The cryostratigraphy of the Yedoma cliff of Sobo-Sise Island (Lena Delta) reveals permafrost dynamics in the Central Laptev Sea coastal region during the last about 52 ka

The present study examines the formation history and cryolithological properties of late Pleistocene Yedoma Ice Complex (IC) and its Holocene cover in the eastern Lena Delta on Sobo-Sise Island. The sedimentary sequence was continuously sampled in 0.5 m resolution at a vertical Yedoma cliff starting from 24.2 m above rivel level (arl). The 20 sequence differentiates into three cryostratigraphic units; Unit A: dated from ca. 52 to 28 cal ka BP; Unit B: dated from ca. 28 to 15 cal ka BP; Unit C: dated from ca. 7 to 0 cal ka BP. Three chronologic gaps in the record are striking. The hiatus during the interstadial MIS 3 (36–29 cal ka BP) as well as during stadial MIS 2 (20–17 cal ka BP) might be related to fluvial erosion and/or changed discharge patterns of the Lena River caused by repeated outburst floods from the glacial Lake Vitim in Southern Siberia along the Lena River valley towards the Arctic Ocean. The hiatus during the MIS 2-1 transition (15–7 25 cal ka BP) is a commonly observed feature in permafrost chronologies due to intense thermokarst activity of the deglacial period. The chronologic gaps of the Sobo-Sise Yedoma record are similarly found at two neighbouring Yedoma IC sites on Bykovsky Peninsula and Kurungnakh-Sise Island, and most likely of regional importance. The three cryostratigraphic units of the Sobo-Sise Yedoma exhibit distinct signatures in properties of their clastic, organic and ice components. Higher permafrost aggradation rates of 1 m ka with higher organic matter (OM) stocks (29±15 kg C m 30 , 2.2±1.0 kg N m, Unit A) and mainly coarse silt are found for the interstadial MIS 3 if compared to the stadial MIS 2 with 0.7 m ka permafrost aggradation, lower OM stocks (14±8 kg C m, 1.4±0.4 kg N m, Unit B) and pronounced peaks in the coarse silt and medium sand fractions. Geochemical signatures of intrasedimental ice reflect the differences in summer evaporation and moisture regime by higher ion contents and less depleted stable δO and δD isotope ratios but lower deuterium excess (d) values during interstadial MIS 3 if compared to stadial MIS 2. The δO and δD composition of MIS 3 35

8 646 BP (SOB18-06-35) was beyond calibration range (Fig. 4). All three ages do, however, support the age-height relation as they are of approximate age.
Each profile was modelled individually (Fig. S2). Section thickness was adjusted in relation to sampling frequency along each profile to balance model performance (Blaauw and Christen, 2019). Sediment profile SOB18-01 was modelled using twelve 14 C dates, a section thickness of 10 cm and a hiatus at 1.75 m bs (22.45 m arl) and 3.25 m bs (20.95 m arl). Profile 230 SOB-03 was modelled using seven 14 C dates, a section thickness of 30 cm and a hiatus at 7.25 m bs (16.95 m arl). This model was extrapolated 0.5 m beyond the uppermost (youngest) dated sample to cover the entire profile. Profile SOB18-06 was modelled using eight 14 C dates, a section thickness of 40 cm and was extrapolated 2.3 m beyond the lowermost radiocarbon age used in the model to cover the lowermost samples of this profile. SOB18-06 represents the oldest deposits of the dataset and infringes on the age limit of radiocarbon dating (Fig. S2). The median of the modelled probability 235 distribution was used to assign an age to each cm along the profiles.
Additional age information was obtained from one mammoth tusk found at beach level below the sediment profile SOB18-01 (Fig. 3 c) and from host sediments of ice wedges SOB18-08 and SOB18-09 (Table 3). Floral and faunal remains from inside wedge ice were dated where available (Table 3). In total, age information for five ice wedges was obtained from 22 radiocarbon dates. Of those, 19 samples were dated at the MICADAS facility mentioned above and three samples were dated 240 The stacked sequence is not continuous and shows three temporal gaps in the record, which are related to changes in the depositional and/or erosional regimes. Those are discussed in detail in section 5.3. One hiatus is obvious within Unit A (in profile SOB18-03) between about 36 and 29 cal ka BP, one hiatus within Unit B (in profile SOB18-01) between 20 and 17 260 cal ka BP and one hiatus between units B and C (in profile SOB18-01) between about 15 and 7 cal ka BP (Fig. 4).
Age inversions are often observed in permafrost chronologies given the effect of cryogenic processes such as cryoturbation within the uppermost thawed active layer before the material enters the perennially frozen state (Bockheim et al., 2007), and the high vulnerability of ice-rich permafrost to thaw and erosion Günther et al., 2015). An obvious example for the latter is seen in the Sobo-Sise record where sample SOB18-03-17 has an age of 15 294 ± 67 BP (18 570 cal 265 BP) while the entire profile SOB18-03 dates from 25 680 to 40 840 cal BP (Table 3). We assume that this age of 18 570 cal BP most likely represents a contamination from thawed sediment, which was redeposited downwards along the cliff and represents a mixed age of Holocene and older OM. To validate this assumption, additional plant material from sample SOB18-03-17 was picked and dated to 40 840 cal BP, in line with the lower age limit of profile SOB18-03.
Three ice-wedge profiles of Unit A were dated. Seven 14 C dates from ice wedge SOB18-09 range from 48 660 to 36 970 cal 270 BP. We found one infinite age of >48 500 BP from the host deposit at the same height level as the sampling transect. Ice wedge SOB14-IW3 shows ages of 43 270 and of 30 930 cal BP, and two infinite ages of >31 000 BP each. Organic material from ice wedge SOB18-08-I yielded a 14 C age of 49 610 cal BP.
The SOB18-02-I wedge ice of Unit B was dated by two ages of 25 350 and of 23 470 cal BP. The host deposits at the same height level as the sampling profile at about 18 to 20 m arl show an age range from 23 170 to 21 940 cal BP that is in general 275 agreement with the assumed SOB18-02-I formation time. The only direct age information from Unit C wedge ice is available for profile SOB14-IW4 with eight ages spanning from 2 290 cal BP to modern (Table 3). Indirect age information is available for SOB18-08-II whose host deposits at the same height level as the sampling transect were dated to 4 480 cal BP (Table 3) implying a middle to late Holocene formation of this ice wedge.
Relocated material might also enter wedge ice when wintertime frost cracks are filled with snowmelt transporting OM and 280 preserving it in vertical ice veins (Opel et al., 2018). This might be the case for the age determination of 49 610 cal BP in IW SOB18-08-I that is, however attributed to Unit A of MIS 3 age by its isotopic composition.

Cryostratigraphy
Each cryostratigraphic unit is characterised by its specific clastic, organic and ice compositions. Those were captured by field observations (Wetterich et al., 2019) and analytical data that are described in detail below and summarised in Fig. 5,285 Fig. 6 and Table 1. Representation of our analytical results is based on the modelled age-height relation for each profile and their stacking by age (Fig. S2).
Stable water isotope records and age information of six horizontal ice wedge profiles sampled at the Sobo-Sise Yedoma cliff are attributed to the cryostratigraphic units A, B and C by position (Fig. 7), isotopic composition ( Table 2) and age (Table 3).
Two of the ice-wedge profiles (SOB18-08 and SOB18-02) exhibited stable isotope compositions pointing to different stages 290 The eastern part of profile SOB18-08 (differentiated as SOB18-08-I) is characterised by more depleted values in δ 18 O (by 4 ‰), δD (by 40 ‰) and d (by 4‰) if compared to the main part of the profile (SOB18-08-II), which is attributed to Unit C.
Mean values of SOB18-08-I are -29.6 ‰ in δ 18 O, -230 ‰ in δD and 6.8 ‰ in d, close to the respective values of the other two ice wedge profiles of Unit A. 325 In summary, the ice wedges of Unit A show most depleted mean values down to -29.9 ‰ in δ 18 O (range from -31.4 ‰ to -26.9 ‰) and -232 ‰ in δD (range from -244 ‰ to -213 ‰). They plot mainly below the GMWL and show low d between 5.2 to 7.4 ‰ in comparison to IWs of units B and C. The slopes in co-isotopic plots of ice wedge data from Unit A vary between 7.2 and 8.3 (Fig. 9 c).

Unit B (MIS 2, Yedoma IC, 28 to 15 cal ka BP) 330
Unit B comprises the uppermost two samples of sediment profile SOB18-03 and most of sediment profile SOB18-01 (except of its uppermost four samples that belong to Unit C; Fig. 4). Unit B is composed of brownish grey poorly sorted sandy silt (mean grain-size 113 ± 64 µm) and occasional sand lenses resulting in a bi-modal GSD and pronounced peaks in the coarse silt and medium sand fractions (Fig. 8 b).
Generally coarser grain-size distributions than in Unit A are characteristic for Unit B and supported by the EMMA results. 335 The middle sand rEM4 (primary mode at 310 µm) is present in the lower part of Unit B while fine silt rEM1 and coarse silt rEM2 dominate the upper part of Unit B (Fig. 5). The fine sand rEM3 is less frequent in Unit B compared to Unit A.
The magnetic susceptibility of Unit B has a mean of 53 ± 9 SI. OM is present as single twig remains (2-4 mm in diameter), dark brown spots, finely dispersed organic remains, and peaty lenses (5 to 25 cm in diameter). The mammoth tusk found at beach level below the profile SOB18-01 most likely originates from Unit B deposits belonging to the faunal component of 340 OM. It was radiocarbon-dated to 16 480 cal BP and thus fits into the age range of Unit B. This finding fits well into the fossil record of the late Pleistocene mammoth fauna in the region (Kuznetsova et al., 2019). The OM content of Unit B is lower compared to that of Unit A with mean values of 2.1 ± 1.3 wt% for TOC and 0.2 ± 0.1 wt% for TN, resulting in mean C/N of 10.5 ± 2.4.
The OM isotopic composition exhibits lower mean values than in Unit A of -26.1 ± 0.6 ‰ for δ 13 C and 1.9 ± 1.0 ‰ for 345 δ 15 N. The DOC content of intrasedimental ice of Unit B is generally lower compared to those of Unit A with values from 85 to 589 mg L -1 (mean of 212 mg L -1 ).
The ice content of Unit B is the lowest of all units with 43 ± 10 wt%. Prevailing cryostructures are lenticular (1-5 cm thick ice layers in 1-20 cm distance), and reticulate (1-2 mm thick ice lenses 4-12 mm long) or wavy parallel (1 mm thick ice lenses 4-10 mm long) between the ice layers. If compared to Unit A, the intrasedimental ice of Unit B shows similar mean 350 values and comparable ranges in δ 18 O of -26.2 ± 2.2 ‰, and in δD of -200 ± 16 ‰ ( Fig. 9 a). The mean d value of about 10 ‰ is much higher than in Unit A, ranging from 3 ‰ to 15 ‰.
The hydrochemical composition of intrasedimental ice of Unit B shows an upward decreasing trend in ion content with electrical conductivity ranging from about 3180 to 1130 µS cm -1 (mean: 1810 µS cm -1 ) (Fig. 6). Ca and Mg cations dominate the cation composition while Cl concentrations decrease upwards. 355 The only IW record of Unit B was obtained in the eastern part of profile SOB18-02. Like SOB18-08-I, the isotopic composition of SOB18-02-I differs from the western part of its profile (SOB18-02-II attributed to Unit C) by more depleted isotopic mean values and a lower d; -28.8 ± 0.5 ‰ in δ 18 O, -225 ± 5 ‰ in δD and d of 5.8 ± 0.9 ‰. The values plot below the GMWL and the co-isotopic plot shows a slope of 9.4 ( Fig. 9 c).

Unit C (MIS 1, Holocene cover, 7 to 0 cal ka BP) 360
The uppermost four samples of sediment profile SOB18-01 represent the cryostratigraphic Unit C including the uppermost seasonally thawed active layer (of 0.2 m on 20 July 2018 at the sampling site) that consists of modern vegetation and peat ( Fig. 4). Below the active layer, grey poorly sorted sandy silt is present. Its grain-size distribution is bi-modal with peaks in the coarse silt and medium sand fractions (mean grain-size of 66 ± 13 µm; Fig. 5, Fig. 8 a). The mean MS is the lowest of all units with 32 ± 23 SI, which corresponds to the highest OM content (present in numerous peaty lenses, 2 to 25 cm in 365 diameter) with mean TOC of 11.3 ± 9.9 wt% and mean TN of 0.6 ± 0.3 wt%. The C/N is highest for all units with a mean value of 18.5 ± 8. The OM stable isotope composition exhibits the most depleted mean value of -28.0 ± 02 ‰ for δ 13 C of all The hydrochemical composition in intrasedimental ice of Unit C was characterised in only one sample, which shows a very low electrical conductivity of 36 µS cm -1 , and major ion concentrations of less than 2 mg L -1 except for Ca and Fe.
One complete IW profile (SOB14-IW4) and two profile parts (SOB18-08-II and SOB18-02-II) belong to the cryostratigraphic Unit C of Holocene cover deposits. We furthermore consider the Holocene ice wedge profile SOB14-IW5 380 from the lowermost part of the Yedoma slope that might represent a former thermokarst basin (alas) level overlying Yedoma  -196 ‰ and -190 ‰ with d values of 13.8 ‰, 13.8 ‰ and 11.2 ‰. All wedge ice records of Unit C are clearly distinguished from those of units A and B by less 385 depleted δ 18 O and δD and d values well above 10 ‰ (Table 2). They plot predominantly above the GMWL. The co-isotopic plot reveals a large range of 7 ‰ in δ 18 O from -30.4 ‰ to -23.4 ‰ and of 53 ‰ in δD from -227 ‰ to -174 ‰. The respective slopes vary between 7.7 and 8.8 (Fig. 9 b).

Permafrost aggradation rates and deposition history
Excluding the chronological gap between about 36 to 29 cal ka BP within Unit A, we assume continuous permafrost aggradation of the MIS 3 Yedoma IC on Sobo Sise from at least about 52 to 36 cal ka BP, which is represented by a 16-mthick permafrost sequence. The resulting aggradation rate of the MIS 3 Yedoma IC amounts to about 1 m per thousand years.
The continuous permafrost aggradation assumed from MIS 2 Yedoma IC between about 28 and 20 cal ka BP excluding the 395 hiatus within Unit B (from 20 to 17 cal ka BP) formed a 5-m-thick sequence at a rate of about 0.7 m per thousand years. If compared to the Bykovsky Yedoma IC record (site Mamontovy Khayata; Schirrmeister et al., 2002a), higher aggradation rates are obvious; about 1.5 m per thousand years for MIS 3 (12-m thick sequence between 46 and 38 cal BP) and about 0.85 m per thousand years for MIS 2 (6-m thick sequence between 28 and 21 cal ka BP). Thus, less permafrost aggradation during MIS 2 than during MIS 3 is also seen on Bykovsky Peninsula. However, it should be noted that the syngenetic growth 400 of ice-oversaturated permafrost such as Yedoma IC is not only controlled by clastic and organic sedimentation, but further triggered by formation of pore and segregation ice that contributes to Yedoma IC on Sobo-Sise 49 ± 10 wt% in MIS 3 and 43 ± 10 wt% in MIS 2 deposits. The volumetric ice content based on the absolute ice content (assuming ice saturation if the ice content is >20 wt%) according to Strauss et al. (2012) amounts to 66 ± 9 vol% and 65 ± 8 vol% for MIS 3 and MIS 2, respectively (Fuchs et al., 2020). At this rather equal volumetric share of intrasedimental ice during MIS 3 and MIS 2, 405 mainly organic accumulation seems to have controlled the difference in permafrost aggradation rates. It should further be noted that growing ice wedges deform the frozen deposits in between by material transport from the polygon center toward the rim and upward push (Mackay, 1981). Thus, the vertical thickness of the sediment layers, determined now, might exceed the initial thickness due to the formation of intrasedimental (excess) ice, but also due to lateral material transport by the growing ice-wedges. The MIS 1 cover deposits accumulated the uppermost 1.4 m since about 6.4 cal ka BP. Due to freeze-410 thaw cycles in the active layer and thaw subsidence on the modern Sobo-Sise Yedoma surface of several centimeters per year (Chen et al., 2018), the aggradation rate for Unit C has not been calculated.
The MIS 3 Yedoma IC of Unit A is characterised mainly by coarse silt and partly by fine sand. This is also seen in the prevalent rEM2 and rEM3 of Unit A and differs from the bi-modal grain-size characteristics of the MIS 2 Yedoma IC of 14 ( Fig. 8; Fig. 10; Fig. S1). Such changes in grain-size distributions of MIS 3 and MIS 2 Yedoma IC may point to different material sources and/or transport processes. A study by Schirrmeister et al. (2020) of Yedoma IC deposition history, sources and material transport mechanisms includes the neighbouring study sites on Bykovsky Peninsula and Kurungnakh-Sise Island ( Fig. 1), but lacks a differentiation into MIS 3 and MIS 2 Yedoma IC as undertaken on data from Sobo-Sise. Therefore, a direct comparison per formation period is of less use to disentangle changes in sedimentation over time, 420 although some general information can be deduced. The medium-sand rEM4 grain-size class of the Sobo-Sise data relates to high-energy transport including saltation in meltwater runoff or fluvial water (rEM2 in Schirrmeister et al., 2020). The finesand rEM3 represents overbank deposits or settled suspensions in temporarily flooded sections during floodplain deposition (rEM4 in Schirrmeister et al., 2020), while the coarse-silt rEM2 (rEM5 in Schirrmeister et al., 2020) might relate to floodplain deposition as well, but could also originate from aeolian sources (Vandenberghe, 2013) or frost weathering 425 processes (Schwamborn et al., 2012). The fine-silt rEM1 (rEM8 in Schirrmeister et al., 2020) reflects low-energy settling of suspended material from aeolian or pedogenic sources under still-water conditions, which is characteristic in low-center polygon ponds. Thus, the Yedoma IC of Sobo-Sise formation during MIS 3 with prevailing fine sand (rEM3) and coarse silt (rEM2) derived mainly from floodplain-related or meltwater runoff alluvial deposition processes, but possibly also includes aeolian and frost-weathering components. The same coarse silt rEM2 dominates the grain-size distributions from MIS 2 430 deposits with a pronounced peak ( Fig. 8; Fig. 10; Fig. S1). We therefore assume a depositional regime similar to that in MIS 3 for this time. MIS 2 deposition, however, shows a second pronounced peak in medium sand (rEM4) pointing to occasional high-energy material transport in alluvial fan environments with strong meltwater runoff and/or fluvial transport. Sparse vegetation cover as deduced from LGM climate conditions in the area  might have promoted the potential for high transport energy in a barren landscape. The differing grain-size compositions at the three location on 435 Bykovsky, Sobo-Sise and Kurungnakh-Sise reflect local diversity in accumulation processes for example with higher fluvial input on Kurungnakh-Sise Island, but generally support the multi-process and multi-source regional Yedoma IC formation (Schirrmeister et al., 2020).

Organic matter stocks and decomposition
The OM characteristics of the Sobo-Sise Yedoma IC differentiate into twofold higher organic carbon (TOC mean of 4.5±2.6 440 wt%) and 50% higher nitrogen (TN mean of 0.3±0.1 wt%) contents in Unit A (MIS 3) if compared to those of Unit B (MIS 2) with mean TOC of 2.1±1.3 wt% and mean TN of 0.2±0.1 wt%. The resulting C/N ratios are slightly higher in Unit A with 12.9 than in Unit B with 10.5 (Table 1). A more productive tundra-steppe environment during MIS 3 (Unit A) with higher OM accumulation at comparable decomposition rates if compared to MIS 2 (Unit B) is deduced.
Furthermore, Unit A has significant higher carbon and nitrogen densities with a mean of 29±15 kg carbon m -3 and 2.2±1.0 kg 445 nitrogen m -3 compared to 14±8 kg carbon m -3 and 1.4±0.4 kg nitrogen m -3 in Unit B. Consequently, the OM input into the Lena River by fast erosion of the Yedoma cliff of Sobo-Sise (up to 22.3 m yr −1 ; Fuchs et al., 2020) is mainly controlled by Unit A that stores twice the amount of carbon compared to Unit B and which is exposed over about two thirds of the cliff height (Fig. 4).
The Holocene cover of Unit C shows highest TOC (mean 11.3±9.9 wt%), TN (mean 0.6±0.3 wt%) and C/N (mean 18.5±8.0) 450 of the record although of large variability (Table 1) mainly due to the low number of samples (n=4) in Unit C. However, for active layer samples in the Holocene cover layer of Sobo-Sise, Fuchs et al. (2018) detected mean values with high variability, too, with 6.7±7.4 wt%, 0.4±0.1 wt% and 15.8±12.3 for TOC, TN and C/N, respectively, indicating a general heterogeneity in OM accumulation in the uppermost soil layer. If compared to the neighbouring Yedoma IC sites on Bykovsky Peninsula (Schirrmeister et al., 2002a) and Kurungnakh-Sise Island (Schirrmeister et al., 2003;Wetterich et al., 455 2008a), the same pattern in OM properties over time from MIS 3 to MIS 1 supports regionally similar variations in palaeoenvironmental conditions.
In relict permafrost, the stable carbon and nitrogen isotope composition of organic matter is strongly controlled by the original botanical composition and further altered by decomposition (Weiss et al., 2016). The latter leads preferentially to loss of isotopically lighter 12 C and 14 N and thus enriches relatively the fraction of the heavier isotopes 13 C and 15 N by 460 leaching and mineralisation processes (Tahmasebi et al., 2018). This fractionation towards less depleted isotopic carbon and nitrogen compositions over time occurs before the OM enters the perennially frozen state. Thus, the permafrost aggradation rate during distinct periods further influences the rate of OM decomposition. However, the differences seen per units in the Sobo-Sise Yedoma record are minor for δ 15 N with mean values of around 2 ‰ for all three units (Table 1, Fig. 11). The δ 13 C Unit mean values vary over about 2 ‰ (between about -28 ‰ and -26 ‰), and are most depleted for Unit C (MIS 1). Due 465 to these only little variations and the range overlap, no significant differences in OM decomposition can be interpreted from the stable carbon and nitrogen isotope composition for the three units.

Intrasedimental ice characteristics
Highest DOC concentrations up to 754 mg L -1 in MIS 3 together with highest average C/N ratios indicate OM preservation ( Fig. 5; Fig. 6). Rapid sediment and OM accumulation rates, as indicated by radiocarbon-based age-depth relationship, lead 470 to effective syngenetic permafrost formation so that particulate and dissolved OM are rapidly incorporated into permanently frozen deposits. Hence, OM degradation is minimised and labile or soluble DOM fractions have not been drained or flushed out from rapidly aggrading permafrost.
The stable water iotope (δ 18 O and δD) and major ion compositions as well as DOC concentrations of Sobo-Sise intrasedimental ice reflect the general cryostratigraphy and have palaeoclimate implications. Preservation of pore water 475 during formation of segregated ice occurs via a wide range of processes. Nevertheless, several studies (e.g. Mackay, 1983;Murton and French, 1994;Kotler and Burn, 2000;Schwamborn et al., 2006;Fritz et al., 2012) have shown that δ 18 O and δD isotopes in intrasedimental ice can still reflect environmental and climatic changes when considered with caution and/or focused on pore ice (Porter et al., 2019;Porter and Opel, 2020). Higher δ 18 O and δD, but lower d values are found in MIS 3 compared to MIS 2 ( Fig. 6; Fig. 9). Relatively warm summers during the MIS 3 interstadial might explain the lower d values 480 in associated intrasedimental ice due to a higher water loss by evaporation (i.e., kinetic fractionation). This would lead to a water reservoir in polygon ponds and soil moisture that becomes successively depleted in 16 O and 1 H compared to the original precipitation. Increased temperature and precipitation amplitudes during MIS 3 Pitulko et al., 2017) may have led to frequent drying and re-wetting in polygon tundra and thus to enhanced kinetic fractionation. Another process of kinetic fractionation producing the same pattern are multiple freeze-thaw cycles of soil moisture in the active 485 layer (Throckmorton et al., 2016).
Elevated ion (Mg, Ca, Na, Cl) concentrations with EC up to 5800 µScm -1 in the MIS 3 record (Unit A; Fig. 6) are likely caused by frequent drying and re-wetting in polygonal tundra in times of higher summer temperature and precipitation amplitudes during the interstadial compared to MIS 2 stadial (Unit B). Meyer et al. (2002a) found similarly elevated EC values of 5500 µScm -1 in MIS 3 deposits on Bykovsky Peninsula. Modern surface waters in Central Yakutia at high 490 continentality show EC values of up to 5710 µScm -1 (Wetterich et al., 2008b) and even up to 7744 µScm -1 (Pestryakova et al., 2018). Ion-rich pore waters have also been found in MIS 3 deposits at Buor Khaya Peninsula (Schirrmeister et al., 2017), but with different composition and including a distinct saline horizon. In contrast, ion composition in the Sobo-Sise Yedoma IC remained stable throughout MIS 2 and MIS 3 and is dominated by Mg, Cl and Ca in both units. Therefore, we assume that water and sediment sources did not change over time, but reflect higher evaporation during warmer summers in MIS 3 if 495 compared to MIS 2.

Palaeoclimatic implications from regional wedge-ice records
Sobo-Sise ice wedge stable isotopes show a complex pattern that at least in parts can be related to the fact that Holocene ice wedges formed epigenetically within older late Pleistocene deposits and penetrated pre-existing ice wedges. This may be related to subsidence and thermo-erosional processes that thaw permafrost, lower the surface and complicate the 500 stratigraphic attribution of the wedge ice. The stable isotope composition of ice wedge profiles sampled in the central (SOB18-02) and western parts (SOB18-08) of the Sobo-Sise cliff allows differentiating late Pleistocene and Holocene wedge ice.
Generally, late Pleistocene wedge ice is characterised by well-depleted δ 18 O and δD values (mean values between -30 ‰ and -29 ‰, and -232 ‰ and -225 ‰, respectively; Fig. 9 c) and low d values (means between 5‰ and 7‰; Table 2). In 505 contrast, a striking feature of Holocene ice wedges are their significantly elevated mean d values between 11‰ and 15‰ (Table 2) accompanied by surprisingly low δ 18 O and δD values (mean values between -28 ‰ and -25 ‰, and -207 ‰ and -190 ‰, respectively; Fig. 9 b). In some instances, Holocene δ 18 O and δD values reach the range of the late Pleistocene ice wedges (Table 2). This is true for both the oldest Holocene ice wedge stage, i.e. the toes of ice wedge SOB14-IW5 at the Ice Complex-Alas slope and the late Holocene to modern ice wedges on the top of the Ice Complex (e.g. SOB14-IW4). Hence, 510 the isotopic difference between late Pleistocene and Holocene ice wedges is more pronounced in d than in δ values.
All co-isotopic regression slopes are highly correlated (R 2 > 0.97) and vary between 7.16 and 9.43 (Table 2). While the ice wedges of units C (MIS 3) and A (MIS 1) show relatively coherent patterns, the Unit B ice wedge SOB18-02-I sticks out with a value of 9.43 (Fig. 9 b and c), likely related to a comparably low internal isotope variability. Hence, we assume that the isotopic composition of all ice wedges carry paleoclimate information for the winter season and is not significantly 515 altered by secondary fractionation processes.
The Sobo-Sise ice-wedge stable isotopes of units A and B fit mostly well into the regional pattern of the Central Laptev Sea coast and the Lena Delta (Fig. 1)  Sise. As the sampled ice wedges likely cover different parts of the MIS 3 and MIS 2 Yedoma IC and therefore different time slices, the slight differences should not be spatially interpreted in terms of winter temperature differences.
The stable isotope compositions of the studied Sobo-Sise ice wedges do not show any significant differences between ice wedges of units A and B, corresponding to MIS 3 and 2, respectively. This might indicate that the globally cold LGM is not reflected in the Sobo-Sise ice wedge-based winter climate record and would be in accordance with both regional scale, when 530 compared to Bykovsky Peninsula (Meyer et al., 2002a) or to other study sites in the Laptev Sea region , and also on Arctic-wide scale (Porter and Opel, 2020). In this context, we observe (1) a depositional gap temporally coinciding to peak LGM conditions for the three sites at regional scale and (2) extremely depleted LGM ice-wedge isotopes have been only found at Bol'shoy Lyakhovsky Island further east ( Fig. 1;  . As such it is not sufficiently resolved yet, whether this is due to a less cold LGM climate in the region or whether the LGM cold period is not 535 captured by the studied ice-wedge profiles that do not preserve a continuous record. In accordance with Holocene ice-wedge records at Bykovsky and Kurungnakh-Sise, the Sobo-Sise ice wedges of Unit C show distinctly warmer winters and significantly changed moisture generation pattern compared to the late Pleistocene records. Overall Holocene mean ice wedge δ values on Sobo-Sise are enriched by about 1.8 ‰ to 2.7 ‰ for δ 18 O and 23 ‰ to 30 ‰ for δD, over MIS 2 and MIS 3 ice wedges, respectively. Mean Holocene ice-wedge d value (14.2 ‰) is about 7 ‰ 540 and 8 ‰ higher compared to MIS 2 and MIS 3, respectively, indicating substantial changes in the moisture generation and transport patterns (e.g. Meyer et al., 2002a). Similar changes have been observed on Bykovsky Peninsula (Meyer et al., 2002a), while Holocene ice wedges at Kurungnakh-Sise show more enriched mean δ values and lower mean d values (Schirrmeister et al., 2003;Wetterich et al., 2008a). It has to be noted that the Holocene ice wedge stable isotope compositions for both Sobo-Sise and Bykovsky exhibit significantly more depleted δ values and significantly higher d values compared to other ice-wedge study sites along the Siberian Arctic coastal lowlands (Opel et al., 2019). Holocene minimum δ values even fit well into the typical MIS 3 and MIS 2 isotopic range. It is, however, unlikely, that this particular region, i.e. the eastern Lena River Delta and the western Tiksi Bay, is characterised by a significantly colder winter climate. Hence, other potential explanations have to be considered, such as regional specifics of the water cycle. The significantly depleted δ values and increased d values of Holocene ice wedges show some similarities to early winter precipitation (October to 550 December) that is, in particular characterised by distinctly increased d values (e.g. Kurita, 2011;Bonne et al., 2020). Hence, the Holocene ice wedge stable isotope composition might be explained by an over-representation of early winter snow to the melt water feeding ice-wedge cracks. This could be related to specific moisture generation and transport patterns influencing the precipitation in this particular region. A second option could be the contribution of moisture from local sources such as evaporation of isotopically depleted and high deuterium excess Lena River water (Juhls et al., 2020) in the period of ice 555 build-up, resulting in a substantial snow cover development in the early winter season. An only little mixed ocean and substantial open water areas with mainly freshwater signature could explain why this low δ and high d values pattern for Holocene ice wedges could so far only be observed in the eastern Lena Delta region.

Chronostratigraphy of the Yedoma IC in the Central Laptev Sea coastal region
The geochronological record of the Sobo-Sise Yedoma IC spans the last about 52 cal ka BP based on the stacked age-height 560 modelling. Older parts of the Yedoma IC are likely to be found up to several meters below the modern river level (Fuchs et al., 2020) as it has also been reported from other sites in the eastern Lena Delta (Pavlova and Dorozhkina, 2000) and from Bykovsky Peninsula . The lowermost sample of the Sobo-Sise record had a finite age of 47 021 ± 646 BP (SOB18-06-35) but is beyond the limit of calibration. However, it supports the modelled age range of the record down to 51.8 cal ka BP. The entire record exhibits three substantial chronological gaps, which are from about 36.7 to 28.4 565 cal ka BP, from about 20.4 to 16.8 cal ka BP and from about 15.5 to 6.4 cal ka BP (Fig. 12).
Taking into account that the exposure conditions and the applied sampling and dating resolution largely define the quality of the resulting geochronological record, we compare our Yedoma IC dataset from Sobo-Sise Island (32 14 C dates over 24 m profile length) to similar ones with a resolution of largely better than 1 m in vertical dimension, i.e. those from site Mamontovy Khayata at Bykovsky Peninsula (51 14 C dates over 37 m profile length; Schirrmeister et al., 2002a;Grosse et al., 570 2007) and Kurungnakh-Sise Island in the central Lena Delta (19 14 C dates over 19 m profile length; Schirrmeister et al., 2003;Wetterich et al., 2008a). However, the sampling approaches differed. On Bykovsky and Kurungnakh-Sise the exposures were sampled during different years at highly dynamic thaw slumps over a rather large lateral extent, i.e. up to several hundreds of meters. Exposed baidzherakhs (thaw mounts of former polygon centers) at different height levels were sampled. In contrast, the permafrost sampling at the vertical Yedoma cliff on Sobo-Sise was performed in three closeby (i.e. 575 within about 120 meters) overlapping profiles resulting in complete coverage of the exposed permafrost sequence.
If compared to nearby studied Yedoma profiles to the east on Bykovsky Peninsula in the Central Laptev Sea and to the west on Kurungnakh-Sise Island in the central Lena Delta a similar pattern is striking. In detail, the Bykovsky record spans from about 60 ka BP and shows the smallest gaps of all considered records from 38 to 32.5 cal ka BP, from 21 to 18 cal ka BP and from 12.5 to 9 cal ka BP (Fig. 12). The Kurungnakh-Sise record shows two large age gaps from 37 to 21 cal ka BP and from 580 20 to 9 cal ka BP (Fig. 12) found in two independent sampling campaigns (Schirrmeister et al., 2003;Wetterich et al., 2008a).
The hiatus overlap recognised at all three Yedoma sites studied in the region, i.e Bykovsky, Sobo-Sise and Kurungnakh-Sise ( Fig. 1), results in three gaps of likely overarching relevance that are found during MIS 3 from 36 to 32.5 cal ka BP, during MIS 2 from 20.5 to 18 cal ka BP and during MIS 2-1 transition from 12.5 to 9 cal ka BP (Fig. 12). 585 To explain the observed gaps in the chronological records, two mechanisms need to be discussed that are (1) no or extremely low deposition during a certain period of time, and/or (2) thaw and erosion of a certain sequence after deposition. Both mechanisms might be related to a variety of processes spanning from global to regional climate variations over time to local geomorphologic disturbance processes that are not necessarily or solely climate-triggered. To disentangle the general hiatus of three time periods at three Yedoma IC sites in the Lena-Laptev region, the following discussion lines can be drawn. 590

Interstadial climate variability and consecutive local disturbance vs. fluvial erosion during MIS 3
The proposed regional overlap hiatus in MIS 3 Yedoma IC deposits spans 3 500 years (36 -32.5 cal ka BP, Fig. 12). The interstadial climatic variability during MIS 3 deduced from permafrost sequences of NE Siberia was subject to previous studies which assume a MIS 3 climatic optimum expressed by warm summer conditions; mainly based on botanic proxy data (e.g. Anderson and Lozhkin, 2001;Andreev et al., 2011;Murton et al., 2015Murton et al., , 2017Pitulko et al., 2017). The proxy record, 595 however, varies in both duration and timing at different Yedoma IC study sites from the Western Laptev coast to the Kolyma lowland (see Fig. 11 in Wetterich et al., 2014). If warmer summer climate conditions during MIS 3 interstadial led to partial Yedoma IC thaw and to the observed overall depositional gap at 36-32.5 cal ka BP by deepening of the active layer and related surface subsidence resulting in degradation rates exceeding aggradation rates, the timing of the MIS 3 climatic optimum is of specific interest. The close-by Yedoma IC site on Bykovsky Peninsula allows for proxy-based reconstruction 600 of MIS 3 interstadial environmental conditions. Here, warm conditions with mean summer temperatures >12 °C and the occurrence of standing water are deduced for around 40-39 cal ka BP from fossil findings of e.g. Callitriche hermaphroditica that is a temperate aquatic plant (Kienast et al., 2005) supported by findings of diverse ostracod faunae that inhabited low center polygon ponds during the same time period (Wetterich et al., 2005). However, all evidence from Bykovsky records for MIS 3 climate optimum predate the observed hiatus by about 7 000 to 6 000 years which makes it 605 unlikely that the observed MIS 3 gap was driven by regional climate. The MIS 3 hiatus in the Kurungnakh-Sise Yedoma IC spans even more from about 37 to 21 cal ka BP -missing substantial parts of the MIS 3 (Schirrmeister et al., 2003;Wetterich et al., 2008a), but again only after the supposed MIS 3 climatic optimum that is likewise in the Bykovsky Yedoma paleontological record reflected by warm summer conditions and the presence of low-centered polygon tundra providing a broad landscape mosaic of ecological niches for e.g. insects and plants indicating dry and warm conditions in drained positions and for aquatic organisms inhabiting polygon ponds (Khazin et al., 2019;Wetterich et al., 2008a). The observed MIS 3 hiatus in the Bykovsky, Sobo-Sise and Kurungnakh Yedoma IC records is not during the MIS 3 climatic optimum, but instead falls to a period of late MIS 3 climate instability as expressed in Dansgaard/Oeschger (D/O) events recorded in the Greenland ice cores (e.g., Dansgaard et al., 1993;NGRIP members, 2004). But it is unknown whether D/O events have impacted the East Siberian Arctic and the lack of deposits or coarse chronology in the Yedoma IC records 615 prevents to draw conclusion on a linkage to these palaeoclimatic events.
Another possible explanation for the MIS 3 hiatus is proposed by Margold et al. (2018) who found evidence for repeated cataclysmic outburst floods from the glacial Lake Vitim in Southern Siberia into the Vitim River valley and further into the Lena River valley towards the Arctic Ocean. The flooding events were dated by multiple techniques including opticallystimulated luminescence dating and cosmogenic nuclide dating (Be-10 bedrock exposures and Be-10 depth profiles). The 620 reconstructed flood chronology spans over the last 60 ka of which the timing of megaflood II at around 34 ka fits into the chronologic gap observed in the Yedoma IC chronology of Bykovsky, Sobo-Sise and Kurungnakh around 36-32.5 cal ka BP (Fig. 12). Thus, fluvial impact by the proposed megaflood event might have affected the continuity of the Yedoma IC chronologies by eroding considerable parts of the sequences. But except for the chronology gaps no direct erosional features such as fluvial sand or pebble layers have been observed in the outcrops. Thus, direct erosion seems unlikely at the studied 625 locations and the flooding events may here have only changed the hydrological regime (e.g. by developing new discharge paths similar to the channels in the today's Lena Delta) for a certain time period and by doing so prevented the deposition of fluvially transported material in the areas of the studied Yedoma IC outcrops. If so, this could have stopped or minimised Yedoma IC accumulation at the study sites.

LGM climate vs. fluvial erosion during MIS 2 630
The second distinct overlapping hiatus in the Yedoma IC chronologies of Bykovsky Peninsula, Sobo-Sise and Kurungnakh-Sise islands occurred during MIS 2 at 20.5-18 cal ka BP (Fig. 12), and falls partly in the last glacial maximum (LGM) period around 26.5-19 cal ka BP (Clark et al., 2009). The LGM environments of the Laptev Sea coastal region are characterised by cold and dry summer conditions and represented in pollen records by grass-dominated communities with Caryophyllaceae, Asteraceae, Cichoriaceae, Selaginella rupestris . Further paleontological evidence for cold and dry 635 summers is provided by plant macrofossils and insect fossil records from the Bykovsky Yedoma IC (Kienast et al., 2005;Sher et al., 2005) while the LGM is almost not captured in the Kurungnakh-Sise Yedoma IC record (Schirrmeister et al., 2003;Wetterich et al., 2008a). The less productive summer conditions most likely hampered OM accumulation while reduced ice-wedge growth might be related to less winter precipitation and stronger wind activity affecting snow drift and sublimation; both leading to reduced Yedoma IC formation during MIS 2 if compared to MIS 3 as also seen in the lower 640 permafrost aggradation rate (see section 5.1.2). However, no permafrost aggradation during MIS 2 at all seems unlikely in the larger study region since it has a good depositional representation in several Yedoma IC sites (Duvanny Yar -Murton et al., 2015;Yana lowland, Pitulko et al., 2004Bol'shoy Lyakhovsky -Wetterich et al., 2011;Mamontov Klyk -Schirrmeister et al., 2008; Fig. 1). On Bol'shoy Lyakhovsky Island (north-east of the Central Laptev Sea region) a shift from accumulation on top of the MIS 3 Yedoma IC to valley positions was found and explained by a lowered erosion base due to 645 LGM sea level lowstand and according changes in the hydrological system of areas with higher relief inclination . As seen in Fig. 12, the MIS 2 chronologic time gap is larger in the central Lena Delta (about 11 ka between about 20 and 9 cal ka BP on Kurungnakh-Sise) if compared to the eastern Lena Delta (about 3 ka between about 20 and 17 cal ka BP on Sobo-Sise) and to Bykovsky Peninsula (about 3 ka between about 21 and 18 cal ka BP at Mamontovy Khayata, site no. 1 in Fig. 1). Additionally, Grosse et al. (2007) reported an observation at the northern end of Bykovsky Peninsula 650 where 22-m thick MIS 3-2 Yedoma IC (dated from about 53 ka BP to 23 cal ka BP) is discordantly covered by 3-m thick sand with organic interlayers of probably shallow fluvial origin dated to about 16 cal ka BP (site B-S in Grosse et al., 2007; site no. 2 in Fig. 1). For the large gap in the Kurungnakh-Sise MIS 2 Yedoma IC record it might also be possible that the MIS 2 gap likely induced by fluvial erosion of megaflood III (Margold et al., 2018) further combines with the deglacial (MIS 2-1) gap and any possible deposition in between got eroded by the latter. 655

Deglacial thermokarst during MIS 2-1
The global lateglacial to early Holocene warming manifested the transition from glacial to interglacial conditions. The effect of warming on permafrost conditions is largely captured by an increase in ground temperature, a deepening of the seasonally thawed active layer, surface subsidence, and activation of thermokarst and thermo-erosional processes (e.g., Wetterich et al., 2009). The resulting ground ice melt and permafrost thaw led to re-organisation of the post-Beringian periglacial landscapes 660 and accumulation areas remaining after the opening of the Bering Strait around 11 cal ka BP (Jakobsson et al., 2017) and the subsequent Holocene sea-level rise and shelf inundation (Bauch et al., 2001;Klemann et al., 2015). The large-scale warming pulse terminated the accumulation of the Yedoma IC during the lateglacial period as it is seen in the age gaps of the Yedoma IC records considered here (Fig. 12) leading to an overlap hiatus at 12.5-9 cal ka BP. Similar lateglacial-Holocene hiatus are found for many other chronostratigraphic records of Yedoma ICs (Fig. 1) such as in the Kolyma lowland at the Duvanny Yar 665 site (Murton et al., 2015), on the New Siberian Islands (Schirrmeister et al., 2011a;Wetterich et al., 2009Wetterich et al., , 2014, on Buor Khaya Peninsula (Schirrmeister et al., 2017) and on Mamontov Klyk (Schirrmeister et al., 2008). During the lateglacial to early Holocene warming intense thermokarst within degrading Yedoma IC created new accumulation areas, i.e. thermokarst basins and thermo-erosional valleys, which dominate the modern surface morphology in Arctic lowlands by more than 50 % of the modern surface on Bykovsky Peninsula (Grosse et al., 2005;Fuchs et al., 2018), on Sobo-Sise (Fuchs et al., 2018) and 670 on Kurungnakh-Sise islands (Morgenstern et al., 2011).
Dated records of thermokarst deposition commonly fit into the hiatus that represents the end of Yedoma IC formation.
Lateglacial to early Holocene thermokarst deposits on Bykovsky Peninsula are dated from about 10 to 1 cal ka BP (Schirrmeister et al., 2002a), on Kurungnakh-Sise Island from about 15 cal ka BP to modern (Morgenstern et al., 2013) and on Sobo-Sise Island from about 7.4 cal ka BP to modern (Fuchs et al., 2018). Thermo-erosional valleys as erosional features cover deposits on top of Yedoma IC are common and also observed on Sobo-Sise where they were dated from 9.8 to 1.3 cal ka BP (Fuchs et al., 2018) and from 6.4 to 2.4 cal ka BP (Unit C in this study). In summary, overall climate warming at the transition from glacial to interglacial conditions promoted extensive Yedoma IC thaw and created new accumulation areas in thermokarst basins and thermo-erosional valleys. Both, the IC degradation and the change in deposition processes caused the 680 hiatus on top of the Yedoma IC of the Laptev Sea coastal region.
Beringian conditions superimposed in its preservation by thaw events that were fluvially-triggered during MIS 3-2 and climate-triggered during MIS 2-1. 710

Data availability
Original data will be available at PANGAEA after final acceptance of the paper: https://doi.org/10.1594/PANGAEA.919470, 2020 .

Author contributions
SW conceptualised the research. SW, MiFr, TO and LS designed the fieldwork, which was performed with help of AK and 715 AA. AK performed the climbing and sediment sampling of the Yedoma cliff, while SW, MiFr, TO and LS performed the wedge-ice sampling. Laboratory work and data analyses were carried out by SW, MiFr, TO, HaMe, LS, GM, JW, JR, MaFu and HeMa. SW wrote the manuscript with input from all co-authors.