A new 30,000 year chronology for rapidly deposited sediments on the Lomonosov Ridge using bulk radiocarbon dating and probabilistic stratigraphic alignment

We present a new marine chronostratigraphy from a high-accumulation rate Arctic Ocean core at the intersection of the Lomonosov Ridge and the Siberian margin, spanning the last ~30 kyr. The chronology was derived using a combination of bulk 14C dating and stratigraphic correlation to Greenland ice-core records. This was achieved by applying an appositely developed Markov chain Monte Carlo algorithm for Bayesian probabilistic alignment of proxy records. The algorithm simulates depositionally realistic alignments that are consistent with the available radiocarbon age estimates and allows 15 deriving uncertainty bands associated with the inferred alignment. Current composite chronologies from this region are reasonably consistent with our age model during the Holocene and the latter part of deglaciation. However, prior to ~14 kyr BP they yield too old age estimates with offsets that linearly increase up to ~40 kyr near the onset of Marine Isotope Stage (MIS) 2. Our results challenge the robustness of previous regional chronostratigraphies and provide a new stratotype for correlation of sediment cores from this sector of the Lomonosov Ridge and East Siberian slope. In particular, they call for a 20 re-interpretation of events in recent sea-ice proxy reconstructions (Xiao et al., 2015) inaccurately attributed to MIS-3 and the Last Glacial Maximum.


Introduction
Sedimentation rates along many of the Arctic margins are an order of magnitude higher (>5-10 cm/kyr) than in the central basins (<1-2 cm/kyr) (Wegner et al., 2015). These higher accumulation rates can provide detailed insights into glacial and 25 Holocene Arctic paleoceanographic variability (i.e. . High-resolution marine sediment cores from Arctic margins are used to constrain deglacial transgression of the shelves (Bauch et al., 2001;Cronin et al., 2017;Jakobsson et al., 2017;O'Regan et al., 2018), reconstruct variability in sea-ice (Stein and Fahl, 2000;De Vernal et al., 2005;Xiao et al., 2015), and fluxes of organic matter from rivers and coastal erosion (Hilton et al., 2015;Martens et al., 2019;Stein et al., 2001;Tesi et al., 2016). 30 2 Like all paleoclimate time series, Arctic Ocean reconstructions must be anchored in a robust chronology (Backman et al., 2004). Although tephra-chronology  and paleosecular variation (Barletta et al., 2010;Lisé-Pronovost et al., 2009;Lund et al., 2016) were successfully applied in the Chukchi and Beaufort seas, radiocarbon dating of calcareous microfossils continues to underpin most Glacial and Holocene chronologies from the Arctic Ocean. However, the rare and often discontinuous occurrence of calcareous microfossils in Arctic Ocean sediments, commonly hinders the development of 35 detailed age models. Altogether, these issues make the reconstruction of depositional histories of Arctic Ocean sedimentary records inherently challenging. One approach to overcome this is through stratigraphic correlation to independently dated sediment cores, or marker horizons (Alexanderson et al., 2014;O'Regan et al., 2019;Sellén et al., 2010). These regional composite stratigraphies provide a valuable tool for dating sediments, but can also result in perpetuating errors in the developed chronologies. As such there is a continued need to revisit, test and improve upon them, as well as to quantify age uncertainties 40 to enable researchers to gauge interpretations and conclusions with respect to their results.
In this study we present a new detailed chronology of the last ~30 kyr for a core retrieved from an area with high sediment accumulation on the Asian end of the Lomonosov Ridge, bordering the east Siberian and Laptev Seas. The need for revising the chronology of sediments from this region of the Arctic emerged from discrepancies between a previously proposed age model and results from 14 C radiocarbon dating of bulk sediments. 45 The new chronology was derived by leveraging the temporal information from bulk radiocarbon dating and stratigraphic correlation to Greenland ice cores using a new probabilistic alignment algorithm appositely developed for this study. The benefits of the algorithm are threefold: it simulates depositionally realistic marine sediment-ice core proxy data alignments that are consistent with the existing independent radiocarbon age estimates; it ensures reproducibility of age-model solutions; it allows deriving age confidence bands inherent to the alignment process via Bayesian uncertainty inference. Our new age 50 model results are considered in the context of published composite chronostratigraphies from this sector of the Arctic Ocean.
Age discrepancies are quantified and implications of these findings are discussed.

Coring and study site
Core SWERUS-C3-31-PC (hereafter referred to as 31-PC) was acquired on Leg 2 of the SWERUS-C3 2014 Expedition on IB 55 Oden, which departed August 21 from Barrow, Alaska, and ended October 3 in Tromsø, Norway. The core was retrieved from the intersection of the southern Lomonosov Ridge and the Siberian continental margin, bordering the East Siberian and Laptev seas (79.91˚ N; 143.23˚ E; 1120 m water depth) (Fig. 1). It was collected with a piston corer (PC) with an outside/inside diameter of 110/100 mm, rigged with a 1360-kg core head. The unsplit sediment cores were allowed to equilibrate to room temperature (20 o C) and subsequently logged shipboard on a Multi-Sensor Core Logger (MSCL). The cores were split and 60 described shipboard, and imaged using a digital line-scanning camera.

Physical properties
Bulk density (ρB) and magnetic susceptibility (Bartington loop sensor) were measured at a downcore resolution of 1 cm on the MSCL (Fig. 2). Porosity (ϕ) was calculated from the MSCL measured bulk density using: (1) 65 where a constant fluid density (ρF) of 1.024 g/cm 3 and grain density (ρG) of 2.67 ±0.01 g/cm 3 were applied. The grain density was determined by 11 shorebased measurements made on freeze-dried sediments from 31-PC using a Micromeritics, AccuPyc 1340, helium displacement pycnometer.
Sediment grain size (<2 mm) was measured at a 2 cm downcore resolution using a Malvern Mastersizer 3000 laser diffraction particle size analyzer. Wet samples were immersed in a dispersing agent (<10% sodiumhexametaphosphate solution) and 70 placed in an ultrasonic bath to ensure full particle disaggregation before analyses.

TOC analysis and 14 C dating
Bulk total organic carbon (TOC) analyses were performed after freeze-drying and homogenizing 80 samples taken from 31-PC at 10 cm intervals (Fig. 2). A split of ~10 mg of sediment was weighed in silver capsules and acidified with 3M HCl to remove carbonates. The TOC of the samples was measured using a Carlo Erba NC2500 elemental analyzer in the Department 75 of Geological Sciences, Stockholm University.
Eight bulk sediment samples and one marine mollusk shell were sent to NOSAMS for 14 C dating (Table 1). Despite shipboard and shore-based efforts, no further carbonate microfossils were found in 31-PC that could be used for radiocarbon dating. In order to remove carbonates from the bulk sediment samples, the samples received HCl vapor treatment. Results were reported as conventional 14 C years (Stuiver et al., 1977) and then converted into 14 C calendar years (cal ka BP) using Calib 7.1 (Stuiver 80 et al., 2018). Bulk 14 C estimates were calibrated using the IntCal13 radiocarbon calibration curve (Reimer et al., 2013), whereas the mollusk shell 14 C estimate was calibrated against Marine13 using DR = 0 &,--./--to conservatively account for unknown changes in the local marine reservoir correction.

Probabilistic algorithm for stratigraphic alignment 85
In the absence of independent means for constraining the chronology of core 31-PC and due to the scarce fossiliferous content, we here rely on a novel combination of i) bulk radiocarbon dating and ii) proxy-to-proxy stratigraphic alignment. Both dating strategies come with limitations and benefits. For instance, bulk sediment dates do not always reflect the true depositional age of the associated stratum and can yield considerably older ages due to varying contamination with 14 C-dead carbon. However, despite its inherent poor accuracy, bulk dating still offers valuable chronological information as it generally provides a 90 maximum estimate of the age of deposition for a particular stratum. On the other hand, correlation of proxy-climate records leans on the assumption that changes in a particular stratigraphic parameter in an undated record correspond to approximately contemporaneous events in a master reference record that contains direct dating information. This method, though theoretically accurate, always has potential for subjectivity. As such, manual identification of tie points across the input and target records introduces subjective constraints on sedimentation rate changes, which are often difficult to validate and do not incorporate 95 uncertainties associated with the proposed alignment.
To surmount these issues and ultimately derive a chronology for 31-PC that leverages the strengths of both dating strategies, we applied a tailor-made hidden Markov model alignment algorithm. The algorithm evaluates alignments probabilistically based on direct observations of sedimentation rate changes from radiocarbon data measured on the sediment core of interest.
This strategy not only improves alignment accuracy, but produces realistic alignments, ensures reproducibility, and allows 100 deriving Bayesian credible bands associated with the alignment procedure.
In our algorithm, radiocarbon dates are used as gateways to bound a relatively large empirical and depositionally-realistic depth-age space that conservatively encapsulates the sedimentation history of the record of interest. This domain is subsequently explored by sampling alignments between an input proxy record and an independently-dated target record in proportion to their probability. More explicitly, in this study we use observed sedimentation rate changes, inferred from the 105 available bulk (and one fossiliferous) 14 C estimates (Table 1), to probabilistically evaluate the alignments between sediment porosity (ϕ) from 31-PC -a robust sedimentological parameter mainly reflecting grain size variations-and GISP2 18 O data (Seierstad et al., 2014) -an indicator of large-scale climate variability. The alignment hinges on the assumption that shifts in sedimentological properties at our coring site and high-latitude climate, as recorded in Greenland ice cores, are virtually synchronous. This assumption is consistent with existing Arctic paleoceanographic reconstructions of the last glacial cycle, 110 which show a good agreement with Greenland ice-core 18 O profiles (Cronin et al., 2012). While considering the probabilities for sedimentation rate changes at our coring site, the algorithm estimates a sample of optimal alignments that relate the depth scale of 31-PC to the timescale underlying the GISP2 18 O data, i.e. the Greenland Ice Core Chronology 2005 (GICC05) (Rasmussen et al., 2006).
Our numerical approach builds upon previous work using Bayesian Markov chain Monte Carlo inversion for probabilistic 115 alignment of paleoclimate records (Malinverno, 2013;Muschitiello et al., 2015;West et al., 2019), which has been successfully applied on a variety of terrestrial and marine archives Sessford et al., 2019;Wohlfarth et al., 2017).
It should be noted that our algorithm, though analogous in spirit to that devised by (Lin et al., 2014), is fundamentally different.
The method by (Lin et al., 2014) weighs the probability of an alignment between a given input and a target record according to a distribution of sedimentation rate change obtained from a spatial network of radiocarbon-based estimates from low 120 5 latitudes. Our method instead, rather than relying on compilation-based observations, employs direct in-situ radiocarbon estimates of depositional variability.

Statistical method
Prior to alignment, the input ϕ and target GISP2 18 O timeseries were scaled between -1 and 1. The alignment is described by random variables Ai, that relate the depth of the i th data point in the input record (i=1, 2,…, n) for all depths to unknown ages 125 on the target GICC05 timescale. We then express an alignment vector A=(A1, A2,…, An) as a series of assignments of GICC05 ages for every depth in the input record d=(d1, d2,…, dn) at which ϕ was measured. Note that this strategy circumvents the shortcomings of using interpolation to process unevenly spaced data and enables alignment of each individual ϕ data point to the GISP2 18 O record. The alignment vector is defined at any point i on the depth scale of 31-PC by a linear interpolation between 11 depth-age nodes ( Fig. 3 and Table 1): one surface node; one node based on a mollusk shell 14 C estimate; eight 130 nodes based on bulk 14 C estimates; and one perturbed node at a random location dk (k=2, 3,…, n-1) to ensure sampling sedimentation rate changes between any two consecutive empirical age constraints. The nodes strictly follow depth-age paths that do not violate the principle of superposition in order to ensure that the depositional age of the input record increases monotonically with depth. Considering a perturbed node at depth dk, sedimentation rates between two adjacent levels dk and dk±1 are allowed to change over a wide range of values spanning ~0.01-100 cm/yr. 135 Since the alignment process is fundamentally uncertain, we apply probability theory in the age assignments of the alignment vector A. Specifically, the requirement of a good match between the ϕ and the GISP 18 O records that at the same time accounts for unobserved changes in sedimentation rates is here derived using Bayes' rule of probabilistic inference. The rule combines the probability that the given ϕ data would be observed for a particular alignment (i.e. the likelihood model), with the probability that a given alignment would be observed independent of any ϕ data (i.e. the prior model). 140 The likelihood is specified by the probability for a given residual misfit between the aligned ϕ record and the target GISP2 18 O. In the alignment problem posed here, it determines the set of nodes that give a good alignment between the input and the target by weighing the competing needs of a small root-mean-square deviation (RMSE) and a high coefficient of correlation (r). The RMSE formula is: where is the number of aligned data points in the input record, 9 is the rescaled ϕ value for i th point in the input record, and 9 is the corresponding rescaled GISP2 18 O value of a proposed alignment.
The prior, on the other hand, is specified by the probability of any given depth-age function. The posterior probability for any given alignment is proportional to the product of likelihood and prior, and can be written as follows: where are the parameters associated with the alignment and θ the parameters used in the likelihood model. The denominator in equation (3) reflects the full likelihood calculated by summing over all possible k mutually exclusive and exhaustive alignments.
The notation ( | , θ) gives the probability that the input ϕ record would be observed for a particular alignment A. The notation ( | ) specifies the probability for the alignment vector A given the parameters used to constrain the depth-age 155 model relationship (i.e. the depth-age nodes). ( | , , θ) gives the posterior probability of a given alignment. Calculation of the posterior probability proceeds by sampling an initial value for each unknown parameter from the associated prior distributions using a reversible jump Markov chain Monte Carlo (MCMC) sampling (Vihola, 2012). The algorithm continues by: 1. Proposing a "candidate" depth-age model and the resulting alignment between the input ϕ data and the target GISP2 18 O 160 record.
2. Accepting or rejecting the candidate depth-age model (and associated alignment) according to its posterior probability using the Metropolis-Hasting algorithm (Hastings, 1970;Metropolis et al., 1953), whereby the posterior probability is higher for alignment functions that yield a closer match between ϕ and GISP2 18 O (i.e. as defined by a relatively smaller RMSE and higher r). 165 3. Repeat from step 1 for 5 x 10 5 iterations.
By convention the sample was divided in two parts. We discarded the initial 10 5 samples ("burn in" period), which in our case corresponds to the MCMC sample length necessary to reach model convergence. The median of the remainder alignment sample was used to infer the posterior optimal correlation while its variability was used to obtain posterior alignment credible bands. Similarly, the sample of accepted depth-age models was used to find posterior median and credible bands for the 170 chronology of 31-PC.

Evaluating the existing regional chronology
Sediment physical property measurements (bulk density, magnetic susceptibility, grain size and TOC) performed on 31-PC 175 allow us to firmly establish a lithostratigraphic correlation to a neighboring sediment core (PS2767-4) collected by RV Polarstern during Expedition ARK-XI/1 in 1995 (Rachor, 1997). PS2767-4 is underpinned by a published composite chronology that was constructed using cross-correlation and the amalgamation of dating information across a number of regional marine sediment sequences (Figs. 1 and 2) (Stein et al., 2001;. Core PS2767-4 is located only 24 km south of 31-PC (Fig. 1). This 8.22 m long core is thought to be younger than 60 kyr BP,180 implying that it contains a complete record of MIS-3 through 1 (Stein et al., 2001;. It has been used to investigate organic matter delivery from the Laptev shelf (Stein et al., 2001;Stein and Fahl, 2000), and more recently for biomarker based reconstructions of sea ice variability during the last glacial cycle Xiao et al., 2015).
The Holocene and postglacial age model for PS2767-4 was dated primarily using 14 C radiocarbon dates on three mollusk shells, while the pre-Holocene stratigraphy is based on correlation to other records where oxygen isotopes, 185 magnetostratigraphy and dinoflagellate biostratigraphy have been used to infer approximate ages (Stein et al., 2001). These older regional age constraints were mapped onto PS2767-4 using a regional correlation between magnetic susceptibility records and sediment lithology (Stein et al., 2001). Stein and Fahl (2012) recognized that in the absence of direct dating, the pre-Holocene stratigraphy of this record remained tentative. (Figs. 1 and 2). Examination of the stratigraphic profiles for 31-PC and PS-2767-4 on their independent depth scale reveals an unequivocal coherence across multiple parameters, which argues 190 in favor of a consistent depositional history at the two sites (Fig. 2).
Although the postglacial and Holocene age model for PS-2767-4 is fairly robust, being variably dated by 14 C measurements in each of the correlated records, the inferred ages for glacial sediments (MIS 2-4) are considerably less well-constrained ( Fig.   2) . Boundaries between MIS 2, 3, and 4 are demarcated mainly by correlative changes in sediment lithology, organic matter content and dated using a dinocyst-based stratigraphy developed for PS2471-1 (Matthiessen et al., 2001) (Fig.  195 1). Results from our bulk 14 C radiocarbon dates illustrate a clear deviation in the proposed age model for PS-2767-4 beyond ~14 kyr BP. The two near-bottom bulk 14 C dates yield an uppermost bound of ~31.8 and ~33.5 kyr BP (Table 1, Fig. 2) for the base of 31-PC. Given that the near-basal depositional age of 31-PC cannot be older than the late MIS-3/early MIS-2, it seems unlikely that sediment from PS2767-4 can date back to 60 kyr BP (i.e. MIS-4), as previously suggested . 200

Proposed age model
The outcomes of the probabilistic alignment and the resulting age model for 31-PC are presented in Figure 3. The correlation between ϕ and GISP2 18 O requires some considerations. Large-scale climate shifts may not be manifested in a similar way in different proxies. In fact, our ϕ record integrates a number of processes, such as sediment grain size, composition, transport and deposition at the coring site. By contrast, GISP 18 O is responsive to changes in Greenland air temperature. Therefore, the 205 two parameters likely do not scale in a linear fashion, and a perfect match should not be expected. Nonetheless, the degree of correlation between the marine and ice-core data -especially during deglaciation and MIS-2-is surprisingly well defined and overall compelling.

8
The results show that 31-PC features relatively linear sedimentation rates with a mean of ~41 ±24 (1s) cm/kyr, with the exception of the last ~9 kyr where sedimentation rates are slightly lower (~15 cm/kyr) (Fig. 3). This is in line with the expected 210 decrease in depositional rate resulting from post-glacial transgression of the Siberian shelves (Bauch et al., 2001;Tesi et al., 2016).
The mean 99% posterior credible intervals for our age model is ~±1.46 kyr (Fig. 3). Mean uncertainties are larger during the Holocene (~±1.85 kyr). The larger age error, reflects on one hand the absence of radiocarbon-based age constraints in the upper 1.4 m, and on the other hand the lack of structure in the ϕ and GISP2 18 O data during the Holocene, which makes the 215 match statistically less robust (Fig. 3).

Implications of the new age model
The implications of these findings are critical for reconstructing and understanding the oceanographic history in this region of the Arctic Ocean. The revised age model reveals substantially faster sedimentation rates during the LGM and deglaciation, compared to the existing regional chronology (Fig. 4). Published proxy reconstructions based on the original chronology of 220 PS2767-4, and putatively assigned to the Last Glacial Maximum (LGM)/MIS-2 and MIS-3, are flawed. Hence, the reconstructed paleoceanographic timeline before deglaciation and related interpretations may not be fully reliable and should be taken with caution. In particular, the attribution of events in recent low-resolution biomarker-based reconstructions of Arctic sea ice conditions (Xiao et al., 2015) require revision for this sector of the Arctic Ocean. This is important, as currently PS2767-4 is the only record from the currently seasonally ice covered southern Lomonosov Ridge where glacial and deglacial 225 reconstructions of sea ice exist (Fig. 4). We suggest that the sea-ice record should be interpreted differently than as outlined in the original publication. According to the new marine chronology presented in this study, variable -though generally more extensive than present day-sea-ice cover characterized the central Arctic Ocean during the early LGM/MIS-2 (rather than during the latter part of MIS-3); the second half of deglaciation (rather than LGM/MIS-2) featured increasing sea-ice growth with permanent sea ice throughout the year; decreased sea-ice cover occurred around the transition between deglaciation and 230 the Holocene (rather than during deglaciation) (Fig. 4).

Conclusions
We present a new chronostratigraphy from the Asian end of the Lomonosov Ridge spanning the last ~30 kyr using a combination of bulk radiocarbon dating and stratigraphic correlation to Greenland ice-core records. The alignment was obtained using a novel probabilistic stratigraphic alignment algorithm. The algorithm simulates correlations of marine and ice-235 core proxy records that are consistent with the observed changes in sedimentation rates obtained from independent radiocarbon dates, and ultimately yields uncertainty bands associated with the alignment process.

9
Stratigraphic comparison with a nearby record constrained by a composite regional age model highlights substantial chronological shortcomings in this region prior to ~14 kyr BP. We identified a linearly increasing age offset that builds up to ~40 kyr at the beginning of LGM/MIS-2 and that question previous attributions of plaeoceanographic events from MIS-3 to 240 the early deglaciation. Specifically, our results allowed us to partly re-interpret the sequence of events observed in a recent sea-ice proxy reconstruction from the central Arctic Ocean (Xiao et al., 2015).
Our new chronostratigraphy constitutes an important regional benchmark that helps revise the paleoceanographic timeline of the central Arctic from MIS-2 to 1, and can serve as a template for future correlations of regional sediment sequences with poor independent age control. 245 Author contribution: FM and MO initiated the study and wrote the manuscript. FM designed the Bayesian probabilistic alignment method and constructed the chronology of core SWERUS-C3-31-PC. JM performed the TOC analysis and did the sampling for 14 C dating. GW performed the physical properties analyses with assistance from MO. OG and MJ participated with the interpretation of the results. All authors were involved in editing the manuscript.