Luminescence properties and dating of proglacial sediments from northern Switzerland

Luminescence properties and dating of proglacial sediments from northern Switzerland Daniela Mueller, Frank Preusser, Marius W. Buechi, Lukas Gegg, Gaudenz Deplazes Institute of Earth and Environmental Sciences, University of Freiburg, 79104 Freiburg, Germany Institute of Geological Sciences, University of Bern, 3012 Bern, Switzerland 5 NAGRA Nationale Genossenschaft für die Lagerung radioaktiver Abfälle, 5430 Wettingen, Switzerland


Introduction
Pleistocene glaciations in the northern foreland of the Swiss Alps have been studied since the early 19 th century (summarised in Preusser et al., 2011). Whereas it appears that a minimum of eight glacial advances have shaped the lowlands, consensus on the exact number and timing is still lacking. In the last three decades, numerical dating turned into a crucial component in 25 the reconstruction of the environmental past of northern Switzerland (e.g. Preusser, 1999;Graf et al., 2007;Kock et al., 2009;Dehnert et al., 2012). In particular, luminescence dating has become a pillar of chrono-stratigraphy and understanding of glacial advances (Preusser et al., 2011). However, luminescence dating of glacial and pro-glacial deposits can be complex (e.g. Duller, 1994Duller, , 2006Spencer and Owen, 2004). For one, in such environments sediment sources and sinks are often within short distance of each other and transport may occur in turbid waters which both reduces the chance for grains to experience 30 sufficient sunlight exposure to reset or bleach any pre-existing luminescence signals. This effect will consequently lead to age overestimations and requires consideration in such deposits. The measurement of single grains or small aliquots has been advised to allow for monitoring of luminescence signal commonly presented as equivalent dose (De) distributions and the https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License.
isolation of a proportion of the De distributions that is considered well bleached (e.g. Olley et al., 2004;Duller, 2006;Trauerstein et al., 2017). The luminescence signal of quartz is more readily bleachable than this of feldspar (Godfrey-Smith et 35 al., 1988;Murray et al., 2012), and is hence favoured as dosimeter (e.g. Lowick et al., 2015). But, to asses partial bleaching of samples from northern Switzerland, Trauerstein et al. (2017) have recommended to compare luminescence signals obtained from both dosimeters. However, quartz and feldspar from the region have found to be equally challenging for dating.
For quartz from the wider region, unstable components were reported for some (e.g. Klasen et al., 2016) but not all samples (e.g. Gaar et al., 2013). Moreover, laboratory dose response curves for quartz were often best fitted with a double saturating 40 exponential function that accounts for high dose responses beyond single exponential behaviour (Lowick et al., 2010;Dehnert et al., 2012). This phenomenon is also known for samples from other regions where luminescence ages are either in agreement (e.g. Murray et al., 2008;Pawley et al., 2010) or disagreement (e.g. Lai, 2010;Timar et al., 2010;Timar-Gabor et al. 2011) with independent age control. However, the physical reason for this behaviour has yet to be identified (Wintle, 2008). In northern Switzerland, quartz luminescence ages of up to ca. 250 ka have been found reliable (Anselmetti et al., 2010;Dehnert 45 et al., 2012;Lowick et al., 2015;Buechi et al., 2017), but not beyond. Hence, this approach is not suitable to establish an independent chronology for the recently modified Mid Pleistocene glaciation history (Graf, 2009;Preusser et al., 2011).
significantly limiting the use of quartz as a dosimeter for the region.
While Feldspar requires longer for bleaching, it is known to saturate at higher doses and therefore often allows to date much older deposits than quartz (cf. Duller, 1997). Yet, the infrared stimulated luminescence signal of feldspar measured at 50 °C 50 (IRSL) often suffers from anomalous fading contributing to age underestimation (Wintle, 1973;Spooner, 1994). Different approaches to determine and to account for the loss of signal over time have been proposed (Huntley and Lamothe, 2001;Auclaire et al., 2003;Lamothe et al., 2003;Huntley, 2006;Kars et al., 2008). These approaches rely on observing signal loss over hours to days within laboratory experiments and deducing from this signal loss that has occurred over geological time scales. In such storage tests, fading has been observed for most samples from northern Switzerland (g-values between 1 and 55 3 % per decade). However, uncorrected IRSL De values were either (1) beyond the linear part of the dose response curve and therefore unsuited for most fading correction approaches, (2) corrected IRSL De values were close to saturation and consequently rejected for age determination or (3) uncorrected feldspar ages were in better agreement with quartz ages and therefore favoured (Dehnert et al., 2012;Lowick et al., 2012Lowick et al., , 2015Gaar and Preusser, 2012;Buechi et al., 2017). Only Gaar et al. (2013) found correcting for fading appropriate and necessary for samples <100 ka. For dating of older samples, Lowick 60 et al. (2012) tested alternative measurement protocols that target signals that are more stable than the IRSL signal, the post infrared infrared stimulated luminescence signals (pIRIR). The pIRIR signals are measured at higher temperatures after an initial readout of IRSL signals at 50 °C (Thomsen et al., 2008;Buylaert et al., 2011). While fading is expected to decrease or become negligible for the more stable pIRIR signals, these take longer to bleach by sunlight during transport. For the used pIRIR225 and pIRIR290 signals, fading was still observed and age overestimations led to the conclusion that these approaches 65 are not beneficial for the investigated waterlain-sediments from northern Switzerland . https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License.
Here, luminescence properties of samples from the Rinikerfeld in the northern foreland of the Swiss Alps are assessed.
Scientific drilling was conducted at a stratigraphic key site within the former glacier forefield as part of a larger campaign. The campaign is aimed to acquire further insights into the long-term glacial and fluvial landscape evolution of Northern Switzerland. Modelling of future erosion scenarios will be based on the gained knowledge and thereby assist in the assessment 70 of the safest and most suitable location for a prospective Swiss nuclear waste depository. For this, the establishment of a chrono-stratigraphy is essential and, hence, dating is a crucial component. Therefore, feldspar, polymineral as well as fine-and coarse-grained quartz are investigated to exploit the dating potential of the samples from this site. Performance tests, signal intensity and composition of quartz as well as fading properties of feldspar are discussed and two approaches to correct for fading are applied. Eventually ages are derived and discussed with regard to problems found for other studies of the region. 75 2 Material and method

Site setting and samples
The study site (Fig. 1) is located within the Rinikerfeld, at the eastern tail of the Jura Mountains, about 30 km NW of Zurich.
Rinikerfeld, is part of an extensive palaeovalley structure and situated in an elevated position, ca. 50 m above the nearby 80 present-day Lower Aare Valley. At the site, the palaeovalley is carved into Mesozoic bedrock (Bitterli-Dreher et al., 2007) and presumed to be of Mid-Pleistocene age (Bitterli-Dreher et al., 2007;Graf, 2009). It was spared from glacial overprint during the Last Glacial Maximum (Bini et al., 2009) and about 40 m of Quaternary sediments were recovered in cores within a scientific drilling campaign (Gegg et al., 2018). The composite core consists of about 4 m of glacial diamicton overlain by ca. 2 m sandy gravel that gradually transition into ca. 24 m of lacustrine clay and silt. The upper lacustrine unit progressively 85 coarsens and is capped by ca. 7.5 m of diamicts and colluvium (Fig. 2).
For luminescence dating, samples were obtained from ca. 10 cm in diameter cores recovered in plastic liners (Table 1); three samples were taken from the gravelly to sandy diamicts unit (RIN1, 2, 3), one sample of lacustrine sands (RIN4) and one of gravelly sand and silt (RIN13). The two latter embrace the lacustrine unit whereof another three samples were taken (RIN5,6,8). Sample material was collected in ca. 10 cm segments whereby the outer rind and the surface, exposed to daylight during 90 core cutting and splitting, were allocated for dose rate determination and the inner core was used for De determination.

Sample preparation and measurement
To determine De values, samples were wet-sieved and treated with HCl and H2O2 to remove carbonates and organic matter, respectively. Of five samples (RIN1 to RIN4, RIN13), coarse grained (150-200 and 200 to 250 µm) potassium-rich feldspar 95 (F) and quartz (Q) were separated using sodium polytungstate at densities of 2.58 g cm -3 and 2.7 g cm -3 . The outer rind of the quartz grains was etched in 40 % HF for 1 h followed by a 32 % HCl treatment for another hour to eliminate any fluoride https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License.
precipitates. For the remaining three samples (RIN5,6,8) and RIN13, the fine-grained fraction (4-11 µm) was separated using settling under Stokes' Law. One half of the sample remained pristine as polymineral fraction (PM) while the other half was treated with hexafluorosilicic acid for seven days in order to obtain purified quartz (fQ). The fine-grained fractions were 100 suspended in acetone and settled onto cups of 9.8 mm in diameter (>1.1*10 6 grains). For the coarse-grained fractions, sample material was mounted onto cups using silicon oil with diameters of 1 mm for feldspar and 4 mm for quartz (ca. 13 and 200 grains, respectively).
De values were obtained with a Freiberg Instruments Lexsyg Research using LEDs with peak emission at 458 nm (BSL) for quartz and at 850 nm (IRSL) for feldspar and polymineral. BSL and IRSL signals were detected by an ET9235QB 105 photomultiplier tube filtered through a 3 mm KG 3 Schott glass combined with a 5 mm BP 365/50 Delta interference filter or a 3 mm BG 39 Schott glass and a 3.5 mm HC 414/46-1 AHF Brightline interference filter (referred to as "410 nm-filter combination), respectively. Additionally, a cardboard barrier was mounted into the filter wheel for IRSL measurements (for details see 3.3). Laboratory irradiation was given by a 90 Sr/ 90 Y beta source mounted into the reader. The beta source was calibrated using Risø calibration quartz batches 108 (4-11 µm) and 118 (180-250 µm) to ~0.10 and 0.11 Gy s -1 , respectively. 110 Notable is the preheat behaviour of the used Lexsyg Research reader (Lexsyg ID 09-0020); at a heating rate of 5 °C s -1 , the socalled 110 °C TL peak emerges at much lower temperatures than expected. For example, for Risø calibration quartz (batch 118) TL counts are peaking at 87 ± 3 °C during preheat. This is in agreement to observations made by Schmidt et al. (2018b) who compared the preheat behaviour of various OSL readers and discovered the appearance of the 110 °C TL peak within a range of 60 °C. This has to be considered when assessing appropriate measurement protocols (see 3.1). 115 Performance tests were conducted on both the coarse-and fine-grained fractions of two representative samples (RIN2Q, RIN2F, RIN5fQ, RIN5PM). Preheat plateau tests were carried out on the natural dose while for thermal transfer and dose recovery tests, sample material was exposed to an daylight lamp for 16 h (Q, fQ) to 30 h (F, PM). All performance tests were conducted using preheating previous to the natural, regenerative and test doses for which the aliquots were heated with 5 °C s -1 to the tested temperature and held for 10 s (Q, fQ) or 60 s (F, PM). 120 Both BSL and IRSL measurements were conducted for De determination following the SAR protocol (Murray and Wintle, 2000) with test doses of ~46 Gy and ~23 Gy administered, respectively. The initial signal was derived from the first 0.4 s of the BSL and the first 15 s of the IRSL signals. A late background subtraction using the last 40 s (BSL) or 50 s (IRSL) was applied. De values were calculated using the numOSL package for R (Peng et al., 2018).
For Q and fQ, a combined recycling and IR depletion ratio (>20 %) step was implemented at the end of each sequence to check 125 the adequacy of the sensitivity correction and for feldspar contamination. A maximum of four aliquots per sample failed this rejection criterion. None of the measured Q, fQ, F and PM aliquots presented dim (<3*BG level) or imprecise (>20 %) test dose signals and were hence accepted. For RIN3 F, one aliquot was rejected as it failed the recycling ratio test (>20 %) and for RIN13 Q, three aliquots were rejected as their De values were in saturation.
For dose rate determination (Table 2), sample material was dried and radio-nuclide content was determined via high-resolution 130 gamma spectrometry at VKTA (Dresden, Germany). A mean alpha efficiency factor of 0.05 ± 0.01 (a-value) was assumed for https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License. all fQ (Buechi et al., 2017) as well as for the F and PM fractions (Preusser, 1999). For the two latter an internal potassium content of 12.5 ± 0.5 % was used (Huntley and Baril, 1997;Gaar et al., 2013). The cosmic contribution was determined for present day depths following Prescott and Hutton (1994). Water content relative to the dry weight of the sample and capacity of water absorption (DIN 18132:2012(DIN 18132: -04, 2016 were determined in the laboratory. Representative long-term water content 135 estimates of 20 to 25 ± 5 % were used for total dose rate (Dtotal) determination. Age calculations were conducted with ADELE 2017 software (www.add-ideas.com; Degering and Degering, 2020).

Performance test 140
For preheat plateau tests on the natural signal of Q (RIN2), statistically consistent (1 σ) De values were obtained on the larger grain size fraction for preheats between 200 and 240 °C (Fig. 3). De values of fQ (RIN5) for this preheat temperature range are consistent at 2 σ (Fig. 3). A given dose of ~130 Gy was fully recovered from both tested samples (RIN2 Q and RIN5 fQ; Fig. 3) at 240 °C with measured-to-given-dose ratios (M/G-ratios) of 1.01 ± 0.03 and 1.05 ± 0.04, respectively. Maximal thermal transfer was detected with <4 Gy which equals to less than 3.5 % of the natural De and is therefore considered negligible. 145 Hence, a preheat temperature of 240 °C was chosen for both grain size fractions of quartz (Q, fQ) for all investigated samples.
For F, poor recovery was obtained for preheats above 190 °C. Considering the poor performance at high temperatures and the early emergence of the 110 °C TL peak on the used Lexsyg Research reader (see 2.2), lower temperatures were tested and a full recovery of the given dose (~130 Gy) was possible for F and PM with a preheat of 170 °C. No plateau but increasing trends of De values from the natural signal were observed between 150 and 210 °C for F and between 170 and 210 °C for PM. 150 Natural De values decrease rapidly for preheats above 210 °C while thermal transfer behaves inversely. However, thermal transfer is with <2 Gy (~1.5 % of the natural De) negligible. To investigate whether thermal quenching induces an underestimation of De values at high preheat temperatures, preheat measurements were conducted on aliquots of RIN2 F following Wallinga et al. (2000). Therefore, the natural signal was bleached with IR LEDs, a fixed dose of ca. 90 Gy was administered and each aliquot was preheated to 50 °C for 10 s followed by IRSL at 50 °C for 0.1 s. This was repeated for 155 preheat temperatures between 50 and 300 °C in 25 °C steps. After the highest preheat, the entire measurement sequence was repeated to allow for normalisation (Fig. 4). An increase in normalised IRSL sensitivity would be expected if the electron trapping probability changes with temperature. Yet, a decrease was observed for temperatures above 200 °C suggesting that it is not a change in trapping probability causing the lower natural De values and dose recovery ratios. Rather the high preheat temperature may cause a removal of signal commonly used for De determination. Therefore, a preheat temperature of below 160 200 °C is most advisable. Best performance of the SAR protocol was obtained at a preheat of 170 °C (Fig. 3) which suggests a much lower preheat than conventionally used. Buylaert et al. (2011) advice against the use of preheats below 200 °C due to thermally unstable signals induced during laboratory irradiation. However, their study was conducted on a Risø reader with https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License. different preheat temperature behaviour. It is stressed that temperature parameters of Risø and Lexsyg readers are not readily comparable. Nevertheless, to investigate a potential contribution from artificially filled unstable traps, short shine pulse 165 annealing tests were conducted on aliquots of RIN2 F with naturally and laboratory irradiated material following Wallinga et al. (2000). Each aliquot was preheated for 10 s respectively and then stimulated with IR LEDs at 50 °C for 0.1 s. Preheats were conducted consecutive in 10 °C steps from 130 to 400 °C. The signal was normalised with a test dose measurement of the same aliquot and renormalized to the first measurement (Fig. 5). Signal distribution of both the natural and laboratory induced irradiation are indistinguishable and it is unlikely that a thermally unstable component from artificially filled traps contributes 170 to the IRSL signal even at low preheat temperatures. Thermal stability of IRSL signals from the here investigated samples is given and thermal quenching was not detected, therefore, preheats of 170 °C are considered appropriate. In absence of preheat plateaus of the natural signals, the dose recovery test results are indicative (RIN2 F 0.97 ± 0.03; RIN5 PM 0.98 ± 0.03) and hence, preheats of 170 °C were used all for measurements.
Yet, the differences in temperature parameters make it necessary to test whether the configuration of the Lexsyg Research 175 reader effects the temperature during stimulation as for preheating. Therefore, De values of the natural signal were obtained at stimulation temperatures of 30, 50 and 70 °C for one representative sample (RIN2 F) (Fig. 6). While the scatter in the natural De values decreases with stimulation temperature, mean De values of all stimulation temperatures are statistically consistent with each other, implying that differences between the reader systems may not be as pronounced in the low temperature range as they are at temperatures above 150 °C. 180

Quartz OSL signal properties
Quartz grains from the investigated site are relatively dim and rarely inherit luminescent behaviour. Tests with an EMCCD camera showed that <1 % emitted a signal when stimulated with BSL. This is common for quartz from Switzerland (Trauerstein et al., 2017) and to allow for bright enough signals during measurement, an aliquot size of 4 mm in diameter was 185 chosen. While this allows for ca. 200 grains to be measured at once, only few rains are expected to significantly contribute to the emitted luminescence signal and, thereby, these are considered as small aliquots.
The signals emitted by Q and fQ are dominated by the fast component and reduce to background after about 15 to 20 s of stimulation (Fig. 7). To further investigate the BSL signals, signal decomposition was conducted (Fig. 8) using the numOSL package in R (Peng et al., 2018) that applies the 'Levenberg-Marquardt algorithm' suggested by Bluszcz and Adamiec (2006). 190 The initial signals consist of 95 % for Q and 90 % for fQ of the fast component while the relative contribution of the medium components to the natural and regenerative doses are almost identical. The latter is considered an indicator for a stable medium  Jain et al. (2003) and Li and Li (2006b). However, Steffen et al. (2009) found that the equation is highly dependent on the photo-ionising cross section of the individual components, the determination of which introduces an array of uncertainties and was therefore deemed unpractical.
De values derived using early background subtraction (0.4 to 1.0 s; cf. Cunningham and Wallinga, 2010) are statistically consistent with those obtained using late background subtraction (2 σ). This indicates an unproblematic slow component. For 205 De determination of Q and fQ, the first 0.4 s of the BSL signal were found appropriate as initial signal and a late background subtraction using the last 40 s was applied. An example of an extended dose response curve is shown for one aliquot Q from RIN 2 in Fig. 9. Laboratory saturation levels exceed 600 Gy (Q) and 800 Gy (FQ) and 2*D0 values are below 400 Gy (Q) and 580 Gy (fQ; Table 1). Dose response curves of both Q and fQ are well fitted with a double saturating exponential as well as a single saturating exponential function. The latter implies that a single type of trap is responsible for the signal (Aitken, 1998). 210

Feldspar IRSL signal properties
In contrast to Q, the natural signal of 1 mm F aliquots (ca. 13 grains) is very bright and cause saturation of the photomultiplier tube, in particular at high doses. Noteworthy is that only few grains inherit an effective luminescence signal with even less being particularly bright as shown by tests with an EMCCD camera (Fig. 11). A Schott NG-11 neutral density filter was tested 215 together with the 410 nm-filter combination but failed in protecting the photomultiplier tube from saturation. A cardboard disc with a hole in its centre was mounted into the filter wheel, reducing the photon passage to 4 mm in diameter which allowed to retrieve a detectable signal. A dose recovery test (given dose of ca. 130 Gy) was conducted with and without using the cardboard disc (Fig. 10). Results of both setups are statistically consistent with each other and suggest that the use of the cardboard is unlikely to impact the reliability of the De determination. Hence, the cardboard was used in all IRSL measurements 220 within this study.
De(t)-plots with 1.5 s intervals show consistent results over the first 200 s of the IRSL signal as presented for both the natural signal and the artificial signal induced during dose recovery tests of RIN2 F and RIN5 PM (Fig. 7). For age determination, De values were derived using the first 1.5 s of the signal and a late background subtraction of the last 50 s. An example of an extended dose response curve is shown for one aliquot from RIN2 F in Fig. 9. Laboratory saturation levels exceed 1500 Gy 225 (F) and 1800 Gy (PM) and 2*D0 values are about 600 Gy (F) and above 425 Gy (PM; Table 1). For both F and PM dose response curves are well fitted with both a double saturating exponential and a single saturating exponential function.

De distributions and ages
The De distributions of the four top Q samples show the tendency towards being positively skewed and have overdispersion 230 (OD) values between 21 ± 3 and 25 ± 3 % (Fig. 12). Arnold et al. (2007) recommend the application of a MAM3 (Galbraith et al., 1999) for De distributions of partially bleached samples that are significantly positively skewed and/or show OD values of >40 %. Following this, skewness of RIN1, RIN2 and RIN4 require for the use of a MAM3. However, the decision tree of Arnold et al. (2007) is based on fluvial samples with divers bleaching histories that are comparably young (<20 ka). For older samples, it is suggested that distributions may vary as time-dependent factors are expected to contribute to the data spread 235 (Galbraith and Roberts, 2012). For example, the impact of beta microdosimetry (Mayya et al., 2006) will increase with the age of the sample. Therefore, the application of a MAM3 was refrained from and derived CAM De values of between 144.2 ± 6.8 and 194.7 ± 7.8 Gy were used for the age determination of the top four samples (Table 1). Q ages for these samples range between 163.5 ± 9.1 (RIN4) and 179.6 ± 11.0 ka (RIN3) and are statistically consistent with each other (1 σ).
The De distributions of RIN13 Q is also positively skewed but presents a much higher OD value, 56 ± 10 %, in comparison to 240 the four top Q samples. the measured aliquots, 12 % are in saturation only allowing for the assessment of a truncated distribution. In this case, the application of the MAM3 is also recommended (Arnold et al., 2007). Besides an expected impact of beta microdosimetry, fitting uncertainties are likely accountable for the skewness and spread of the distribution. For RIN13 Q, the natural doses are projected on the high proportion of the dose response curves emphasising any fitting uncertainty.
Therefore, a MAM3 De value is unlikely to represent the distribution most appropriately and the CAM De value is considered 245 to represent a more conclusive lower limit of the distribution. The CAM De value of 255.4 ± 34.2 Gy is consistent with the 2*D0 limit of 274.1 ± 16.3 Gy and it is arguable whether an obtained age at this range is still reliable. However, the minimum CAM age of 185.8 ± 25.6 ka is consistent with ages of the top four samples and from this and the stratigraphy it can be deduced that RIN13 is of similar age or older than RIN1 to RIN4.
The four fQ samples (RIN5, RIN6, RIN8, RIN13), representing depths of 10 to 36 m, show normally distributed De values 250 ( Fig. 2, Fig. 12, Table 1). This is to be expected for fine-grained samples as the number of grains on each aliquot will have an averaging effect that will mask any skewness of the De distributions. CAM De values of 195.0 ± 4.4 to 388.0 ± 12.0 Gy were derived which are all well below 2*D0 (Table 1). For the lowest samples (RIN13), the obtained age RIN13 with 216.8 ± 12.8 ka is close to the upper limit of reliable quartz ages in the region (Lowick et al., 2015). Obtained fQ ages for the two top most lacustrine samples are 108.0 ± 5.6 ka (RIN5) and 113.5 ± 5.9 ka (RIN6). These ages are between 30 and 40 % lower than ages 255 derived for the Q fractions of RIN1 to RIN4. This is from a chrono-stratigraphy perspective unreasonable and can be ascribed to either issues with the De value or the dose rate determination. The latter will be discussed in detail in section 4.2. However, if fQ ages are equal or older than Q ages higher De values are expected for fQ. This is due to higher Dtotal given for fine-grained sediments. But, De values of RIN4 and RIN5 are almost equal (ratio 1.00 ± 0.05) resulting in a much lower age of RIN5 with a negative offset to RIN4 by 56 ± 11 ka. 260 https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License.
None of the F and PM De distributions show a trend towards being skewed and OD values are between 0 and 17 % (Fig. 12). This is to be expected for the fine-grain PM aliquots that each represent a De value derived from over one million grains. For F, each aliquot consists of only about 13 grains and the De distributions are not affected by skewness. The CAM was applied for all F and PM samples following the decision tree of Arnold et al. (2007). CAM De values of between 140.9 ± 3.1 (RIN1 F) and 412.8 ± 14.1 Gy (RIN13 F) were derived for the F fraction and between 148.2 ± 1.6 Gy and 332.7 ± 7.2 Gy for the PM 265 fraction ( Table 1). The offset between RIN4 F and RIN5 PM is with 19 ± 3 Gy less pronounced than in the quartz fractions.

Fading attributes
As the IRSL signal commonly is subject to a loss of signal over time, fading tests were conducted following Preusser et al. Fading properties (ρ'-value of 2.21*10 -6 , g-value of 3.1 ± 0.6 % per decade) were obtained for RIN2 as a representative sample 275 for the F fractions and fading correction was applied to the F measurements, providing corrected De values and ages (Table 1).
Following Kars et al. (2008), the corrected F De values, except for RIN13 F, are about 16 % higher than those corrected using Lamothe et al. (2003). 'Kars' corrected De values are between 247.2 ± 6.5 Gy (RIN1) and 297.3 ± 5.5 Gy (RIN4) and well below 2*D0 (RIN1 560 ± 6 Gy, RIN2 603 ± 8 Gy, RIN3 590 ± 8 Gy, RIN4 612 ± 24 Gy). For RIN13 F, a significant number of aliquots 12 (Lamothe correction) and 19 ('Kars' correction) out of 30 measured samples are in saturation. Minimum De 280 values of >694.5 and >973.7 Gy are obtained, respectively, which are beyond the 2*D0 limit (650.5 ± 31.8 Gy). However, for all other samples the 'Kars' corrected F ages are in good agreement with those of the Q fraction (statistically consistent at 1 σ, Table 1). It cannot be excluded that this effect is due to an overestimation of both 'Kars' corrected F and Q ages. Due to the lack of evidence for the prior, the Kars et al. (2008) fading correction approach is considered most appropriate for samples of this study. 285 For PM, fading properties (ρ'-value of 1.89*10 -6 , g-value of 2.4 ± 0.6 % per decade) were obtained from one representative polymineral sample (RIN5). Following Lamothe et al. (2003), the corrected ages are 20 to 100 % lower than those corrected using Kars et al. (2008). As the latter have shown good agreement for F and Q ages, ages calculated using 'Kars' corrected PM De values are presented in the following. Corrected PM ages of RIN5 and RIN6 are, with 138.4 ± 6.6 and 169.0 ± 8.9 ka, 28 and 50 % higher than the associated fQ ages. While RIN5 is younger than to be expected from the Q and fading corrected 290 F ages of the overlaying samples (RIN1 to RIN4), the derived fading corrected age is still consistent at 2 σ with the latter. The corrected PM age of RIN6 is statistically consistent (at 1 σ) with the ages of the overlaying samples. Consequently, an age https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License. offset as observed between Q and fQ cannot be reported for F and PM and confidence of the chosen 'Kars' fading correction is given.
For RIN8 PM and RIN13 PM, corrected ages of 471.2 ± 47.1 and 525.8 ± 53.2 ka were calculated, respectively. Fading 295 corrected De values are with 1100.5 ± 96.7 Gy (RIN8) and 955.4 ± 85.6 Gy (RIN13) far beyond the according 2*D0 values (RIN8 480.1 ± 39.5 Gy, RIN13 514.5 ± 38.6 Gy). For RIN13 PM, this is in agreement with the minimum age of >436.6 ka obtained for the F fraction of the same sample. There are three possible explanations for the high De values of RIN8 and RIN13; 1) the sediment was deposited about half a million years ago, 2) the applied fading correction overestimates for doses derived from the high dose range of the response curve or 3) these De values are the result of an averaged signal with a greater 300 contribution of partially bleached, high residual signals. The latter is an effect that is to be expected for glacier-proximal deposits that are likely to have had limited light exposure during short transport times. Samples RIN8 and RIN13 are possibly from such an environment. However, no drastic increase in De values was observed for the fQ fraction, but the OSL signal of quartz bleaches faster than feldspar. Nevertheless, corrected De values are higher than the 2*D0 values. This measure has been introduced as a dating limit for quartz (Wintle and Murray, 2006) but was found to be as well an appropriate measure for 305 adequately dating feldspars (Zhang and Li, 2020). Therefore, it has been refrained from using the presented results for finite age determination of samples RIN8 and RIN13.

Fading correction 310
One fading property, the g-value, determined for this study is in good agreement with those obtained for comparable studies in the investigated region (2-3 % per decade; Lowick et al., 2012Lowick et al., , 2015Gaar et al., 2013;Buechi et al., 2017). However, most studies have abstained from relying on fading correction for several reasons. Buechi et al. (2017) concluded that already uncorrected PM IRSL ages exceeded the reliable range for dating. Lowick et al. (2012) found that corrected IRSL ages overestimated what was to be expected from independent age controls while uncorrected IRSL ages were in good agreement. 315 In Lowick et al. (2015), fading correction was rejected as most corrected IRSL ages were higher than quartz ages obtained for the same samples. Gaar and Preusser (2012) found lower g-values of between 0.5 and 1.9 % and argued that corrected PM ages lead to overestimation and much higher ages than those derived for fQ. In contrast, Gaar et al. (2013) recommended fading correction for one well-bleached sample as corrected F and Q ages were consistent with each other. However, those studies used a fading correction following Huntley and Lamothe (2001), a correction based on the loss of luminescence signal 320 measured on artificially irradiated aliquots and extrapolated to a decadal percentage (g-value) and designed for natural values that are projected onto the linear part of the dose response curve. For most investigated samples, this is not the case and two different approaches to account for anomalous loss of signal over time were applied here. Lamothe et al. (2003) found a dose rate correction equation that incorporates differences of irradiation in nature and those of the laboratory irradiation source. This equation is also applicable for geologically old samples. Kars et al. (2008) published an approach that requires the 325 estimation of the sample specific density of recombination centres (ρ'-value). This approach may allow correction beyond the linear part. Therefore, these two fading correction measures were found most appropriate for the here investigated samples and were tested. However, 'Kars' corrected F ages have been found to be in good agreement with those of the Q fraction (see 3.5, Table 1) and are therefore considered most appropriate for samples of this study.

Age comparison
For the top four samples, Q and fading corrected F ages range between 159.6 ± 6.6 (RIN4) and 179.6 ± 11.0 ka (RIN3) and are, thereby, statistically consistent with each other at 2 σ. For the two samples (RIN5, RIN6) below, a discrepancy between fQ and PM ages (10 to 30 %) was reported. Also, fQ ages of these two samples are up to 40 % lower than those of the overlying samples RIN1 to RIN4. This leads to an age offset of ca. 35 ka between Q and fQ that is unaccountable from a chrono-335 stratigraphic perspective. The observed age offset equals 30 to 40 % difference and is likely to be emphasised through the actual age calculation which in terms relies onto derived Dtotal values. The used Dtotal values are all based on the assumptions that firstly, (1) the long-term average estimates of the water content are accurate, and that secondly, (2) used alpha efficiency values are appropriate in this context. However, in contrast to the top four coarse-grained samples (RIN 1 to RIN4), De values of RIN5 and RIN6 were determined on the fine-fraction. Therefore, (3) the effect of grain-size dependent luminescence 340 characteristics has to be assessed as well.

Water content and age determination
A water content of 20 ± 5 % was assumed for all but two samples which is within 5 % uncertainty of the measured field water content (Table 2). RIN5 and RIN6 have a slightly higher field water content (23 to 24 %) due to their argillaceous character 345 and hence, an average long-term water content of 25 ± 5 % was found more appropriate for these samples. For comparable sedimentary facies, values between 20 and 30 % were used (Anselmetti et al., 2010;Dehnert et al., 2012;Lowick et al., 2015;Buechi et al., 2017). However, the assessment of an appropriate long-term water content is complex but important to avoid age over-or underestimations. It is well known that moisture has an attenuation effect of beta and gamma radiation (Zimmermann, 1971) which may lead to drastic changes in Dtotal and thereby age determination (Nathan and Mauz, 2008). 350 Between 34 and 49 % of water, in relation to the samples dry weights, were absorbed over a 24 h period by unconsolidated sample material in this study. The highest amounts of water were absorbed by RIN5 and RIN6 which also presented the highest field water contents. For the purpose of comparison, an absolute water content maximum can be set to 50 % which is similar to the highest observed absorption capacity for unconsolidated sample material ( Table 2). The Dtotal values are calculated for water contents between 10 and 50 %, in 10 % steps. In addition, a step with 25 % water content, as assumed appropriate for 355 age determination of RIN5 and RIN6, is included in the comparison. Re-calculated ages (Fig. 13 A. and B.) show a difference https://doi.org/10.5194/gchron-2020-15 Preprint. Discussion started: 29 June 2020 c Author(s) 2020. CC BY 4.0 License.
in ages for the different water contents are up to 40 % for Q, 15 to 25 % for F and 40 % for PM and fQ. A water content for Q and F (RIN1 to RIN4) of 10 % results in statistically consistent ages at 2 σ with fQ and PM (RIN5, RIN6) at 50 %. However, a 10 % moisture for the coarse-grained samples is unlikely as under current conditions ground water penetrates these layers and oxidation features are present. Also, 50 % of moisture for fQ and PM are at the limit of maximum water absorption capacity 360 for unconsolidated material of these samples. It is unlikely that for the consolidated deposits, as found in nature, the maximum absorption capacity will be equally high and that these samples were saturated to a maximum for over ca. 150 ka. Therefore, neither 50 % nor 10 % water content are considered representative as a long-term average. Consequently, assumptions of the water content may well have an effect onto the age discrepancy between RIN5 and RIN6 and the coarse-grained samples from above (RIN1 to RIN4) but they are not the sole cause for this offset. 365

Alpha efficiency values and age determination
A further cause for the observed discrepancy may be found in the chosen alpha efficiency values (a-values) that account for the efficiency of alpha particles to produce luminescence which are grain size dependent (cf. Mauz et al, 2006). For the coarse Q fraction, the need to consider alpha radiation is circumvented by HF etching the outer rind of the grains and thereby removing 370 the sphere penetrated by alpha particles. However, fQ grains cannot be etched due to the possibility of total dissolution and therefore an appropriate a-value has to be chosen. In this study, an a-value of 0.05 ± 0.01 was used for fQ following Buechi et al. (2017). In the literature, a-values between 0.02 ± 0.01 and 0.05 ± 0.01 have been presented for fine-grained quartz from several continents (e.g. Rees-Jones, 1995;Mauz et al., 2006;Lai et al., 2008) and a-values of 0.03 ± 0.02 to 0.05 ± 0.01 were used in studies of northern Switzerland (e.g. Gaar and Preusser, 2012;Lowick et al., 2015;Buechi et al., 2017). For the 375 polymineral fractions, a-values of up to 0.10 ± 0.01 and 0.11 ± 0.01 (e.g. Rees-Jones, 1995;Lang et al., 2003, Schmidt et al., 2018a are present in the literature while a-values of between 0.05 ± 0.01 and 0.07 ± 0.02 (Preusser, 1999;Preusser et al., 2001;Gaar and Preusser, 2012;Gaar et al., 2013;Lowick et al., 2015) are commonly used for polymineral and coarse-grained feldspar in the studied region. Here, an a-value of 0.05 ± 0.01 is used.
The impact of a-value estimates onto the age calculation, can be observed for ages of the investigated samples when re-380 calculated for different a-values (0.01 to 0.05, Fig. 13 C. and D.). For fQ and PM, a reduction of the a-value leads to a 13 to 16 % age increase, while for corrected F the a-value has an insignificant impact onto the finite ages (~2 ka). Re-calculated ages are consistent with each other at 2 σ over the a-value spectrum. Between corrected F and PM samples, lower a-values (e.g. 0.01) will decrease the age differences. However, the chosen a-value of 0.05 is at the lower limit of values proposed in the literature. A sample specific determination of a-values is needed to justify the use of a lower a-value. However, in absence 385 of sample-specific a-values and limited gain of age precision, the initially chosen a-value will be retained. Also, with the lowest possible a-value of 0.01 for fQ, an offset to the Q ages of the top four samples and RIN5 and RIN6 is with ca. 20 ka still apparent. A further increase of the used a-values will amplify the age offset. Consequently, neither the chosen a-value nor water content are the sole causes for the observed age offset.

Quartz grain-size dependency
A further component, capable of causing discrepancies in ages is the effect of grain-size dependent luminescence characteristics. Here, two different grain size ranges (200-250 and 4-11 µm) were used for age determination and the offset is seemingly marked by the two grain size groups.
A comparison of fine-and coarse-grained quartz of waterlain sediments from the studied region reported ages in agreement 395 for both fractions or older ages for the fine-grained fractions (Lowick et al., 2015). Similarly, in several studies older ages are presented for the fine-grained fractions of alluvial samples, even though, bleaching conditions were expected to be preferential for the fine fractions (e.g. Olley et al., 1998;Colls et al., 2001). Nevertheless, a pattern has been observed for numerous sites from around the globe where fine-grained quartz ages of ~40 ka underestimate ages derived using coarse-grained quartz and/or independent age controls (cf. Timar-Gabor et al., 2017). A dose dependency (>100 Gy) was linked to this phenomenon and 400 thermal instability was rejected as a potential cause, though, a finite explanation is still lacking. Given the internal consistency of Q ages for RIN1 to RIN4 and findings summarised by Timar-Gabor et al. (2017), fQ ages of RIN5 and RIN6 should be regarded as minimum age estimates. This is also applicable for fQ of RIN8 and RIN13.

Conclusions 405
The investigated luminescence properties of coarse-and fine-grained quartz, feldspar polymineral fractions of eight samples from a palaeovalley in northern Switzerland were assessed and six were found appropriate for finite dating. For the chosen dose rates (a-value of 0.05, water content of 20 to 25 %), derived age estimates are in good agreement with each other and the chrono-stratigraphic context (Fig. 2). For the upper four samples (RIN1 to RIN4), Q and 'Kars' fading corrected F ages are in good agreement with each other and with fading corrected PM ages of two samples directly underneath (RIN5, RIN6). 410 Minimum fQ for the two latter are no contradiction. This indicates a deposition of at least 16.6 m during MIS 6 with a rapid transition from lacustrine to colluvial-dominated environments. For the two lowest samples, F and PM measurements exceeded the dating limit at 2*D0 but minimum fQ and Q ages indicate that these samples are of similar age or older than the top units.

Author contributions 415
Manuscript conceptualisation and experiment design was conducted by DM with support by FP. DM obtained and analysed the presented data and prepared the manuscript with contributions from all co-authors. Sample context and material was provided by MB and LG. LG contributed Fig. 1 and parts of Fig. 2 while GD was instrumental for funding acquisition. Steffen, D., Preusser, F., Schlunegger, F.: OSL quartz age underestimation due to unstable signal components, Quat.

710
A Overdispersion was calculated using the CAM (Galbraith et al., 1999). B A g-value (normalised to 2 days) of 3.1 ± 0.6 was used for F of RIN1 to RIN4 and RIN13 and of 2.4 ± 0.6 for PM of RIN5 to RIN13. C A ρ'-value of 2.21*10-6 was used for F of RIN1 to RIN4 and RIN13 and of 1.89*10-6 for PM of RIN5 to RIN13. 715 D Ages were calculated using the weighted mean of the uncorrected De values derived with the CAM (Galbraith et al., 1999). E Out of 30 measured aliquots, 12 (Lamothe) and 19 (Kars) aliquots are in saturation following fading correction and, therefore, the values shown are derived using a CAM on a truncated distribution.