A 180 ka record of environmental change at Erdut (Croatia): a new chronology for the loess–palaeosol sequence and its implications for environmental interpretation

While there are numerous thick loess–palaeosol sequences preserved across the Carpathian Basin, well dated sites that provide terrestrial palaeoenvironmental records extending beyond last glacial–interglacial cycle are scarce. Robust chronologies are essential for correlations of loess with other long‐term Quaternary records and to further understanding of the palaeoenvironment and climate of this important region beyond the last 125 ka. Here a new geochronology based on 13 post‐infrared infrared stimulated luminescence ages focused on the lower part of the loess–palaeosol sequence at Erdut is presented. The results show that the lower part of the Erdut profile spans the penultimate glacial cycle (MIS 7 to MIS 5). The considerable sediments overlaying the investigated part of the profile suggest that this section spans two glacial cycles, rather than the previously suggested one. The most likely source of the discrepancy is the use of uncorrected infrared stimulated luminescence signal, which can cause age underestimation if not accounted for. This study demonstrates the need to revisit sites such as Erdut, re‐date them using updated measurement protocols, and update existing palaeoenvironmental interpretations.


Introduction
Loess-palaeosol sequences are important terrestrial archives often preserving near-continuous and high-resolution sedimentary records. They often span multiple glacial-interglacial cycles and offer long-term Quaternary records in regions where other longterm archives are frequently absent. Therefore they provide a unique insight into the Earth's past and an opportunity to compare and correlate terrestrial signals with ice-and marine-core records (Kukla, 1977;Antoine et al., 2009;Stevens et al., 2018).
Despite their potential as palaeoarchives, loess-palaeosol sequences often suffer from a lack of robust, high-resolution, absolute chronologies. Traditionally, researchers have relied on broad stratigraphy, e.g. position and number of palaeosols (Marković et al., 2015), geomagnetic polarity reversal tie points Sümegi et al., 2018), and sedimentological correlations with distant ice cores (An, 2000;Rousseau et al., 2002;Antoine et al., 2009). However, these methods assume continuous deposition and the absence of erosion, which has since been shown to be erroneous (Stevens et al., 2006. In addition, stratigraphic correlation has proven more complicated than originally thought, for example, single soils corresponding to multiple soils at other sites (Marković et al., 2008), and has led to ambiguous links (cf Thiel et al., 2014). A reliable and independent chronology is therefore accepted as a critical underpinning of studies covering the last glacial-interglacial period, and by extension wherever possible the same requirement should be applied to studies extending beyond 125 ka.
In theory, infrared stimulated luminescence (IRSL) dating of feldspar provides the potential to date 300-500-ka-old material (Thomsen et al., 2011), because of feldspars' higher saturation limit. They therefore provide the opportunity to construct chronologies for loess profiles extending beyond the last interglacial period (e.g. Novothny et al., 2002;Schmidt et al., 2010;Wacha and Frechen, 2011). However, some feldspar signals are athermally unstable, a phenomenon termed 'anomalous fading' (Wintle, 1973). This process causes feldspars to 'leak' signal and results in age underestimation if not corrected for. Although corrections for the signal loss are possible (Lamothe and Auclair, 1999;Huntley and Lamothe, 2001), they can be problematic and are mainly applicable to younger samples. In 2008, Thomsen et al. published a post-infrared infrared (pIRIR) measurement protocol for potassium feldspar signals, which allows a more stable signal to be accessed. Subsequently, a number of studies across the region have tested the pIRIR protocol and when compared with independent age controls such as tephra (Thiel et al., 2011b), radiocarbon ages within the last 50 ka (Thiel et al., 2011a;Újvári et al., 2014), or semi-independent quartz optically stimulated luminescence (OSL) ages (Buylaert et al., 2011;Schatz et al., 2012), they showed the ability to accurately date loess sites extending beyond 70 ka.
Here a new pIRIR luminescence geochronology based on 13 ages for the loess-palaeosols section at Erdut, Croatia, is presented. A nearby section was investigated by Galović et al. (2009), but their study relied on only five luminescence ages, calculated using an additive-dose IRSL protocol. This protocol can result in large age uncertainties (Novothny et al., 2010;, and in their study, Galović et al. (2009) did not investigate or correct their IRSL signals for fading, generating the potential for age underestimation.

Geological setting and site description
Loess deposits in eastern Croatia are primarily found along the Danube River valley where a number of thick sequences have been reported (Wacha and Frechen, 2011;Banak et al., 2012;Wacha et al., 2013). Erdut (45°30'49.78"N,19°4′39.072″E, 97 m a.s.l.) is located on the western bank of the Danube (Fig. 1), 150 m from an abandoned channel and 1.2 km from the modern river. Loess-palaeosol sequences at the site are associated with a very small (~30 km 2 ) plateau-like feature, which follows the bend of the Danube River. The feature is heavily incised by gullies on the northern side, while on the southern side, exposures are not as extensive or thick. The loess outcrops on the southern side are situated on a gentle slope towards the river, with individual unit thickness decreasing in the same direction.
Investigated outcrop comprises three palaeosol and three loess units, one of which is punctuated by a laminated sediment (Table 1 and Fig. 2). The top four units (Unit Imodern soil, Unit II, Unit III and Unit IV) were not sampled in this study. Unit V comprises a pale brown soil (10YR 6/3) with small, granular peds and numerous small shells. Unit VI, a thick loess layer, has a high concentration of well-preserved gastropods, 1 mm -2 cm in size, especially in its upper 2.5 m (Fig. 3B). The sediment is pale yellow (2.5Y 7/3), though it gradually becomes yellowish brown (2.5YR 6/4) within the first metre. The field assessment suggests that the unit coarsens downwards from silt to sandy silt. The laminated sediment, Unit VII, comprises three sub-layers: horizontal lamination (30 cm), cross lamination (10 cm) and horizontal lamination (80 cm) again. In both cases, horizontal laminations comprise sheets a few millimetres thick of alternating silt and very fine sand (Fig. 3A). The cross laminations are also a few millimetres in thickness.
The final light yellowish brown loess (2.5Y 6/4) layer (Unit VIII) is mottled throughout and full of calcified root holes. The section culminates in a light reddish brown to lightly yellowish brown clayish palaeosol (10YR 6/4 to 2.5YR 6/3). The unit grades over 1 m from loess to a full soil. The total thickness of this unit could not be established as the profile could not be excavated fully.

Sampling, preparation and equipment
Due to access restrictions, only the lower parts of the Erdut profiles were accessible. Thirteen samples for grain size analysis ( Supplementary Fig. S1) and age determination were collected (Fig. 2) using light-tight plastic cylinders. All samples for luminescence dating were handled under subdued orangelight conditions at the Oxford Luminescence Dating Laboratory, University of Oxford. The light-exposed sediment from both ends of the cylinder was removed and retained for dose rate and particle size analysis. Prior to sieving, all samples were pre-treated with 32% HCl to remove carbonates, and 30% H 2 O 2 to remove organic matter. Samples were sieved to retain fine sand (63-90 μm), and density separated using sodium polytungstate to obtain potassium feldspar (ρ < 2.58 g cm -3 ). Small (2 mm) coarse-grained aliquots were made by mounting a monolayer of sediment on the disc using silicone oil. All measurements were carried out using a Risø TL/OSL luminescence reader with a calibrated 90 Sr/ 90 Y beta source, and a bialkali photomultiplier tube. Potassium feldspar grains were stimulated with infrared-emitting diodes emitting at 870 nm, and signals were detected in the blue-violet region of the electromagnetic spectrum through a combination of Schott BG39/Corning 7-59 filters. Thomsen et al. (2008) and Buylaert et al. (2012) have shown that the pIRIR signal measured at an elevated temperature is more stable over Quaternary timescales, and typically fades substantially less than the IR signal. Therefore, a pIRIR single aliquot regenerative dose (SAR) protocol was used for all measurements and D e determination (Supplementary Table  S1). After a 250°C preheat, IR 50 and pIRIR elev signals were measured for 200 s, followed by the measurement of a test

Dose rate determination
Environmental dose rates were calculated from radionuclide concentrations (U, Th, K and Rb) measured using inductively coupled plasma mass spectrometry at the British Geological Survey, Keyworth. Secondary standards were periodically measured to monitor analytical precision and accuracy. Radionuclide concentrations for each sample were converted to infinite-matrix dose rates using the Guérin et al. (2011) conversion factors. The infinite-matrix dose rates were adjusted for alpha efficiency using an a-value of 0.15 ± 0.05 (Balescu and Lamothe, 1994), grain size attenuation using the data set of Brennan et al. (1991) for alpha dose rates and Mejdahl (1979) for beta dose rates, and a commonly used loess moisture content of 15 ± 5% (Novothny et al., 2009;Stevens et al., 2011;Schatz et al., 2012;Újvári et al., 2014). Cosmogenic radiation contributions were corrected for sediment thickness and altitude (Prescott and Hutton, 1994). The internal dose rate was calculated in addition to the external dose rate using an internal potassium concentration of 12.5 ± 0.5% (Huntley and Baril, 1997). Dose rate calculation was carried out using DRAC (v1.2) software (Durcan et al., 2015).

Equivalent dose determination
Preheat test/temperature selection It has been suggested that the first IRSL stimulation temperature may impact pIRIR signal stability (Jain and Ankjaergaard, 2011). To test this, a prior-IR stimulation plateau test was conducted , to ascertain the dependence of pIRIR D e on the first IRSL measurement. Five temperatures (50-170°C) were tested for the first IRSL stimulation while the pIRIR  Palaeosol: At both profiles (Erdut 3 and 4) unit's structure is made of small granular peds (Fig. 3C). Palaeosol is pale brown (10YR 6/3), clayey and compacted. Small (<1 cm) unidentified gastropod shells (Fig. 3C) abundant throughout, while small insect burrows become abundant from 3.7 m. Both boundaries are irregular but quite sharp. Erdut 3: 3.2-4 At Erdut 4 the unit is slightly thinner (70 cm). Top and bottom boundaries are sharp and irregular. VI Erdut 4: 4-9.2 L2 Loess: At Erdut 3 the loess is pale yellow (2.5Y 7/3) dry and silty. The sediment is very compacted and cemented in the top 10 cm. Sediment shifts to light yellowish brown (2.5YR 6/4) at~5.7 m. Erdut 3: 4-6.5 An assemblage of large gastropods (2-3 cm; Fig. 3B) is observed between 4 and 5 m. While a second assemblage of much smaller gastropod shells (max 1 cm) is noted between 5.6 and 6.1 m. Erdut 2: -At Erdut 4 the same colour changes from pale yellow (2.5Y 7/3) to light yellowish brown (2.5YR 6/4) are observed. As with Erdut 3, gastropod assemblages of varying sizes (mm to 2 cm) are abundant between 5.7 and 6.4 m. Additionally, between 7.5 and 8.2 m small (1 cm) to large (>30 cm) carbonate concretions are observed. Erdut 1: -While visible at Erdut 1 and 2 these units have only been observed at a distance. Therefore thickness and sedimentological description is unavailable. VII Erdut 2: 2.4-4 N/A Sandy and laminated complex: At Erdut 1 a 50 cm unit of horizontally laminated silt and very fine sand with sharp top and bottom boundary is observed. The unit's thickness is not uniform laterally and shifts to 70 cm away from the river. At Erdut 2 Unit VII is 1.6 m thick and comprises four subunits: sandy layer (VIIa), horizontal laminations (VIIb), cross laminations (VIIc), and horizontal laminations (VIId). 2.4-2.8 m. VIIa subunit is 40 cm thick of very fine, light olive brown (2.5Y 5/4). Erdut 1: 3.1-3.8 2.8-3.1 m. VIIb preserves 30 cm thick horizontal laminations of silt and very fine sandmm in size and pale yellow (2.5Y 7/4) in colour. Sharp boundary with subunit above and below. 3.1-3.2 m. VIIc comprise cross-laminated silt and very fine sand (10 cm). Sharp boundary with subunit above and below. 3.2-4 m. VIId subunit contains 80 cm thick horizontally laminated silt and very fine sand, pale yellow (2.5Y 7/4). At~3.7 m multiple large (20-30 cm) multiple calcrete precipitations are seen. Sharp boundary with loess below. VIII Erdut 2: 4-6.6 L2 Loess: At Erdut 1 profile unit comprises predominantly lightly yellowish brown (2.5Y 6/4) silt though very fine sand is also present especially within the first 10 cm from the boundary. Dark Fe/Mn and lighter CaCo 3 mottles present throughout the unit though they increase gradually down section. Erdut 1: 3.8-7.1 At Erdut 2 Unit VIII is also predominantly silty with no obvious structures. Below 5 m insect burrows become very abundant.
stimulation temperature was kept at 225°C. Fig. 4 demonstrates that D e is insensitive to the prior-IR stimulation temperature. However, because signal intensity diminishes with an increased IRSL temperature (Fig. 4, inset), 50°C was selected for the IRSL stimulation in this study. Roberts (2012) showed the importance of testing the preheat temperatures, as they can have an effect on the final D e values. They argued for conducting a preheat plateau for feldspar to determine the most appropriate preheat and measurement temperature combination for the samples, as in the case of quartz. This step seems particularly imperative in central European loess as a number of researchers have reported problems with dose recovery when applying only pIRIR 290 protocols (Schatz et al., 2012;Murray et al., 2014;Thiel et al., 2014;Újvári et al., 2014). Therefore, six preheat (205-320°C) and elevated pIRIR temperature (25-30°C lower than the preheat) pairings were tested. Fig. 5 shows the results of the preheat plateau conducted on sample CRO16/2/10. A plateau is observed at the three lower temperatures (205-250°C), with an average D e value of approximately 590 Gy. D e s increased with higher preheat temperatures (>280°C), with the pIRIR 290 protocol producing D e values more than twice as large (1108 ± 129 Gy, Fig. 5). This is consistent with the trends observed by both Roberts (2012) and Zhang et al. (2015), who also saw a tendency for D e to increase as higher preheat temperatures were applied. Given the consistency in D e values measured using lower temperatures, a pIRIR 225 protocol with the 250°C preheat was selected.

Dose recovery
The suitability of the selected temperature paring was tested through the dose recovery experiments, which examined the ability of the pIRIR 225 protocol to recover an administered laboratory dose. The test was applied to four samples: CRO16/ 2/9, CRO16/2/10, CRO16/2/17 and CRO16/2/27. Twelve aliquots per sample were used for the first three samples and six aliquots for the last sample due to sediment availability. The aliquots were bleached in daylight, irradiated in the laboratory with a beta dose approximating their natural dose and measured using the pIRIR 225 protocol (Supplementary Table S1). The average residual signal (see Residual test and signal resetting section) from the 42 discs was subtracted from the measured equivalent dose prior to calculating the dose recovery ratio. In initial dose recovery tests, the test dose constituted~20% of the given dose, which resulted in an underestimation of the given dose by 12%. A similar trend was reported by Colarossi et al. (2018), who showed that low test doses (as a percentage of the given dose) supress the shape of the dose-response curve, causing D e underestimation. To improve the performance of our SAR protocol the test dose was increased to~30% (170 Gy) of the given dose. All discs passed the dose recovery test within uncertainty (Supplementary Figure S2). The average dose recovery ratio for pIRIR 225 protocol was 1.04 ± 0.07 (n = 42, Table 2).
The 2D 0 value has been suggested as a prudent upper limit for dating (Wintle and Murray, 2006  used for the age calculation are not saturated, the D 0 values of each disc were checked. Fig. 6 shows a typical dose-response curve for sample CRO16/2/5, which has a 2D 0 of 842 Gy. Taking into consideration the average dose rate at this site (3 Gy ka -1 ), this suggests that sediment at this site can be dated to~280 ka.

Anomalous fading
It has been suggested that all feldspars suffer from signal loss termed anomalous fading (Wintle, 1973), which can result in D e underestimation and therefore erroneous ages. However, the pIRIR protocols have been shown to utilise more geologically stable, non-fading signal in feldspars (Thomsen et al., 2008;Buylaert et al., 2012). To assess the extent to which the pIRIR 225 signals from Erdut samples fade, and whether fading corrections are required (Huntley and Lamothe, 2001;Kars et al., 2008) fading tests were performed. To measure the fading rates (g 2days ) three aliquots per sample were irradiated with a fixed dose of 170 Gy repeatedly and the luminescence signal measured following a series of pauses (0, 1, 10, 100 and 1000 h) following the protocol of Auclair et al. (2003). Sensitivity-corrected luminescence was plotted against hold time, and data were fitted with an exponential function to determine the fading rate (%/decade normalised). The range of the fading rates for the IR 50 signal varies between 1.14 ± 0.96%/decade and 6.91 ± 1.76%/decade (mean of 3.02 ± 1.25%/decade), whereas pIRIR 225 results range from 0.06 ± 0.85%/decade and 7.52 ± 1.81%/decade with a mean of 1.68 ± 1.13%/decade (Table 2 and Fig. 7). The results show that pIRIR 225 fades less than the IR 50 signal. This relationship also holds true for sample CRO16/2/9 for which the average pIRIR 225 fading rate is slightly higher than the other samples (3.30 ± 1.37%/decade). The fading rate for this sample is a result of an unusually high fading rate recorded for one aliquot, 7.52 ± 1.81%/decade (Fig. 7), which increases the average fading result for this sample. Out of the remaining two aliquots, one is found within the accepted fading range for the pIRIR protocol (0.09 ± 0.86%/decade), while the other one is only borderline (2.29 ± 1.43%/decade). Given the generally low fading rates calculated, fading correction was not applied to the majority of the samples as the recorded pIRIR 225 fading is in line with previously reported laboratory fading rates for non-fading quartz signals . Nonetheless, as sample CRO16/2/9 was found to be borderline acceptable, the age was corrected for fading and presented alongside the uncorrected age. The correction was performed in the R Luminescence package using the calc_FadingCorr script (Kreutzer, 2019) which is based on the Huntley and Lamothe (2001) calculation.

Residual test and signal resetting
Sunlight exposure during aeolian transport should almost completely bleach signal in feldspar (Buylaert et al., 2011). Nonetheless, testing for the size of the residual component is important in aeolian samples because feldspars may have an unbleachable 'residual' component which, if large enough, could cause age overestimation . Sets of 6-12 aliquots from samples CRO16/2/9, CRO16/2/10, Figure 4. D e as a function prior-IR stimulation temperature for sample CRO16/2/10. The average and standard deviation of three aliquots is shown. Inset, the pIRIR 225 response normalised to the natural test dose plotted against the first stimulation temperature.  Given that the IR signals are much harder to bleach in nature than quartz (Godfrey-Smith et al., 1988), samples were further investigated to eliminate the possibility that the signals were not fully reset prior to the burial. To test this, IR 50 signals, which bleach more rapidly in nature, were compared with pIRIR signals (Supplementary Fig. S3). Fading corrected IR 50 ages are slightly lower than pIRIR 225 ages, with an average ratio between the two of 0.86 ± 0.09 σ, although if the individual ratio of 0.58 for sample CRO16/2/25 is excluded, the overall ratio increases to 0.89 ± 0.04 σ. Given that in all cases the IR 50 fading correction is applied to relatively old ages, and therefore the non-linear portion of the dose-response curve these results were accepted as indicative of reasonable signal resetting of both the IR 50 and pIRIR 225 signals. For sample CRO16/2/25 the underestimation is much greater (~58%), which could suggest that the signal was not fully reset in nature and therefore the age of this sample may be overestimated. However, the age is stratigraphically consistent with the samples above and below, and therefore is accepted.

D e calculation
Luminescence signals were screened using typical rejection criteria, i.e. signal recuperation (<5% of the natural), and recycling ratios of 1 ± 10% (Murray and Wintle, 2000). For all measured signals (n = 193), all discs passed the recycling ratio (1.03 ± 0.05), and recuperation was less than 0.57 ± 0.03%. Overdispersion values recorded varied between 0 and 6.7%, with the majority of samples recording no overdispersion. Therefore, equivalent doses were calculated by applying the Central Age Model (Galbraith et al., 1999).

Age-depth model and mass accumulation rates
To assist with the investigation of the age change with depth, and calculation of accumulation rates, an age-depth model based on the pIRIR 225 ages was constructed using Zeeden et al.
(2018) R package. It uses Bayesian and inverse modelling to separate systematic and random errors of luminescence age by creating probability density functions for both, and include only the latter in the uncertainty calculations. All ages discussed in the text refer to modelled ages (Supplementary Table S2).
To investigate dust fluxes and interpret sedimentary deposition, mass accumulation rates (MARs) were calculated based on modelled luminescence ages, following the equation of Kohfeld and Harrison (2003):

MAR gm a SR
where, SR denotes the sedimentation rate (m a -1 ), ƒ eol is the fraction of the sediment that is aeolian in origin, and ρ dry represents the bulk density of dry sediment (g m -3 ). The sediment was interpreted to be aeolian in origin, therefore a ƒ eol value of 1 was used. The commonly applied loess bulk density value of 1.5 g cm -

A new chronology for the Erdut loess-palaeosol section
Thirteen unmodelled and modelled pIRIR 225 luminescence ages from Units V to IX are presented in Figs 8 and 9, Table 3 and Supplementary  Therefore, while underestimation is observed, there are a number of factors that could be contributing to this discrepancy, including, for example, use of an uncorrected IR 50 signal for the equivalent dose estimation (Galović et al., 2009) or use of different grain size fractions (63-90 μm potassium feldspar herevs 4-11 μm polymineral (Galović et al., 2009)) between two studies. The uncorrected IR 50 ages (Supplementary Table S3) for Unit VI show ages in a similar range (53.19-98.96 ka) to those reported by Galović et al. (2009). This suggests that the use of uncorrected IR 50 signal for the equivalent dose estimation is probably the largest source of age underestimation. However, this requires further investigation and is beyond the scope of this paper. Galović et al. (2009) did not obtain ages from the palaeosol (Unit V); however, they attributed ages 46.9 ± 4.8 and 46.5 ± 4.7 ka to the loess above it. While these ages cannot be verified, in view of the rest of the chronological results, they are likely to be underestimated.
At profiles 1 and 2 the luminescence ages provide chronology for laminated sediment (Unit VII), loess (Unit VIII) and bottom palaeosol (Unit IX). Sample CRO16/2/10 (203.73 ± 12.8 ka) from the top of the laminated unit, appears to be significantly overestimating the depositional age and is the only age inconsistent with the rest of the chronological record (within uncertainties). Based on stratigraphy, field observations and grain size analysis (Figs 2 and 3 and Table 1), this sample is interpreted as fluvial in origin. Therefore, ã 60 ka age reversal could potentially be explained by incomplete bleaching of the sediment. Partial bleaching is not unusual in fluvial sediments as complete exposure to light may be restricted in sediment-lain water (Wallinga, 2002). However, it is unclear why the discrepancy is recorded only in one sample and without further investigation using single grain approaches it is difficult to interpret this age.
Four ages from the lowest loess unit (VIII), at Erdut 1 and 2, fall in the 148.69 ± 6.98-158.42 ± 7.77 ka age range, covering a period of 10-16 ka (based on mean and minimum ages). As with Unit VI it is not possible to correlate more precisely the depth at which Galović et al. (2009) sampled from this Unit, and therefore compare it with the results of this study. Nonetheless, their results provide an uncorrected IR 50 age of 61.5 ± 6.2 ka which corresponds to 85 ± 5.6-92.9 ± 6.2 ka IR 50 age range (Supplementary Table S3). As highlighted above, there are other potential sources of the discrepancy, including mineral fraction and grain size, though the largest   170.14 ± 11.13 driver appears to be the use of uncorrected IR 50 signal. The Erdut 1 profile culminates in a thick palaeosol (Unit IX), dated to 231.29 ± 40.63 ka (uncorrected 168.37 ± 11.27 ka). These results suggest that this section spans two glacial-interglacial cycles, rather than just one as was previously proposed. Despite a large overlap between ages from two parts of the profile (Figs 8 and 9), the lateral disappearance of the palaeosols and the laminated fluvial sediments prevents this record from being consolidated into one. Further, the chronological gap recorded at the bottom of profile 4 requires further attention.

Mass accumulation rates
High-resolution loess chronologies extending beyond the penultimate interglacial are limited. As a result, so are mass accumulation rates for this period. Fig. 8C shows the MARs based on the modelled pIRIR 225 ages for the Erdut outcrop. While the most commonly taken approach relies on calculation of MARs based on only the mean age, it has been shown that in an aeolian context this may lead to misleading interpretations of luminescence ages (Leighton et al., 2014). Therefore, the results presented here also show the minimum MAR values.
Palaeosol Unit IX records the lowest accumulation rate in the whole section: 6 g m −2 a −1 and 4 g m −2 a −1 , mean and minimum respectively, though Unit V's rates are also low (19-23 g m −2 a −1 ). Conversely, rates within loess are not only high but also quite variable. The top part of the loess Unit VI shows very rapid accumulation 94-1265 g m −2 a −1 . However, between 143 and 140 ka MARs drop to the average 213 g m −2 a −1 (minimum 33 g m −2 a −1 ). These shifts could indicate changes in a number of factors including sediment production at source, sediment transport (either by the Danube River or the wind), preservation potential, wind direction, and activation/deactivation of an additional or new source.
Importantly, Fig. 8 shows that loess deposition has been neither uniform nor continuous, which are common assumptions in loess research. By nature, loess is a very homogeneous sediment in which depositional gaps or erosional events are not obvious when assessing the stratigraphy. Furthermore, the homogeneity also means that even if an erosional event did occur, the new sediment is likely to have a similar grain size and magnetic susceptibility to the one deposited before. Therefore, unless a change in source or material delivered to a site occurred as well, hiatuses will be difficult to identify from the proxy record. At Erdut the erosional gap is represented by fluvial sediments in profiles 1 and 2. However, fluvial sediments are not preserved in profile 4, yet the dates suggest a sedimentary gap. This hiatus should be further investigated as its size may vary between 1.85 and 47.5 ka, depending on whether a minimum or maximum error is used.
With more high-resolution dating studies conducted, the size of hiatuses in loess records could be better constrained and investigated in the context of climatic and/or local influences on the loess deposition. Whether climatically or locally driven, a number of gaps in loess records have been reported at multiple locations (e.g. Dong et al., 2015;Stevens et al., 2018), highlighting the danger of the loess continuity assumption. The profiles at Erdut provide a unique view of the issue as profile 4 shows a hiatus that would not be recognised due to the lack of lithological changes. However, 100 m away profiles 1 and 2 preserve the likely cause of the gap. This not only shows a need for high-resolution chronological approaches but also investigation of sedimentological sections in a wider spatial context.

Environmental implications
Samples CRO16/2/9 and CRO16/2/7, from the middle of the gradational boundary and the top of Unit IX at Erdut, provide modelled ages of 231.29 ± 40.63 ka and 163.05 ± 7.77 ka, respectively (Fig. 9). This thick, dark and well-developed palaeosol reportedly has relatively high total organic carbon (TOC) content and a large proportion of clay and fine silts (Galović et al., 2009). Although the corrected age for the palaeosol has a very large uncertainty, it falls well within the bounds of MIS 7 and is therefore identified as S2 soil. The luminescence age from the stratigraphic boundary between loess and palaeosol likely represents the transition period between MIS 6 and MIS 7 stages. Other detailed and dated climatic archives for this period are scarce, with the majority located near the southern Balkans and Mediterranean Sea.
As only the transitional zone and top of this palaeosol is exposed, it is not possible to determine the full thickness and appearance of Unit IX palaeosol in Erdut. In the Danubian loess, S2 is represented as a double or triple palaeosol pedocomplex with a strongly developed S2S1 palaeosol that has enhanced magnetic susceptibility values (Marković et al., 2006(Marković et al., , 2009Buggle et al., 2009;Wacha et al., 2013). Currently it is not possible to determine how many palaeosols are preserved at Erdut and therefore how this site fits into the regional loess stratigraphy.
Based on the modelled luminescence ages, Units VIII and VI likely represent the same period of loess deposition (Fig. 9). However, at present it is impossible to precisely combine them into a single stratigraphic unit. Loess accumulation at Erdut began at 170.14 ± 13.49 ka and concluded soon after 139.18 ± 8.75 ka, showing that the site experienced semicontinuous loess deposition for 9-53 ka, based on minimum and maximum ages, respectively (Fig. 9). The chronology suggests that these Units provide a record of MIS 6 and are therefore classified as L2 loess. Further, a drop in the TOC associated with decreased biological productivity, as well as an increase in the proportion of coarse silt and sand is noted in this unit (Galović et al., 2009). These changes suggest increased windiness and/or a change in wind direction, a change in sediment supply, or new source activation; all likely possibilities during glacial periods.
The sites extending beyond the last interglacial are often located near the Danube (Fig. 1). Along the Danube the L2 loess is not as well preserved or exposed as the L1 (the record of the last glacial period). In many cases just the top of the L2 unit is exposed at the base of profiles (cf Fitzsimmons et al., 2012). Nonetheless, the thickness of the exposed L2 unit at Erdut, Vukovar (Wacha and Frechen, 2011), Ruma (Marković et al., 2006), Batajnica (Marković et al., 2009), Süttő , Costineşti (Constantin et al., 2014) and Harletz (Lomax et al., 2018), indicates large quantities of sediment available for transport and deposition along the Danube corridor. This is not surprising as the MIS 6 stage is considered to be one of the most climatically severe periods in Europe, with the farthest extent of the glaciers (Ehlers and Gibbard, 2007;Francke et al., 2016). The glacial action producing the sediment, increased fluvial and aeolian transport, in combination with a cold and dry climate and predominantly dry steppe and/or forest steppe-like conditions (Marković et al., 2006(Marković et al., , 2007(Marković et al., , 2018 supports increased sediment availability in the system. Fig. 8 shows that this period is marked by highly variable accumulation rates. All Erdut's profiles indicate that in the early parts of MIS 6, rates were relatively low: 32-60 g m -2 a -1 (Erdut 1 and 2) and 23-97 g m -2 a -1 (Erdut 3 and 4). This is followed by a temporary drop to 75 g m -2 a -1 (mean) between 158.42 ± 7.34 and 152.43 ± 6.11 ka and rise to~120 g m -2 a -1 (mean). At Erdut 1 and 2, rates increase rapidly (mean 1104 g m -2 a -1 and 125 g m -2 a -1 ) after 145.47 ± 9.33 ka, drop to 213gm -2 a -1 (mean) at 143.29 ± 7.76 ka, and increase to their highest mean value (1265 g m -2 a -1 ; minimum 94 g m -2 a -1 ) between 140.48 ± 7.52 and 139.18 ± 8.75 ka, after which the rates drop until the beginning of MIS 5. This further supports the suggestion that MIS 6 can be divided into two parts at around 155 ka (Margari et al., 2010), with the first part resembling more of an interstadial, and the full glacial conditions beginning after 155 ka, as the global ice volume increases to reach glacial maximum. The division at Erdut is seen in both minimum and mean MARs and occurs after 152 ka in agreement with Margari et al. (2010). This again highlights that loess accumulation is not uniform and continuous during glacial periods and suggests that the sediment dynamics were highly susceptible to change and continued to respond to climatic and/or geomorphological changes.
At the profiles closest to the modern river channel (Erdut1 and 2), the L2 is punctuated by a series of horizontally and crosslaminated sand and silt. These are absent from profiles 3 and 4, only 100 m further away. Laminated deposits were previously interpreted as fluvial (Galović et al., 2009;Galović, 2014), which is supported by field observations conducted during this research. The deposition of Unit VII commenced between 148.69 ± 6.98 ka and 143.36 ± 7.43 ka. However, its termination is unclear as the unmodelled luminescence age from the top of the unit dates it to 203.73 ± 12.8 ka. This age is not consistent with the rest of the stratigraphic and chronological record at Erdut and is probably an overestimation based on the sample's stratigraphic position (see discussion in A new chronology for the Erdut loess-palaeosol section). While profiles 3 and 4 do not contain laminated sediments, a hiatus in ages (170.14 ± 13.49-145.47 ± 9.33 ka), corresponding to the onset of fluvial sediment, suggests that the whole site may have been subject to fluvial influence.
This temporary shift to fluvial sedimentation could indicate changes to the Danube's main channel or a development of a small branching channel. Fluvial sediments also punctuate the L2 unit at Stari Slankamen  and Zmajevac (Galović et al., 2009;Banak et al., 2012) either as a series of laminated sediments or as an erosional feature. Further fluvial features at Šarengrad (Galović et al., 2009) could also be of the same age. However, this cannot be confirmed until the chronologies for these sections are verified and more detailed sedimentological investigations are conducted. Nonetheless, this could indicate a basin-wide hydrological event, potentially restricted to the sites found along the main Danube corridor. However, the exact timing and cause of this event cannot be resolved due to the uncertainty on luminescence dates.
The relatively thin palaeosol (Unit V) dates to 75.01 ± 5.44 ka, with the loess sample below providing a bracketing age of 139.18 ± 8.75 ka, suggesting that this palaeosol represents the last interglacial (MIS 5) (Rasmussen et al., 2014). The bottom age estimate is in line with the onset of Termination II (Moseley et al., 2015), the transition from MIS 6 into MIS 5, supporting its identification as the last interglacial soil (S1).This is also supported by an increase in clay particles and TOC to the highest values in the section (Galović et al., 2009). It is possible that Erdut experienced locally driven, favourable conditions for the pedogenesis to occur over~50 ka (Fig. 9), that resulted in low MARs (Fig. 8). Therefore, Unit V could be preserving a record of the whole interglacial period, including all of the substages in a single palaeosol.
However, the S1 palaeosol at Erdut is relatively thin in comparison with palaeosols of that age across the region (cf Fitzsimmons et al., 2012), e.g. the three-times thicker S1 palaeosol at Vukovar (Wacha and Frechen, 2011), a few kilometres downstream. Furthermore, other S1 palaeosols in the Carpathian Basin, e.g. Dunaszekcső (Újvári et al., 2014), Süttő  and Dolní Věstonice (Fuchs et al., 2013) are represented by double or triple palaeosols. Consequently, partial erosion of the other the pedocomplex at Erdut is likely.
Nevertheless, S1 pedogenic complexes are heterogeneous across the region (cf Fitzsimmons et al., 2012), and palaeosols at Erdut could reflect pedogenic processes driven by local conditions such as palaeotopography (including drainage), vegetation, proximity to the river, and micro-climate (Vandenberghe et al., 2014;Lehmkuhl et al., 2016). This could be pointing to regional and/or local drivers preserved in the loess records on the sub-orbital level. Therefore, the S1 palaeosol at Erdut requires further high-resolution chronological analysis combined with a detailed sedimentological and micromorphological investigation.

Conclusions
A review of the chronology for the loess-palaeosol sequence at Erdut is presented, based on coarse-grain (63-90 μm) potassium feldspar and a pIRIR 225 measurement protocol. The modelled uncorrected pIRIR 225 ages range from 75.01 ± 5.44 to 231.29 ± 40.63 ka, which suggest a semicontinuous loess accumulation at this site since MIS 7. The results show that Erdut preserves much longer climatic and environmental records than originally thought and could therefore greatly contribute to our understanding of the longterm Pleistocene changes in the region. A fluvial event, which disturbed L2 loess accumulation, is also reported. This could be pointing to a wider hydrological that affected loess deposits along the main Danube channel (e.g. Stari Slankamen (Murray et al., 2014)). Other sites in Croatia (Zmajevac and Šarengrad) that preserve fluvial sedimentary sequences in L2 were dated using an uncorrected IR 50 protocol, which this research shows is likely to be underestimating ages. Therefore, it is not possible to compare and provide a more detailed picture of the fluvial event. Consequently, chronologies at Zmajevac and Šarengrad should be revisited. Finally, this research shows that more highresolution approaches based on the pIRIR protocols are needed to better understand the timing of the climatic and environmental changes during the penultimate glacial-interglacial stage.

Supporting information
Additional supporting information may be found in the online version of this article at the publisher's web-site.
Supporting information.