Distinguishing between old and modern permafrost sources in the northeast Siberian land – shelf system with compound-specific δ 2 H analysis

Pleistocene ice complex permafrost deposits contain roughly a quarter of the organic carbon (OC) stored in permafrost (PF) terrain. When permafrost thaws, its OC is remobilized into the (aquatic) environment where it is available for degradation, transport or burial. Aquatic or coastal environments contain sedimentary reservoirs that can serve as archives of past climatic change. As permafrost thaw is increasing throughout the Arctic, these reservoirs are important locations to assess the fate of remobilized permafrost OC. We here present compound-specific deuterium (δ2H) analysis on leaf waxes as a tool to distinguish between OC released from thawing Pleistocene permafrost (ice complex deposits; ICD) and from thawing Holocene permafrost (from near-surface soils). Bulk geochemistry (%OC; δ13C; %total nitrogen, TN) was analyzed as well as the concentrations and δ2H signatures of long-chain n-alkanes (C21 to C33) and midto long-chain n-alkanoic acids (C16 to C30) extracted from both ICD-PF samples (n= 9) and modern vegetation and Ohorizon (topsoil-PF) samples (n= 9) from across the northeast Siberian Arctic. Results show that these topsoil-PF samples have higher %OC, higher OC /TN values and more depleted δ13COC values than ICD-PF samples, suggesting that these former samples trace a fresher soil and/or vegetation source. Whereas the two investigated sources differ on the bulk geochemical level, they are, however, virtually indistinguishable when using leaf wax concentrations and ratios. However, on the molecular isotope level, leaf wax biomarker δ2H values are statistically different between topsoil PF and ICD PF. For example, the mean δ2H value of C29 n-alkane was −246± 13 ‰ (mean±SD) for topsoil PF and−280± 12 ‰ for ICD PF. With a dynamic isotopic range (difference between two sources) of 34 to 50 ‰; the isotopic fingerprints of individual, abundant, biomarker molecules from leaf waxes can thus serve as endmembers to distinguish between these two sources. We tested this molecular δ2H tracer along with another source-distinguishing approach, dual-carbon (δ13C–114C) isotope composition of bulk OC, for a surface sediment transect in the Laptev Sea. Results show that general offshore patterns along the shelfslope transect are similar, but the source apportionment between the approaches vary, which may highlight the advanPublished by Copernicus Publications on behalf of the European Geosciences Union. 1880 J. E. Vonk et al.: Distinguishing between old and modern permafrost sources tages of either. This study indicates that the application of δ2H leaf wax values has potential to serve as a complementary quantitative measure of the source and differential fate of OC thawed out from different permafrost compartments.

Abstract.Pleistocene ice complex permafrost deposits contain roughly a quarter of the organic carbon (OC) stored in permafrost (PF) terrain.When permafrost thaws, its OC is remobilized into the (aquatic) environment where it is available for degradation, transport or burial.Aquatic or coastal environments contain sedimentary reservoirs that can serve as archives of past climatic change.As permafrost thaw is increasing throughout the Arctic, these reservoirs are important locations to assess the fate of remobilized permafrost OC.
We here present compound-specific deuterium (δ 2 H) analysis on leaf waxes as a tool to distinguish between OC released from thawing Pleistocene permafrost (ice complex deposits; ICD) and from thawing Holocene permafrost (from near-surface soils).Bulk geochemistry (%OC; δ 13 C; %total nitrogen, TN) was analyzed as well as the concentrations and δ 2 H signatures of long-chain n-alkanes (C 21 to C 33 ) and midto long-chain n-alkanoic acids (C 16 to C 30 ) extracted from both ICD-PF samples (n = 9) and modern vegetation and Ohorizon (topsoil-PF) samples (n = 9) from across the northeast Siberian Arctic.
Results show that these topsoil-PF samples have higher %OC, higher OC / TN values and more depleted δ 13 C-OC values than ICD-PF samples, suggesting that these former samples trace a fresher soil and/or vegetation source.Whereas the two investigated sources differ on the bulk geochemical level, they are, however, virtually indistinguishable when using leaf wax concentrations and ratios.
However, on the molecular isotope level, leaf wax biomarker δ 2 H values are statistically different between topsoil PF and ICD PF.For example, the mean δ 2 H value of C 29 n-alkane was −246 ± 13 ‰ (mean ± SD) for topsoil PF and −280 ± 12 ‰ for ICD PF.With a dynamic isotopic range (difference between two sources) of 34 to 50 ‰; the isotopic fingerprints of individual, abundant, biomarker molecules from leaf waxes can thus serve as endmembers to distinguish between these two sources.We tested this molecular δ 2 H tracer along with another source-distinguishing approach, dual-carbon (δ 13 C-14 C) isotope composition of bulk OC, for a surface sediment transect in the Laptev Sea.Results show that general offshore patterns along the shelfslope transect are similar, but the source apportionment between the approaches vary, which may highlight the advan-tages of either.This study indicates that the application of δ 2 H leaf wax values has potential to serve as a complementary quantitative measure of the source and differential fate of OC thawed out from different permafrost compartments.
The release of OM from thawing permafrost into aquatic sediments varies over time and space.A recent study showed that at the end of the last glacial, the surface active layer of terrestrial permafrost released about 4.5 Tg organic carbon (OC) per year from just the Lena River watershed onto the nearby shelf, whereas current annual OC release is estimated to be only about a tenth of this (Tesi et al., 2016b).In addition to active layer material, OM from deeper and older permafrost sources can also thaw and be released into the environment (Shakhova et al., 2007(Shakhova et al., , 2014)).This process currently dominates the delivery of terrestrial material onto the East Siberian Arctic Shelf (Vonk et al., 2012;Semiletov et al., 1999) and is expected to increase due to accelerating coastal erosion rates (Günther et al., 2013).
Different permafrost OC stocks exhibit variable vulnerabilities to thaw remobilization (Schuur et al., 2015).In addition to a subsea permafrost OC stock, soils and sediments of the terrestrial northern permafrost zone store about 1300 ± 200 Pg OC, with separate upscaling approaches applied for soil stocks (0-3 m depth), deltaic sediments (full depth) and Yedoma sediments (full depth; Hugelius et al., 2014).Yedoma sediments, also known as ice complex deposits (ICD), are polygenetic, ice-rich Pleistocene-age deposits that are present in the unglaciated parts of Siberia and Alaska (Schirrmeister et al., 2011).These deposits contain roughly a quarter of the OC stored in permafrost terrain, but estimates vary from ca. 200-400 Pg C (Strauss et al., 2013;Schuur et al., 2015).The presence of massive ice wedges in ICD causes landscapes to collapse upon thaw, exposing deeper stocks of OC.This type of relatively abrupt thaw is increasing in many parts of the arctic landscape (Schuur et al., 2015).At the same time, deepening of the active layer causes gradual thaw that occurs across entire landscapes (Shiklomanov et al., 2013).
With a tool to detect and monitor different types of permafrost OM in coastal environments, one could assess (historical and spatial) variability in permafrost source input, degradation, and thaw, as well as the relative degradation of different permafrost types.For example, the relative release of OC from ICD versus topsoil permafrost has been distinguished and quantified previously through the use of dual-carbon isotopes (δ 13 C and 14 C) on bulk OC in the shelf environment of the Laptev and East Siberian Sea.It was shown that topsoil permafrost OC dominates in suspended particulate matter (Karlsson et al., 2011(Karlsson et al., , 2016;;Vonk et al., 2012) and ICD permafrost OC dominates in the surface sediments (Vonk et al., 2012;Semiletov et al., 2011Semiletov et al., , 2012)).Vonk et al. (2014) further showed that topsoil OC is actively degraded during horizontal transport whereas ICD permafrost OC rapidly settles.Winterfeld et al. (2015) showed, using dual-carbon isotopes on riverine material, that suspended particulate OC in the Lena River delta mostly consists of Holocene material instead of material from ICD permafrost.This δ 13 C-14 C dual-carbon isotope approach carries the strong advantage that it operates on the bulk OC level, thereby circumventing the "molecular bulk upscaling challenge".This challenge relates to issues associated with upscaling from the molecular isotope level to the bulk level.These issues relate to the relative concentration (n-alkanes and n-alkanoic acids represent only a fraction of the total OC) but also to processes such as selective degradation, differences in physical association or dispersion differences.However, the δ 13 C-14 C approach also has drawbacks, such as a weak distinction between the δ 13 C endmember values of topsoil PF versus ICD PF.Also, the marine δ 13 C endmember values in coastal Arctic shelf waters are uncertain and may be more depleted than at mid-latitudes due to uptake of relatively depleted dissolved CO 2 values caused by cold polar water (Meyers, 1997;Tesi et al., 2016b) or degradation of terrestrial matter (Anderson et al., 2009(Anderson et al., , 2011;;Semiletov et al., 2013Semiletov et al., , 2016)), generating a potential overlap between marine and topsoil δ 13 C endmembers.
Here we propose a complementary tool to trace permafrost OC release into the coastal environment based on molecular δ 2 H analysis on leaf waxes.We will evaluate the performance of this tool using additional geochemical data as well as the bulk δ 13 C-14 C mixing approach.Isotopes in water molecules (δ 2 H or δ 18 O) in glacial ice cores as well as in massive ground ice in the Northern Hemisphere have been used for reconstructing paleotemperatures (e.g., Kotler and Burn, 2000;Johnson et al., 2001;Opel et al., 2011;Meyer et al., 2015;Wetterich et al., 2016) as the isotopic value of local precipitation is a function of local climate (Craig, 1961;Sachse et al., 2004;Smith and Freeman, 2006).Higher plants use water as their primary source of hydrogen during photosynthesis (Sternberg, 1988).The δ 2 H isotope values of leaf wax n-alkanoic acids or n-alkanes are therefore reflecting the δ 2 H isotopic value of local precipitation (e.g., Sachse et al., 2004;Sessions et al., 1999), after correction for the net fractionation during biosynthesis, and evapotranspiration (Leaney et al., 1985).Global precipitation values can vary immensely (Dansgaard, 1964) with values up to +50 ‰ in eastern Africa but approaching −200 ‰ near the North Pole (www.iaea.org)or even below −400 ‰ in Antarctica (i.e., SLAP2 standard, Standard Light Antarctic Precipitation, is −427.5 ‰).Additionally, the fractionation between source water and plant wax molecules varies both in time and space and can be up to −170 ‰ (Smith and Freeman, 2006;Sachse et al., 2004;Polissar and Freeman, 2010) but appears relatively small at higher latitudes (between −59 and −96 ‰; Shanahan et al., 2013;Wilkie et al., 2013;Porter et al., 2016).Differences in δ 2 H signatures of leaf wax molecules from terrestrial regions with different (past) climates could therefore potentially be applied to derive the relative proportion of different types of thawing permafrost in nearby coastal settings.Despite the plant fractionation associated with kinetics and plant physiology (Sachse et al., 2012), we hypothesize that δ 2 H signatures of leaf wax n-alkanoic acids and n-alkanes are more depleted in OC from permafrost deposits formed during the colder Pleistocene (generally correlating with 2 Hdepleted precipitation), compared to more enriched values in OC from active layer or surface permafrost formed during the warmer Holocene.
This study investigates a source-specific δ 2 H signature for both ICD permafrost and recent, surface soil permafrost in northeast Siberia.Furthermore, we explore the possibilities of using these isotopic endmember values in regional sourceapportionment calculations that aim to quantify the relative contribution of different sources of permafrost OC.As permafrost thaw progresses, particularly in ice-rich permafrost such as ICD, it is increasingly important to trace the fate of remobilized and decomposing OC in the Arctic environment.Our proposed tool may be used to trace these temporal and spatial differences in OC release from permafrost thaw as well as the extent of burial of OC in sedimentary reservoirs.

Sampling
A total of 18 samples were collected throughout the Siberian Arctic.Recent surface soils (n = 7) and vegetation (n = 2) samples were analyzed and (from here on) referred to as the "topsoil" permafrost (topsoil-PF) sample set, whereas ICD-PF samples were obtained from ICD soil profiles (n = 7) and suspended particulates from ICD formations (n = 2; Fig. 1 and Table 1).Eight offshore sediments along a shelf-slope continental rise transect in the Laptev Sea were collected in 2014; further marine sampling details can be found in Bröder et al. (2016b).
The topsoil-PF samples represent O and A soil genetic horizons in sites with active soil formation.The sites were chosen to represent typical soil and vegetation types in the investigated permafrost landscapes, including both taiga and tundra sites.Samples were collected by depth or soil horizon increments from open soil pits using fixed-volume sampling procedures.
The ICD-PF samples were collected from vertical exposures that were excavated to expose intact permafrost.Fixedvolume samples were collected by coring horizontally into the frozen sediments to extract ICD-PF samples from consecutive depths.
For more details about sampling sites, including location, vegetation and soil types see Table 1 (terminology following the USDA Soil Taxonomy; Soil Survey Staff, 2014).Sampling was done in late summer near the time of maximum annual active layer depth; in July 2010 (ICD-8 and ICD-9; Vonk et al., 2013) and August 2011 (Palmtag et al., 2015) for the Kolyma River region; in August 2012 for the lower Lena River and Indigirka River (Siewert et al., 2015;Weiss et al., 2015); and in August 2013 for the upper Lena River (Siewert et al., 2016).For more detailed descriptions of sample collection we refer to these references.The vegetation samples TS-8G (grass) and TS-9G (grass) were obtained from the tundra near the Medvezhka River and a birch forest near the Y4 stream, respectively, in July 2012.
Samples ICD-8 and ICD-9 were obtained in July 2010 at the Duvannyi Yar ICD exposure along the Kolyma River (Vonk et al., 2013).The particulate sediment samples were taken from thaw streams that were freshly formed from thawing ICD (transport time from thaw to sampling is estimated to be less than 1 h).

Analytical methods
Freeze-dried samples were extracted using an ASE 200 accelerated solvent extractor (Dionex Corporation, USA) using DCM-MeOH (9 : 1 v/v) at 80 • C (5 × 10 6 Pa; Wiesenberg et al., 2004).After the extraction, solvent-rinsed activated copper and anhydrous sodium sulfate were added to the extracts to remove sulfur and excess water, respectively.After 24 h, extracts were filtered on pre-combusted glass wool and concentrated with the rotary evaporator.Extracts were transferred into glass tubes, evaporated to complete dryness and redissolved in 500 µL of DCM.Lipid fractionation was performed via column chromatography using aminopropyl Bond Elut (500 mg/3 mL) to retain the acid fraction and Al 2 O 3 to separate the hydrocarbon and polar fractions (Vonk et al., 2010).
Prior to the analyses, saturated n-alkanes (hydrocarbon fraction) were further purified using 10 % AgNO 3 coated silica gel to retain the unsaturated fraction.The acid fraction was methylated using a mixture of HCl, MilliQ water and methanol at 80 hexane and further purified using 10 % AgNO 3 coated silica gel.The hydrocarbon and FAME fractions were quantified via gas chromatography mass spectrometry (GC-MS) in full scan mode (50-650 m/z) using the response factors of commercially available standards (Sigma-Aldrich).The GC was equipped with a 30 m × 250 µm DB-5ms (0.25 µm thick film) capillary GC column.Initial GC oven temperature was set at 60 • C followed by a 10 • C min −1 ramp until a final temperature of 310 • C (hold time 10 min).
The hydrogen-isotopic composition of hydrocarbon and FAME fractions was measured with continuous-flow gas chromatography coupled to continuous-flow isotope ratio mass spectrometry (IRMS).Purified extracts were concentrated and injected (1-2 µL) into a Thermo Scientific ™ Trace Ultra GC equipped with a 30 m × 250 µm HP5 (0.25 µm thick film) capillary GC column.Oven conditions were similar to the setting used for the quantification.The conversion of organic biomarkers to elemental hydrogen was accomplished by high-temperature conversion (HTC) at 1420 • C (Thermo Scientific ™ GC Isolink).After the HTC, H 2 was introduced into the isotope ratio MS (Thermo Scientific ™ Delta V ™ IRMS) for compound-specific determination of δ 2 H values via a Thermo Scientific ™ ConFlo IV.Following a linearity test, we only used peaks with amplitude (mass 2) between 1500 and 8000 mV for the evaluation.The δ 2 H values were calibrated against saturated high-molecular-weight (HMW) n-alkanes using the reference substance mix A4 (Biogeochemical Laboratories, Indiana University).The H + 3 factor (Sessions et al., 2001) was determined every day and stayed constant (< 3 ‰/V throughout our analyses period.Each purified extract was injected three times.FAMEs were further corrected to account for the methylation agent by comparing the hydrogen abundance of lauric acid (C 12 -FA; i.e., 12 carbon atoms) as acid and corresponding methyl ester.The average methylation correction for lauric acid was 23.97 ± 3.9 ‰ (n = 4).This correction was, normalized to chain length (i.e., increasing chain lengths result in lower corrections), applied to all the FAMEs.δ 2 H values of n-alkanes and FAMEs are reported as mean, standard deviation and weighted average.Details of the analytical methods for extraction, workup, and purification of the eight offshore sediment samples for biomarker analysis that are included in our source- apportionment comparison (Sect.4.3) can be found in Bröder et al. (2016b).The δ 2 H analysis on the shelf sediments was performed in parallel with the ICD-PF and topsoil-PF samples, according to the method described above.

Source apportionment
The compound-specific δ 2 H signatures in this study were used to differentiate between the two major sources (endmembers), topsoil PF and ICD PF, using an isotopic massbalance model.We used a Markov chain Monte Carlo (MCMC) approach to account for the endmember variability (Andersson et al., 2015;Bosch et al., 2015).The endmembers were represented by normal distributions, with mean and standard deviations obtained from our analysis (δ 2 H on TS and ICD samples) and from literature (δ 13 C and 14 C on endmembers).For each Laptev Sea station, the isotope signatures from three different terrestrial molecular markers (long-chain n-alkanes C 27 , C 29 and C 31 ) were used jointly to improve source-apportionment precision.The δ 2 H signatures for the two endmembers were based on our topsoil-PF and ICD-PF samples.
The compound-specific δ 2 H-based source apportionment was compared to 14 C-δ 13 C-based analysis of bulk OC using analogous MCMC techniques (e.g., Vonk et al., 2012).The 14 C-δ 13 C approach allows estimation of the relative contribution of a third source, marine, which does not af-fect the presently investigated (terrestrial) compounds.Accounting for the marine component contribution to OC allows direct comparison of the Holocene and Pleistocene contributions.All MCMC calculations were made using Matlab scripts (version 2014b) using 200 000 iterations, a burn-in phase (initial search period) of 10 000 and a data thinning of 10.
The spatial extent of ICD in the Lena River basin was calculated by overlaying the extent of the drainage basin (from Watersheds of the World published by the World Resources Institute, www.wri.org/publication/watersheds-world) with the extent of the Yedoma region (digitized from Romanovsky, 1993) in an equal area map projection.It was assumed that 30 % of the Yedoma region consists of intact ICD (following Strauss et al., 2013).

Bulk geochemistry
The investigated topsoil-PF and ICD-PF samples are, on a bulk geochemical level, very different.Mean organic carbon contents (as %OC) and total nitrogen content (as %TN) are 25 ± 12 and 1.1 ± 0.67 for topsoil-PF samples and 1.6 ± 0.31 and 0.17 ± 0.058 for ICD-PF samples, respectively (Table 1).This gives OC / TN ratios of 25 ± 8.0 for topsoil-PF samples  1).Radiocarbon ages were unfortunately only available for a few ICD samples and ranged between 17 and 28 14 C ka (Table 1).
Long-chain n-alkanoic acids (C 22 and above) were abundant in concentrations between 0.122 and 2670 µg gOC −1 for individual homologues in topsoils, with the sum of HMW n-alkanoic acids (> C 22 ) being 6397 7454 3167 µg gOC −1 (median and IQR) and the most abundant n-alkanoic acids (sum of C 22 -C 24 -C 26 -C 28 ) adding up to 4700 6092 2670 µg gOC −1 (Tables 3, 4 and Fig. 2b).ICD-PF samples contained individual long-chain n-alkanoic acids in 2.17 and 18 700 µg gOC −1 (Table 2), a sum of HMW n-alkanoic acids of 8290 11 430 6290 µg gOC −1 and the sum of most abundant, even n-alkanoic acids of 6630 8790 5285 µg gOC −1 (Table 4).Topsoil-PF and ICD-PF samples had average chain lengths of 24.1 ± 1.1 and 24.3 ± 0.59 and CPI (C 22 -C 28 ) values of 5.9 ± 2.7 (average ± standard deviation) and 5.0 ± 1.6, respectively (Table 4).Shorter-chain n-alkanoic acids C 16 and C 18 are produced in basically all types of life in soils or , where X is concentration.Stars indicate that the two compared values are statistically significant (95 % confidence).Note that panel (a) and (b) are reported as median with IQR (interquartile range), and the other panels are reported as average ± standard deviation.aquatic environments and are not specific for higher plants.Topsoil PF contained C 16 and C 18 homologues in concentrations between 220 and 4600 µg gOC −1 and ICD-PF samples between 200 and 10 400 µg gOC −1 (Table 3).
Degradation of organic matter involves the loss of functional groups, e.g., the loss of carboxylic acids (Meyers and Ishiwatari, 1993).A high ratio of HMW n-alkanoic acids over HMW n-alkanes in a sample therefore implies a relatively fresh, less degraded, status (i.e., relatively more functional groups present).For topsoil-PF samples, the HMW nalkanoic acid / HMW n-alkane ratio varied between 5.6 and 25, with an average value of 13 ± 7.6, whereas ICD-PF samples varied between 7.6 and 140, with an average value of 29 ± 43 (Table 4, Fig. 2f).

Molecular isotopic composition
We measured δ 2 H values in long-chain n-alkanes and nalkanoic acids between −119 and −313 ‰ (Fig. 3, Table 5).Mean values for HMW n-alkanes (C 25 -C 27 -C 29 -C 31 ) were between −201 and −247 ‰ for topsoil-PF samples and between −221 and −297 ‰ for ICD-PF samples, with consistently lower δ 2 H for longer chain lengths.For HMW nalkanoic acids (C 22 -C 24 -C 26 -C 28 ) mean δ 2 H values were between −203 and −236 ‰ for topsoil-PF samples and between −261 and −278 ‰ for ICD-PF samples (Table 5).The decrease in δ 2 H values with increasing chain length is less distinct for n-alkanoic acids, but one can observe a decrease of around 25-30 ‰ from C 22 to C 26 (Fig. 3).For ICD-PF samples, the isotopic depletion for the average of the three most abundant n-alkanes is comparable to the average for nalkanoic acids, whereas, in topsoil-PF samples, the isotopic depletion for the three most abundant n-alkanes is a bit larger than for n-alkanoic acids (Fig. 4).

Using bulk geochemistry and molecular proxies
Bulk geochemical and isotopic analysis as well as analysis of molecular proxies remained inconclusive in distinguishing between the two investigated sources in this study.Topsoil-PF samples have a higher organic content, higher OC / TN values (representing fresh, higher plant material; Meyers, 1994) and more depleted δ 13 C values (indicative for terres-trial C3 plants; Meyers, 1997) than ICD-PF samples, suggesting that these samples indeed trace a fresh soil and/or vegetation source (Table 1).The δ 13 C values of a larger ICD-PF and topsoil-PF dataset have previously been summarized (Vonk et al., 2012 and references therein;Schirrmeister et al., 2011), giving values of −26.3 ± 0.67 ‰ (n = 374) and −28.2 ± 2.0 ‰ (n = 30), respectively.Our values (Table 1) are in a similar range.Despite the differences between these two sources in their bulk geochemistry, it is hard to use these parameters for source distinction as their variability is fairly high and their behavior in the environment is not conservative but is affected by degradation processes.On a molecular geochemical level the two investigated sources are virtually indistinguishable as there is a considerable variation in molecular concentrations and proxy values (Fig. 2).Only one of the tested parameters, the CPI C 23 -C 31 of n-alkanes, showed a statistically significantly different value for the two investigated sources.

Evaluation of molecular δ 2 H values as a source endmember
To alleviate the difficulty in distinguishing between topsoil PF and ICD PF with just bulk and molecular geochemical characteristics, we explore the δ 2 H values of leaf wax molecules (i.e., long-chain n-alkanoic acids and nalkanes) to differentiate between their relative source contributions.The overall mean δ 2 H of the four most abundant n-alkanoic acids is −231 ± 29 and −271 ± 13 ‰ for topsoil-PF and ICD-PF samples, respectively.These values compare well with available literature (Fig. 5).The mean isotopic difference between the most abundant n-alkanoic acids of the two investigated sources is around 40 ‰ (δ 2 H values of −231 ± 29 and −271 ± 13 ‰ for topsoil-PF and ICD-PF samples, respectively).Despite the relatively large standard deviations, the isotopic differences are statistically significant for each of the n-alkanoic acids individually (C 22 , C 24 , C 26 , C 28 ; Fig. 3).The isotopic difference between the two sources for the mean value of the four most abundant n-alkanes is 35 ‰, with a mean value of −229 ± 33 and −264 ± 34 ‰ for topsoil-PF and ICD-PF samples, respectively.Here, the individual n-alkane isotopic  signatures are statistically significantly different for C 27 , C 29 and C 31 (Fig. 3) in topsoil-PF and ICD-PF samples.The selection and application of individual chain length δ 2 H values as endmembers, in contrast to mean chain length values, might be more appropriate for several reasons: (i) to reduce variability (δ 2 H ranges for C 29 and C 31 n-alkanes and C 22 and C 24 n-alkanoic acids are relatively low; Fig. 3), (ii) to target the most abundant species (C 29 and C 31 n-alkanes are generally more abundant in soils and ICD PF compared to shorter chain lengths; Table 2) and (iii) to make use of the largest dynamic range between source endmember values (C 31 n-alkane δ 2 H values of topsoil PF and ICD PF differ by 50 ‰).Based on these arguments, the C 28 n-alkanoic acid and the C 29 or C 31 n-alkanes are most appropriate to use for source apportionment.The available previous studies (Fig. 5) have also selected these chain lengths (C 28 n-alkanoic acid and C 29 n-alkanes) for proxy development.
The use of molecular δ 2 H values as tracers of terrestrial material in a marine or coastal setting has the advantage that it avoids uncertainty issues related to the definition of the marine endmember.On the other hand, the inherent bulk upscaling challenge of any molecular proxy is a disadvantage of the δ 2 H approach as it introduces unknowns related to the molecular bulk upscaling effort (e.g., taking into account sorting and recalcitrance; discussed in depth in Sect.4.3).We also want to emphasize that δ 2 H leaf wax values in the two studied endmember sets (topsoil PF vs. ICD PF) largely depend on the climate (warm vs. cold) and continentality (near the coast vs. further inland) during plant formation and associated differences in fractionation mechanisms.Consequently, when δ 2 H values in samples are used for source apportionment, this may represent the fraction leaf wax produced in cold vs. warm conditions (as well as degree of continentality), and not necessarily the fraction topsoil PF vs. ICD PF.
Finally, we realize that the amount of soil and ICD samples analyzed in this study is limited, and we want to point out that the results may change when more data are analyzed in the near future.Additionally, studies have shown that the δ 2 H signature of ice within ICD permafrost deposits can range from roughly −150 to −260 ‰, depending on the type of ice (wedge ice vs. pore or texture ice) as well as the period of formation (different Pleistocene cold stages; Opel et al., 2017 and references therein).The source of water (i.e., type of ice) and age of the deposit will therefore influence the n-alkane or n-alkanoic acid δ 2 H signal.However, regardless of the natural variability associated with the processes mentioned above, both ICD and texture-ice isotopic compositions appear to reflect long-term climate changes (Opel et al., 2017;Schwamborn et al., 2006;Dereviagin et al., 2013;Porter et al., 2016) which, likely, were also captured in the n-alkane or n-alkanoic acid δ 2 H signal.Unfortunately, we do not have 14 C ages available for all ICD samples, so cross-referencing to published stratigraphies in the region is not possible.Coastal sediments, however, will represent a mixture of material released from different depths, outcrops and stratigraphies within the catchment or coast.For sourceapportionment applications, we reason that a growing body of leaf wax δ 2 H endmember data from the ICD region can overcome the variability issues highlighted above. 13C-14 C source apportionment: a case study
For δ 2 H source apportionment there is no need to include a marine endmember as marine organisms do not produce long-chain n-alkanes or n-alkanoic acids.We were unfortunately only able to analyze n-alkanes in the shelf-slope transect samples, and no n-alkanoic acids, due to limitations in sample volume.We used the δ 2 H values of the C 27 , C 29 and C 31 n-alkanes, individually.In other words, these three chain lengths are taken as independent markers, providing an overdetermined system (i.e., two sources defined with three different markers).This is more representative than using the average (concentration-weighted) δ 2 H value for these nalkanes as the endmember values for each chain length are different.For topsoil PF we used −215 ± 39, −246 ± 13 and −247 ± 23 ‰ for C 27 , C 29 and C 31 n-alkanes, and for ICD PF we applied −259 ± 18, −297 ± 15 and −282 ± 13 ‰ for C 27 , C 29 and C 31 n-alkanes, respectively (see also Table 5).Afterwards, we averaged the three endmember contributions derived from the three calculations for each station, thereby taking the variability introduced by the endmembers into account.
The source apportionment of OC from topsoil PF and ICD PF to surface sediments along the Laptev Sea transect differs between the bulk δ 13 C-14 C and leaf wax δ 2 H approaches (Table 6).The former approach suggests topsoil-PF contributions between 21-70 %, generally decreasing offshore, and, consequently, ICD-PF contributions of 30-79 %, generally increasing offshore.The latter (leaf wax δ 2 H) approach results in a more extreme division of sources with topsoil-PF contributions of 83-91 % and ICD-PF contributions of 9-17 %, with similar patterns nearshore and offshore (Table 6).A contribution of 9-17 % may seem more in line with the estimated extent of ICD in the Lena River basin: 12 % of the basin falls within the Yedoma region (as defined by Romanovsky, 1993) and about 3 % consists of intact ICD (see Sect. 2.3).However, the cross-shelf sites are also strongly influenced by coastal and/or subsea erosion (Karlsson et al., 2011;Vonk et al., 2012;Semiletov et al., 2012Semiletov et al., , 2016)), so the catchment characteristics are only one part of the story.It is challenging to interpret the differences between the two proxies but we elaborate below on potential reasons.
Assumptions in the bulk δ 13 C-14 C approach may affect these results.First, the outcome of the bulk δ 13 C-14 C approach is sensitive to the definition of the marine endmember.
www.the-cryosphere.net/11/1879/2017/The Cryosphere, 11, 1879-1895, 2017 Changes in the currently used δ 13 C and 14 C value of the marine endmember of the East Siberian Arctic Shelf (n = 10; Panova et al., 2015) would likely alter the relative topsoil-PF and ICD-PF contributions.The currently used standard deviation for the δ 13 C marine endmember is 2.6 ‰, which is much higher than the values for the terrestrial endmembers.Second, lateral transport time enroute the shelf-slope transect (> 600 km) causing potentially significant aging of sediments and their organic carbon is not accounted for in the source apportionment.Lateral transport time results in older surface OC ages on the shelf, compared to those at the initial coastal deposition.Without correcting for this factor, the source apportionment will generate lower contributions of the (younger) topsoil-PF component.In an attempt to estimate this effect, we recalculated (similar to Bröder et al., 2016a) the relative source contributions of topsoil PF, ICD PF and marine component with the bulk δ 13 C-14 C approach with the assumption that the topsoil-PF 14 C age would be subject to a cross-shelf lateral transport time of 5000 years.We assumed a linear aging along the transect based on distance from the coast, with a maximum value of 5000 years aging at station SW-01.This resulted in topsoil-PF contributions that were up to 20 % higher (for the deepest stations) compared to the source apportionment where lateral transport time was unaccounted for (Table 6; Fig. 6).Assumptions in the leaf wax δ 2 H source-apportionment approach could potentially also impact the outcomes and hence differences with the bulk δ 13 C-14 C results.First, there is an inherent assumption related to the molecular to bulk level upscaling challenge.We assume that the physical association of n-alkanes in different source endmembers (topsoil PF vs. ICD PF) as well as their fractionation in the coastal system is similar.However, previous research has shown that n-alkanes behave rather differently upon their release into coastal waters; n-alkanes originating from surface soil or vegetation debris are not bound to minerals and remain in suspension during transport while being actively degraded, whereas n-alkanes originating in deeper mineral soils settle quickly and are protected from extensive degradation (Vonk et al., 2010).It is possible that most of the n-alkanes in the Laptev Sea sediment transect originate in (deeper) mineral soils.An effect of physical association as well as the potential effect of hydrodynamic sorting patterns (Tesi et al., 2016) on the leaf wax δ 2 H values of both sources could impact the source-apportionment. Another factor that can introduce a bias in our leaf wax δ 2 H approach is a proton exchange of the C-bound H atoms in n-alkanes with environmental water, either from in situ sources (soil water) or during transport (river, ocean, or sediment pore water).As there is no evidence for such exchange in young (< 1 million years) cold sediments (Sessions et al., 2004), we suspect this process may be minimal in our samples (and endmembers).
When accounting for an estimated lateral transport time, the difference in estimates of source contribution by the two different approaches (bulk δ 13 C-14 C and leaf wax δ 2 H) increases offshore, from about a 25 % difference near the coast to a 40 % difference at stations SW-01 and SW-03.This increasing offset between the results of the two endmember mixing methods may be caused by several factors such as variability in the marine endmember (e.g., due to changes in seasonal ice cover), a selective degradation (of the topsoil OC) enroute that introduces a source bias or isotopic fractionation, or remaining factors related to the lateral transport time (incorrect assumption of 5000 years, nonlinear aging along transect).These differences highlight that both sourceapportionment tools still could be fine-tuned further by (i) increasing the sample size of sources to reduce endmember uncertainties, (ii) continuous adjustments in endmember values and Markov chain Monte Carlo calculations based on latest knowledge, and (iii) assuring regional testing and verification of the method when applied to new environments.

Conclusions
Leaf wax δ 2 H values in samples from aquatic recipient environments can be used to source-apportion the incoming terrestrial OC into two endmembers: a Pleistocene ICD permafrost source and a younger, Holocene, topsoil source.Mean isotopic values of the C 29 n-alkane, C 31 n-alkane and C 28 n-alkanoic acid showed a dynamic, statistically significant range of 34, 50 and 46 ‰ between topsoil-PF and ICD-PF samples, respectively, with ICD-PF samples being consistently more depleted, which is indicative of formation during the colder and drier Pleistocene.
A case study where we tested two isotopic proxies (leaf wax δ 2 H and bulk δ 13 C-14 C) to calculate the relative terrestrial source contribution of topsoil PF and ICD PF along a Laptev Sea surface sediment transect showed that the two proxies yield variable results but overall generate similar trends offshore.We reason that variability is caused by factors such as lateral transport time, remaining uncertainties in endmember definition, or environmental factors such as physical association.
Both methods (leaf wax δ 2 H and bulk δ 13 C-14 C) bring along their inherent disadvantages and advantages.The molecular approach has the distinct advantage that it circumvents the uncertainties that are associated with marine endmember definition in the case of bulk OC mixing model analysis.However, application of molecular δ 2 H in sourceapportionment studies brings along challenges related to the molecular bulk upscaling step.Bulk δ 13 C-14 C source apportionment, on the other hand, has the advantage of operating on a bulk and perhaps more representative level but is hampered by remaining uncertainties associated with the marine endmember.
This study shows that δ 2 H of leaf wax molecules has the potential to be used in quantitative source-apportionment studies of thawing permafrost in coastal or marine settings.It can serve as an alternative or complementary approach to the commonly applied bulk δ 13 C-14 C method.We recommend continuing data collection and optimization of endmember definition and calibration.Refining the molecular δ 2 H proxy presented here will be beneficial in pinpointing the location and extent of OC release from thawing permafrost in the coastal or fluvial environment.With enhanced Arctic warming and associated intensification of permafrost thaw, constraining the amount and fate of permafrost OC release will help to assess the magnitude of the permafrost carbon feedback to climate warming.
Data availability.All data are available in Tables 1 through 6 and in Table S1 in the Supplement.
Author contributions.Land-based samples were collected by GH and JEV, and ship-based samples were collected by IS, OD, ÖG, TT, LB and JEV.Laboratory analysis was performed by LB, TT and HH.Markov chain Monte Carlo simulations were run by AA.The manuscript was written by JEV with input of all co-authors.
Competing interests.The authors declare that they have no conflict of interest.Special issue statement.This article is part of the special issue "Climate-carbon-cryosphere interactions in the East Siberian Arctic Ocean: past, present and future (TC/BG/CP/OS inter-journal SI)".It is not associated with a conference.has been provided by the Dutch NWO (Veni no.863.12.004),US-NSF (Polaris Project no.1044610), the Bolin Centre for Climate Research, the Knut and Alice Wallenberg Foundation (SWERUS-C3 Program; KAW no.2011.0027), the Swedish Research Council (VR nos. 621-2004-4039 and621-2007-4631), the Russian Science Foundation (no. 15-17-20032 to Oleg Dudarev), the Nordic Council of Ministers Cryosphere-Climate-Carbon Initiative (project Defrost, no.23001), the European Research Council (ERCAdG project CC-TOP no.695331 to Örjan Gustafsson).This study was supported by the Delta Facility of the Faculty of Science, Stockholm University.Gustaf Hugelius would like to acknowledge funding from ESF-CryoCarb and EU FP7-PAGE21 projects for topsoil and ICD sample collection.The ISMAR publication ID is no.1931.
Edited by: Nina Kirchner Reviewed by: Trevor Porter and two anonymous referees

Figure 1 .
Figure 1.Map of coastal northeast Siberia showing the extent of ice complex permafrost (ICD PF, red) overlaid with the location of ice complex (n = 9; black diamonds) and topsoil samples (n = 9; green squares).The shelf-slope Laptev Sea transect is shown with yellow stars.

Figure 3 .
Figure 3. Molecular isotopic signature against chain length of longchain n-alkanoic acids (a) and n-alkanes (b) for Holocene topsoil samples (green) and Pleistocene ice complex deposits samples (ICD, blue).Stars indicate that the two compared values are statistically significant (95 % confidence).Standard deviations are represented as vertical bars and are smaller than the sample circles when not visible.

Figure 4 .
Figure 4. Concentration-weighted mean δ 2 H values of C 27 -C 29 -C 31 n-alkanes plotted against concentration-weighted mean δ 2 H values of C 24 -C 26 -C 28 n-alkanoic acids to illustrate the fractionation differences between these two leaf wax markers.Dashed line indicates an identical fractionation.

Figure 5 .
Figure 5.Comparison of δ 2 H values of C 28 n-alkanoic acid (a) and C 29 n-alkane (b) in modern topsoil PF (green circles) and ICD PF for this study (blue circles) and available literature, with crosses from Zech et al. (2011; glacial and interglacial paleosols from permafrost bluff exposure at Tumara River northeast Siberia), black triangles from Yang et al. (2011; C3 plants and trees from Canada and Alaska), light grey triangles from Wilkie et al. (2013; C3 plants from the El'gygytgyn lake basin, Siberia), white triangles from Pautler et al. (2014; modern and paleosols from Yukon, Canada) and dark grey triangles from Porter et al. (2016; muck deposits from Yukon, Canada).

Figure 6 .
Figure6.Contribution of OC from topsoil-PF (green) and ICD-PF (blue) sources to surface sediments along a shelf-slope transect in the Laptev Sea (see alsoBröder et al., 2016b for further transect information), calculated with a δ 13 C-14 C (triangles) and leaf wax δ 2 H mixing model (circles).Stations are plotted against log water depth (m; see also Table6) following the transect order from the coastal, nearshore, zone in the South (furthest left; TB-46, 6 m depth) towards the continental rise in the North (furthest right; SW-01, 3146 m depth).Topsoil 14 C endmember values are corrected for cross-shelf transport time (see Sect. 4.2).

Table 1 .
Site characteristics and geochemical properties of eight topsoil and eight ice complex deposit samples.A table with more detailed sample descriptions can be found in TableS1.

Table 2 .
Long-chain n-alkane concentrations (in µg gOC −1 ) of topsoil Holocene samples (modern vegetation and O horizon) and Pleistocene ice complex samples.

Table 3 .
Long -chain n-alkanoic acids concentrations (in µg gOC −1 ) of topsoil Holocene samples (modern vegetation and O horizon) and Pleistocene ice complex samples.

Table 4 .
Sum of most abundant long-chain n-alkanoic acids and n-alkanes (concentrations in µg gOC −1 ), and characteristic ratios of nalkanoic acids and n-alkanes of topsoil Holocene (modern vegetation and O horizon) and Pleistocene ice complex samples.
a HMW: high molecular weight.b CPI: carbon preference index for chain lengths C 22 -C 28 , for calculation see caption of Fig. 2. c CPI: carbon preference index for chain lengths C 23 -C 31 , for calculation see caption of Fig. 2.

Table 5 .
δ 2 H signatures (in ‰) of n-alkanoic acids and n-alkanes of topsoil Holocene (modern vegetation and O horizon) and Pleistocene ice complex samples.

Table 6 .
Bröder et al., 2016b)pth and isotopic values of samples along a surface sediment transect in the Laptev Sea (data fromBröder et al., 2016b), with percentage topsoil (TS) and ice complex deposit (ICD) OC contributions to the samples based on source-apportionment calculations with δ 2 H leaf wax endmembers versus δ 13 C-14 C endmembers (endmember values are described in the text).