Sample description

Samples were collected from two loess sections (A and B) at the Jingbian loess site, which is located on the northern margin of the Chinese Loess Plateau, adjacent to the Mu Us Desert (Fig. 1a). Paleoclimatic reconstructions and OSL dating have previously been conducted at the site (e.g. Ding et al., 2005; Buylaert et al., 2015; Stevens et al., 2018). Fine sand grains in this section are relatively abundant, ensuring sufficient coarse K-feldspar grains for the measurements. The well-bleached nature of loess samples makes them well suited for testing dating protocols. The samples were collected using stainless steel tubes to avoid light exposure (Fig. 1b). In section A, two samples were collected from the S1 and S2 paleosol layers, respectively; and in section B, four samples were collected from the layer which was assumed to be S2 based on field observation (e.g. colour, moisture).

Fig. 1

a) Locations of the Jingbian and Luochuan loess sections. b) Photograph of the sampling of section B at Jingbian.

Sample preparation and measurements

Sample pretreatment was performed in a dark room under subdued red light. Sample material from the two ends of the tubes was removed, and then the inner part was wet sieved to separate the >63 μm fraction, which was then successively treated with 10% HCl and 30% H₂O₂ solutions to remove carbonates and organic matter, respectively. The samples were then rinsed several times with water, oven-dried at 50°C overnight, and then dry sieved. The 90–125 μm fraction was used to separate K-feldspar grains, using a sodium poly-tungstate solution (ρ <2.58 g/cm³). K-feldspar grains were immersed in 10% HF solution, with magnetic stirring, for ∼15min, in order to remove the outer layer exposed to alpha irradiation. According to Duval et al. (2018), this treatment is sufficient to remove the outer 10 μm of K-feldspar grains. Etched K-feldspar grains were then mounted on steel discs with silicone oil. The IRSL measurements were performed using a Risø-TL/OSL-20 reader, with a filter package comprising a Schott BG-39 and Corning 7–59. The attached ⁹⁰Sr/⁹⁰Y beta source had a dose rate of 0.14 Gy/s on the steel discs. The first 10 s of the decay curve was integrated as the signal, with the final 10 s integrated as background. Details of the protocols are listed in Table 1.

In addition, the low-frequency magnetic susceptibility (MS) of dried samples was measured using a Bartington Instruments MS2 meter. Measurements were repeated three times, and the mean values were normalised by the sample weights to obtain the mass-specific MS (10⁻⁸ m³/kg).

D_e and dose rate estimations

Standard growth curves (SGCs) were constructed to save measurement time. SGC constructions for the SAR protocols applied the least-squares normalisation (LS-normalization) method implemented with the R package ‘numOSL’ (Li et al., 2016; 2017a; 2018; Peng and Li, 2017). In order to obtain D_e with the SAR SGC, aliquots were measured using two cycles, with the first cycle measuring the sensitivity-corrected natural signal (L_n/T_n), and the second cycle measuring the sensitivity-corrected signal (L_x/T_x) of a regenerative dose. Details of the calculation of D_e have been presented previously (Li et al., 2015a, 2015b; 2017a; Zhang and Li, 2019). As the size of the test dose affects the shapes of growth curves, test doses were fixed at 300 Gy for SGC construction and D_e measurements.

For the “MAR with heat” protocol, samples from both the Jingbian section and the Luochuan section were used to construct the MAR SGC. Test doses were fixed at 100 Gy. Aliquots were first bleached using the ORIEL solar simulator (Newport Corporation, model 94042A) with a 1000W xenon arc lamp for more than 8 hr. Experiments showed that the residual doses of the MET-pIRIR₃₀₀ signals are reduced to ∼10 Gy after bleaching for 8 hr (Fig. 2). Bleached aliquots were divided into several groups, with different doses administered. For each sample, a group of aliquots was given a dose of 500 Gy. The average L_x/T_x of the aliquots with 500 Gy was used to re-normalise the L_x/T_x values of all the aliquots for each sample, and the re-normalised L_x/T_x values were plotted against the doses and fitted with a single or double saturating exponential function (Fig. 3). The growth curve with the single saturating exponential function can be expressed as y = 0.100+1.755*(1–exp(–x/695)). In order to measure D_e with the MAR SGC, two groups of aliquots were prepared. One group of aliquots was used to measure the natural signal (L_n/T_n), and the other group of aliquots was bleached for more than 8 hr and then given a specific dose that was close to D_e, and the signals were measured (L_x/T_x). The average L_n/T_n of aliquots in the first group was normalised by the average L_x/T_x of aliquots in the second group, in order to calculate D_e. Outliers of the individual L_n/T_n or L_x/T_x values were identified and rejected by comparing the absolute deviation from the median value and the median absolute deviation using the criterion of 2.5 (Rousseeuw et al., 2006; Hu et al., 2019). The error in D_e was estimated using the error propagation formula.

Fig. 2

Residual doses of IRSL signals stimulated at different temperatures using the SAR MET-pIRIR protocol. After 8 hr of bleaching with the solar simulator, the residual doses were reduced to ∼10 Gy for the MET-pIRIR₃₀₀ signal.

Fig. 3

Standard growth curve (SGC) of the ‘MAR with heat’ protocol. The re-normalisation dose was 500 Gy. SGC curves were fitted with the single saturating exponential (SSE) function: y = y₀ + A*(1-exp(-x/D₀)) and the double saturating exponential (DSE) function: y = y₀ + A1*(1-exp(-x/D₁)) + A2*(1-exp(-x/D₂)), respectively. The fitted results are shown in the graph. Note that the two SGCs only begin to diverge when the dose exceeds 2000 Gy.

Environmental dose rates were also estimated. Thick source alpha counting was performed to quantify the contribution of U and Th to the dose rate, and the K content of the whole rock was measured by X-ray fluorescence (XRF) analysis. Conversion factors were adopted from Adamiec and Aitken (1998). The internal K and Rb concentrations of the K-feldspar were assumed to be 13 ± 1% and 400 ± 100 ppm, respectively (Huntley and Baril, 1997; Zhao and Li, 2005). The dose rate of cosmic rays was calculated according to the sampling altitude, depth, and geomagnetic latitude (Prescott and Hutton, 1994). For the Jingbian section, the water content was generally set as 15 ± 5% for samples in paleosol layers, and 10 ± 5% for samples in loess layers (Stevens et al., 2018). However, for sample B-610, where an MS peak was located, the water content was measured to be 19.3% for sample material from the inner part of the sampling tube, and we adopted the measured water content for dose rate calculation. Because of the close similarity of depths and MS values between samples B-610 and B-640, we assumed that sample B-640 also had a water content of 19.3 ± 5%.

Possible cause of D_e underestimation with the SAR protocol

It was reported that the pIR₅₀IR₂₉₀ signal was not applicable for dating samples with D_e values exceeding 500 Gy, because the effect of anomalous fading cannot be adequately removed; whereas the pIR₂₀₀IR₂₉₀ and MET-pIRIR₂₅₀ signals were sufficiently stable to date older samples (Li and Li, 2012a). Here, we found that the pIR₂₀₀IR₂₉₀ and MET-pIRIR₂₅₀ or MET-pIRIR₃₀₀ signals also failed in dating samples with D_e exceeding ∼750 Gy, as long as the SAR protocol was used. With the identical signal, the ‘MAR with heat’ protocol always had a larger D_e than the SAR protocol (Fig. 5). Therefore, the failure of the pIR₂₀₀IR₂₉₀, MET-pIRIR₂₅₀, MET-pIRIR₃₀₀ signals with the SAR protocol for D_e estimation was not related to the signal stability but rather related to the SAR protocol itself.

Fig. 5

D_e values with different stimulation temperatures for the two SAR protocols with MET-pIRIR stimulation to 250°C or 300°C, and the ‘MAR with heat’ protocol. Note that the MAR D_e values always exceed the SAR D_e values.

Previous studies have shown that the pattern of sensitivity changes of K-feldspar in the first cycle of the SAR protocol may be dissimilar to the pattern of sensitivity change in the subsequent cycles and that the failure of sensitivity correction of the first cycle would result in the underestimation of D_e for low-temperature IRSL signals (Wallinga et al., 2000; Chen et al., 2013; Kars et al., 2014; Li et al., 2017a; Qin et al., 2018; Zhang, 2018). The difference between the patterns of sensitivity change in the first and subsequent cycles was initially attributed to the increase in the electron trapping probability caused by the first preheat treatment (Wallinga et al., 2000; Kars et al., 2014; Zhang, 2018). However, a later study found that it was due to the combination of the decrease in the electron trapping probability and the increase in recombination probability (Qin et al., 2018). For low-temperature IRSL signals, such as IR₅₀, the failure of sensitivity correction would always result in D_e underestimation when the preheat temperature was high (e.g. 200°C) (Wallinga et al., 2000; Kars et al., 2014; Zhang, 2018). However, for high-temperature pIRIR signals, the sensitivity correction of the first cycle is generally acceptable (Kars et al., 2014; Li et al., 2017a; Zhang, 2018), and dose recovery ratios are close to unity (Kars et al., 2014; Zhang, 2018). Nevertheless, a slight underestimation or overestimation of dose recovery ratios can occur, depending on the type of pIRIR signal, the magnitude of the regenerative dose and test dose, the preheat temperature (Qin et al., 2018), as well as the sample origin (Zhang, 2018). In order to study the influence, sample B-1000 was used to perform the dose-recovery-like experiments which were the same as those used by Qin et al. (2018). The ORIEL solar simulator bleached about 15 aliquots for 20 hours and then divided into five groups, each group comprising ∼3 aliquots. The given doses for the five groups were 0, 150, 300, 600, 900 Gy, respectively. The test dose was 300 Gy, the same as that used in the SAR De measurements. Two repeating cycles were carried out for each group with identical dose. The sensitivity-corrected MET-pIRIR₂₅₀ signal of the first cycle (L₁/T₁) can be treated as the natural signal, and that of the second cycle (L₂/T₂) can be treated as the regenerative dose signal. Growth curves of the two kinds of signals were constructed with all of the aliquots from the five groups (Fig. 6a). The growth curve of L₂/T₂ was in fact constructed using multiple aliquots, but it was equivalent to the laboratory growth curve of a single aliquot in the SAR protocol, because of the consistency between different aliquots and the excellent recycling ratios (near unity) in SAR cycles for K-feldspar. In order to calculate the recovery ratios, the given doses of the first cycle were taken as unknown, and the corresponding ‘D_e’ values were calculated based on the growth curve of L₂/T₂, using the gSGC method (Li et al., 2015a, 2015b; Zhang and Li, 2019). The mean recovery ratio was calculated for each group and plotted against the recovery dose (Fig. 6b). The results show that with the test dose of 300 Gy, the recovery ratio became slightly overestimated when the recovery dose exceeded 600 Gy. Therefore, the failure of the sensitivity correction of the first cycle in the SAR protocol cannot explain the SAR D_e underestimation observed in this study.

Fig. 6

Results of dose-recovery-like experiments for sample B-1000. (a) The growth curves of the L₁/T₁ and L₂/T₂ signals fitted with a single saturating exponential (SSE) function: y= y₀+a*(1-exp(-x/D₀)). (b) Regenerative doses of the first cycle are taken as unknown and recovered by the growth curve of L₂/T₂. The recovery ratios are plotted versus the recovery doses. Note that the recovery ratios are slightly overestimated when the doses exceed 600 Gy.

With the SAR protocol, the growth curve of K-feldspar has D₀ of ∼ 400 Gy (Fig. 6a). The beginning point of SAR D_e underestimation is ∼750 Gy, which is very close to 2D₀. For quartz, it has been proposed that reliable D_e estimation should be within the upper limit of 2D₀ (Wintle and Murray, 2006). Age underestimation has been observed when applying quartz OSL dating in marine isotope stage 5, with D_e values close to the 2D₀ limit (e.g. Murray and Funder, 2003; Lai, 2010). It appears that the empirical 2D₀ limit also exists for K-feldspar. A recent study suggested that D_e underestimation for quartz when exceeding the 2D₀ limit, was caused by the rejection of ‘saturated’ grains or aliquots, and an ‘average L_n/T_n’ method was thus proposed to overcome the problem (Li et al., 2017b). Successful application of this method has been performed on archeological cave sediments with an age range of 170–80 ka (Hu et al., 2019). For the K-feldspar samples used in the present study, no ‘saturated’ aliquots were identified for the SAR protocols. With the application of the ‘average L_n/T_n’ method, D_e became slightly smaller than the values obtained using the conventional ‘average D_e’ method.

We can provide no credible hypothesis to explain the SAR D_e underestimation in the high dose range in the present study. It is possible that the natural growth curve and laboratory growth curve of K-feldspar begin to diverge in the high dose range, similar to the case for quartz (Timar-Gabor and Wintle, 2013). However, it cannot explain the different behaviour of the SAR protocol and the ‘MAR with heat’ protocol and therefore, further studies are needed in the future.

Implications for dating

As shown in Fig. 3, the growth curve of the ‘MAR with heat’ protocol saturates much later, with D₀ of ∼700 Gy (Li et al., 2013; Chen et al., 2015). Application of the 2D₀ limit would enable reliable dating to ∼1400 Gy. In addition, with the MAR protocol, the discrepancy between the sensitivity changes between the first and subsequent cycles within the SAR protocol is not a concern, since all the aliquots were run for only one cycle. The ‘500°C heat’ inserted before the test dose was designed to erase pre-dose memory (Li et al., 2013; Chen et al., 2015). Although a large sensitivity change would also result from such a high-temperature treatment, it would not influence the reliability of D_e measurements, as the signal of the subsequent test dose is now solely related to the size of the aliquots. The test dose signal is used for mass normalisation between different aliquots, rather than sensitivity correction. Thus, for old samples, the D_e values obtained with the ‘MAR with heat’ protocol are the most accurate.

The SAR ages of the four samples from section B range from 160–190 ka with the pIR₅₀IR₂₉₀ signal, and 190–240 ka with the pIR₂₀₀IR₂₉₀, MET-pIRIR₂₅₀, MET-pIRIR₃₀₀ signals, which correspond to loess layer L2 and paleosol layer S2, respectively. The ages obtained using the ‘MAR with heat’ protocol are systematically 30–40 ka older than the SAR ages with the same MET-pIRIR₃₀₀ signal. Considering the errors of ∼20 ka for the MAR ages, it is uncertain whether the samples are located in S2 layer or L3 layer.