Abstract

In situ migration experiments using different radiotracers have been performed in the HADES Underground Research Facility (URF), built at a depth of 225 m in the Boom Clay formation below the SCK–CEN nuclear site at Mol (Belgium). Small-scale experiments, mimicking laboratory experiments, were carried out with strongly retarded tracers (strontium, caesium, europium, americium and technetium). Contrary to europium, americium and technetium which are subjected to colloid mediated transport, the transport of strontium and caesium can be described by the classic diffusion retardation formalism. For these last two tracers, the transport parameters derived from the in situ experiments can be compared with the laboratory-derived values. For both tracers, the apparent diffusion coefficients measured in the in situ experiments agree well with the laboratory-derived values.

In the large-scale experiments (of the order of metres) performed in the URF, non-retarded or slightly retarded tracers (HTO, iodide and HC314) were used. The migration behaviour of these tracers was predicted based on models applied in performance assessment calculations (classic diffusion retardation) using migration parameter values measured in laboratory experiments. These blind predictions of large-scale experiments agree well in general with the experimental measurements. Fitting the experimental in situ data leads to apparent diffusion coefficients close to those determined by the laboratory experiments. The iodide and HC314 data were fitted with a simple analytical expression, and the HTO data were additionally fitted numerically with COMSOL multiphysics, leading to about the same optimal values.

The Belgian Nuclear Research Centre (SCK–CEN) and the Belgian Agency for Radioactive Waste and Enriched Fissile Materials (ONDRAF/NIRAS) pioneered studying the geological disposal of high-level nuclear waste in clay formations. Boom Clay has been studied as a potential host formation and is present under the facilities of SCK–CEN in Mol at a depth between approximately 190 m and 290 m. Due to its very low hydraulic conductivity (around 2.5 × 10–12 m/s; Aertsens et al., 2004; Wemaere et al., 2008), self-sealing properties (Bernier & Bastiaens, 2004; Bastiaens et al., 2007) and a strong sorption capacity for many radionuclides (De Cannière et al., 1996), Boom Clay is envisaged as a potential host formation for nuclear waste disposal in Belgium. The radiological consequences of the release of hazardous radionuclides from the disposed waste are estimated by performance assessment studies (e.g. Marivoet & Weetjens, 2009) for which the transport parameters in Boom Clay need to be known. These are mainly determined in laboratory experiments using clay cores sampled in the vicinity of the URL. In order to demonstrate that they are also applicable to larger scales of space and time, in situ experiments were performed.

In 1980 the construction of the HADES Underground Research Facility (URF) within the Putte Member of the Boom Clay formation (Aertsens et al., 2004) was started. Since then, HADES has been expanded several times (Fig. 1). The facility allows development of appropriate techniques for constructing a repository, waste emplacement (and possible retrieval), construction of engineered barriers, as well as in situ investigations of the disposal environment (geochemistry, hydromechanics, radionuclide behaviour, etc.). In situ experiments have the advantage that they represent the disposal conditions as closely as possible. They can be used to confirm behaviour on a large scale and over long durations (Wersin et al., 2008) using laboratory experiments (Aertsens et al., 2008a) as demonstrations.

This paper provides an overview of the in situ migration experiments performed at HADES, starting by describing the different types of experiments and the corresponding models, followed by the presentation of the modeling results. The transport parameters values derived by laboratory experiments are compared to those obtained from the in situ experiments.

Types of Experiments

Two types of experiments were performed. The first type concerns retarded tracers. The experiments were performed on the same small scale as laboratory experiments (of the order of centimetres). This type of experiment is merely used for a long-term demonstration. In Fig. 2, a scheme for the experimental set-up is given. An aliquot of strongly retarded tracer, homogenously distributed by spiking on a filter paper or directly on the clay surface, was sandwiched between two back-to-back clay cores. The clay cores were confined in an open-ended diffusion cell mounted on a hollow cylindrical tube (piezometer) installed within the clay surrounding HADES. The open end of the diffusion cell was equipped with a small tube enabling water sampling. This allows collection of the water seeping out of the clay for a radiological assay. After retrieval from the borehole (by overcoring), the tracer profile in the clay cores was determined. This type of experiment is similar to a laboratory percolation experiment (Aertsens et al., 2008a) but, in the in situ version, advection is not provided by an externally applied pressure but is caused by the pressure difference between the water pressure at the depth of the URF and the atmospheric pressure (being the pressure in the tunnel of the URF). The tracers used in these experiments were europium, strontium, caesium, americium and technetium. The measurement error (and, consequently, also the error in the mass balance) is below 10%.

The second type of experiment concerns non-retarded or slightly retarded tracers and was conducted on a large scale (of the order of metres). This type of experiment was also used for testing blind predictions. It consists of one or more multi filter piezometers installed in HADES. After drilling a borehole, the piezometer was inserted. Nitrogen gas was flushed through the filters to prevent oxidation of the clay. Due to the plastic behaviour of the clay, convergence is fast and, after several days, the remaining gap between the piezometer and the borehole wall was completely sealed, preventing the existence of fast transport paths due to voids along the piezometer.

The filters allow water exchange with the Boom Clay and are connected by feeding lines to and from the URF. These feeding lines allow initial injection of tracer into a central filter, as well as permitting regular sampling of the pore water in all filters and measurement of its tracer concentration. The average time between consecutive samplings is a compromise between having enough measurements and avoiding the fact that the filters act as a sink for the water flow around the URF.

Three different experimental set-ups are used. In the CP1 experiment (Put et al., 1993), HTO is injected in the central filter of a horizontal piezometer (Fig. 3). The distance between the centres of consecutive filters is one metre. The Tribicarb-3D experiment (tracers HTO and HC314) involves three piezometers, of which two are horizontal and one is inclined (Fig. 4 and Table 1). Each piezometer comprises four to eight filters. The TD41HV set-up consists of a vertical and a horizontal piezometer (Fig. 5), each comprising nine filters. This set-up was used in two consecutive experiments: the first with iodide, the second with 14C labelled Natural Organic Matter. Due to the limited half life of 125I (60 days), the source concentration after three years is below detection limit. Consequently, the experiment was stopped after that time. The remaining experiments are still on-going (some for more than 20 years).

Measurement errors are typically around 15% at low tracer concentrations, decreasing to 5% at higher concentrations. In clay water, HC314 is unstable and tends to escape the clay water after transforming to the gaseous form of 14CO2. In order to prevent the loss of 14CO2 due to evaporation (which leads to erroneous measurements), the HC314 solution injected in Tribicarb-3D as well as the sampled solution were stabilized with sodium hydroxide (NaOH). In the Tribicarb-3D experiment, the activity of both isotopes HTO and HC314 is determined by liquid scintillation counting. The activity spectra of both isotopes overlap, causing problems when the activity of one isotope dominates the other. Initially, HC314 concentrations are low (lower than 1000 Bq/l) and the HTO activity is one to two orders of magnitude higher, due to which the HC314 activity measurements cannot be considered as very reliable.

Modelling

Transport of a decaying tracer in porous media is described by solutions of the dispersion/advection equation (Crank, 1975) complemented with a decay term:  
Ct=Dappi2Cx2VappCxλC
(1)
with C(x,t) the tracer concentration in the pore water (Bq/m3), t time (s), Dappi the apparent dispersion coefficient (m2/s), x position (m), Vapp the apparent tracer velocity (m/s) and λ the decay constant (1/s). Recent reviews of transport in clays are in (Bourg, 2004; Altmann et al., 2012).

Modelling the small-scale in situ percolation experiments (retarded tracers)

All tracer is initially in the source (at x = 0) which is approximated to be infinitely thin. If zero or only a minor fraction of the tracer has reached the inlet/outlet of a clay core, the tracer distribution C(x,t) in the clay pore water is given by the standard expression (Crank, 1975)  
C(x,t)=Q0SηRC1(x,t,Dappi,Vapp)exp(λt)
(2)
with Q0 the injected tracer quantity (Bq), S the cross section of the clay cores (m2), η the diffusion accessible porosity (dimensionless), R the retardation factor R (dimensionless), and  
C1(x,t,Dappi,Vapp)=12πDappitexp((xVappt)24Dappit)
(3)

The product ηR of the diffusion accessible porosity η (dimensionless) and the retardation factor R (dimensionless) is the ratio VDarcy/Vapp of the Darcy velocity VDarcy (m/s) and the apparent velocity Vapp. Because the apparent velocity is determined by the shift of the peak position of the tracer profile (equation 3), which is nearly zero for strongly retarded tracers, the error in the fitted value is large. This type of experiment only leads to accurate values for the apparent dispersion coefficient Diapp. The apparent velocity Vapp is sufficiently low to approximate the apparent diffusion coefficient Dapp (m2/s) by the apparent dispersion coefficient (Dapp = Diapp – αVappDiapp with α the dispersion length (m)). The tracer profile in the clay is the bulk concentration ηR C(x,t).

Introducing the notation L0 for the distance between the source and the outlet side of the clay core, the tracer quantity Qout(t) (Bq) (the decay is recalculated to zero time) percolated out of the clay core can be estimated as the amount of tracer stored in a semi-infinite clay core at positions larger than L0:  
Qout(t)=ηRSL0C(y,t)exp(λt)dy
(4)
Substituting equations 2, 3 in equation 4 leads to  
Qout(t)=Q02erfc(L0Vappt2Dappit)
(5)

In this way, neither the amount of tracer in the filter, nor the tracer in the thin tube connecting the filter to the URF (see Fig. 2), are taken into account. Because the filter and the clay cores are inside the diffusion cell at the end of the piezometer (contrary to the large-scale experiments) there is no water and tracer exchange with the clay surrounding the URF.

Modelling the large-scale in situ percolation experiments: the MICOF model

The injection filter is approximated as a bar shaped source with dimensions 2a, 2b and 2c. The origin x = y = z = 0 is chosen in the middle of this volume and the x axis is parallel to the apparent velocity. The initial condition is  
C(x)>a,y>b,z>c,t=0)=0C(x)<a,y<b,z<c,t=0)=Q0(ηR8abc)
(6)
Solving the three dimensional version of the dispersion/advection/decay equation 1 leads to the tracer concentration  
C(x,y,z,t)=18Q0ηR8abcexp(λt)(erf(xVappt+a2Dapp,xit)+erf(ax+Vappt2Dapp,xit))(erf(y+b2Dapp,yit)+erf(by2Dapp,yt))(erf(z+c2Dapp,zit)+erf(cz2Dapp,zit))
(7)
with Diapp,x, Diapp,y, Diapp,z, the apparent dispersion coefficient in the x, y and z directions. If the thickness 2a becomes very small, or at sufficiently long times, then |y0| >> a0 
erf(xVappt+a2Dapp,xit)+erf(ax+Vappt2Dapp,xit)=2πy0a0y0+a0exp(y2)dy2aπDapp,xitexp((xVappt)24Dapp,xit)
(8)
with  
y0=xVappt2Dappita0=a2Dappit
(9)
Similar rearrangements for the other directions lead to the point source approximation for the concentration (7):  
C(x,y,z,t)(Q0ηR)C1(x,t,Dapp,xi,Vapp)C1(y,t,Dapp,yi,0)C1(z,t,Dapp,zi,0)exp(λt)
(10)
with C1 given by equation 3. Comparing equations 10 and 2 shows that basically the large-scale percolation experiments can be considered as a three dimensional version of the small-scale experiments.

In the large-scale experiments, the filter dimensions (e.g. in CP1 the filter length is 8.5 cm and the diameter 4.6 cm) are much smaller than the distance between two consecutive filters (in CP1, one metre), justifying the point source approximation in all filters except the source filter. In the source filter, the filter dimensions are numerically only significant for small times (similar to the small-scale model, where the source is sufficiently small to omit its width). For the filter dimensions the approximation a = b = c is used. The remaining value a = b = c is calculated by putting the bar volume 8abc = 8a3 equal to the water volume inside a filter.

MICOF assumes that all around the URF the apparent velocity Vapp is the same. This is not correct; the water flow around the URF (and therefore also the apparent velocity) is directed towards the centre of the gallery and is not parallel at all sites in the clay. Similarly, due to mass conservation of water flow towards a cylindrical gallery, the Darcy velocity (and thus the apparent velocity) depends on the distance from the centre of the gallery; its value is not constant, as supposed in MICOF. Although not correct, the approximations just mentioned hardly influence the MICOF predictions as advection in Boom Clay is weak and transport is dominated by diffusion (Henrion et al., 1985). By applying a more sophisticated and more complex model for the large-scale experiments, where the mathematics of the model are solved numerically by the COMSOL code (Weetjens et al., 2011; Weetjens & Maes, 2013), these assumptions are avoided, but it leads numerically to nearly the same predictions as with MICOF.

The MICOF model (equation 7) has five transport parameters: the apparent dispersion coefficients Dapp,xi, Dapp,yi, Dapp,zi, the capacity factor ηR and the apparent velocity Vapp. The three apparent dispersion coefficients are assumed to be equal to the corresponding diffusion coefficients. Due to the layered structure of the clay, two of these diffusion coefficients are equal and two values remain: the apparent diffusion coefficient parallel to the bedding plane Dapp,// and the apparent diffusion coefficient perpendicular to the bedding plane Dapp,⊥. The apparent velocity Vapp is the ratio VDarcyR, and the Darcy velocity VDarcy is the product of the hydraulic conductivity and the pressure gradient around the URF (e.g. Aertsens et al., 2008b). For CP1 a value VDarcy = 6 × 10–11 m/s is used; for Tribicarb-3D and TD41HV, VDarcy is 7.6 × 10–11 m/s. Fitting the experiment, only the diffusion coefficients Dapp,// and Dapp,⊥ are varied. Due to the simple set-up of the experiment, the modelling is relatively simple (Yi et al., 2012; Naves et al., 2012; Samper et al., 2010).

The Small-Scale in situ Percolation Experiments

Experiments are carried out with strontium, caesium, europium, americium and technetium. Because europium, americium and technetium transport in Boom Clay is affected by the interaction with organic colloids (Maes et al., 2011), the classic diffusion retardation formulation cannot be used to model successfully the corresponding tracer profiles (Put et al., 1989; Noynaert, 2000; Baston et al., 2002).

Fitting the strontium tracer profile (duration of the experiment 260 days, Fig. 6) leads to an accurate value for the apparent dispersion coefficient: Dappi=7×1012 m2/s (Aertsens et al., 2009). The percolated amount of strontium (≈5 cps) is much smaller than predicted by the model (≈240 cps), using the parameter values fitted from the profile. As in Opalinus Clay (Wersin et al., 2008), the in situ measured value for the apparent dispersion coefficient is about the same as the values measured in laboratory experiments, which are: 9 × 10–12 m2/s, 6 × 10–12 m2/s, 8 × 10–12 m2/s, 7 × 10–12 m2/s and 6 × 10–12 m2/s (Aertsens et al., 2009a).

Similar to strontium, a clear profile (after a duration of 6.93 years) was observed for caesium (Fig. 7). No tracer leached out of the clay core. The profile leads to an apparent dispersion coefficient Dappi=1.0×1013 m2/s (and ηR ≈ 1100). The value of the apparent dispersion coefficient is similar to the values measured in two laboratory percolation experiments (1.9 × 10–13 m2/s and 1.6 × 10–13 m2/s; Maes et al., 2008) and an electromigation experiment (1.2 × 10–13 m2/s; Maes et al., 2008).

The Large-Scale in situ Percolation Experiments

Except for the 14C labelled Natural Organic Matter injection in TD41HV, blind predictions were made for each experiment. Table 2 lists the transport parameter values used for the blind predictions, the values determined from laboratory experiments and the values obtained by fitting. Since the start of the in situ experiments, new laboratory experiments were performed, resulting in laboratory values for Dapp,//, and Dapp,⊥ and ηR which can differ slightly from the values used at the time the blind predictions (see Table 2) were done. Table 2 lists for each experiment the Dapp,// and Dapp,⊥ values fitted with the ηR value used for the blind prediction as well as the Dapp,// and Dapp,⊥ values fitted with the most recent ηR value. Both HTO and iodide are considered as unretarded. HCO314 is retarded at high ionic strengths (Aertsens et al., 2009b) and possibly slightly retarded under Boom Clay conditions (Aertsens et al., 2008b).

HTO migration in CP1 and Tribicarb-3D

For CP1, the scatter in the experimental data is small and the agreement with the blind prediction is very good (Fig. 8). As the CP1 piezometer is placed horizontally (Fig. 3), it is evident that the diffusion coefficient parallel to the bedding plane Dapp,// is the dominant fit parameter. This follows from the very small fit error (<1%), but also from Table 2, showing that all fits lead to the same Dapp,// value, while the difference between the optimal Dapp,⊥ values is larger. Despite its simplicity and its crude approximations, the MICOF model leads to the same optimal Dapp,// value as the more sophisticated COMSOL code, as well as to a similar Dapp,⊥ value. Clearly, the agreement between the Dapp,// and Dapp,⊥ values determined in the laboratory and the fitted ones from the CP1 experiment is very good. The fitted data coincide very well with the experimental results (Fig. 9). For a comparison of the MICOF and COMSOL results, both codes use the same Vapp and ηR values.

The scatter in the Tribicarb-3D HTO data is higher than for CP1, but still limited (Fig. 10). The disagreement between the MICOF prediction and the experimental data is larger than in CP1, but still acceptable. In general, the prediction is similar to, or lower than the experimentally measured concentration. Fitting leads to the same optimal parameters for the MICOF model as for the COMSOL code. The optimal Dapp,⊥ value agrees very well with the value determined in the laboratory. The value for Dapp,// is about 25% higher than the laboratory value, which is almost identical to the CP1 optimal Dapp,// value. The agreement between the fitted and experimental data is very good (Fig. 11).

HCO314 migration in Tribicarb-3D

The large scatter at low concentrations (Fig. 12) is consistent with the explanation already given that this tracer is difficult to measure accurately in that case. Thus, it is not surprising that the disagreement between the experimental data and the MICOF prediction is large. After longer times, for the piezometers R32-3 and R34-1 the agreement between the prediction and experimental data is better.

The ηR value used for the blind prediction (ηR = 0.33) differs considerably from the most recently determined laboratory value (ηR = 0.26) (Table 2). Fitting with ηR = 0.33 leads to the same optimal Dapp,// value (Table 2) as fitting with ηR = 0.26. The optimal Dapp,// value is about the same as the lab value. The optimal Dapp,⊥ is ∼50% higher for ηR = 0.26 than for ηR = 0.33, where the optimal value is close to the laboratory value. At ηR = 0.26, the optimal Dapp,// and Dapp,⊥ are about equal, which, considering the anisotropy of the clay, is not realistic. Both fits (ηR = 0.33 and ηR = 0.26) lead to about the same predicted concentrations, which agree well with the experimental data (Fig. 13).

Iodide migration in TD41HV

Iodide was injected simultaneously into the central filters of both the vertical and horizontal piezometers. For both injections, the tracer concentration was only measured in the nearest neighbouring filter. The predicted tracer concentration in the source filter is systematically significantly lower than the measured data (Fig. 14). In the nearest neighbouring filter, the agreement between experimentally determined and predicted concentration is reasonable.

Although the agreement between experimental data and the prediction by the fit is not always good (Fig. 14), the optimal values agree well with the values used for the blind prediction: an apparent diffusion coefficient in the direction parallel to the bedding plane Dapp,// ≈ 2.6 × 10–10 m2/s and an apparent diffusion coefficient in the direction perpendicular to the bedding plane Dapp,⊥ ≈ 1.4 × 10–10 m2/s. Due to the good agreement between the parameter values of the blind prediction and those fitted, we do not add a figure showing the experimental data next to the fit.

Natural Organic Matter migration in TD41HV

In order to understand the migration behaviour of Organic Matter in Boom Clay better, large-scale and long-term in situ migration experiments were initiated using the TD41HV piezometers. Transport of organic matter is complicated due to its colloidal nature which may result in filtering processes related to the Boom Clay porosity. Starting from laboratory-derived data interpreted by the classical diffusion retardation, extra complexity was introduced to account for the colloidal behaviour. This first attempt, which is not discussed further here, resulted in a reasonable fit of the experimental data (Martens et al., 2010) but further fine tuning is needed.

Conclusions

With respect to the small-scale in situ migration experiments, only strontium and caesium can be described by the classical diffusion retention formalism. Modelling the tracer profiles in the clay leads to apparent diffusion coefficients (7 × 10–12 m2/s for strontium, 1 × 10–13 m2/s for caesium) which are consistent with the values determined via laboratory experiments on clay cores sampled from the vicinity of the URF.

Large-scale (of the order of metres) in situ experiments were performed with non-retarded or slightly retarded tracers (HTO, iodide and HCO314) and with 14C labelled Natural Organic Matter. The tracer migration behaviour of HTO, iodide and HCO314 was predicted at the start of the experiments based on migration parameters measured in laboratory experiments, and compared afterwards with the in situ experimental data. In general, the blind predictions provide excellent (CP1) to reasonable (Tribicarb-3D and TD41HV) agreement with the experimental data. Fitting the experimental in situ data leads to a diffusion coefficient parallel to the clay bedding plane of ∼4 × 10–10 m2/s and 5 × 10–10 m2/s for HTO, 1 × 10–10 m2/s for HCO314 and 3 × 10–10 m2/s for iodide and to a diffusion coefficient perpendicular to the bedding plane of 2 × 10–10 m2/s for HTO, 1 × 10–10 m2/s for HCO314 and 1.5 × 10–10 m2/s for iodide. These values are close the ones determined in lab experiments. All data were fitted with a simple analytical expression, and the HTO data were additionally fitted with COMSOL multiphysics, leading to about the same optimal values.

This work is performed in close co-operation with, and with the financial support of ONDRAF/NIRAS, the Belgian agency for radioactive waste and fissile materials, as part of the programme on geological disposal of high-level/long-lived radioactive waste.