The studied piston core SWERUS-L2-13-PC1 (hereafter referred to as 13PC; Figure 1) was collected from the Arlis Plateau (76.18° N, 179.29° E) during the Swedish-Russian-US expedition on the Swedish icebreaker Oden in 2014, from a water depth of 1119 m. The core length was 7.68 m. Samples for organic carbon and sterols were taken onboard after splitting, with 10 cm³ scoops and frozen at −20°C. The core was stored at 4°C before the archive halves were subsampled with standard palaeomagnetic cubes (7 cm³) at Stockholm University in 2016.

Bulk density (BD, ρ) and magnetic susceptibility (κ) were measured shipboard at 1 cm downcore resolution on whole core sections approximately 24 h after the core was recovered. Measurements were performed using a GeoTek multi-sensor core logger (MSCL). Line scan images were taken from the archive halves after splitting. A fall cone was used to measure the undrained shear strength of the sediment on the working halves. Sediment colour is often used in the western Arctic Ocean as an indicator for glacial-interglacial cycles and cross-core correlation (Jakobsson et al., 2000). We measured sediment colour of the archive halves with a handheld Minolta spectrophotometer (model CM-700d) at 2 cm intervals at Stockholm University. Colours are expressed as components of lightness (L), green-red (a^∗) and blue-yellow (b^∗). Radiographic images were obtained from the archive halves using an ITRAX core scanner (Stockholm University) with a Mo tube at 30 kV, 45 mA and 200 μm increments.

Relative elemental composition was measured by means of X-Ray Fluorescence (XRF) with the ITRAX core scanner on the archive halves of the core. Measurements were made with a Mo tube (30 kV, 45 mA and 20 s dwell time) at 2 mm increments. Elements with low average counts and/or noisy signals (Al, P, S, Ga, Y, Ba, Ta, W) and elements that exhibited large concentration changes at section breaks (Ar, Ni, Pb), were disregarded. The remaining elemental counts (Si, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Cu, Zn, Br, Rb, Sr, Zr) were normalised to the ratio of incoherent to coherent scattering and log normalised (base e).

Total organic carbon (TOC), the stable isotopic composition of organic carbon (δ¹³C_org) and sterol concentrations were measured at 10 cm resolution at the Department of Geological Sciences, Stockholm University. Samples were freeze-dried and homogenised in the laboratory before subsequent analyses. For TOC and δ¹³C_org, the homogenised samples were first treated with a 1M HCl-solution to remove any carbonates, and then measured using a Carlo Erba NC2500 elemental analyser coupled to a Finnigan Delta V Advantage mass spectrometer. For the sterols, Androstanol was added as an internal standard before extraction to allow quantification of the concentration of sterols in the sediments. The homogenised samples were subjected to three cycles of 15 min sonication with dichloromethane and methanol (DCM/MeOH; 2:1 v/v), followed by centrifugation and decanting to obtain the total organic extracts. The polar fraction containing the sterols was eluted with DCM:MeOH (1:1) using glass pipette columns packed with pre-combusted silica gel and cotton wool. The fraction was silylated in BSTFA and pyridine (1:1) at 60°C for 20 min. The samples were then further separated using a Shimadzu QP2010 Ultra Gas Chromatography-Mass Spectrometry (GC-MS) with a ZB-5-HT column using a temperature increase of 10°C/minute from 40 to 300°C. The abundances of brassicasterol and ß-sitosterol were quantified as trimethylsilyl ethers (TMS) in selected ion mode (SIM) at m/z 470 and m/z 486, respectively.

Anisotropy of magnetic susceptibility (AMS) and frequency dependent magnetic susceptibility were measured with an AGICO MFK1-FA Kappabridge equipped with a sample rotator for spinning specimen measurements in three positions (Pokorný et al., 2011) at the Department of Earth Sciences, Uppsala University, Sweden. AMS was measured to assess potential sampling induced fabric deformation and identify resulting unreliable palaeomagnetic records (Marino and Ellwood, 1978). The degree of anisotropy (P_j) and the shape parameter (T; both after Jelinek, 1981) describe the fabric. We further utilise the parameter 90-K_3Inclination, where K₃ is the axis of minimum susceptibility, to identify disturbed/unreliable portions of the core (Hrouda and Kahan, 1991). Where this parameter is >15° the palaeomagnetic directional data are considered unreliable (Snowball et al., 2019). To quantify the proportion of superparamagnetic (SP) particles we calculated χ_fd1,₁₆ (after Hrouda, 2011) using bulk magnetic susceptibility measurements at 976 Hz and 15616 Hz.

NRM, anhysteretic remanent magnetisation (ARM) and isothermal remanent magnetisation (IRM) were measured with a 2G Enterprises 760 superconducting rock magnetometer (SRM) at the University of Bremen, Germany. NRM was measured and subsequently demagnetised with 15 incremental alternating field (AF) demagnetisation steps from 5 to 100 mT. Characteristic remanence magnetisation (ChRM) inclination and prolate maximum angular deviation (MAD_p; Love, 2007) were calculated by principal component analysis (PCA) in the 20 to 60 mT demagnetisation range (Kirschvink, 1980).

ARM was imparted using an AF of 100 mT and a 0.05 mT DC bias field and subsequently measured and demagnetised using the same steps as for the NRM. Susceptibility of ARM (χ_ARM) was calculated by normalising ARM to the bias field. IRM was first imparted with a 2.5 T field and measured as the saturation IRM (SIRM). Afterward, 20 increasing fields from 10 to 300 mT were imparted in the opposite direction as the SIRM. Based on the IRM data we calculated the average coercivity of remanence (B_cr) for each sample by interpolation of IRM values to a resolution of 0.1 mT, as well as the S-Ratio and Hard IRM (HIRM) at 100 and 300 mT backfields to characterise the magnetic mineralogy. The S-Ratio gives the relative concentrations of minerals with B_cr higher than 100 and 300 mT, whereas HIRM is the absolute magnetisation of minerals with B_cr higher than 100 and 300 mT (e.g., Liu et al., 2007). ARM, SIRM and HIRM were mass normalised using the cube weights, corrected for the weight of an empty cube.

To identify magnetic domain states, hysteresis properties (Coercive force, B_c; Saturation magnetisation M_s; Remanent magnetisation, M_rs) were measured on 20 selected samples by subsampling bulk material from the palaeomagnetic cubes. Measurements were made with a Lakeshore VSM at the Ångström Laboratory, Uppsala University, Sweden. A maximum field of 1 T was used to magnetise the samples. Measurement results were processed using HystLab (Paterson et al., 2018).

To find relationships between different parameter types that relate to (1) sediment input or (2) diagenetic changes, we calculated Pearson correlation coefficients using the pandas python library (McKinney, 2010). To correlate values that were measured at different sampling intervals, data were resampled to match the resolution of the palaeomagnetic samples using NumPy (van der Walt et al., 2011). Corresponding values for organic carbon content and sterol concentrations were linearly interpolated. XRF values were averaged across the length (2.2 cm) of palaeomagnetic cubes.

In order to differentiate sediments based on their physical, chemical, biological and magnetic properties, we employed fuzzy c-means clustering (FCM) on data matrices with the SciKit fuzzy logic toolbox for python (Warner et al., 2017). Before clustering, data were standardised using

where z_i is the standardised value, x_i the original value and s the standard deviation. Data were transformed into standard normal form to eliminate different measurement units across parameters. Clustering was performed based on Euclidean distances with a fuzzy exponent chosen according to the data matrix dimensions (after Schwämmle and Jensen, 2010). The optimal number of clusters was determined based on a combination of different validation indices (Schwämmle and Jensen, 2010).

FCM was run on a combined data set of physical (ρ, κ, L, a^∗, b^∗), chemical (Ca, Fe, Mn) and biological (TOC, δ¹³C_org, brassicasterol, β-sitosterol) data (denoted SC) and rock magnetic data (denoted MC). In the following text and figures, clusters are referred to by their denomination (SC or MC), a subscript that identifies the cluster in question (1, 2, 3, etc.) and the downcore occurrence of aforementioned cluster (a, b, c, etc.).

The sediments of 13PC consist of alternating dark greyish brown to olive brown (2.5Y 4/2 – 4/3) and dark grey to dark greenish grey (GLEY1 4/- 4/1) layers. Density (1.5 to 1.7 g/cm³) and shear strength (∼2 to 10 kPa) show a downcore trend to higher values (Figure 2), characteristic of compaction associated with dewatering as burial depth increases.

Brown layers comprise dark brown to greyish brown mud with slightly darker layers/mottles and signs of bioturbation that extend into underlying grey layers. The brown layers are characterised by red (a^∗) and yellow (b^∗) colour components, minimum organic carbon and sterol content, high δ¹³C_org values, and high Ca and Mn. In 13PC, two distinct brown layers are found at 0 to 0.72 m and 2.5 to 3.5 m. A third layer that exhibits similar characteristics is found between 6.3 and 7.3 m. However, this third brown layer lacks high Ca concentrations, displays a larger density and lightness (L) and is a different shade of brown compared to the two other brown layers (cf., Figure 2).

Grey layers comprise olive grey to greenish grey mud with yellowish-brown layering and mottles. They display low density, low magnetic susceptibility and elevated β-sitosterol and brassicasterol concentrations. Additionally, grey layers exhibit low δ¹³C_org values and low Mn and Ca concentrations.

Analysis of vector end-point diagrams (Zijderveld, 1967) reveals multi component demagnetisation behaviour (Figure 3). One set of samples is characterised by the removal of a low coercivity (viscous) magnetisation between 0 and 15 mT and a more stable component of magnetisation above 15 mT. A second set of samples is characterised by a curved demagnetisation behaviour indicative of at least two components with overlapping coercivity spectra. Both sets contain normal and reversed directions. NRM averages 4.92 × 10^–3 ± 3.31 × 10^–3 A/m, with slightly higher values and variability in the brown layers (6.31 × 10^–3 ± 4.12 × 10^–3 A/m) compared to the grey layers (4.08 × 10^–3 ± 2.35 × 10^–3 A/m). We attribute the shallow inclination component found in most samples (c.f., Figure 3) to a viscous component acquired during horizontal storage in Stockholm between sample retrieving and sub-sampling.

The MDF_NRM (Figure 4) displays a high variability within the uppermost 2 m of the core. Below 2 m, MDF_NRM drops to less than 10 mT on average with high variability between 4 and 6 m. The very high values of MDF_NRM around 3 m are associated with low NRM intensity, which we attribute to the presence of anti-parallel directional components that have overlapping coercivities. Additionally, samples from the uppermost 2 m of the core exhibit a predominantly inverted AMS fabric as identified by the 90-K₃ parameter.

PCA isolated ChRM inclinations in the range of 20 to 60 mT show average MAD_p values of approximately 8° with 28% being steeper than 10° (Figure 4). Almost 20% of the samples show an inclination below 0° that is prevalent in, but not restricted to, brown layers. Using rejection criteria to filter ChRM inclination based on high MAD_p values (>10°) and the detection of potentially disturbed sediment (90-K₃ > 15°) discards about half (149 out of 283) of the data set.

Values of χ, χ_ARM and SIRM suggest high concentrations of magnetic minerals in brown layers (Figure 5) although the most prominent peak in concentration, between 2.5 and 3.0 m, is slightly offset across the parameters. The peak in χ_ARM is approximately 0.1 m deeper than in χ and SIRM, although the peak in χ_ARM coincides with a peak in χ_fd. The HIRM reveals that magnetic minerals with coercivities > 100 mT but lower than 300 mT are more prominent in the uppermost 4.25 m of the core. Accordingly, magnetic mineralogical dependent parameters (S-Ratio, B_cr) indicate a distinct change to a lower average coercivity at 4.25 m. This change is accompanied by an increase in degree of anisotropy (P_j) and shape (T) of the magnetic susceptibility ellipsoid below 4.25 m.

The ratios of ARM/SIRM, χ_ARM/χ and SIRM/κ, which are commonly interpreted as proxies for magnetic domain state or magnetic mineralogy (Peters and Dekkers, 2003) are dominated by magnetic mineralogy changes in 13PC (Figure 6A). The peaks at 0.15 and 2.8 m in ARM/SIRM and χ_ARM/χ coincide with peaks in χ_ARM and SIRM and lows in B_cr and S-Ratio₁₀₀ (Figure 5), which indicates a higher concentration of low coercivity magnetic minerals rather than changes in magnetic domain state of a single mineral. Accordingly, SIRM/χ follows trends in SIRM, S-Ratio₁₀₀ and HIRM₁₀₀, which indicates a strong influence of compositional changes in the magnetic mineral assemblage. Results from hysteresis measurements (Figure 6B) confirm this influence by showing a low correlation of B_cr/B_c vs. M_rs/M_s values with χ_ARM/χ. The interpretation of the Day plot (Figure 6; Day et al., 1977) is, however, complex because the magnetic hysteresis parameters are influenced by a large number of factors, such as mineralogy, magnetocrystalline anisotropy and particle interactions to name just three (Roberts et al., 2018).

The optimal number of clusters was determined to be 3 for both FCM runs (SC and MC) with a fuzzy exponent of ∼1.3. Results from SC confirm our initial division into brown and grey layers with SC₁ correlating with brown layers while SC₂ and SC₃ correlate with grey layers (Figure 2). The subdivision of the grey layers in the FCM results is mainly based on differences in β-sitosterol, brassicasterol and Mn concentrations.

MC results indicate that there is no simple connection between magnetic properties of 13PC and its other physical and chemical properties (Figure 5). Based on its magnetic properties, the core can clearly be divided into an upper part (0–4.25 m, MC₁ and MC₂) and a lower part (4.25 m and below, MC₃) that display different magnetic properties. The division is mainly based on the difference in high coercivity minerals (i.e., >100 mT, based on S-Ratio₁₀₀) because a clear separation is not apparent in FCM results from physical and chemical properties (i.e., the SC clusters). MC₁ is characterised by high values in χ, χ_fd, χ_ARM, SIRM and HIRM₁₀₀ while MC₂ and MC₃ are characterised by lower values in χ_fd and χ_ARM. MC₂ and MC₃ are distinctly different in their χ, SIRM, HIRM₁₀₀, S-Ratio₁₀₀ and B_CR values. SC and MC correlate within the first 4 m of 13PC, but with offsets in the location of boundaries. MC₁ correlates with SC₁ but is thinner, whereas MC₂ and MC₃ are not uniquely bound to SC₂ and SC₃ but merely separate the core into an upper and lower part.

The lithostratigraphy of core 13PC is similar to other cores from the region, specifically nearby cores ARA03-28A and HLY0503-8JPC (hereafter 8JPC; Adler et al., 2009; Schreck et al., 2018; Figures 7, 8). The lithostratigraphic framework presented by Schreck et al. (2018) follows the nomenclature used in the western Arctic with dark brown layers labelled as Bx and yellowish to greyish layers labelled Gx, where x is a downcore increasing number that indexes the occurrence of successive layers. The dark brown layers are interpreted to represent interglacial intervals while the grey layers represent glacial intervals. We find a good correlation of brown layers B1 and B2 in 13PC and ARA03-28A, with the characteristic double peak in Ca in layer B2 (Figure 7). Three radiocarbon ages from core 13PC (Table 1; Jakobsson et al., 2016) confirm a Holocene age for layer B1 and a minimum MIS 3 age for B2. Jakobsson et al. (2016) further suggest an MIS 5 age for B2 based on the rare occurrences of E. huxleyi and Gephyrocapsa spp. at 2.86 m. However, based on the stratigraphic correlation an MIS 3 age of B2 would be more consistent with reported basal ages of 40 to 45 ka for this layer (Schreck et al., 2018). The much older ¹⁴C ages we obtain from our two samples from B2 (>43 and >48 ka) may be a result of reworking of older material and a small sample size, which can bias the results toward older ages (Lougheed et al., 2018). There still remains considerable uncertainty in age-depth models for Arctic Ocean sediments, and beyond the limit of radiocarbon dating it becomes difficult to differentiate between substages in glacial cycles, or define the boundaries between glacial and interglacial marine isotope stages. Here we simply acknowledge the difference in the age models reported by Jakobsson et al. (2016) and Schreck et al. (2018), and do not present any new data to support or refute either one.

FCM results confirm the general division in brown and grey layers but they do allow for a more detailed distinction of the sediment sequence. While cluster SC₁ is closely associated with the interglacial brown layers (B1 and B2), SC₂ and SC₃ differentiate the glacial layers (G1 and G2). The differences between SC₂ and SC₃ can indicate changing depositional environments and different (glacial) conditions or they may be the result of diagenetic alterations and dewatering due to compaction. SC_1c shows a distinctly different colour compared to the rest of layer G2 and may represent glacially reworked (brown) sediments as suggested by Schreck et al. (2018) for core PS72/342-1. Alternatively, SC_1c could represent diagenetically altered sediments as proposed by März et al. (2012) which would open the possibility that it is layer B3.

The higher (organic) carbon concentrations in the glacial or stadial grey layers in 13PC contrast with Polyak et al.’s (2004) results for cores from the Mendeleev Ridge, where more organic carbon is found in interglacial sediments (NP26 cores; Figure 1). Increased erosion and remobilisation of sediments and permafrost carbon from the Siberian Shelf during glacial or deglacial periods can explain the high values in the glacial and stadial sediments in 13PC (Martens et al., 2018). The proximity of 13PC to the shelf and the stable carbon isotope signature support this interpretation. The intervals of increased TOC and sterol concentrations are characterised by more depleted stable isotopic composition of the organic carbon, which suggests a terrestrial source (Goñi et al., 2005). While brassicasterol and β-sitosterol can have both terrigenous or marine origins, brassicasterol is more often considered to be of marine origin whereas β-sitosterol is considered to be of terrigenous origin (Fahl and Stein, 2012). Further support for our interpretation is the δ¹³C composition of the organic matter, where lower values (−24 to −25‰) are typically indicative of land-derived carbon, while pure marine organic carbon displays higher values (−21 to −22‰). The low concentrations of organic carbon in the brown layers can also be the result of immediate seafloor consumption of dominantly marine organic matter and release of carbon (as carbon dioxide) due to high microbial activity, which can be linked to increased Mn input (Löwemark et al., 2014).

High Mn concentrations in brown layers in 13PC indicate oxic bottom water conditions (Calvert and Pedersen, 1993, 1996), but we can not infer anoxic or suboxic conditions during the deposition of grey layers. It has been shown that sediment Mn content in the Arctic Ocean is controlled by variable input of Mn that correlates with glacial/interglacial cycles and that bottom waters remained oxic during glacial times (e.g., Löwemark et al., 2014). Elevated concentrations of Mn in pore waters in Arctic Ocean cores show that buried Mn rich layers undergo diagenesis (Meinhardt et al., 2016). Mn (oxyhydr)oxides are dissolved under suboxic conditions and dissolved Mn migrates upwards to reprecipitate in oxic pore waters. The effect of this process is clearly seen within some of the late Quaternary diamicts identified in sediment cores on the central Lomonosov Ridge. Decimeter scale intervals within these diamicts are dark grey, contain higher concentrations of TOC than surrounding interglacial sediments, with unusual spikes in Mn concentration found at the top and/or base of these intervals (see Figure 4 in Löwemark et al., 2014). The lowermost brown layer in 13PC, classified as SC_1c, may thus be layer B3 that underwent extensive Mn reduction. Mn variations have been used in the Arctic Ocean for cross-core correlation and cyclostratigraphy (Jakobsson et al., 2000; Ye et al., 2019; Park et al., 2020), however, the possibility of asynchronous Mn diagenesis in some areas and downcore intervals necessitates considerable scrutiny before using the approach (März et al., 2011; Meinhardt et al., 2016).

Correlative patterns in downcore inclination changes in Arctic Ocean sediments, whether they are a genuine palaeomagnetic field recording or a diagenetic chemical remanence, are often assumed as evidence of synchronicity (e.g., Frederichs, 1995; Jakobsson et al., 2001; Adler et al., 2009; Polyak et al., 2009). However, when cores are aligned using independent lithological parameters it becomes clear that inclination changes fall within different intervals (O’Regan et al., 2008; Wiers et al., 2019).

The changes in inclination (based on ChRM, 20–60 mT) in core 13PC are difficult to correlate to known geomagnetic field changes in the time frame provided by our (litho)stratigraphic correlation. Some excursional features coincide with globally accepted geomagnetic excursions, e.g., the negative inclination interval at approximately 3 m which dates at around 40 ka (i.e., the age of the Laschamps excursion). On the other hand, given the frequent occurrence of anomalous palaeomagnetic directions in sediments that were probably deposited during periods when no accepted excursions took place, we cannot rule out that apparently genuine features simply coincide with accepted excursions.

Channell and Xuan (2009) and Xuan and Channell (2010) proposed that negative inclinations are carried by a reversed CRM caused by seafloor oxidation of (titano)magnetite to (titano)maghemite, similar to the explanation proposed for organic-poor pelagic sediments in the north Pacific (Kent and Lowrie, 1974; Johnson et al., 1975). However, according to O’Reilly and Banerjee (1966) Ti-content and oxidation state in titanomaghemite need to be large (e.g., for x = 0.7, z > 0.8) to generate reversed CRM’s.

By placing 13PC into the stratigraphic framework of Schreck et al. (2018) (cf., Figure 7) we can compare it with core 8JPC (Figure 8; Adler et al., 2009; Schreck et al., 2018), one of the cores studied by Xuan and Channell (2010). Here we base core-to-core correlation on sediment physical and geochemical properties, not on the palaeomagnetic inclination records. In comparing 13PC and 8JPC it is evident that 13PC contains many inclination changes in a depth interval that exhibits mostly normal inclinations in the corresponding interval in 8JPC. This is an important observation, as the two cores should show a more coherent pattern if widespread seafloor oxidation, caused by low (organic) carbon influx and low sedimentation rates, is the common cause of the inclination changes. Based on the available age estimates, 13PC exhibits higher sedimentation rates than 8JPC but more intervals of negative inclination. The differences in the Ti content and oxidation state of (titano)maghemite between the two cores may explain the incoherency. However, high sedimentation rates in 13PC should reduce the exposure time of (titano)magnetite to oxic bottom waters and hence should lead to a lower degree of oxidation. Additional complications arise from the fact that 8JPC’s record was generated using measurements of u-channels, which smooths the signal.

Based on FCM results on rock magnetic parameters (MC), the magnetic properties of 13PC seem to deviate somewhat from lithological changes (SC), especially below 4.15 m, which is due to the lower concentration of minerals with B_cr of 100–300 mT. The shift in magnetic properties may reflect a subtle diagenetic process, since there is no comparable prominent feature in the lithostratigraphy (Figure 2). The sharp increase in Fe/κ just above 4.15 m (Figure 5) is indicative of (titano)magnetite dissolution (Funk et al., 2003; Korff et al., 2016) and may mark the current or a former Fe-Redox boundary. Changes in Fe/κ between 4.15 and 3.0 m are concurrent with minima in χ, χ_ARM and a downcore decrease in high coercivity magnetic minerals (HIRM₁₀₀). Above the indicated dissolution front, we find a correlation of χ_fd1,16 (r = 0.54; Figure 9) and χ_ARM (r = 0.59; Figure 9) with Mn that may be the result of iron (oxyhydr)oxide

enrichment within a narrow band in brown layers SC_1a and SC_1b (MC_1b) and potentially to a lesser extent in layer SC_1c (MC_1c). This enrichment may be facilitated by the upward migration of released iron from the underlying dissolution front (Myers and Nealson, 1988; Figure 6). The strong correlation of χ_fd1,16 and χ_ARM (r = 0.88; Figure 9) supports the interpretation of mineral authigenesis within the sediment column.

The low concentration of high coercivity minerals below 4.15 m may be due to dissolution of one or more diagenetic-authigenic magnetic phases. Given previous work on Arctic Ocean sediments (Channell and Xuan, 2009; Xuan and Channell, 2010), a potential candidate for this phase may be coatings of (titano)maghemite that formed during seafloor oxidation. Smirnov and Tarduno (2000) show that pelagic sediments from the western equatorial Pacific contain (titano)magnetite coated in (titano)maghemite, and the coating is dissolved below the Fe-Redox boundary. The changes in hysteresis parameters Smirnov and Tarduno (2000) show are comparable to the ones in 13PC (Figure 6). The migration of redox fronts due to changing organic matter and Mn input across glacial-interglacial cycles (März et al., 2012) may facilitate dissolution of previously precipitated or deposited iron minerals and upward migration of released iron. To test this hypothesis, pore water ion concentration of Mn and Fe are necessary, which do not exist for 13PC. Mn pore water concentrations from nearby core PS72/343-1 (Figure 1) rise below B3 or approximately 3 m, and display the reduction of Mn (oxyhydr)oxides below this depth (März et al., 2011).

Henshaw and Merrill (1980) showed that the anomalous palaeomagnetic record in the north Pacific is more likely due to the formation of an authigenic ferromanganese phase than to maghemitisation. Substitution of Mn into the spinel lattice of (titano)magnetite can lead to self-reversals (Henshaw and Merrill, 1980), although this mechanism has not received much attention. In the Arctic Ocean, the enhanced deposition of Mn-oxides and enrichment of trace metals during interglacials shifts the Mn-Redox boundary upwards, close to the sediment/water interface. This shift triggers Mn cycling (März et al., 2011) and associated enrichment of iron (oxyhydr)oxides (Fe_ox1 and Fe_ox2 in März et al., 2012). Reduced Mn input during glacials results in potential diagenetic Mn redistribution in deeper sediment layers that may lead to post-depositional biogeochemical Fe cycling and (titano)magnetite/maghemite formation (März et al., 2012).

März et al. (2012) did not find a direct relationship between Fe_mag [(titano)magnetite, (titano)maghemite] and χ, hinting at a different magnetic carrier in the Mn-rich (brown) layers. Similarly, in 13PC Mn shows weak to zero correlation with magnetic susceptibility (r = 0.21; Figure 9) even though Mn has been shown to be linked to iron reduction (Myers and Nealson, 1988; Smirnov and Tarduno, 2000). The precipitation of ferromanganese oxides, congruent to what was proposed for north Pacific sediments (Henshaw and Merrill, 1980), is conceivable in the Arctic Ocean. The formation of these oxides would be linked to Mn influx and changing redox conditions (Figure 6) and may present an alternative hypothesis to seafloor oxidation or genuine recordings of the palaeomagnetic field, and would explain the unusual palaeomagnetic record in 13PC. März et al. (2012) used the same extraction method for the Fe_ox1 phase as Henshaw and Merrill (1980) used to dissolve a magnetic manganese-iron phase. The Fe_ox1 phase could, therefore, contain an unidentified magnetic ferromanganese phase and the correlation of Fe_ox1 to Mn content found by März et al. (2012) supports our hypothesis of a link between the anomalous palaeomagnetic records and the presence of ferromanganese oxides.

Neither lithology nor magnetic mineral composition show a strong correlation to the ChRM inclination. On the other hand, there is a weak correlation between ChRM inclination and Mn (r = −0.29; Figure 9). As expected, NRM intensity is more strongly affected by changes in the composition of the magnetic mineral assemblage (SIRM – NRM, r = 0.64; χ – NRM, r = 0.61; χ_ARM – NRM, r = 0.58; Figure 9). The low correlation coefficient between Mn and ChRM inclination may be a result of the dominantly bimodal nature of the inclination data and ongoing processes driving Mn redistribution within the sediments. A more robust approach to test the correlation was applied by calculating the circular-linear correlation coefficient (e.g., Mardia and Jupp, 2009), however, this results in the same coefficient (r = −0.29) for Mn and ChRM inclination. Yet, correlation of the filtered ChRM inclination with Mn yields a larger coefficient (r = −0.47). On the other hand, visual inspection of Mn and ChRM inclination reveals some coherence between negative inclinations and increased Mn values (Figure 10).

Non-steady-state diagenesis (Kasten et al., 2004), driven by changes in sediment input and bottom water ventilation, may facilitate upward migration of released ions that further distort the magnetic record. Changes in sea level across interglacial/glacial cycles that move the position of (Arctic) Atlantic water may also impact the position of redox boundaries, especially in water depths of <1500 m where the lower boundary of the Atlantic water layer generally resides today (Rudels, 2009).