above; the mesh was held in place with epoxy. 46 total wells were made with lengths ranging from 5.8 cm to 52.6 cm. The void volume of the deepest wells was

b0.3 mL. Small sampling wells minimized alterations to the flow regime caused by sampling and the physical presence of the wells. 2.2. Sediment preparation Uranium contaminated sediment from the Naturita field site (Davis and Curtis, 2003) was used in this experiment. Field procurement of the sediment is addressed in the previous paper (Miller et al., 2013). The field derived b2 mm fraction was separated into three different particle sizes. To create these different fractions, large amounts (~1100 kg) of the b2 mm fraction were allowed to air dry in the lab. The majority of this dried sediment was used directly as b2 mm composite material in the tank. The remainder was dry sieved using a 0.250 mm screen. This created the b 0.250 mm and 0.250– 2.0 mm size fractions; the 0.250–2 mm fraction is referred to as >0.250 mm throughout this paper. Approximately half of the b0.250 mm fraction was re-sieved using a 0.125 mm screen. What remained on the screen became the 0.125– 0.250 mm fraction; the fines (b 0.125 mm) were not used in this experiment. The 4–12 mm material had a high degree of fines adhering to the larger particles, and it was contaminated with large, dried clumps of clay which were not separated in the field. This required the 4–12 mm fraction to be wet-sieved. A small portion (~30 kg) of sediment was placed into a portable cement mixer with approximately 12–20 L of DI water. The cement mixer was turned on, and the water and sediment were allowed to mix for several minutes. Immediately upon shutting off the mixer, the water and sediment were dumped onto a 4 mm screen. What remained on the screen was further rinsed with 0.25–0.5 L of DI water to remove the remaining fines. The water was captured for reuse. The sediments were allowed to air dry before packing. The total mass of each sediment used was: b2 mm composite — 602 kg, b0.250 mm — 83 kg, >0.250 mm — 231 kg, 0.125– 0.250 — 62 kg, and 4–12 mm — 179 kg. This gives a total of 1157 kg of sediment and an average bulk density of 1.59 g/cm 3. Labile uranium (Kohler et al., 2004) of the b2 mm, b0.250 mm, and >0.250 mm fractions was determined to be 1.8×10−2 μmol U/g sediment, 1.8×10−2 μmol U/g sediment, and 1.7×10−2 μmol U/g sediment, respectively (Miller et al., 2013). Assuming the 4–12 mm fraction had negligible sorbed uranium due to the wet sieving, and the 0.125–0.250 mm fraction also had 1.8×10−2 μmol/g sediment, 17.6 mmol labile uranium were added. 2.3. Sediment packing

Fig. 1. Photograph of the tank as constructed. The length is 2.44 m, the width and height are both 0.61 m. The tubing connects each individual well to the pressure monitoring system. The black uprights are welded metal brackets which act as the major load bearing feature. Not shown in the photograph are weight bearing brackets that encompass both ends.

Sampling well and particle size spatial orientations were determined through premodeling of several different orientations. The flow domain was discretized into 500 cells with dimensions of 11.2 cm × 11.7 cm × 11.2 cm (L × H × W). The 500 cells were assigned physical properties of the 5 different fractions in 5 different orientations which were believed to cause significant non-Fickian behaviors. From the simulated breakthrough curves, the packing orientation selected exhibited a prominent shoulder on the main peak and mild tailing (for further details see Miller, 2010). It was assumed the least Fickian

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

flow regime would give the steepest chemical gradients within the tank, leading to the most significant chemical scaling effects. The tank was packed in five 11.2 cm layers. An aluminum frame was constructed to divide each layer into 100 cells with dimensions identical to the premodeling simulations. Size separated sediment was poured into each cell until it was nearly full. Juxtaposed cells with two different particle sizes were filled simultaneously to minimize leakage through the aluminum frame. Sediment compaction was completed using a concrete vibrator. When an entire layer was packed, the aluminum frame was pulled out of the packed sediment and set on top of the packed layer to permit packing of the next layer. When adjacent layers had the same particle size, the concrete vibrator was extended through the layers to minimize packing density variation between the layers. When the 0.125–0.250 mm fraction was immediately adjacent to the 4–12 mm fraction, the smaller particle size flowed into the pore

53

space of the larger sized material. This necessitated surrounding the 0.125–0.250 mm cells with the same #750 mesh used for the endfilters. The hydraulic conductivity of the screen is significantly higher than the sediment itself. This same behavior was not observed for any other particle size pairings. The wells were placed in the tank during the packing of the sediments (well location is depicted in Fig. 2A–E, exact dimensional locations are in Miller, 2010). Wells were placed at 9 different depths, with each depth corresponding to the middle of a single layer or to the boundary between two layers. As the sediment was added and the desired well depth was reached, a plastic stabilizing slat with a pre-drilled hole was placed across the top of the tank at the well location. Once the bottom of the well was in place, sediment was poured around it to hold it in place through the compaction process. The stabilizing slat was left in place until enough sediment had been added to support the well (about a layer and a half);

Fig. 2. Vertical cross sections through the tank showing packing orientation of the size fractions as well as the location of the sampling wells; upgradient end is on the left, downgradient end is on the right. The dimensions of each cell are 11.2 cm × 11.7 cm × 11.2 cm (x, y, z). The (0,0,0) reference point is given in Fig. 2A; the y-axis goes into the plane of the page. The y-dimensions of each slice are as follows: A= 0–11.7 cm, B = 11.7–23.4 cm, C = 23.4–35.1 cm, D = 35.1–46.8 cm, and E = 46.8–58.5 cm. Square symbols are wells present in the middle of the cross section; triangles are wells which are on the boundary between cross sections. The number next to the well is the well number. The wells shown in each slice are the wells used to create the kriged plots in the Results section.

54

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

the slat was then removed. This process was repeated for all 46 of the wells.

This sampling procedure effectively sampled a total volume of 30 mL within the porous media, equating to a sphere with diameter of ~4 cm.

2.4. Water flow, tracer injection, sampling and analysis An artificial groundwater (AGW) was used in the tank experiment. The composition, seen in Table 1, was intended to mimic that found at the Naturita field site, and is distinct from AGW's used in the companion paper by the addition of dissolved silicon. Average flow rates through the tank were measured to be 3.95 L/day (±0.75), and the spatially averaged groundwater velocity was 3.34 cm/day. AGW flowed through the tank for 9 months (~3.9 PV). After 1.0 PV (~ 270 L) had flowed through the tank, a pulse of bromide was injected into the tank as an inert tracer. The injection occurred by switching flow to an identical parallel head boundary through which the tracer solution was being circulated. A total of 9.69× 10−3 mol (774 mg) of bromide was injected over a period of 25.5 h. The bromide concentration was 1.95× 10−3 M KBr (155.5 mg Br −/L) which has been shown to avoid density induced sinking (Barth et al., 2001a). Bromide was monitored for about six months until effluent values were below detectable limits (~0.1 mg/L). Over the entire time of water flow through the tank, spatial samples were removed on a weekly basis from all of the groundwater wells, and influent/effluent samples were taken twice a week. Influent samples were removed directly from the upgradient head boundary. Effluent samples were removed from a T-joint in the effluent line. This line was purged with ~ 25–50 mL of solution before taking a sample. At the top of each ground water well, a three-way valve was put in place to allow for pore water removal, and for connection to a digital sensor array (DSA 3207, Scanivalve Corp.) to measure local pressure head. These wells were sampled by connecting a syringe to the three-way valve. Each well was purged with 5 mL of volume, which was discarded. Then, using a clean syringe, ~7 mL was removed for analysis. Once the effluent and spatial samples were taken, they were treated in an identical manner. Samples were split for analysis as before (Miller et al., 2013). Table 1 Composition of the artificial groundwater; AGW with Si was used exclusively in tank 3. Concentrations for ions are in molar, pH is in standard pH units, alkalinity is in meq/L, and the saturation index is unitless. Constituent

AGW-atmospheric

AGW-2%

AGW with Si

Na+ K+ Mg2+ Ca2+ SiO2

2.43 × 10−2 2.50 × 10−4 2.67 × 10−3 5.73 × 10−3 0

2.11 × 10−2 2.50 × 10−4 2.67 × 10−3 7.30 × 10−3 0

Cl− SO42− Ionic strength pH Alkalinity Equilibrium Gas

1.65 × 10−2 1.21 × 10−2 5.11 × 10−2 7.85 0.61 Atmospheric

Calcite Saturation Index (calculated)

0.09

1.33 × 10−2 1.21 × 10−2 6.54 × 10−2 7.15 3.85 2%CO2, 20%O2, 78%N2 0.08

2.47 × 10−2 2.42 × 10−4 2.46 × 10−3 5.65 × 10−3 4.72 × 10−4/ 7.35 × 10−4 (before/after 1.26 PV) 1.44 × 10−2 1.34 × 10−2 5.13 × 10−2 7.85 0.61 Atmospheric

0.06

3. Results 3.1. Water chemistry results Figs. 3 through 5 show the spatial relationships for pH and uranium for three different, roughly equidistant, time points in the tank (note the changing uranium concentration scales between the plots). In each plot pH values tend to drop rather precipitously in the upgradient half of the tank, and maintain lower values throughout the flow domain. Also, the pH range is fairly small and similar for the different time points shown. Uranium concentrations tend to increase as the water moves downgradient. Concentration gradients for uranium were generally smooth, but well #10 tended to have much higher uranium concentrations than the surrounding areas (compare Figs. 2A/B with 3A/B, 4A/B, and 5A/B). However, pH around well #10 is consistent with other local values. Local uranium values generally decreased with time, but the spatial gradients ranging from high to low were static. Many more kriged plots are available in the Supporting Information. The spatial trends of these ions are similar to uranium with both alkalinity and calcium being positively correlated with uranium (see below). In comparing the pH and uranium plots there is a visual relationship between pH and uranium where lower pH correlates with higher uranium values. This is shown explicitly with a correlation plot in Fig. 6. Each point in Fig. 6 represents the analysis of a single sampling well for the number of pore volumes shown. The data presented cover the entire experimental range. At early time points, 0.37 PV and 1.06 PV, the correlation is quite weak, but at later time points, the relationship becomes much stronger. This may represent a flushing of highly soluble salts created in the sediment drying process as well as an initial time period required to mix and equilibrate influent water with the sediments. A notable exception to the trend is well #10, where the uranium concentrations were always much higher than other wells with similar pH values. Wells #2 and #7 are also anomalously high, but not as high as well #10 (wells #2 and #7 are 1–2 standard deviations above the mean uranium value for a given sampling day, and well #10 is 4–5 standard deviations above the mean). A similar time dependent relationship was observed between Ca and pH (Fig. 7). Fig. 8 shows the uranium plotted as a function of calcium concentration. Here there is a time invariant relationship between all the data points. Speciation calculations show that the Ca2UO2(CO3)0 dominates uranium chemistry over time and space. The trend in Fig. 8 is consistent with these calculations, although the slope of the line and the associated stochiometric ratio are apparently concentration dependent. For smaller values of uranium and calcium the slope is near 0.5 as the speciation calculations would predict. At the higher concentrations, which are earlier points in experimental time, the slope is closer to one indicating a uranium concentration control beyond aqueous complexation. Also seen in Fig. 8 is the behavior of well #10. At early time points well #10 plots above the quasi-linear relationship of all the other points. However, as experimental time progresses, the relationship falls

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

55

Fig. 3. Spatial profiles for pH (left) and uranium (right) after 1.06 PV of water flow. Uranium scale is micromolar; pH is in standard units. Each plot corresponds to the vertical cross sections in Fig. 2A–E.

on the same line as the other data points, just at higher concentrations. Fig. 9 is a correlation plot of silicon with pH which shows a time independent relationship. Silicon is distinct from calcium and uranium as the net silicon flux is from the solution phase to the solid phase. There is an apparent equilibrium relationship between pH and silicon, although it does not appear to be caused by the solubility of any well-characterized silicon bearing mineral phase. Solubility calculations for many silicon bearing minerals were completed using an average pore water composition, and varying assumptions regarding aluminum fate. Aluminum is not detectable in the pore waters, but it is present in clay minerals; clay solubilities are directly related to

assumptions made regarding aluminum. The closest fit to the silicon concentration data assumes equilibrium with chlorite, excess aluminum and a magnesium concentration of 3.1 × 10 − 3 M, a median value found in the tank. In general the solubility curve is parallel to the data but offset by about 30%. When the calculations are repeated with an amorphous Al(OH)3 solid phase controlling the aluminum concentration (a more likely scenario given the pH) the relationship is much worse; at pH b 7.4 the calculated silicon concentration is much higher, and at pH> 7.4 the calculated silicon concentration is much lower (data not shown). At all points in the tank, aqueous chemistry is supersaturated with respect to all major silicon bearing phases (kaolinite, albite, anorthite, K-feldspar,

56

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

Fig. 4. Spatial profiles for pH (left) and uranium (right) after 2.54 PV of water flow. Uranium scale is micromolar; pH is in standard units. Each plot corresponds to the vertical cross sections in Fig. 2A–E.

K-mica, Ca-montmorillonite, illite, chlorite, chalcedony, and quartz) with the exception of amorphous silica. The equilibrium concentration for amorphous silica for the entire pH range shown is 1.92–1.96 mM.

3.2. Uranium transport behavior Fig. 10 shows the flux averaged uranium concentration in the effluent from the tank. One pore volume is approximately two months of real time. Of the 17.6 mmol uranium added, only 4.80 mmol eluted giving a mass recovery of 27%. Fig. 11 shows the uranium data from a selection of wells as a function of

the total number of pore volumes eluted. In an attempt to normalize behaviors from the overall physical and chemical heterogeneities in the tank, only wells surrounded by b2 mm material are shown in this plot. Figs. 10 and 11 imply that although there is a large amount of local variability, the variability does not translate up in scale to the tank effluent breakthrough curve.

3.3. Bromide transport The effluent bromide breakthrough curve is shown in Fig. 12. In general the breakthrough curve shows non-Fickian

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

57

Fig. 5. Spatial profiles for pH (left) and uranium (right) after 3.94 PV of water flow. Uranium scale is micromolar; pH is in standard units. Each plot corresponds to the vertical cross sections in Fig. 2A–E.

behaviors observed at other scales including a relatively early breakthrough, and a long tailing period. Integrating the area under the curve gives a bromide mass recovery of 85%. Less than 100% recovery is attributable to a small amount of mass left in the tank when the effluent values were non-detectable; wells #2, 3, 7, 10, 13, and 33 all had detectable bromide when there was no more detectable bromide in the effluent. Several of the monitoring wells had detectable bromide for considerable periods of time (2 months or more). Local breakthrough curves for three individual wells are shown in Fig. 13. Wells 7, 19, and 32 were chosen as they encompass the overall range of bromide responses in the tank. Well 7 is located in the b2 mm composite material, wells 19 and 32 are in the 4–12 mm

fraction. Breakthrough curves for wells 1, 2, 3, 6, 13, 29, 31, 32, and 33 can be found in Miller, 2010. Well #7 is non-Fickian, with a smaller peak concentration and longer tail continuing beyond the monitoring range. Well #19 is Fickian, with a sharp peak and a relatively short to non-existent tail. Well #32 has two separate peaks. 4. Discussion At the core of upscaling methodologies is the idea that small scale behavior needs to be combined to a larger scale global output. It is well known local physical and chemical heterogeneities can cause unexpected system behavior. What

58

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

25.0

Uranium Conc. (µM)

Well #10 20.0

15.0

10.0

5.0

0.0 7.00

7.20

7.40

7.60

7.80

8.00

2.98 PV

3.94 PV

8.20

pH 0.37 PV

1.06 PV

2.05 PV

Fig. 6. Correlation plot between uranium and pH. Each point represents the analysis from a single sampling well for the number of pore volumes indicated. The dates shown encompass the majority of the experimental timeframe.

remains unknown is how the interaction between several local scale systems interact to create global scale behavior. The results from these experiments exemplify this core problem. 4.1. Uranium transport In Figs. 11 and 13 the local scale uranium and bromide transport behavior is shown. In both plots, the local scale behavior is quite divergent. For uranium, the range of behaviors from this narrow range of the dataset is quite broad. The most dynamic wells, 10 and 14, show very high initial uranium concentrations with well 14 quickly decreasing and well 10 maintaining a higher concentration for about two pore volumes before declining. Wells 2, 22, and 29 all show intermediate initial uranium concentrations with more moderate rates of decline, whereas, wells 41 and 42 show a very low initial concentration

with no significant decline for the timescale shown. Some of this behavior is attributable to the locations within the tank. Wells 41 and 42 are very close in the tank and have similar uranium responses. However, well 22 has a similar uranium breakthrough curve to well 29, but it is located at the downgradient end while well 29 is at the upgradient end (see Fig. 2A–E). Considering only this data, it appears that despite the fact these wells are physically/chemically similar (same particle size, assumed narrow groundwater velocity range) there is still spatial variation in the observed rate and extent of uranium desorption. This type of behavior has been used as a conceptual model at smaller scales (Szecsody et al., 1998), and bolsters the use of distributed reactivity models to upscale chemical information (e.g., Liu et al., 2008). Several wells exhibit disparate behavior to the extreme, specifically wells #2, 7, and 10. Each of these wells has

18.0

Calcium Conc. (mM)

16.0 14.0 12.0 10.0 8.0 6.0 4.0 7.00

7.20

7.40

7.60

7.80

8.00

8.20

pH 0.37 PV

1.06 PV

2.05 PV

2.98 PV

3.94 PV

Fig. 7. Correlation plot between calcium and pH. Each point represents the analysis from a single sampling well for the number of pore volumes indicated. The dates shown encompass the majority of the experimental timeframe. Influent calcium concentration is 5.65 mM.

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

59

25.0

Well #10 Uranium Conc. (µM)

20.0

15.0

10.0

5.0

0.0 5.0

7.0

9.0

11.0

13.0

15.0

17.0

19.0

Calcium Conc. (mM) 0.37 PV

1.06 PV

2.05 PV

2.98 PV

3.94 PV

Fig. 8. Correlation plot between uranium and calcium. Each point represents the analysis from a single sampling well for the number of pore volumes indicated. The dates shown encompass the majority of the experimental timeframe.

considerably higher concentrations of uranium and other major ground water ions compared to the mean values for a given day, while pH and silicon values are not particularly different. The area around these wells appears especially reactive, although the reason for this is entirely unknown. One possibility is that random packing variations put highly soluble salts near these wells, and the relatively low groundwater velocities did not carry away the dissolution products. Despite the local scale complexity in terms of observed uranium breakthrough curves as a function of space, the global effluent breakthrough curve is fairly simple. If there had not been some buffering of uranium transport, the breakthrough curve would have had sharp peaks and valleys as each chemical variation contributes its own signal to the global breakthrough curve. The dramatic differences in local uranium behavior are seemingly integrated over the flow domain and small scale

variability is not being transmitted to the tank scale. Admittedly, some of the homogenized behaviors are due to flux averaging in the downgradient endfilter within the tank and the extent of this flux averaging is not easily quantifiable. Despite this, if the tank is representative of an individual grid cell at the field scale, this simple breakthrough curve gives a representative description of behavior at the grid scale. 4.2. Bromide transport As with uranium, local bromide behaviors were quite varied, but the global effluent breakthrough curve was non-Fickian with an early breakthrough and long tail. This suggests a smoothing of behaviors caused by local physical heterogeneities to create a simple global response. Because of the dense spatial characterization within the flow domain, part of the non-Fickian inert

0.80 0.37 PV

0.70 1.06 PV

0.60

Silicon (mM)

2.05 PV

0.50 2.98 PV

0.40 3.94 PV

0.30 Chlorite Equil.

0.20 Influent through 1.26

0.10 Influent after 1.26 PV

0.00 7.10

7.60

8.10

pH Fig. 9. Correlation plot of silicon as a function of pH. Each point represents the results from a single well for the number of pore volumes indicated. Also shown are the influent concentrations before and after 1.26 PV, as well as the solubility limit for silicon assuming equilibrium with chlorite (Mg5Al2Si3O10(OH)8).

60

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

12.0

Uranium Conc. (µM)

10.0

8.0

6.0

4.0

2.0

0.0 0.0

1.0

2.0

3.0

4.0

5.0

Number of pore volumes eluted from the tank Fig. 10. Uranium breakthrough curve as a function of pore volumes.

transport can be directly attributed to the physical heterogeneity packed into the tank. The bromide behavior is consistent with a dual domain response with the b 2 mm sediment acting as the diffusion dominated domain; five of the six wells that still had bromide present when the effluent value was non-detectable are located in the b2 mm sediment. Other intermediate scale experiments have exhibited far more Fickian inert transport behaviors. Using variously sized lab sands, Barth et al., 2001a packed a 2-D heterogeneous system. In this system, a breakthrough curve constructed from within the flow domain had no tailing but a pronounced shoulder. Likewise, Istok et al. (1999) used a homogenous packing of field sediment, and also observed Fickian bromide behavior at several points within the flow domain. Contrasting with what was seen in this work, at many of the points within the flow domain, the tracer response ranged from fairly Fickian to completely non-Fickian. In this work the elapsed time from bromide injection to non-detect effluent values was approximately six months. In other intermediate scale work the total elapsed time

of a single replicate ranges from 5 h (Levy and Berkowitz, 2003) to 7 days (Barth et al., 2001a,b; Levy and Berkowitz, 2003). One possible explanation of the Fickian responses is that larger water fluxes limit the extent tracers can diffuse into lower flow zones. In the slower water fluxes used in this experiment, the time scale of advection approaches the time scale of diffusion allowing the tracer to diffuse into low flow zones, leading to the apparent dual domain behavior both within the flow domain and in the effluent. Alternatively, it could also be a more fundamental difference between the sediments used in these experiments compared to the lab sands more commonly used. 4.3. System geochemistry In examining Figs. 3 through 5 the first observation is that within a water flowpath of only about 2 m, there are still considerable fluctuations in local pH values and uranium concentrations both spatially and temporally. Some fluctuations

Uranium Conc. (micromolar)

25.0

20.0

2 15.0

10 14 22

10.0

29 41 5.0

0.0 0.0

42

1.0

2.0

3.0

4.0

5.0

# Pore Volumes Fig. 11. Uranium concentrations as a function of total pore volumes eluted for a selected set of wells. The wells selected are all located in b2 mm composite material.

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

61

3

Bromide Conc. (mg/L)

2.5 2 1.5 1 0.5 0 0

0.5

1

1.5

2

2.5

3

# Pore volumes since injection Fig. 12. Bromide breakthrough curve. Samples removed from effluent end of the tank.

are clearly linked to the physical heterogeneity induced in the tank by particle size placement. This is the most pronounced in comparing spatial plots for both uranium and pH where y~28.0 cm (Figs. 3C, 4C, and 5C) with Fig. 2C. A sharp vertical (perpendicular to flow) gradient occurs in both pH and uranium where the 4-12 mm material is placed. Since the upgradient uranium concentrations are b 0.5 μM, the source of the uranium is most likely from adjacent cross sections where the uranium concentrations are much higher. Thus local to these sharp gradients both uranium diffusion/advection from lower to higher flow zones, as well as a significant amount of mixing within the 4–12 mm material appears to be occurring. There is also a clear correlation between uranium concentration gradients and physical heterogeneity in the slice where y~39.1 cm (Figs. 3D, 4D, and 5D). In this slice (compared to Fig. 2D), the uranium concentrations are higher in areas with anticipated larger flow velocities. Specifically, flow lines from the premodeling show water movement through the upgradient 4–12 mm section, down through the b2 mm composite and under the 0.12– 0.250 mm fraction into the downgradient 4–12 mm section and 8.0

Bromide Conc. (mg/L)

7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 0.0

0.5

1.0

1.5

2.0

2.5

3.0

Number of pore volumes since injection Well #7

Well #19

Well #32

Fig. 13. Bromide breakthrough curves for three wells, plotted as a function of pore volumes eluted since bromide injection. The lines connecting the points are presented only as visual aids, and are not connected to any quantitative model.

then out of the tank. This preferential flow line has higher uranium concentrations in the y~39.1 cm kriged plots. Uranium desorption is known to be sensitive to many chemical variables including uranium concentration, pH, alkalinity, and mineral type. These variables also played a role in the observed distributions of uranium in the tank. The importance of this role is seen clearly in the correlation plots, Figs. 6 through 8. For many cations, sorption to a mineral surface decreases with pH through both site competition and charge repulsion. This same behavior was seen throughout the tank. In areas where the pH is low, uranium concentrations were relatively high. When the pH/U relationship was well described by a line (r2 ≥0.7, from 1.61 PV through 3.94 PV) and omitting the data from well #10, the x-intercept of linear data fits ranged from 7.70 to 7.86; above this pH there should be near 100% uranium sorption for this system. For samples with a pH above this range, the measured uranium values were among the lowest measured (0.033– 0.122 μM, n=9). In the literature, uranium sorption behavior in this pH range is mixed depending on the minerals involved and other experimental conditions. For many iron based mineral phases, as the pH increases from 7 to 8 there is a general decrease in the amount of sorption when CO2 is present (Ho and Miller, 1986; Hsi and Langmuir, 1985; Waite et al., 1994) and also when Ca and CO2 are present (Fox et al., 2006). In certain iron mineral cases uranium sorption remains at 100% sorbed for this pH range (Lenhart and Honeyman, 1999). For many silicon based minerals in single mineral systems there is also a general decrease in uranium sorption from pH 7 to 8 when CO2 is present (quartz, Fox et al., 2006; quartz, chlorite, muscovite, and albite, Arnold et al., 1998, 2001; and Schmeide et al., 2000; quartz and clinoptilolite, Prikryl et al., 2001). In mixed mineral and field based mineral systems the data is mixed ranging from a marked decrease in uranium sorption over this pH range (Barnett et al., 2002), to no changes or a slight decrease in uranium sorption (Zheng et al., 2003), to a marked increase in uranium sorption (Benes et al., 1998; Dong et al., 2005). The degree that the sorption behavior changes is often dependent on specific conditions such as Ca concentrations, partial pressures of CO2, and soil:solution ratios. However, interpreting the tank data in terms of equilibrium, pH based desorption edge, the data is most in agreement with Dong et al.,

62

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

2005. Both Barnett et al. (2002) and Dong et al. (2005) used silty–clayey sediments similar to those used in the tank. However, Barnett et al. (2002) used sediments largely devoid of calcite while Dong et al. (2005) used sediments with and without added calcite. There is calcite in the sediments used in the tank and evidence of calcite dissolution, so it is not surprising the tank data is phenomenologically consistent with the batch data given in Dong et al. (2005). Fig. 9 shows the pH–silicon relationship as a function of time. As opposed to the pH–U and pH–Ca relationships, the pH–Si relationship is fairly constant over the entire time span studied. The influent concentrations changed as the AGW needed to be replenished halfway through the experiment, and it was difficult to get an exact dissolved silicon concentration. However, the measured values in the tank were always less than in the AGW; silicon is being removed from solution. On the plot, points above the quasi-linear relationship are generally from the most up-gradient wells (wells #1, 4, 26, 36, 41). Points comprising the quasi-linear relationship are from more down gradient wells. The reaction path in Fig. 9 is from the upper right to the lower left in terms of silicon removal and pH suppression. The abrupt halt in the reaction path is indicative of a solubility control; however, as previously mentioned, there is no well characterized silicon bearing mineral phase whose silicon concentrations match the observed silicon–pH measurements. The reason for this relationship remains unclear, but it would appear that: 1. either silicon removal reactions are the overall control on pH in the tank or pH changes are forcing silicon removal from solution, 2. there is some sort of kinetic hindrance to this removal as the relationship becomes more consistent with increasing water–sediment contact time, and 3. this relationship is constant over the experimental timeframe. For both silicon and uranium the observed behavior appears to be beyond what could be described using only thermodynamic considerations. Kinetic relationships appear to be influencing uranium desorption and silicon removal. In transport studies with comparable groundwater velocities kinetic formulations limiting the rate of desorption have been necessary to describe the uranium data (Kohler et al., 1996; Qafoku et al., 2005), but local quantitation of bulk ions was not possible. 5. Inter-tank comparisons In this pair of papers three different experiments were completed encompassing three levels of heterogeneity examining uranium migration in saturated groundwater systems. In this section, similar data from all three tanks is plotted to explicitly examine the similarities/differences between behaviors in the tanks. The goal of this analysis is to compare the differences based on the known decimeter scale macroscopic heterogeneities placed in the tanks. This comparison ignores smaller scale processes which may be occurring. In this sense, tank #1 is the most homogenous (1 particle size, uniformly packed), tank #2 is more heterogeneous (2 particle sizes, non-uniformly packed), and tank #3 is the most heterogeneous (5 particle sizes, non-uniformly packed). In terms of solid phase chemical heterogeneities, tanks #1 and #2 are similar (subsets of a single size fraction with identical mass ratios were used with different packing orientations, Miller et al., 2013). Tank #3 introduces more chemical heterogeneity with the addition of two particle sizes not used in the previous tanks.

Furthermore, the sediment mass ratios, and subsequently the initial uranium distribution, in the third tank were distinct and non-uniform relative to the first two tanks. In terms of aqueous heterogeneity, tank #2 was the only one to have a temporally variable influent composition. Fig. 14 shows uranium breakthrough curves for all three tanks. The largest differences appear to occur within the first two pore volumes of elution. Tank #1 starts at an intermediate concentration and declines along an exponential decay curve. Tank #2 has the highest initial concentration and declines the most sharply. Tank #3 starts at a very low value, increases to an intermediate value, and then declines nearly linearly when plotted on this scale. However, after two pore volumes, the differences in the tank behaviors approach a minimum. Tailing behavior for all three tanks occurs at a uranium concentration of about 2 μM, although this behavior is dependent on the influent chemistry (for tank #2 the AGW-2% was being injected from 2.4 to 3.7 PV). Fig. 15 shows the cumulative mass fluxes after being normalized to retention time and the amount of sediment added to the tanks. Plotted in log–log space, linear cumulative mass fluxes imply mixing limitations on uranium transport. Tanks #1 and 2 appear to be mixing limited when AGW-atmospheric is being injected. When the influent was switched to AGW-2% in tank #2, the slope changes, but the relationship remains linear. Tank #3 is less linear in log–log space, so despite a slower average groundwater velocity processes beyond mixing may be controlling uranium output. Fig. 16 shows the normalized breakthrough curves for bromide in each of the tanks. Similar to tank #3, pulse bromide injections were also completed in tanks #1 and #2. In both cases a Br− concentration of 2.02×10−3 M was used. In tank #1 the injection occurred over a period of 13.25 h and a total of 2.52×10−3 mol (201 mg) of Br− were injected. In tank #2 the injection lasted 18.25 h and a total of 1.91×10−3 mol (153 mg) of Br− were injected. The mass recoveries were 79% and 85% for tanks #1 and #2, respectively. Tank #1 is quite ideal with a large Cmax and no tailing, although breakthrough is somewhat earlier than expected. Tank #2 shows non-Fickian behavior with an even earlier breakthrough, lower Cmax, and prolonged tailing. Tank #3 shows behavior between the first two with a slightly later breakthrough relative to tank #2, a Cmax occurring at about the same number of pore volumes as tank #1, but a tail matching the length and relative concentration of tank #2. Since only physical heterogeneities need to be considered, the increasing heterogeneity from tank #1 to tank #3 again leads to relatively simplified behavior. Tank #1 acts as a homogenous control. Tank #2 has slightly more physical heterogeneity and has a breakthrough curve consistent with what is seen at other experimental scales with increasing heterogeneity. But then, tank #3, the most heterogeneous system, exhibits breakthrough behavior between tanks #1 and #2. Thus the addition of more heterogeneity (size, number and orientation of size fractions, dimensionality) does not make for more complex tracer breakthrough behavior; instead, for these systems, the homogenous tank #1 and slightly heterogeneous tank #2 act as end members, and the most heterogeneous system has attributes of both. Besides comparing effluent values, spatial sample analyses from within the flow domain of each tank can also be compared. Table 2 shows the range of values for a selected set of dates for all

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

63

14.0

Uranium Conc. (µM)

12.0 10.0 8.0 6.0 4.0 2.0 0.0 0.0

2.0

4.0

6.0

8.0

# Pore Volumes Tank #1

Tank #2

Tank #3

Fig. 14. Uranium breakthrough curves for all three tanks. Samples removed from the effluent line of each individual tank.

three tanks. The range of pH values changes with increasing heterogeneities (tank #1 is the smallest, tank #3 is the largest), the range of alkalinity values is not remarkably different between the tanks, while the range of ion concentrations in tank #1 are 2.4–7.5 times greater than in tanks #2 and #3. These relationships are shown as correlation plots in Fig. 17. All of the constituents are plotted as a function of pH with uranium being shown in Fig. 17A, alkalinity in Fig. 17B, calcium in Fig. 17C and silicon in Fig. 17D. In Fig. 17A, the uranium–pH correlation is the strongest for tanks #2 and #3, and it is weaker for tank #1. Tank #1 has the widest range and the weakest relationship between these two variables. Tank #2 has a narrower range and tighter correlation, and tank #3 again falls between these two end members. The same relationship between the tanks is seen in Fig. 17B correlating alkalinity and pH. Here, tank #3 has the tightest correlation, while tanks #1 and #2 have weaker

correlations, and much of tank #3 overlaps considerably with tank #1. In stark contrast to this behavior is the calcium-pH relationship shown in Fig. 17C. In Fig. 17C, tanks #2 and #3 are practically indistinguishable with exceedingly strong correlations, while tank #1 has no real correlation between these variables. This relationship implies a solubility control on calcium in tanks #2 and #3 which is not occurring in tank #1. Speciation calculations showed calcite supersaturation in tank #1 over the majority of the space and time domains. Also, although the average retention time within the tanks is slightly different, the samples shown represent the entire flow domain. Each individual point has a unique water–sediment contact time, and these times overlap between the three tanks. Therefore, assuming calcite is uniformly distributed in the tank to act as nucleation sites, it would appear there is some other barrier to precipitation in tank #1 that is not present in tanks #2 and #3.

1.0E+00

Cumulative Uranium Flux (µmol U/kg Sed./tr(days)

1.0E-01

1.0E-01

1.0E-02

1.0E-02 1.0E+00

1.0E+01

1.0E-03

1.0E-04 1.0E-03

1.0E-02

1.0E-01

1.0E+00

1.0E+01

# Pore Volumes Tank #1

Tank #2

Tank #3

Fig. 15. Normalized uranium fluxes from all three tanks. The inset shows tank 2 alone with a specific focus around the time of the AGW-2% injection. The space between the red arrows denotes the approximate time when AGW-2% was being injected into tank 2.

64

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

Normalized Br- Conc. (C/C0)

0.06 0.05 0.04 0.03 0.02 0.01 0 0

0.5

1

1.5

2

2.5

3

# Pore Volumes Since Injection Tank #1

Tank #2

Tank #3

Fig. 16. Normalized bromide breakthrough curves for all three tanks as a function of pore volumes since the bromide was injected. Samples removed from effluent line of the individual tanks.

Fig. 17D shows the silicon–pH relationship between the tanks. Just as with calcium, tanks #2 and #3 plot in very similar space with tight correlations, while tank #1 is distinctly different with no correlation. This behavior is even more interesting because the steady-state relationship seen in tanks #2 and #3 are approached from both high and low silicon concentrations. In tank #2 there was no silicon in the influent, while in tank #3 the influent concentration was either 0.472 mM or 0.735 mM. Thus in tank #2 the net flux of silicon is from the solid phase to the aqueous phase, while in tank #3 the net flux of silicon is from the aqueous phase to the solid phase. In both tanks the pH is higher at the upgradient end. Therefore the reaction path moves from lower right to upper left for tank #2, and from upper right to lower left for tank #3. This is indicative of a solubility control, but the mineral controlling solubility has yet to be identified. The silicon–pH relationship is distinctly different and more complex than the calcium-pH relationship. Weathering of clay and other silicates is known to proceed at very slow rates which are difficult to measure in laboratory conditions (Kohler et al., 2005). Precipitating clay minerals can also control the dissolution of primary minerals in the same system (Maher et al., 2009; Zhu, 2005). In this case the kinetics associated with this dissolution/ precipitation appear fairly similar and relatively fast; in both Table 2 Minimum, maximum, and the range of values for master variables within all three tanks. Data from the following number of pore volumes were used: Tank 1: 0.77, 2.71, 4.09, 4.55, Tank 2: 1.07, 1.83, 3.10, 3.95, 5.86, and Tank 3: 1.06, 2.05, 2.98, 3.94. This table corresponds to the data shown in Fig. 17A–D.

Tank #1

Tank #2

Tank #3

Minimum Maximum Difference Minimum Maximum Difference Minimum Maximum Difference

pH

Alkalinity (meq/L)

Uranium (μM)

Calcium (mM)

Silicon (mM)

7.494 8.004 0.510 7.066 7.778 0.712 7.203 8.052 0.849

0.77 3.07 2.30 0.89 2.91 2.02 0.39 3.43 3.04

0.277 25.403 25.126 0.009 14.562 14.554 0.033 10.584 10.550

5.647 40.103 34.455 5.607 10.229 4.622 4.836 10.859 6.023

0.092 1.164 1.071 0.027 0.583 0.556 0.172 0.573 0.401

tanks #2 and #3 the apparent equilibrium was reached in the upgradient half of the tank. The equilibration timeframe weakens the solubility control argument, and instead suggests a steady state relationship of some sort. What ultimately controls this steady state is unknown. In terms of tracer behavior there was no clear indication of either precipitative clogging of pore spaces in tank #3 (Tartakovsky et al., 2008) or of an increase in pore space due to dissolution in tank #2. Although, measurement of such processes using the experimental apparatus described may not be particularly feasible. Understanding the dramatic differences between calcium and silicon behavior may be a critical first step in determining the extent of the scaling effect in these systems. It is likely that mineral phases involving one or both of these elements are controlling the pH and alkalinity distributions in the tanks. In turn, the pH and alkalinity variations are controlling the release and transport of uranium. 6. Conclusions Do more heterogeneous systems exhibit more complex behavior? The evidence from these systems is not conclusive; for certain constituents, emergent properties appear evident, for others, they do not. All three tank experiments can encompass more heterogeneities than any column scale experiment, and yet the general breakthrough curves are not demonstrably different in terms of major features. Although they do not perfectly co-locate, all three tanks exhibit an early peak in uranium concentration, a quick decrease to a much lower tailing concentration, and a prolonged tailing period. Also, using bromide as an inert tracer, the breakthrough curves were not demonstrably different than column scale breakthrough curves. The bromide curves ranged in behavior from Fickian (tank #1) to non-Fickian (tanks #2 and #3); this change in inert transport is attributable to the known heterogeneities in each tank experiment. When considering chemical relationships within the flow domain, many more similarities were observed. Local relationships between pH, alkalinity, and uranium were all fairly linear and consistent with

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

65

Fig. 17. Correlation plots for A — Uranium, B — Alkalinity, C — Calcium, and D — Silicon versus pH. Tank 1 is represented by the x symbols, tank 2 the + symbols and tank 3 by the ¤ symbols. Points are the water chemistry results from spatial samples for all three tanks. The samples were taken from within the flow domain. Timepoints included are: Tank 1: 0.77 PV, 2.71 PV, 4.09 PV, 4.55 PV, Tank 2: 1.07 PV, 1.83 PV, 3.10 PV, 3.95 PV, 5.86 PV, Tank 3: 1.06 PV, 2.05 PV, 2.98 PV, 3.94 PV.

thermodynamic arguments which have been observed at smaller scales. These relationships are phenomenologically similar to relationships observed at much larger scales. If the tanks are considered as a series of experiments with increasing heterogeneity, other similarities emerge. Increasing the physical heterogeneities correlates with chemical equilibrium of dissolved calcium with calcite; similar increases in complexity correlate to an observed aqueous silicon steady state relationship. All of these observations point to potential simplifications that can be made to uranium transport simulations which may ignore reactions known to contribute to transport behavior at smaller scales. However, in many ways the increase in complexity also led to disparate behaviors. There was a significant scale discrepancy in using an equilibrium, batch derived surface complexation model to describe uranium concentrations in tanks #1 and #2. The model generally over calculated the observed uranium concentrations. Also, within the flow domain there were apparent kinetic variations in major ion behavior. In tank #1 the U–pH relationship was always linear, but the slope of the line changed with time. In tank #2, the U–pH relationship appeared constant as a function of time. While in tank #3, the relationship ranged from very weak at early timepoints to increasingly linear as time proceeded. Si–pH relationships were simplified with increasing heterogeneity; tanks #2 and #3 had almost identical behavior. However, the behavior is not attributable to the solubility of a known silicon bearing phase, and the rates of dissolution/precipitation appear to be considerably faster than reported elsewhere for similar mineral classes. In tank #3, both uranium and bromide breakthrough

curves, when considered on a well by well basis, showed a considerable range of behaviors which are not particularly attributable to local, known heterogeneities. The study of larger, heterogeneous systems allows for the interplay of variables which are unknowable using reductionist experimental techniques. These variables can enhance, negate, or have no effect on each other. In contaminated subsurface systems, the interactions span the physical and chemical domains, as well as a range of scales not generally considered in flow and inert transport. The concept of upscaling from the atom to the field scales using deterministic modeling is technically and pragmatically impossible using current computing systems and technologies. Another approach is to use complex experimental systems to determine similarities and differences as a function of scale. These similarities and differences can be used to create scalable conceptual models which will simplify non-critical fundamental behaviors while including representations of critical behaviors. The completion of column scale experiments limits the total complexity possible in a given system, and also limits the scale of chemical and physical processes incorporated into an experimental system. Thus intermediate scale systems are necessary to allow for the inclusion of effects caused by larger scale chemical and physical processes. To generalize the results presented here, much more experimental data is needed at the intermediate scale including: different mineral systems, different metals/radionuclides of interest, different levels of physical heterogeneity including geologically relevant tank packings, and different flow rates. As was seen in the experiments presented here, it is expected that in other complex systems,

66

A.W. Miller et al. / Journal of Contaminant Hydrology 148 (2013) 51–66

metal/radionuclide behavior will be somewhat simplified as smaller scale processes are integrated over larger travel distances and times. Acknowledgments This material is based upon work supported by the Department of Energy under Award Number: DE-FG0206ER64233. This work would not have been possible without ongoing collaborative efforts from James Davis, Gary Curtis, Matthias Kohler, and Carl Steefel. Field sediment collection and preparation over nine days in 35–40 °C heat would not have been possible without: Patricia Fox, Jennifer Joye, Kelly Johnson, Linda Figueroa, Ana Ruiz, and Jason Deardorff. Tank construction, sediment sieving and tank sampling were greatly aided by Emily Lesher, Seth Davis, Jason Peterson, and Andrea Koenig. Joern Larson performed all of the alkalinity titrations. Also, the manuscript benefited greatly due to a tremendously in-depth and unambiguous review by Michael Hay, and three anonymous reviewers. Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.jconhyd.2012.12.011. References Arnold, T., Zorn, T., Bernhard, G., Nitsche, H., 1998. Sorption of uranium(VI) onto phyllite. Chemical Geology 151, 129–141. Arnold, T., Zorn, T., Zanker, H., Bernhard, G., Nitsche, H., 2001. Sorption behavior of U(VI) of phyllite: experiments and modeling. Journal of Contaminant Hydrology 47, 219–231. Barnett, M.O., Jardine, P.M., Brooks, S.C., 2002. U(VI) adsorption to heterogeneous subsurface media: application of a surface complexation model. Environmental Science and Technology 36 (5), 937–942. Barth, G.R., Illangasekare, T.H., Hill, M.C., Rajaram, H., 2001a. A new tracerdensity criterion for heterogeneous porous media. Water Resources Research 37, 21–31. Barth, G.R., Hill, M.C., Illangasekare, T.H., Rajaram, H., 2001b. Predictive modeling of flow and transport in a two-dimensional intermediatescale, heterogeneous porous medium. Water Resources Research 37 (10), 2503–2512. Benes, P., Kratzer, K., Vlvkova, S., Sebestova, E., 1998. Adsorption of uranium on clay and the effect of humic substances. Radiochimica Acta 82, 367–373. Davis, J.A., Curtis, G.P., 2003. Application of Surface Complexation Modeling to Describe Uranium (VI) Adsorption and Retardation at the Uranium Mill Tailings Site at Naturita, Colorado. U.S. Nuclear Regulatory Commission, Washington D.C.(NUREG/CR-6820). Dong, Wenming, Ball, William P., Liu, Chongxuan, Wang, Zheming, Stone, Alan T., Bai, Jing, Zachara, John M., 2005. Influence of calcite and dissolved calcium on uranium(VI) sorption to a Hanford subsurface sediment. Environmental Science and Technology 39 (20), 7949–7955. Fox, P., Davis, J.A., Zachara, J.M., 2006. The effect of calcium on aqueous uranium(VI) speciation and adsorption to ferrihydrite and quartz. Geochimica et Cosmochimica Acta 70, 1379–1387.

Ho, C.H., Miller, N.H., 1986. Adsorption of uranyl species from bicarbonate solution onto hematite particles. Journal of Colloid and Interface Science 110 (1), 162–171. Hsi, Ching-Kuo Daniel, Langmuir, Donald, 1985. Adsorption of uranyl onto ferric oxyhydroxides: application of the surface complexation sitebinding model. Geochimica et Cosmochimica Acta 49, 1931–1941. Istok, Jonathan D., Amonette, J.E., Cole, C.R., Fruchter, J.S., Humphrey, M.D., Szecsody, James E., Teel, S.S., Vermeul, V.R., Williams, M.D., Yabusaki, Steven, 1999. In situ redox manipulation by dithionite injection: intermediate-scale laboratory experiments. Ground Water 37 (6), 884–889. Kohler, M., Curtis, G.P., Kent, D.B., Davis, J.A., 1996. Experimental investigation and modeling of uranium(VI) transport under variable chemical conditions. Water Resources Research 32, 3539–3551. Kohler, M., Curtis, G.P., Meece, D.E., Davis, J.A., 2004. Methods for estimating adsorbed uranium(VI) and distribution coefficients of contaminated sediments. Environmental Science and Technology 38, 240–247. Kohler, S.J., Bosbach, D., Oelkers, E.H., 2005. Do clay mineral dissolution rates reach steady state? Geochimica et Cosmochimica Acta 69 (8), 1997–2006. Lenhart, John J., Honeyman, Bruce D., 1999. Uranium(VI) sorption to hematite in the presence of humic acid. Geochimica et Cosmochimica Acta 63 (19/20), 2891–2901. Levy, M., Berkowitz, B., 2003. Measurement and analysis of non-Fickian dispersion in heterogeneous porous media. Journal of Contaminant Hydrology 64, 203–226. Liu, C., Zachara, J.M., Qafoku, N.P., Wang, Z., 2008. Scale-dependent desorption of uranium from contaminated subsurface environments. Water Resources Research 44, W08413 http://dx.doi.org/10.1029/2007WR006478. Maher, K., Steefel, C.I., White, A.F., Stonestrom, D.A., 2009. The role of reaction affinity and secondary minerals in regulating chemical weathering rates at the Santa Cruz Soil Chronosequence, California. Geochimica et Cosmochimica Acta 73, 2804–2831. Miller, A.W., 2010. Ph.D. Thesis. Homogenized Behavior from increasingly heterogeneous systems: uranium transport experiments at the intermediate scale. Colorado School of Mines, Golden, CO. Miller, A.W., Rodriguez, D.R., Honeyman, B.D., 2013. Simplified behaviors from increased heterogeneity: I. 2-D uranium transport experiments at the decimeter scale. Journal of Contaminant Hydrology 148, 39–50. Prikryl, J.D., Jain, A., Turner, D.R., Pabalan, R.T., 2001. Uranium(VI) sorption behavior on silicate mineral mixtures. Journal of Contaminant Hydrology 47, 241–253. Qafoku, N.P., Zachara, J.M., Liu, C., Gassman, P.L., Qafoku, O.S., Smith, S.C., 2005. Kinetic desorption and sorption of U(VI) during reactive transport in a contaminated Hanford sediment. Environmental Science and Technology 39 (9), 3157–3165. Schmeide, K., Pompe, S., Bubner, M., Heise, K.H., Bernhard, G., Nitsche, H., 2000. Uranium(VI) sorption onto phyllite and selected minerals in the presence of humic acid. Radiochimica Acta 88, 723–728. Szecsody, J.E., Zachara, J.M., Chilakapati, A., Jardine, P.M., Ferrency, A.S., 1998. Importance of flow and particle-scale heterogeneity on Co(II/III) EDTA reactive transport. Journal of Hydrology 209, 112–136. Tartakovsky, A.M., Redden, G., Lichtner, P.C., Scheibe, T.D., Meakin, P., 2008. Mixing-induced precipitation: experimental study and multiscale numerical analysis. Water Resources Research 44, W06S04 http://dx.doi.org/10.1029/ 2006WR005725. Waite, T.D., Davis, J.A., Payne, T.E., Waychunas, G.A., Xu, N., 1994. Uranium(VI) adsorption to ferrihydrite: application of a surface complexation model. Geochimica et Cosmochimica Acta 58 (24), 5465–5478. Zheng, Zuoping, Tokunaga, Tetsu K., Wan, Jiamin, 2003. Influence of calcium carbonate on U(VI) sorption to soils. Environmental Science and Technology 37 (24), 5603–5608. Zhu, C., 2005. In situ feldspar dissolution rates in an aquifer. Geochimica et Cosmochimica Acta 69 (6), 1435–1453.