Introduction and methods. A multitracer (3H, 3He, 85Kr, 39Ar and 14C) approach is used to investigate the age structure of groundwater in the semiconfined Fontainebleau Sands aquifer that is located in the shallower part of the Paris Basin (France). The hydrogeological situation - which is characterized by spatially extended recharge, large screen intervals and possible leakage from deeper aquifers - leads to expect a wide range of residence times and pronounced mixing of different water components. The commonly adopted approach for dating and quantifying the portion of young (post-bomb) water components is the combination of 3H with either 3He or 85Kr. However, these tracers preferentially detect the "young tail" of age distribution which may include significant amounts of pre-bomb water. Hence, to verify to which extent the extrapolation of the interpretation with these tracers to the "old tail" fits the real situation in the Fontainebleau sands aquifer an intermediate-age (< 1000 years) dating technique is required, which is what 39Ar enables in this study. This tracer has been proposed for dating groundwater for its ideal characteristics (Loosli, 1983): a constant and well known atmospheric input concentration, no local contamination, an isotope ratio (39Ar/Ar) that is insensitive to degassing or incomplete gas extraction yield, and an important dating range for groundwater hydrology (100- 1000 years). We examine the use of 39Ar, a noble gas radioisotope with a half-life of 269 years, to constrain the age distribution of groundwater in this timescale range. The limited number of sampling sites in the project area requires the application of multiple groundwater dating tracers, and the use of lumped parameter approaches for the assessment of groundwater dynamics where the choice of an appropriate age weighting function that appropriately represents the hydrogeological situation can be validated using the measured tracer data. Recharge depths vary between 20-40 m below the soil surface; therefore a transport model of unsaturated zone flow (one dimensional advection-diffusion-decay transport model: 1D-ADDTM) is coupled with the lumped parameter models (LPM) for saturated flow at each individual well in the saturated zone for interpreting the 3H, 3He and 85Kr data. Some parameters of the 1D-ADDTM are known from the literature. Others like a mean unsaturated zone thickness (for all wells) and the recharge rate are included in the inversion procedure as global free parameters. LPM parameters for each well (Mean residence time-Tm, Mixing ratio or fraction of young water - m; and dispersion - Pe) and global parameters (mean recharge depth Z, and recharge rate ql) are then estimated by a χ2 fitting routine (Corcho Alvarado et al., 2007). This procedure is different from the commonly adopted way of interpretation were tracer ages are compared. Here we characterise transit time distributions and/or model parameters by fitting to tracer concentrations. The LPM parameters which best describe the data are then checked for consistency with the 39Ar measurements.
Results and discussion. Several box models (e.g. Piston Flow, Exponential, Dispersion and Exponential-Piston Flow models) were tested and checked for consistency with the data. Best fits for 3H, 3He and 85Kr, according to the χ2 fitting routine, are obtained with the Exponential Model (EM); while other models lead to larger deviations. Tracer input functions at the water table assuming a mean unsaturated zone thickness of 35 m yielded the best agreement between modelled and measured tracer concentrations. The estimated recharge rate in the area of investigation of 150 mm/yr is in excellent agreement with previous values given in the literature. Relatively low residence times Tm and mixing fractions m of the young water components following the EM are responsible for the observed concentrations of 3H, 3He and 85Kr. This suggests that a considerable portion of the water is older than 50 years. This can be explained by a two (or multi) component mixing of water originating from different sources or by an age distribution with a more pronounced tailing towards older ages. In some cases it is not possible to constrain Tm and m because these parameters are not independent. This is e.g. the case for the sample LRN10. This results in large errors of the fitting parameters although χ2 is excellent. This emphasizes the need of an additional tracer. In Figure 1, 39Ar is shown as function of 85Kr, 3He and 3H concentrations. The curves plotted in this Figure correspond to the theoretical relation when the whole water mass (m=1) follows the Exponential Model (EM). The calculations are based on the atmospheric input (broken line) of the tracers (85Kr activity concentrations in air, and 3H fallout) and the input values calculated according to the 1D-ADDTM at a depth of 35 m below the soil surface (solid line). The tracer results of the samples LRN10 and SLP5 are within uncertainties consistent with the one component EM. In the fitting routine a high correlation between Tm and m caused large errors of the estimated model parameters for these two samples. The 39Ar measurements reduce this large range of solutions and result in mean EM ages Tm of 116 and 373 years, respectively. The assumption of a one-component EM distribution is obviously not valid for samples SA and SM. At least two young residence time indicators (85Kr, 3H or 3He) predict concurrently, in comparison with 39Ar, an at least bi modal age distribution for SA and SM. The "young" end members of the mixing lines indicated in Figure 1 are based on the age of the young components estimated by the fitting routine. The extrapolated age of old component is about 300-400 years. At least for SM this multimode age distribution could be the result of the separated screen intervals in this well (Corcho Alvarado et al, 2007). Because both wells are situated in the same section of the area of investigation it is also possible that permeability variations in the sands in this section cause a separation of water bodies with different ages. In the frame of a pure one component EM scenario 3Hetrit tends to be too low for the wells SLP4, IMR and CGEB. Two explanations appear to be the most plausible for this observation: (i) 3He is the most sensitive tracer for variations of the thickness Z of the unsaturated soil zone as can be seen in Figure 1 from the large difference of the two model curves (Z=0 and Z=35 mbgl). In comparison the 3H and 85Kr model curves vary only slightly for 39Ar activities below 85% modern. Z was assumed to be similar for all of the wells and was therefore selected as a global fitting parameter in the calculations. Adjusting Z for each individual well would improve in particular the agreement of the 3He data in the frame of a pure EM. (ii) Incomplete confinement of 3He in the saturated zone. This would cause depleted concentrations compared to the less volatile tracers 3H and 85Kr. The lowest value of 3Hetri to be expected for pre19 bomb water is 4-5 TU if 3H delay in the unsaturated zone and mixing are neglected, and 1 TU if a decay time of 25 years in the unsaturated zone is considered. Hence, at least at well SLP4 with a 3Hetrit concentration of 0.7 TU some loss of 3He has to be assumed. The Fontainebleau sands aquifer is recharging at a rate of about 100-200 mm/yr (vertical water velocity of 0.4-0.8 m/yr); therefore a high confinement of 3Hetri can be expected (Schlosser et al., 1989). The analyses of all five noble gas concentrations (He, Ne, Ar, Kr and Xe) measured in the aquifer, including 22Ne/20Ne and 40Ar/36Ar isotope ratios, indicated that some partial degassing may have led to a maximum loss of 28% of the originally dissolved excess air 3He. In this case, the calculated 3Hetri of SLP4 and IMR would be increased by about 0.6 TU; but the deviation of the measured 3Hetrit concentration from the expected values is as high as 3 TU. Hence the 3He loss from the saturated zone appeared to be only partly responsible for the relatively depleted 3He concentrations observed. Strong mixing and a mean residence time exceeding 50 years reduce the fraction of water that can be "seen" by the young residence time indicators. It raises the question in how far dating results from transient tracers can be extrapolated to the old part of the age distribution and how unimodal and multimodal age distributions can be distinguished. In the present study, the transient tracers failed to describe the whole age distribution of the sampled groundwater. In five boreholes, they predicted bimodal age distributions which were not confirmed with 39Ar. Reliable dating of groundwater with mean residence times older than 100 years requires a tracer with a longer half live.