A 62-ka geomagnetic palaeointensity record from the Taymyr Peninsula, Russian Arctic

. This work represents the first palaeomagnetic study carried out on the sedimentary record of lake Levinson-Lessing, which is the deepest lake in northern Central Siberia. Palaeomagnetic analyses were carried out on 730 discrete samples from the upper 38 m of the 46 m-long core Co1401, which was recovered from the central part of the lake. Alternating field demagnetisation experiments were carried out to obtain the characteristic remanent magnetisation. The relative palaeointensity is determined using the magnetic susceptibility, the anhysteretic remanent magnetization and the isothermal remanent 25 magnetization for normalization of the partial natural remanent magnetization. The chronology of Co1401 derives from accelerated mass spectrometer radiocarbon ages, optically stimulated luminescence dating, and correlation of the relative palaeointensity of 642 discrete samples with the GLOPIS-75 reference curve. This study focuses on the part >10 ka but although includes preliminary results for the upper part of the core. The record includes the geomagnetic excursions Laschamps and Mono Lake, and resolves sufficient geomagnetic features to establish infill of this lake is a target for palaeoenvironmental and paleoclimatic studies since the 1990's. Analyses of a first 22.4 m-long sediment core PG1228 revealed continuous sedimentation with nearly constant sedimentation rates of ~70 cm ka -1 within the past ~32 ka (Ebel et al., 1999; Andreev et al., 2003). Seismic surveys suggesting the presence of up to 115 m of undisturbed sediments (Niessen et al., 1999; Lebas et al., 2019) motivated a second coring campaign in the central part of the lake in 2017, which yielded a 46-m-long record at site 60 Co1401 (Fig. 1b). Scheidt et al. (2021a) showed that only the upper 38 m of the associated composite core Co1401 are undisturbed and that this upper part is highly suitable for palaeomagnetic analyses. Here, we present the first paleomagnetic study of lake Levinson-Lessing sediments and compare the relative palaeointensity (RPI) variations of core Co1401 with the most recent version of GLOPIS-75 palaeointensity stack (Laj et al., 2014; Laj and Kissel, 2015). The chronology is supported by accelerated mass spectrometer (AMS) radiocarbon ( 14 C) ages and optically 65 stimulated luminescence (OSL) dating. This work provides new information on the behaviour of the Earth's magnetic field in the Eurasian Arctic during the Late Quaternary and forms the chronostratigraphic backbone for upcoming palaeoenvironmental studies in the region. been detected within cores. Here, core parts are probably twisted against each other. Because of these problems, a pole wander curve was not calculated for Co1401. by doubling of the value for symmetric lock-in functions (e.g., rectangular or Gaussian), but not for other shapes (linear, cubic or exponential). Besides the choice of the lock-in function, the used geomagnetic field reference curves influence the final estimates. Additionally, Mellström et al. (2015) explain the appropriateness of their exceptionally deep 330 lock-in depth values with the relatively low wet density and high organic content (45-50 % loss-on-ignition; LOI) of the lake`s sediment. Following this argumentation and assuming a LOI to TOC ratio of approximately two (Vereș, 2002; Haflidason et al., 2019), the lock-in depths in lake Levinson-Lessing must be well below those specified by Mellström et al. (2015) for both the dipole component of the EMF dominates the secular variation at the Levinson-Lessing site. It remains an open question whether the latitude of Levinson-Lessing Lake of about 74°3' is not close enough to the magnetic North pole to record significant non-dipolar features expected inside the tangent cylinder. As mentioned earlier, gradual lock-in of magnetic minerals over long periods of time would also be a plausible explanation for the low variability of the PSVs. However, considering the determined sedimentation rates, a uniform lock-in depth of 18 cm and a corresponding lock-in width 470 of the half of the lock-in depth, we gain a smoothing constant <220 years, which would probably not erase all PSV. Further insights into the magnetic field on the tangential cylinder may arise from future studies combining the results of this study with RPI and PSV data from other sites, or from modelling the behaviour of the EMF in the Eurasian Arctic using the data presented. 18 O record. This value suggest a smoothing constant of <220 years for the complete sedimentation record. Although the course of the characteristic declination is obscured by the lack of overlap of core segments of core Co1401, the PSV data may be used in the future, if data of the general course of the declination from close-by study sites become available. At present, 495 the lacustrine sediment core Co1401 provides the only high-resolution RPI and PSV dataset extending to ~62 ka within a radius of more than 1500 km around lake Levinson Lessing.

75-GICC05 will be used in the following to distinguish between GLOPIS-75 records with the former and the new time 155 scale, respectively.

Magnetic mineralogy and selected geochemical properties
The magnetic mineralogy of Co1401 was investigated in Scheidt et al. (2021a) to determine the suitability of this core for palaeomagnetic investigations (Tauxe, 1993;Stoner and St-Onge, 2007). Results suggest that the main remanence carriers of 200 Co1401 are ≤5 µm pseudo-single domain (titano-) magnetite and maghemite. Additionally, up to 8 % of the saturation remanent magnetization is carried by single domain particles. The contribution of multi-domain particles is generally low. The magnetic mineralogy appears to be exceptionally homogeneous between 6.7 and 38 mcd. Above 6.7 mcd, bulk hysteresis parameters suggest smaller magnetic grain sizes, but still in the PSD range. As documented by first-order reversal curves, this apparent fining is due to the additional contribution of single domain greigite. The maximum contribution of greigite to the 205 saturation remanence is negligible in the 0-20 mT range but reaches ~50% above ~50 mT at ~3 mcd. Based on these results, samples collected from Co1401 are generally well suited for RPI studies, although caution should be exercised in the upper meters.
After careful inspection of the measurement data, 89 samples were discarded (red dots in Fig. 2), for the following reasons.
First, sections with erratic, discontinuous ChRM directions were assumed to be affected by core disturbances. However, if 210 inconsistent ChRM directions occurred only near the cut edges of the cores, the corresponding RPI values were not discarded, because directional changes may occur at the top and bottom of the core section due to the division of the cores and compression of the core ends during transport. Second, samples with unstable demagnetization behaviour and those that were almost completely demagnetized in ≤30 mT were considered unreliable. Third, samples at the end of core sections with significantly lower NRM values than adjacent samples were discarded because likely affected by oxidation of unstable remanence carriers. 215 Fourth, samples with particularly elevated NRM values (≥2 times larger than adjacent samples) or Fe/Ti ratios were assumed to be lithologically different from the surrounding samples. Such differences could be caused, for example, by turbidite horizons that were overlooked during sample selection.
Samples that passed the above selection criteria are characterized by consistently low and stable values in the Fe/Ti ratio (Fig. 2), which are generally interpreted as an expression of relatively stable redox conditions (cf., van der Bilt et al., 2015). It is 220 important to note, however, that initial greigite formation in the uppermost meters is not detected by this proxy, because the onset of redox conditions first affect the smallest particles of the magnetic mineral fraction, which represent only a small portion of the total Fe content. Sample selection resulted also in a reduction of the already low scatter of NRM, χ, ARM, and IRM values (Fig. 2). Interestingly, downcore variations of the overall low total organic carbon (TOC) content of Co1401 are anticorrelated with concentration-dependent magnetic parameters (Fig. 2). In the upper 6.7 mcd a change in the environmental 225 conditions is likely to cause the decrease in these magnetic parameters and the increase of TOC. Below 6.7 mcd, the anticorrelation might be explained by a sorting effect, with finer detrital material originating from more vegetated areas.

Basic chronology of core Co1401
OSL (Table 1) and the AMS 14 C (Table 2) dating results suggest a quite regular sediment accumulation with an age of ~60 ka 235 at the base of the succession (Fig. 3). Overall, OSL yields slightly younger ages than 14 C dating. Age differences between the two dating methods may arise from redeposition of old organic macro remains or hard-water effects on 14 C ages. On the other hand, OSL results may be influenced by unaccounted changes of the water content, due to insufficient correction of sediment https://doi.org/10.5194/gchron-2021-12 Preprint. Discussion started: 3 June 2021 c Author(s) 2021. CC BY 4.0 License. compaction, or incomplete luminescence signal resetting prior to sedimentation (Lang and Zolitschka, 2001;Buylaert et al., 2013). 240

Palaeosecular variations and relative palaeointensity
AF demagnetisation to 80 mT removes, on average, 92.8 % of the NRM left after the initial 15 mT whole core demagnetization.
A small viscous overprint, which is fully removed after the 15 mT AF step, was observed only in a limited number of samples. 255 Additionally, a few samples around 3 mcd acquired a gyroremanence in fields >60 mT, which is typically related to the presence of greigite (e.g., Scheidt et al., 2015). In the PCA, the ChRM is defined by the steepest part of the AF demagnetization curves, which was generally at higher AF fields in the upper part of the core than in the lower part ( Fig. 4a, 4b). Overall, one stable magnetic direction was extracted in all samples. In case of steep inclinations, the horizontal component frequently clusters close to the origin (Fig. 4c). The extracted characteristic inclination varies between 89.7° and -66.1° (Fig. 2), with a 260 mean of 74.5° and a median of 77.3°, and thereby is slightly less than the 82.1° of the geomagnetic axial dipole (GAD) expected at the site. This deviation may be due to inclination shallowing, but could also originate from sampling procedures. Besides, the sections with lower inclination values between 16 and 20 mcd and the field reversal at ~25 mcd reduce the mean and median characteristic inclination value by 2.3° and 0.6°, respectively. Due to the lack of orientation of the cores with respect to the north direction, and because the individual 2 m-long cores do not overlap, the consecutive rotation of each core section 265 to make the declination curve continuous add a cumulative error that increases with depth. Thus, only relative changes of the characteristic declination within core sections can be considered reliable. In a few cases, sudden changes in declination also have been detected within cores. Here, core parts are probably twisted against each other. Because of these problems, a pole wander curve was not calculated for Co1401.

275
The RPI of Co1401 shows a clear decreasing trend between 0 and ~16 mcd, followed by two intervals with relatively constant average values at ~16-25 and ~25-38 mcd, respectively. Short-term fluctuations are superimposed to these trends. Very similar results are obtained with all three normalizers, as expected from the homogeneous magnetic mineralogy (Fig. 2). However, although all normalizers show the same pattern, there is a strong increase in RPI fluctuations from about 7.4 mcd upwards.
Especially in the upper 6 mcd the largest scatter and highest RPI values occur. This upward increase in variability is probably 280 due to a change in recording efficiency related to a change in magnetic mineralogy reported at about 6.7 mcd (Scheidt et al., 2021a). Alternatively, or in addition, a misorientation of magnetic minerals caused by compaction of the sediment in the lower part could result in lower RPI values.
The lowest RPI values are displayed around 25 mcd and coincide with the negative values of the characteristic inclination (−66.1°). Age and pattern of this event coincide with those of the Laschamp geomagnetic excursion (e.g., Laj et al., 2006;Li 285 et al., 2018;Simon et al., 2020). It is worth noting that samples around ~25 mcd also show the largest PCA scattering (Fig.   4d), which is probably caused by the higher influence of measurement noise at low magnetization values. An additional RPI low in ~21.3 mcd is associated with shallower inclination values rather than true excursional directions and can be attributed to the Mono Lake geomagnetic excursion. The lack of reversed directions for the Mono Lake excursion has been reported before (e.g., Lund et al., 2017). The assigned ages of the two identified geomagnetic excursions are within 2-sigma uncertainty 290 of the absolute dating results (Fig. 3). At ~19.4 mcd a weakened RPI is combined with shallower inclination values. This https://doi.org/10.5194/gchron-2021-12 Preprint. Discussion started: 3 June 2021 c Author(s) 2021. CC BY 4.0 License. feature of the EMF has been recognized similarly in the sediments of the Black Sea at about 31 ka (Jiabo et al., 2019) and, also softened, at about 28 ka in the core PS 2138-1 SL from the Arctic Ocean (Nowaczyk and Knies, 2000). Finally, the lower inclination values at around 17.5 mcd do not correspond to lower RPI values. However, there are no correspondences with a large turbidite, a core end, or a change in the magnetic properties of the rock, nor is there any evidence that this element is a 295 drilling-or sampling-induced artefact.

Remanence acquisition in lake Levinson-Lessing
The chronostratigraphic relevance of a palaeomagnetic sediment sequence depends crucially on the NRM acquisition mechanism. Mineral magnetic analyses of core Co1401 do not suggest significant changes of the recording mechanism. Only in the uppermost part of the succession greigite formation has started. Consequently, it is possible that the magnetic signals of 300 the individual samples from this part of the core integrate over the primary magnetization and a secondary magnetization that arose at any time. Since the greigite formation seems to be initial, we assume the main part of the magnetic signal to be primary.
However, this needs to be analysed in greater detail, so the range up to approx. 6.7 mcd should only be considered cautiously at present.
For the remaining part of the core, between 6.7 and 38 mcd, further comparison with other records require some considerations 305 about the lock-in depth. The lock-in depth is defined as the depth below the sediment-water interface where a post-depositional remanent magnetization (pDRM) becomes locked and contributes to the permanent record. The occurrence and extent of this effect depends on several factors, including water content, grain size, accumulation rate, and bioturbation intensity (e.g., Tauxe et al., 2006;Roberts et al., 2013;Egli and Zhao, 2015;Valet et al., 2017), but the underlying processes are poorly known.
Therefore, the lock-in depth is described by an empirical lock-in function representing the fraction of total NRM that is blocked 310 above a given depth (e.g., Nilsson et al. 2018, and references therein). The median and width of the lock-in function, divided by the accumulation rate, give the mean delay and time constant of the recording process, respectively. Large accumulation rates, as in Co1401, ensure that delay and smoothing associated with the lock-in function are minimized. A small delay is essential for establishing a chronology with small uncertainties.
Co1401 provides a rare continuity in TOC (Fig. 2) grain size, and water content in most parts of the core. Due to the fine-315 grained composition and soft consistency in the upper meters, a significant lock-in effect is expected. Since an accurate and precise chronology (e.g., varve chronology) for a systematic evaluation of the dimension of the lock-in effect is not available for lake Levinson-Lessing, we refer to recent studies of varved sediments from two lakes in Sweden to obtain a rough estimate of the potential acquisition delay. Similar to our study, the age-depth relationships of lakes Kälksjön and Gyltigesjön indicate high and overall nearly constant sedimentation rates of approximately 59 cm ka -1 during the past c. 3 ka (Mellström et al., 320 2015) and approx. 75 cm ka -1 during the past 6 ka (Snowball et al., 2013), respectively. Furthermore, the preservation of Swedish lakes (cf., section 4.4). Assuming a lock-in depth range of 10-30 cm and a sedimentation rate of 70 cm ka -1 for lake Levinson-Lessing, the expected magnetic record delay is of the order of 240-430 years. 335

RPI correlation and implications for Co1401
Absolute ages from OSL and 14 C dating (Table 1, 2; Fig. 3) provide important tie points for the comparison of the RPI record of Co1401 with reference data. Due to the problems with reference datasets described above (section 3.4), we correlate the RPI(ARM) of core Co1401 with the GLOPIS-75 stack and with the six northern, high-resolution records included in this stack, and rely exclusively on its recently updated age model (Fig. 5). Thus, we focus on the time interval covered by these records, 340 which is >10 ka. We decided not to include additional reference records for the younger part of Co1401 for two reasons. First, the presence of greigite associated with the intensification of RPI variation in Co1401 requires a more detailed investigation of the magnetic mineralogy of the individual samples before the RPI can be considered reliable. Second, a large number of RPI records are available for the Holocene providing many and complex perspectives for the discussion, which would go beyond the scope of this publication. As a preliminary result, the range <10 ka is therefore only correlated with the virtual 345 axial dipole moment (VADM) of GLOPIS-75-GICC05 that show a general trend of the intensity variation of the last 10 ka (Fig. 5). For the older part of the record, 12 additional tie lines were defined. Due to the natural variability of the EMF, the influence of sedimentological properties on records, and possible inaccuracies of the applied age models of the reference records, not all of these tie lines connect identically shaped features.
https://doi.org/10.5194/gchron-2021-12 Preprint. Discussion started: 3 June 2021 c Author(s) 2021. CC BY 4.0 License. The correlation procedure started with the Laschamp geomagnetic excursion, which is tie point (TP) 7 (Table 3). TP 6 was set 355 to the RPI low of the Mono Lake geomagnetic excursion. In contrast to Co1401 and almost all of the individual sites shown in Figure 5, the RPI record of the GLOPIS-75 stack does not have pronounced drops in RPI between these two geomagnetic excursions. The 14 C sample G (35.78±0.59 ka cal. BP) provides an age for the decline in the RPI at lake Levinson-Lessing.
The OSL sample O2 (32.6±2.1 ka) is slightly younger than expected from the age model but overlaps within its 2-sigma https://doi.org/10.5194/gchron-2021-12 Preprint. Discussion started: 3 June 2021 c Author(s) 2021. CC BY 4.0 License. uncertainties (Fig. 3). The position of TP 5 in Co1401 was chosen to represent the RPI minimum before the pronounced RPI 360 increase visible between ~30-32 ka in GLOPIS-75 and MD95-2009, and to a lesser extend in P013 and PS2644-5. TP 5 was placed on the RPI low best supported by the 14 C sample F (30.26±0.77 ka cal. BP). Similarly, but starting from the RPI feature at ~15 m in Co1401, TP 4 was positioned in GLOPIS-75 considering 14 C sample E (25.01±0.66 ka cal. BP). If Co1401 would be solely correlated to GLOPIS-75, the minima below 15 mcd in Co1401 would likely be matched to the minima at ~21 ka in GLOPIS to delineate areas with similar trends (Fig. 5). In such a correlation scheme, TP 4 would be shifted to younger ages. 365 The resulting age model (dashed line in Fig. 6a) would be supported by the age of OSL sample O1 (17.6±1.0 ka), but at least 14 C sample E and, depending on downward correlation, possibly 14 C sample F would be too old. Therefore, we tentatively follow the correlation scheme that fit to the ages of the 14 C samples E and F, which is still within a 2-sigma error of the age of O1. This decision is supported by the similarity of the RPI of Co1401, MD95-2009 and SU90-24 in the section between TP 4 and TP 5. 370 The depth interval between TP 3 and TP 4 is the largest between the tie points, because no distinctive correlation features are present in Co1401. The age of 14 C sample D (24.26±0.37 ka cal. BP) was not considered, because admixture of tiny coal fragments has been detected, which might result in erroneously high ages. The upward increasing trend in RPI visible in Co1401 is also present in GLOPIS-75, PS2644-5, MD95-2009 andSU90-24, albeit in a more attenuated form. TP 3 ties the depth level of Co1401 to the age of GLOPIS-75, at which the amplitude of variations gets larger towards the top. Beside 375 GLOPIS-75, the course of the RPI curve of Co1401 between TP 3 and TP 2 is similar to P012, P013, ODP984, and SU90-24. TP 2 is set to an RPI minimum below the significant increase at~4.8 mcd. By this arrangement, the 14 C sample B (6.09±0.17 ka cal. BP) is slightly younger than the expected age (Fig. 6a). However, TP 2 and TP 1 are preliminary, as a detailed analysis of the varying magnetic carriers is necessary and will follow in a succedent study. Thus, we do not further discuss TP 2 and TP 1, nor the deviation from the age of 14 C samples B and A (3.35±0.10 ka cal. BP) suggested by the age model (Fig. 6a). 380 the ages constrained by the geomagnetic excursions. Thus, TP 10 was correlated to an RPI high supported by 14 C sample I before TP 8 and TP 9 were placed in the Co1401 to an RPI minimum and maximum, respectively. The resulting age-depth model is supported by the mean value of OSL sample O3 (45.1±3.1 ka). The course of the RPI of Co1401 between TP 7 and TP 10 is similar to all records shown in Figure 5. TP 11 assigns an age of 55.35 ka to the RPI peak in 34.88 mcd (Table 3). 390 The associated RPI feature is also expressed in PS2644-5 and ODP984 and appears to be slightly shifted to the younger in MD95-2009, P012 and P013 (Fig. 5). This correlation is still within a 2-sigma error of the OSL sample O4. Finally, TP 12 and TP 13 reflect a downcore decrease in RPI as observed in a number of reference records in Figure 5. Although the RPI between TL 11 and TL 13 of Co1401 appears to mimic that of SU90-24 and ODP984, the correlation was complicated due to large data gaps and is thus, associated with a certain degree of uncertainty. 395

400
The resulting age-depth model of Co1401 gives an age of ~62 ka at ~38 mcd (Fig. 6a). The age-depth model suggests two changes in sedimentation rates. From an average of 62 cm ka -1 below 14.8 mcd, the sedimentation rate increases to 95 cm ka -

Conclusion
We conducted a high-resolution study of the RPI of composite core Co1401 from lake Levinson-Lessing using ARM, IRM, and χ as normalizers and obtained consistent results throughout the core. The results for the upper 6.7 mcd indicated a change 485 in magnetic mineralogy including initial greigite formation, and the RPI fluctuations substantially increase in the upper part of the core. The provided correlation of the data <10 ka is thus, preliminary. Palaeomagnetic events recorded by Co1401 include recordings of the geomagnetic excursions Laschamps and Mono Lake. The age-model for core Co1401 was established by a combined approach using the correlation of the RPI(ARM) to the GLOPIS-75 stack at the GICC05 time scale, AMS 14 C and OSL dating. The age-depth model suggests an age of ~62 ka in ~38 mcd. Average sedimentation rates shift from 45 cm ka -1 490 down to ~7.6 mcd (17 ka), via 95 cm ka -1 between 7.6 and 14.8 mcd, to 62 cm ka -1 below 14.8 mcd. The lock-in depth was tentatively set to 18 cm based on a single tie point obtained from correlation of the downcore changes in TOC with the NGRIP δ 18 O record. This value suggest a smoothing constant of <220 years for the complete sedimentation record. Although the course of the characteristic declination is obscured by the lack of overlap of core segments of core Co1401, the PSV data may be used in the future, if data of the general course of the declination from close-by study sites become available. At present, 495 the lacustrine sediment core Co1401 provides the only high-resolution RPI and PSV dataset extending to ~62 ka within a radius of more than 1500 km around lake Levinson Lessing. https://doi.org/10.5194/gchron-2021-12 Preprint. Discussion started: 3 June 2021 c Author(s) 2021. CC BY 4.0 License.