Water samples for this study were collected in August 2012 and July 2013 mainly from the southeastern part of the desert where lakes are concentrated. Samples were collected from various lakes, including the Sumu Jaran Lake (∼ 0.6 km2), Sumu Barun Jaran Lake (∼ 1.3 km2), Yindeertu Lake (∼ 0.9 km2), East Badain Lake (0.03 km2) and West Badain Lake (0.08 km2), as well as from domestic wells and a spring (Fig. 1). The Sumu Barun Jaran Lake is the second-largest saline lake in the BJD whereas the East Badain Lake is one of a few small lakes with relatively fresh lake water. The maximum depths of water in the lakes range between 1 and 13 m. We collected groundwater samples from four wells around the Sumu Jaran Lake and Sumu Barun Jaran Lake (Fig. 1c). The wells are shallow (maximum depth is 16 m) and screened with stone blocks on the wall, being utilized every day to supply drinking water for the local people. In addition, two sets of samples were collected from different depths in these two lakes (Fig. 2a). The deep samples from the lakes were collected with a peristaltic pump with a tube penetrating into the lake and another tube discharging water. The water was pumped through a layer of gauze and a layer of paper filter into 10 mL glass bottles. Before a water sample was collected into a sample bottle, 5 min of pumping was performed to flush the sampling system, and each sampling bottle was rinsed 3 times with the water to be sampled. For groundwater in wells, sampling was carried out after a 20 min pumping to flush the sampling system. Groundwater samples were collected in 600 mL Pyrex Brand square bottles for radiocarbon (14C), [HCO3-] and tritium (3H) analyses. Each bottle was sealed carefully to avoid the existence of bubbles within the water sample.

Each groundwater sample for 14C analysis was dosed with a drop (∼ 50 µL) of 20 M NaOH to poison the samples to inhibit microbial activity before the bottle was sealed. The samples were sent to Beta Analytic Radiocarbon Lab in Miami (Florida) for the radiocarbon analysis. Because the desert area is sparsely settled, there are only a few wells from which we could obtain samples of the groundwater. The existing wells are generally less than 20 m deep and the depths to water table are less than 4 m. We have also collected a water sample (SY-1) from a spring near the Yindeertu Lake (Fig. 1c).

There was a precipitation event during the week of our field exploration in the Sumu Jaran Lake area in August 2012. It rained at night and continued into the early morning. A rain water sample was collected from a puddle on the ground. This sample, combined with previously published precipitation isotope data (Gates et al., 2008a) and data from Zhangye – the nearest IAEA-GNIP (International Atomic Energy Agency Global Network of Isotopes in Precipitation) station located approximately 170 km to the southwest of the study area at the altitude of 1400 m, (IAEA/WMO, 1986–2003), was used to approximately represent the isotopic composition of the local precipitation in the BJD.

To better understand the effects of strong evaporation on the isotopic characteristics of water in the desert, we conducted surface water evaporation experiments and soil water infiltration–evaporation experiments on the shore of the Sumu Jaran Lake (Fig. 2b) in July 2013. The observation site and experimental methods are schematically presented in Fig. 2b and c.

Evaporation from open water was monitored in two experiments using local groundwater (Pan-1) and lake water (Pan-2), respectively (Fig. 2c). The two plastic pans, which have the same cylinder shape with a diameter of 21 cm and height of 10 cm, were used in the evaporation experiments. They were filled with water to an initial depth of 6 cm and then put on a flat ground surface to allow water to evaporate. With continuous evaporation, the water levels in the pans gradually decreased. We measured the water depths every day at 07:00 and 19:00 LT. Isotope samples from both pans were collected at 19:00 LT every day after the measurements of the water level. These experiments continued for 6 days. The air temperature and humidity at the height of 2 m above the ground surface were also measured every 2 h in daytime. A total of six water isotope samples were obtained from each of the experiments. During the period of the experiments, the air temperature varied between 17 and 42 ∘C and the humidity varied between 20 and 70 %.

Evaporation of soil water was observed in three experiments, ES-1, ES-2 and ES-3 (Fig. 2c). Three pits were excavated to different depths below a flat ground surface to install the evaporation–infiltration systems. Each of the evaporation–infiltration systems was composed of a sandbox with a funnel and a bottle attached to the bottom. The sandbox is 0.5 m × 0.5 m wide, with a depth of 9, 15 and 22 cm, respectively, for ES-1, ES-2 and ES-3. The sands were initially dry but were wetted slightly with freshwater from a nearby well (WS-4) using a small sprinkler when the sands were put into the box layer by layer. Artificial rainfall with 250 mL volumetric water in 6 min was made twice a day during the daytime using a watering can (the mean “rainfall” intensity is 0.167 mm min-1) for each sandbox to increase the soil moisture content. After the artificial rainfall, some of the soil water was lost due to evaporation and some became infiltration water that passed through the sandbox and was collected in the bottle below the sandbox. After 6 days, the bottles were dug out and the water samples from the bottles were immediately transferred into sampling vials for isotope analysis later in the lab. We obtained three samples of infiltration water (ES-1, ES-2 and ES-3) for isotope analysis.

The water samples were collected for various chemical analyses. Among them, 21 were prepared for stable hydrogen and oxygen isotope analyses, including 12 lake water samples, 1 spring sample, 1 precipitation sample and 7 groundwater samples. The depths of water tables and the conductivities of water bodies in the desert were measured using a TLC (temperature, level, conductivity) meter (Model 107 TLC, Solinst). Seven groundwater samples were analyzed for radiocarbon contents of dissolved inorganic carbonate (DIC) and four samples were analyzed for tritium (3H).

The radiocarbon contents of DIC samples were analyzed using an accelerator mass spectrometer. The 13C / 12C ratio was also measured relative to the VPDB (Vienna Pee Dee Belemnite) standard for the radiocarbon age correction. Radiocarbon dating was performed by Beta Analytic, Inc. (Miami, Florida, USA). The radiocarbon age represents the measured radiocarbon age corrected for isotopic fractionation using the δ13C. The HCO3- and Tritium contents of the groundwater samples were measured using Metrohm automatic titration apparatus and a low background liquid scintillation spectrometer (Quantulus 1220-003), in the Analytical Laboratory of Beijing Research Institute Of Uranium Geology. The analytical uncertainty of the Tritium analysis has a level from ±0.4 to ±0.7 TU. Stable isotope samples were analyzed using a Finnigan GasBench II Auto carbonate and water device interfaced to a Finnigan MAT DELTAplus XP stable isotope ratio mass spectrometer in the Stable Isotope Laboratory at Florida State University. The stable isotopic results are reported in the standard notation as δD, δ18O and δ13C values Eq. (1): δ=RSARSTD-1×1000‰,δ=RRVSMOW-1×1000‰, where δ is the isotopic concentration of a sample, RSA is the isotope atom ratio D / H, 18O / 16O or 13C / 12C and RSTD is the corresponding isotope atom ratio of the international standard VSMOW (Vienna Standard Mean Ocean Water) for hydrogen and oxygen and VPDB for carbon.

The δ2H and δ18O values are normalized to the VSMOW SLAP scale, with a reproducibility (±1σ) of about ±1 and ±0.1 ‰, respectively. The precision for DIC–δ13C is ±0.3 ‰ (±1σ).

The δD and δ18O values of the water samples collected from the pan evaporation experiments and evaporation–infiltration experiments are shown in Table 1 and Fig. 3. The residual volume of water in the pans after 5 days of evaporation decreased to less than 50 % of the initial volume while the water isotope contents became progressively enriched along an evaporation line EL1 (Fig. 3a). The local evaporation line (EL1) established through our pan evaporation experiments has a slope of 4.6 (Fig. 3a). As a result of evaporation, the δ18O (δD) of water in Pan-1 increased from -3.5 to 23.2 ‰ (-46 to 66 ‰ for δD), and the δ18O (δD) of water in Pan-2 increased from 7.6 to 18.2 ‰ (10 to 70 ‰ for δD). The isotopic enrichment factor is ∼ 0.2 ‰ for δ18O and 1 ‰ for δD per 1 % water loss, respectively. The d-excess values (=δ2H – 8δ18O) of remaining water, on the other hand, decreased with the evaporation (Fig. 3b) from -18 to -120 ‰ in Pan-1 and from -51 to -76 ‰ in Pan-2. Note that Pan-1 and Pan-2 were initially filled with groundwater and lake water, respectively. The electric conductivity (EC) results shown in Table 2 indicate fresh groundwater (EC < 1 mS cm-1) and saline lake water (EC is generally higher than 100 mS cm-1). Similarly, heavy isotopes in the water samples collected from the evaporation–infiltration experiments (i.e., ES-1, ES-2 and ES-3) were progressively enriched with evaporation (Fig. 3). The d-excess values of infiltrated water (minimum -35 ‰) were also significantly lower than that of the artificial rainfall water (-18 ‰).

The δ2H and δ18O values of natural water samples are shown in Table 2. Water samples collected from different depths in both of the Sumu Jaran Lake (LX-1 to LX-4) and the Sumu Barun Jaran Lake (SX-1 to SX-4) exhibited few isotopic variations with depth. For Sumu Jaran Lake, the δ18O of water collected from 0, 1.5, 3 and 5 m depths varied from 7.0 to 7.4 ‰, with a mean of 7.3 ± 0.2 ‰ (1 standard deviation from the mean) while the δD ranged from 1 to 5 ‰, averaging 2 ± 2 ‰. For the Sumu Barun Jaran Lake, the average δ18O and δD values of water samples from 0, 2, 4 and 6 m depths are 5.9 ± 0.1 ‰ (ranging from 5.7 to 6.0 ‰) and -1 ± 1 ‰ (ranging from -3 to 1 ‰), respectively. The relatively small variation of δ18O and δD with depth for such groundwater-fed lakes suggests that the lake water was well mixed, which is confirmed by the measurements of temperature and electric conductivity profiles in the summer (Chen et al., 2015). Similar results were reported by Wu et al. (2014) for the Nuoertu Lake in the same desert.

The radiocarbon activities in groundwater samples, reported as a percentage of modern carbon (pMC) range from 35.81 to 72.71, and the δ13C values vary from -11.1 to -6.7 ‰ (Table 3). The lowest and highest radiocarbon activities are found in the samples from wells near the Sumu Jaran Lake (WS-1 and WS-3). The tritium activities in the groundwater samples range from 0.4 TU (WS-1) to 5.4 TU (WS-3) and are positively correlated with the radiocarbon activities. The tritium levels in our samples are significantly less than that of the single groundwater sample W34 (43.2 TU) collected in 2004 from the Yabulai Mountain (Fig. 1a) reported in Gates et al. (2008a).

It is well known that evaporation preferentially removes the light isotopes from the liquid phase, resulting in a progressive isotopic enrichment in the remaining water following an evaporation line in a δ18O vs. δD plot (e.g., Gonfiantini, 1986). The slope of the evaporation line varies from 2 to 5 depending on local climatic conditions. The evaporation line in the study area established through our evaporation experiments has a slope of 4.6 (i.e., EL1 in Fig. 3a). The δ18O and δD values of the infiltration water samples collected from the infiltration–evaporation experiments also fall on the evaporation line established through pan evaporation experiments (Fig. 3a). This agreement indicates that evaporation during infiltration occurs and imparts a characteristic isotopic trend reflecting local climatic conditions. These experiments demonstrate that in the present-day arid environment infiltration water could undergo significant isotope enrichment due to evaporation as rainwater descends through the atmosphere and infiltrates into the ground to recharge the groundwater. The similarity of the evaporation effects between the pan water and infiltration water samples may be caused by two factors. First, surface evaporation is the dominant evaporation process in the sand boxes without significant influences from pore vapor transport and vegetation uptake. Second, the stable isotope fractionation in evaporation processes within the sands is similar to that in the surface evaporation. Laboratory experiments also show that the isotope fractionation of water evaporation in unsaturated sands is similar to surface water evaporation (Sun et al., 2009). However, we had only three samples in the in situ infiltration–evaporation experiments, which may be not enough to represent the general behaviors of water in the desert; more experimental studies are required to confirm this finding. For groundwater recharged by direct filtration through the unsaturated zone in arid regions, isotopic enrichment due to evaporation during infiltration needs to be considered when interpreting water isotope data. Our evaporation experiments also show that d-excess values of water decrease progressively with increasing evaporation and were negatively correlated with the δ18O values (Fig. 3b). Although the d-excess values are often used to infer atmospheric vapor sources, the evaporation experiments show that the d-excess values of water in the study area are primarily controlled by evaporation and decrease significantly but systematically with the extent of evaporation, providing another fingerprint for tracing the locally recharged water. Linear regression of the experimental data yielded an evaporation trend line with a slope of -3.4 in the d-excess vs. δ18O plot for the study area (Fig. 3b).

The evaporation experiments were conducted in the summer and may not represent evaporation conditions in other seasons. However, the similarity between the evaporation line determined through our evaporation experiments (Fig. 3a) and those derived from measurements of natural water samples (Fig. 4) implies that seasonal variations in meteorological conditions do not significantly alter the evaporative δD–δ18O pattern in the desert.

Water samples collected from various lakes in the BJD are significantly enriched in the heavy isotopes 18O and D relative to the groundwater samples (Fig. 4). For example, the mean δD value of the samples in lakes is about 46 ‰ higher than that of the groundwater samples. However, all the δ18O and δD values of groundwater and lake water samples fall below the Global Meteoric Water Line (GMWL) along with a trend line (Fig. 4), indicating that they have been affected by evaporation (Gonfiantini, 1986). This trend line (i.e., EL2 in Fig. 4) is nearly identical to the local evaporation line EL1 (Fig. 3a) established through the evaporation experiments, indicating that lakes and groundwater in the study area have evolved from meteoric precipitation under modern or similar to modern climatic conditions. This rules out the hypothesis that the lakes were recharged by groundwater that derived from precipitation during the early Holocene when the climate was much wetter than today (Yang and Williams, 2003; Yang et al., 2010). This also confirms that the lake water is originated from groundwater but has evolved into a more 18O / D-enriched state due to higher degrees of evaporation.

Previous studies have also determined the evaporation line in the BJD (Chen et al., 2004, 2012; Wu et al., 2014). The evaporation line reported by Wu et al. (2014; i.e., δD = 4.2 δ18O - 31.4; sample number n=110, trend-line fit R2=0.92) is very similar to the EL2 (Fig. 4). Although Chen et al. (2004) reported an EL that had a slope of 5, higher than that of the EL2 (δD = 4.5 δ18O – 27.7), they did not provide sufficient information about their samples and δD–δ18O results to allow for comparison with our results. However, their later study is more informative as it reported the sampling sites and the δD–δ18O results (Chen et al., 2012), which showed similar results to ours in the hinterland area of the BJD where our data were collected. It appears that the higher slope and lower intercept can be attributed to the samples that were collected from areas located on the northern and western margins of the BJD, Wentugaole and Gurinai (Chen et al., 2012). These two sites, especially the Wentugaole, is too far away from our sampling sites (Fig. 1a).

The intersection of the evaporation line and the GMWL in the δ18O–δD plot suggests that the meteoric precipitation recharging the local groundwater system had average δ18O and δD values of -10.8 and -76 ‰, respectively (Fig. 4). They are significantly lower than those of the sole rain sample that we collected in the BJD in August of 2012 (Fig. 4). This does not necessarily mean that groundwater and lake water in the BJD were not recharged or affected by local precipitation, because the number of precipitation samples is too small to capture the full range of isotopic variation of local precipitation. The precipitation isotope data for the period of 1986–2003 from the nearby IAEA-GNIP station in Zhangye show that the isotope composition of precipitation in the area could vary considerably due to variations in atmospheric conditions, with summer precipitation having higher δD and δ18O values than winter precipitation (Fig. 4). The single rainwater sample that we collected in the summer yielded δD and δ18O values similar to those of summer precipitation measured at the IAEA station in Zhangye (Fig. 4). The weighted mean δ18O and δD values of the annual precipitation (1986–2003; 70 samples) in Zhangye are also very close to the inferred δ18O and δD values of meteoric precipitation feeding the shallow aquifer in the BJD (Fig. 4). This is consistent with the view that the groundwater in the BJD originated from modern precipitation in the region (Hoffmann, 1996; Jäkel, 2002).

Although a number of studies have used the natural abundances of stable isotopes to infer the origin of water in the BJD (Chen et al., 2004; Ding and Wang, 2007; Gates et al., 2008a, b; Ma and Edmunds, 2006), the effects of evaporation on the isotopic ratios and the d-excess value of water were not systematically analyzed. Here, we compare our new isotope data with previously published data (Gates et al., 2008a; Zhang et al., 2011; Zhao et al., 2012) and discuss their implications for the origin of water in the BJD.

As shown in Fig. 5a, the isotope data from the BJD and the Qilian Mountain do not support the hypothesis regarding the Qilian Mountain being the recharge area for groundwater in the BJD. The evaporation line EL3, which was obtained through linear regression of all available groundwater and lake water isotope data from the BJD, is very close to the local evaporation line EL1 obtained from the evaporation experiments in this study. However, water samples from the Qilian Mountain are all plotted above the evaporation lines in the BJD (i.e., EL1 and EL3 in Fig. 5a). In particular, the meltwater and groundwater samples from the Qilian Mountain have δD and δ18O values higher than those determined by the intersection of the GMWL and local evaporation lines (Fig. 5a). Furthermore, the d-excess values of various waters in the Qilian Mountain are significantly higher than those of waters in the BJD and deviate significantly from the evaporation trend line for the study area (Fig. 5b). If groundwater in the BJD came from the Qilian Mountain as fast and deep subsurface flows in large fractures or faults underground, as suggested by some studies (Chen et al., 2004), evaporation impacts would have been negligible, and the δ18O, δD and d-excess values of groundwater in the BJD would be similar to those of the source waters in the Qilian Mountain. However, this is not the case (Fig. 5). Thus, the stable isotope data do not provide support for the remote-source hypothesis proposed by Chen et al. (2004) and Ding and Wang (2007).

Figure 5a also shows that the shallow groundwater samples from the Yabulai Mountain area fall above the evaporation lines in the BJD, but follow a trend line (EL Yabulai: y=4.2x-24.1) that extends through some of the lakes (Fig. 5a). This suggests that shallow groundwater in the Yabulai Mountain may be a source for some of the lakes in the BJD or the evaporation–infiltration conditions in the mountain areas were different from those in the desert. Two deep groundwater samples from Xugue and Gurinai (144 m deep) on the northwestern and eastern margins of the BJD (Gates et al., 2008a) have similar δ18O and δD values to the groundwater in the BJD (Fig. 5a). Because of the limited number of deep groundwater samples, it is not yet possible to determine the relationship between them, but these aquifers may be connected, as suggested by their similar isotope compositions (Fig. 5a).

Based on the hydrogeological conditions and the water isotope data, we propose a new conceptual model for the origination of water in the BJD, which is schematically shown in Fig. 6. The sources of groundwater and lake water in the BJD are mainly derived from meteoric precipitation in the desert and adjacent mountains. These mountains include the Zongnai Mountain in the east, the Yabulai Mountain in the southeast and the Beida Mountain in the south, providing lateral groundwater flow toward the lakes in the BJD, especially the fresh and brackish lakes near the margin of the desert. Groundwater feeding most of the lakes, particularly the ones located farther away from the mountains in the BJD, however, receives at least partial recharge from infiltration water originated from local precipitation on the highly permeable sand dunes in the area. Similar results were reported from arid central Australia (Tweed et al., 2011). Based on the stable isotope (δ18O and δ2H) data, Tweed et al. (2011) suggested that groundwater in that arid region was principally recharged during larger and intense rainfall events, but over longer time frames, and groundwater recharge was predominantly via diffuse processes rather than infiltration of floodwaters, even though the recharge may locally vary with distance from the floodplain. Our data show that the d-excess value of water decreases progressively from precipitation, infiltration, groundwater and lake water due to evaporation (Figs. 3b and 5b). For groundwater originating from the mountain areas, the d-excess value is generally higher than that originating from the local sand dunes because of more rapid infiltration and less evaporation under cooler and higher precipitation conditions in the mountains. The groundwater originating from local precipitation, on the other hand, inherits the d-excess value of the infiltration water in the dunes, which depends on the long-term environmental conditions in the area. Evolution of the lakes from brackish to saline also reflects the long-term accumulation of salts (Gong et al., 2016) and increase in the d excess driven by lake surface evaporation.

Only a few other studies also reported d-excess values and δ18O values of groundwater from similar arid areas, such as Lake Eyre Basin (LEB), Australia (Tweed et al., 2011), and Jabal Hafit in the United Arab Emirates (UAE; Murad and Mirghni, 2012). Analysis of these previously published d-excess values and δ18O values of groundwater from these arid areas also reveals strong relationships between the two (Fig. 7), suggesting similar recharge processes to those observed in the BJD. This implies that previous interpretation in terms of the origin of groundwater may need to be revised. For example, Wood (2011) interpreted the negative d-excess (-34.27 to -10.8 ‰) values of paleogroundwater as indicative of an influx of evaporated runoff into the Red Sea during the last wet period, resulting in the negative d-excess values in the moisture source. However, the strong relationship between the d-excess and δ18O values of Gachsaran aquifer indicates that the water was affected by evaporation.

Assuming that water in the lakes originated from the local precipitation in the desert, the water balance of the lakes can be analyzed using the following mass-balance equation: Alake(E0-P)=α(Azone-Alake)P, where Alake and Azone are the total water surface area of the lakes and the catchment (or recharge) area for the lakes in the desert, respectively; E0 is the mean annual potential evaporation of open water; P is the mean annual precipitation; and α denotes the ratio of total evaporative water loss from lakes to the total precipitation in the catchment. From Eq. (3), the following relationship can be derived: required recharge area, Azone, is controlled by the lakes and recharge ratio as follows. AzoneAlake=1+1αE0P-1 The potential evaporation rate (E0) of the lake water has been estimated as 1200–1550 mm yr-1 from the observation data of the Sumu Barun Jaran lake (Wang et al., 2014). Using E0=1550 mm yr-1 (to account for the maximum effect of the evaporation loss) and P=100 mm yr-1, the estimated E0 / P ratio is 15.5. If the infiltration recharge of groundwater is only 5 mm yr-1 (α=0.05), the recharge area that is required to balance the evaporative water loss from the lakes would be 291 times the lake's area (Azone/Alake=291). Since the area of the lakes in the desert is about 22.3 km2, the estimated catchment area Azone would be ∼ 6489 km2 in this scenario. As mentioned in Sect. 2, most of the lakes are concentrated in an area of ∼ 2500 km2, which is smaller than the estimated recharge area. The estimated recharge area with respect to α=0.05, however, is much smaller than the area of the southeastern region of the BJD (∼ 14 000 km2). In a preliminary modeling simulation, Wang et al. (2014) suggested that the infiltration recharge would be 31 mm yr-1. Using an α value of 0.31 the calculated Azone/Alake ratio would be 47 and the required recharge area would decrease to ∼ 1048 km2, which is significantly less than the lake concentration area where most of lakes occur. Thus, it is highly possible that the evaporative water loss from the lakes could be balanced by local recharge in the desert area where lakes are located, without the need for external water sources.

The radiocarbon ages of DIC in groundwater are shown in Table 3, which range between 2 and 9 ka. This does not indicate very old water since groundwater ages in deserts are generally higher than 10 ka (Bentley et al., 1986; Kronfel et al., 1993; Sultan et al., 1997; Edmunds et al., 2006; Hagedorn, 2015). These ages do not represent the residence time of water due to the input of 14C-deficient DIC derived from dissolution of old carbonates. Gates et al. (2008a) estimated that it would take 1000–2000 years for groundwater to flow to the East Badain Lake from Yabulai Mountain (∼ 25 km apart) – the presumed recharge area for groundwater in the BJD. The wells that we sampled in this study are located within 1 km of each other near the Sumu Jaran Lake (Fig. 1). However, the radiocarbon ages of DIC in groundwater varied greatly from 2560 yr BP at location WS-3 to 8250 yr BP at WS-1 (Fig. 8). Although both the tritium and 14C activities in groundwater decreased progressively along the flow path from the sand dune to the lake (Fig. 8), the rate of 14C change is much greater than that of tritium. The HCO3- in groundwater also increased along the flow path from WS-4 to WS-1 (Table 3; Figs. 1c and 8). Thus, the old 14C ages are most likely the result of the addition of old DIC or 14C-deficient DIC from the dissolution of old carbonates in the aquifer.

The DIC of groundwater is likely composed of two sources: soil CO2 in the recharge area and carbonates in the aquifer. The δ13C of DIC derived from dissolution of soil CO2 in the recharge zone can be estimated using equilibrium isotope fractionation factors for carbonate-water system (Deines et al., 1974). The desert is sparsely vegetated with shrubs and grasses. Although one C4 grass (Agriophyllum squarrosum) and two C4 shrubs (Haloxylon ammodendron and Calligonum alaschanicum) have been found the desert, the biomass in the region is dominated by C3 plants including Caragana korshinskii (C3), Pugionium cornutum and Psammochloa villosa (Yan et al., 2001; Wang et al., 2007; Ramawat, 2009). The shrubs are distributed on the dunes, but the lowland areas near the lakes are covered by grasses. The δ13C of soil CO2 in soils hosting dense C3 vegetation is about -23 ‰ (Cerling et al., 1991). In soils with < 60 % vegetation cover, δ13C of soil CO2 is > -21 ‰, due to mixing with atmospheric CO2 resulting from low soil respiration rates (Cerling et al., 1991; Quade et al., 1989). Assuming that the δ13C of soil CO2 is -20 ‰, the pH of the infiltration water in the soil zone in the recharge area is 5.3 and carbonates in aquifer have a δ13C value +2 ‰. We can calculate the fraction (F) of soil CO2-derived DIC in groundwater from the measured DIC–δ13C value using the following mass-balance relationship: F=δ13C-δ13Ccarbδ13CDIC-CO2(soil)-δ13Ccarb, where the δ13Ccarb is δ13C of DIC derived from dissolution of carbonates in the aquifer, and δ13CDIC-CO2(soil) represents δ13C of DIC derived from soil CO2 in the recharge zone.

For example, the calculated F value for sample WS-1 is 58 %. This suggests that ∼ 58 % of the DIC in this sample was derived from soil CO2, and the radiocarbon age (8250 yr BP) of this sample could be a result of 42 % dilution by DIC derived from dissolution of 20 000-year-old carbonates in the aquifer (Table 3). That demonstrates that just a small amount of DIC from the dissolution of old carbonates can yield an erroneous DIC radiocarbon age that could be several thousand years (or more) too old. Various types of carbonate have been found in the lake area including tufa deposits, lacustrine carbonates and calcareous cementation (Yang et al., 2003). These carbonates provide possible sources of old DIC in groundwater.

Furthermore, these groundwater samples contained detectable amounts of tritium, indicating at least some component of modern recharge (Fig. 8; Table 3). This is because tritium is a radioactive isotope with a short half-life of 12.43 years and its concentration in a sample should fall below detection limit after 7–10 half-lives (∼ 100 years). Tritium concentration in the atmosphere was elevated considerably during the nuclear era but has been declining steadily toward natural background levels after the test ban agreement in 1963 (Clark and Fritz, 1997). The yearly average tritium concentrations of precipitation in the region, measured at the IAEA-GNIP station in Zhangye, have decreased from 101 TU in 1986 to 17 TU in 2003. Gates et al. (2008a) found significant amounts of tritium caused by bomb-testing in groundwater samples collected in 2004 and 2005 from Yabulai Mountain and the BJD, also indicating recent recharge by meteoric precipitation. However, it is important to note that the number of 14C and 3H samples that have been analyzed from the BJD is very limited and less than 10. Analyses of more groundwater samples and application of innovative dating techniques are required in order to better constrain the residence time of groundwater in the region. Our limited data show that radiocarbon dating of DIC in groundwater in the unconfined aquifer in the BJD is unreliable as dissolution of old carbonates in area contributes old DIC to groundwater, leading to erroneous old DIC radiocarbon ages (Table 3).