Sediment coring took place in April 2022. As the objective was to determine the transformation of organic matter during permafrost thaw along a thaw and salt transect, coring positions were chosen from the least thaw and salt affected location, the upland, to the most thaw and salt affected sites, the lagoons. Fieldwork aimed at sampling near-surface and deep (>1 m) sediments, which was reached with the exception of coring at West Twin Lake (23 cm) and the semi-open thermokarst lagoon (46 cm). Detailed characteristics of core sites and metadata are provided in Table S1 in the Supplement. Depending on the sediment conditions, three different coring devices were used. Frozen material from the upland and the DLB sites were cored with a “Snow, Ice and Permafrost Establishment” (SIPRE) corer. Since bedfast ice was present at Elson Lagoon and the semi-open thermokarst lagoon, sediments were cored with a SIPRE corer, too. For transportation the cores were wrapped in plastic foil and stored frozen in thermoboxes.

Unfrozen sediments of West Twin Lake were cored using a push corer (UWITEC gravity corer). Additionally, to get deeper, East Twin Lake was sampled with a vibration coring system (Livingstone-type drive rod piston corer). These cores were kept unfrozen in their PVC (push core) and aluminum (vibra core) tubes in thermoboxes for transportation to the Alfred Wegener Institute (AWI, Potsdam, Germany).

In order to assess sediment characteristics along thaw and salinity gradients, a multiproxy approach was applied as individual organic matter proxies can be influenced by source heterogeneity, post-depositional alteration, and mixing of modern and reworked deposits (Diefendorf et al., 2011; Jong et al., 2024; Wu et al., 2022). Combining complementary proxies therefore allows robust interpretations of carbon sources, preservation states, and transformation processes.

In preparation for the investigation of individual biogeochemical parameters, SIPRE cores were processed in a climate chamber at -8 °C. The lake cores were prepared at 4 °C. All cores were cut in halves, cleaned, photographed, and described (sediment structures, color, cryostructures, Tables S2–S7 in the Supplement). Subsequently, sediment cores were subsampled in adjusted steps (approx. 5 cm intervals) following the sediment structure, and weighted. With the exception of East Twin Lake, no artificial disturbances of the cores were observed.

During vibracoring of the East Twin Lake, especially sediments in the upper part of the lake deposits experienced compaction due to high water content and extended vibration time, while the lower sediments were more consolidated and subjected to less compaction. To correct the recovered core depths for compaction, a normalized exponential depth correction model was applied, following Athy's law of exponential porosity–depth decay (Athy, 1930) and approaches used in ocean drilling studies (Lisiecki and Herbert, 2007). The true sediment depth zt was calculated from the measured depth zm (in the core liner) using the formula: 1zt=Dp×1-e-zm/λ1-e-Lc/λ where Dp is the total penetration depth of the coring barrels (259 cm in this case), Lc is the total recovered core length (here 138 cm), and λ (lambda) is the exponential e-folding depth scale. The e-folding depth (λ) is defined as the characteristic depth over which a quantity decreases to 1/e (approximately 37 %) of its initial value, assuming an exponential compaction. In this study, λ represents the depth scale at which sediment compaction effectively reduces porosity or thickness by a factor of e due to vibration-induced consolidation. Based on previous studies of lakes, thermokarst properties were adapted (especially water content, grain size, and porosity, e.g., Walter Anthony et al., 2014), and λ=50 cm was assumed, reflecting the highly compressible, water-rich upper layers. This correction ensures that the calculated true depth equals Dp at the base of the core and follows an exponential compaction trend consistent with observed porosity–depth profiles in similar sediments.

All samples were kept at 4 °C. Porewater was extracted with Rhizon samplers (0.12–0.18 µm membrane pore size). The extracted waters were measured with an Orion VERSASTAR PRO (Thermo Fisher) for electrical conductivity (mScm-1). Salinity stages were adapted from Cahyadi et al. (2018) as follows: Freshwater <1.5 mScm-1, brackish 1.5–15.0 mScm-1, saline 15.0–50.0 mScm-1, and hypersaline >50 mScm-1. For subsequent sample treatment, the sediment remains were freeze-dried, and the water/ice contents (in weight %, wt %) were calculated by taking into account the wet and dry sample weights.

For the investigation of the grain size distribution, subsamples were treated with hydrogen peroxide for the removal of organic matter for four weeks. The grain sizes were measured with a Malvern Mastersizer 3000 (0.01–1000 µm grain size range) and evaluated with GRADISTAT 8.0 (Blott and Pye, 2001).

Using further subsamples, samples were milled and homogenized for the elemental analyses using a planetary mill (Fritsch PULVERISETTE 5). Using a soliTOC and a rapid N exceed element analyzer (Elementar), total organic carbon (TOC) and total nitrogen (TN) contents were quantified through the combustion and the analysis of the resulting gases. The detection limits of TOC and TN are 0.001 wt % and 0.05 wt %, respectively, however, the detection limit was artificially set to 0.1 wt % for both elements. TOC and TN contents below the detection limit (0.1 wt %) were replaced by half of the detection limit (0.05 wt %). This is to prevent a bias and is according to Strauss et al. (2022) to make the point that low measurements are not the same as no measurements, avoiding zero inflation. For the TN samples below the detection limit, no C:N ratio calculation was done to avoid artificial artifacts. The C:N ratio is a common indirect indicator of organic matter degradation if assuming a same source signal, with lower values reflecting higher decomposition (e.g., Fuchs et al., 2019; Mu et al., 2025). It also helps to infer source material, distinguishing algal input (C:N<10) from terrestrial plant matter (C:N>20) (Meyers, 1994).

Samples were decarbonized with hydrochloric acid before quantification of δ13C at AWI's ISOLAB facility. δ13C ratios were measured using internationally certified reference materials obtained from the IAEA, including USGS-24 (-15.99 ‰) and IAEA-CH-7 (-31.80 ‰). In addition, internal laboratory standards (peptone (-24.09 ‰) and nicotinamide (-32.73 ‰)) were analyzed for quality control. Isotope values are reported in ‰ relative to Vienna Pee Dee Belemnite (Coplen et al., 2006), and analytical precision was better than ±0.15 ‰ based on repeated standard measurements (Schwamborn et al., 2023). Measurements were run on a ThermoFisher Scientific Delta-V-Advantage gas mass spectrometer equipped with a FLASH elemental analyser EA 2000 and a CONFLO IV gas mixing system. Stable carbon isotope ratios are commonly applied as a proxy for organic matter origin and degradation in permafrost regions (e.g., Alewell et al., 2011; Strauss et al., 2015).

Radiocarbon dating was conducted on 21 selected samples by accelerator mass spectrometry at the MICADAS radiocarbon laboratory at AWI Bremerhaven (Mollenhauer et al., 2021). Preferably, plant remains were picked for dating, however, bulk sediment was analyzed for four samples due to the lack of macrofossils (Table S8 in the Supplement). Macrofossil material was subject to a standard acid–base–acid pre-treatment (Mollenhauer et al., 2021). Subsequently, samples were rinsed to neutral pH, dried, and combusted using an elemental analyzer. Owing to limited sample mass, all but two macrofossil samples (AWI MICADAS sample ID 12532.1.1 and 12535.1.1, Table S8) were analyzed in gas mode rather than as graphite targets. Radiocarbon measurements were normalized and blank-corrected using size-matched OxAII and PhA standards. Data quality was monitored using wood reference material (IAEA-C5) in parallel (Mollenhauer et al., 2021). The bulk solid phase samples were acidified three times in silver boats with 6 M HCl to remove inorganic carbon. Radiocarbon dating was performed on 1 mgC graphite targets. Sample treatment and blank determination was performed following existing protocols by Mollenhauer et al. (2021). Raw radiocarbon ages were calibrated with Calib 8.20 using the IntCal 20 calibration curve (Reimer et al., 2020; Stuiver and Reimer, 1993).

The total lipid extract from freeze-dried, milled and homogenized sediment was extracted with a Dionex ASE 350 Accelerated Solvent Extractor using dichloromethane/methanol (DCM/MeOH, 99:1 v/v, heating phase 5 min, static phase 20 min at 75 °C and 10 MPa). Next, the internal standard 5α-androstane was added to the extract. Asphaltenes were precipitated to avoid complications during the next step, where the n-alkane including aliphatic fraction was separated via medium pressure liquid chromatography using n-hexane over silica columns (Radke et al., 1980). This fraction was then measured using a Thermo Scientific ISQ 7000 Single Quadrupole Mass Spectrometer in combination with a Thermo Scientific Trace 1310 Gas Chromatograph (capillary column from BPX5, 2mm×50m, 0.25 mm). With a total run time of 118 min, the MS transfer line temperature was set to 250 °C and the ion source temperature to 230 °C (ionisation energy 70 eV at 50 µA). Targeted screening of mid- and long chain n-alkanes (C23–C33) was realized with the software Xcalibur (Thermo Fisher). The mass spectra (m/z 50–600 Da, 2.5 scanss-1) and the internal standard were used for compound specific identification and quantification.

The total n-alkane content (TAC) is the first proxy that is indicative of (paleo) environmental conditions and degradation effects (e.g., Brittingham et al., 2017; Thomas et al., 2021). TAC is provided for the following chain lengths in µgg-1 Sediment (Sed) and µgg-1TOC: 2TAC=∑C23–C33 The average chain length (ACL), developed by Poynter and Eglinton (1990), is a proxy that is indicative of the organic matter source. Algae and microorganisms are typically dominated by short-chain n-alkanes (<C22), aquatic (submerged/floating, e.g., Sphagnum moss) vegetation can be recognized by a dominance of mid-chain n-alkanes (C23/C25) while terrestrial vegetation is dominated by long-chain n-alkanes (>C25) (Baas et al., 2000; Ficken et al., 2000; Killops and Killops, 2005; Otto and Simpson, 2005). Terrestrial vegetation can further be distinguished between shrubs and trees (C27/C29 predominance) and grasses and herbs (C31/C33 predominance) (Maffei, 1996; Schäfer et al., 2016). This chemotaxonomic “fingerprint”, as Eglinton et al. (1962) put it, can therefore be used as a tool for past environmental changes (e.g., Schwark et al., 2002; Zech et al., 2010). ACL limitations concern the blindness towards gymnosperms, overlapping chemotaxonomic patterns of different source material and potential post depositional alteration through degradation (Diefendorf et al., 2011; Jongejans et al., 2020; Struck et al., 2018; Zech et al., 2021). The ACL is calculated following Brittingham et al. (2017) and others: 3ACLC23-C33=∑i×Ci∑Ci Odd chain alkanes predominate over even chain numbers (Eglinton et al., 1962). The ratio of odd against even chain lengths, called CPI, was first developed by Bray and Evans (1965) and further developed by Marzi et al. (1993). With advancing organic matter decay this ratio decreases (e.g., Andersson and Meyers, 2012; Schäfer et al., 2016). The CPI values of algae, bacteria, and highly degraded substances such as oil approach one while less degraded organic matter retains values >5 (Haugk et al., 2022; Killops and Killops, 2005; Tipple and Pagani, 2010). The CPI calculation is based on the formula introduced by Marzi et al. (1993) for the chain length interval C23–C33: 4CPIC23-C33=∑oddC23-31+∑oddC25-332×∑evenC24-32 To differentiate terrestrial vegetation between trees and shrubs (approaching 0), and grasses and herbs (approaching 1), the endmember model of Schäfer et al. (2016) is applied: 5n-alkane ratio=C31+C33C27+C31+C33 Ficken et al. (2000) developed a proxy (Paq) which allows to more specifically quantify the share of aquatic macrophyte input into the sedimentary record. Submerged/floating macrophytes reveal relatively high values (Paq>0.4) compared to emergent and terrestrial plants (0.1<Paq<0.4; Paq<0.1; respectively). To evaluate aquatic influences in the continuum from the permafrost upland, over the thermokarst lakes and the drained basin, towards the lagoons, Paq was applied accordingly: 6Paq=C23+C25C23+C25+C29+C31

The combination of the biogeochemical and hydrochemical parameters was used to interpret units along the sediment cores (described top-down, labelled in Roman numerals) to enhance clarity and facilitate understanding. Mean values are reported alongside their standard deviations. Correlations between geochemical variables were assessed using Pearson correlation analysis.

In order to statistically quantify the characteristics of organic matter parameters (TOC, C:N, δ13C) in different sediment regimes, groups of thermal conditions (seasonally frozen, perennially frozen, unfrozen, refrozen, lagoon) and salinity stages (freshwater, brackish, saline, hypersaline) were created. Due to violations of normality in multiple groups, as determined by Shapiro–Wilk tests, the non-parametric Kruskal–Wallis test was used to assess differences in TOC, C:N, and δ13C among groups. For subsequent pairwise comparisons of the categories with the parameters, Mann–Whitney U tests were applied, coupled with the Benjamini–Hochberg (BH) p-value adjustment method. n-Alkane proxies were not included in the statistical tests due to limited sample counts in individual groups.

To quantify if any group statistically differs from another based on the combination of TOC, C:N, and δ13C, a Permutational Multivariate Analysis of Variance (PERMANOVA) was performed. Also, unique groups based on both their thermal and salinity condition were created and tested with a PERMANOVA. Following a significant overall result, post-hoc pairwise PERMANOVA tests were conducted with p-value adjustment (BH) to identify which specific groups differed. All statistical analyses were carried out in RStudio and AI tools have been utilized for R coding.

The upland core from the tundra site is characterized by silt-dominated ice-rich permafrost in the first two meters of the soil column (on average 66.4 wt % ± 13.1 wt % ground ice; Fig. 2a). The electrical conductivity of the porewater indicates freshwater conditions (<1.5 mScm-1) and a trend towards saline conditions (max. 17.3 mScm-1) below 140 cm depth (Unit VI). At 75 and 102 cm depth, radiocarbon dating reveals similar ages with 6.98 and 6.97 cal ka BP, respectively, while a lower sample at 186 cm depth has an age of 13.3 cal ka BP. This suggests that Holocene deposits with a thickness of at least 126 cm (7.3 cal ka BP) cover the underlying late Pleistocene sediments. The TOC contents range between 35.3 wt % (Unit I) and 5.6 wt % (Unit VI). Between 70 and 100 cm depth, TOC contents are increased (Unit IV) compared to sediments above and below. TN contents relate to TOC contents and range between 0.4 wt % and 1.7 wt % (on average 0.7 wt % ± 0.3 wt %). Overall, the C:N ratio is relatively homogenous along the core (on average 17.1 ± 2.9). In the upper meter (Unit I–IV), the ratio is increased (19.7 ± 1.5) compared to the lower meter (14.4 ± 0.5). Carbon isotopic values range from an average of -26.8 ‰ ± 0.6 ‰ in the upper 48 cm (Units I, II) to -27.7 ‰ ± 0.4 ‰ between 53 and 198 cm depth. C:N and δ13C (r=0.78, p<0.001) correlate significantly (Fig. S1 in the Supplement).

Total n-alkane contents in the upland indicate a decreasing trend with depth (average 42.0 ± 37.1 µgg-1Sed; 268.7 ± 160.7 µgg-1TOC; Fig. 3). The maximum content is found at 10 cm depth with 113.3 µgg-1Sed (524.7 µgg-1TOC) and the minimum at 186 cm depth with 5.1 µgg-1Sed (89.3 µgg-1TOC). Long-chain n-alkanes dominate (average ACL 26.9 ± 0.6) and the degradation proxy CPI (average 15.1 ± 4.4) reveals a distinct pattern. First, the CPI increases with depth from 10.6 to 21.5 (75 cm depth). Then, it drops to 10.0 (126 cm depth) and increases again to 15.6 (186 cm depth). The n-alkane ratio, indicative for different terrestrial vegetation inputs, increases along the core from 0.1 (uppermost sample) to 0.4 (lowermost sample). Paq indicates aquatic influences with values between 0.4 and 0.7.

The pushcore of the West Twin Lake talik is 23 cm long (4 subsamples) and therefore represents the surface sediments of the thermokarst lake (Fig. 2b). The sediment consists of fine silt (mean grain size 4.8 ± 0.2 µm) with porewater of slightly brackish conditions (1.9 mScm-1). TOC contents are on average 16.1 wt % ± 0.4 wt %, and TN values range from below detection limit to 0.9 wt %. This results in C:N ratios of 17.5 (8 cm depth), 17.4 (16 cm depth), and 18.9 (21 cm depth). The δ13C ratio is on average -28.6 ‰ ± 0.1 ‰.

One biomarker sample was analyzed at the West Twin Lake core at 8 cm depth (Fig. 3). The total alkane content is 192.2 µgg-1Sed (1214.3 µgg-1TOC) and is therefore higher than in the upland. On the other hand, the ACL (25.5) and CPI (6.9) are lower compared to the upland. The n-alkane ratio, signalizing terrestrial input into the lake, is 0.2, while the Paq value is 0.8.

The sediments in the vibracore from the East Twin Lake are sandy to silty (Fig. 2c). Until 187 cm depth (Unit I, II) mean grain sizes increase to 68.6 µm, and below that they average 16.4 ± 6.1 µm. The electrical conductivity continuously increases from brackish (14.1 mScm-1 at 8 cm depth) to saline (43.3 mScm-1 at 235 cm depth). Subsequently, the conductivity decreases but remains in saline conditions. At 41 cm depth, radiocarbon dating revealed a Holocene age of 2.9 cal ka BP. At 215 and 252 cm depth, the sediment is of late Pleistocene age, 27.7 and 42.6 cal ka BP, respectively (Unit III). The surface sediments have a TOC content of 16.0 wt % (Unit I). Subsequently, TOC decreases and remains <3.0 wt % below 123 cm depth (Unit II, III). Only six out of 26 samples showed TN values above the detection limit. The TN content in these samples is 0.1 wt %–0.9 wt %. The resulting average C:N ratio is 9.9 ± 6.1. Carbon isotopic values increase from a minimum of -28.9 ‰ at 41 cm depth to -25.7 ‰ at 197 cm depth. Below that, δ13C values remain relatively homogenous until the core bottom at 258 cm with an average of -25.8 ‰ ± 0.1 ‰. C:N and δ13C (r=-0.91, p=0.01) reveal a very strong negative correlation (Fig. S1).

The surface sediment sample of East Twin Lake has a TAC of 305.4 µgg-1Sed (2179.2 µgg-1TOC), while at 75 and 121 cm depth, it decreased to 1.9 µgg-1Sed (129.7 µgg-1TOC) and 17.0 µgg-1Sed (1020.0 µgg-1TOC), respectively (Fig. 3). Long-chain n-alkanes predominate (average ACL 26.0 ± 0.5). CPI (average 6.2 ± 0.8), n-alkane ratio (average 0.3 ± 0.1), and Paq (0.7 ± 0.1) are similar as in West Twin Lake.

The DLB core comprises frozen silty sand (Fig. 2d). The permafrost is less ice-rich compared to the upland (on average 48.8 wt % ± 20.6 wt % ground ice), and the electrical conductivity indicates predominantly brackish porewater conditions until 110 cm depth (on average 3.7 ± 3.1 mScm-1). The basal surface peat (21 cm depth, Unit I), an indication of the drainage event date, has an age of 0.7 cal ka BP. Between 74 and 108 cm depth, an age-depth inversion occurs with 10.5 and 9.9 cal ka BP, respectively. The first late Pleistocene age is found at 140 cm depth with 34.1 cal ka BP. TOC values range between 1.5 (129 cm depth) and 42.4 wt % (undecomposed surface peat). Between 74 and 113 cm depth (Unit IV), a cryoturbated sediment structure is found, which is characterized by organic-rich sediments (33.5 wt % TOC at 113 cm depth). The TN contents are in the range of contents found in the upland, but are on average slightly lower (0.5 wt % ± 0.4 wt %). The C:N ratios (on average 18.6 ± 5.1) reflect the TOC and TN patterns along the core however, peak values (0–15 cm and 113 cm depth) are less dominant. The δ13C values increase with depth and range between -29.4 ‰ and -25.0 ‰ at 3 and 137 cm depth, respectively. No significant correlation is found between C:N and δ13C (r=-0.35, p=0.08; Fig. S1).

Compared to all investigated landforms, n-alkane specific parameters indicate the strongest heterogeneity in the DLB (Fig. 3). TAC ranges between 0.9 µgg-1Sed (38.6 µgg-1TOC) and 519.8 µgg-1Sed (3419.6 µgg-1TOC), which is the lowest and highest observed value in our study, respectively. The minimum ACL (25.6 at 31 cm depth) is similar to the surface samples of the lakes. At 108 cm depth, extreme values are reached for ACL (30.2), CPI (51.3), the terrestrial n-alkane ratio (0.9), and Paq (0.02).

The thermokarst lagoon core contains silty sediments (Fig. 2e). The average water content is 44.4 wt % ± 3.6 wt % and the porewater is characterized by hypersaline conditions with a peak value of 119.0 mScm-1 at 34 cm depth. Radiocarbon dating of plant macrofossils shows an age-depth inversion with 7.2 cal ka BP at 19 cm depth and 4.6 cal ka BP at 44 cm depth. The TOC and TN contents average 5.8 wt % ± 0.6 wt % and 0.4 wt % ± 0.0 wt %, respectively. This results in an average C:N ratio of 16.1 ± 0.6. Relatively homogenous values are also found for δ13C with an average of -27.6 ‰ ± 0.1 ‰. C:N and δ13C (r=-0.50, p=0.17) are not correlating (Fig. S1).

The two analyzed biomarker samples in the thermokarst lagoon are similar (Fig. 3). A TAC of 40.0 µgg-1Sed (784.8 µgg-1TOC) and 31.5 µgg-1Sed (534.7 µgg-1TOC) is present at 19 and 44 cm depth, respectively. The average ACL (26.4 ± 0.1) and CPI (8.3 ± 0.1) lie between the observed values of the thermokarst lakes and the terrestrial sites. Also, the terrestrial n-alkane ratio (average 0.3 ± 0.0), and Paq (average 0.6 ± 0.0) are within the range of values observed in the other cores.

The Elson Lagoon core comprises silty sediments underlain by sand (Fig. 2f). The maximum and minimum porewater contents are 55.8 wt % and 16.6 wt % at 7 and 109 cm depth, respectively. Compared to the thermokarst lagoon, Elson Lagoon is less saline, yet hypersaline conditions prevail (on average 76.2 ± 8.7 mScm-1). Radiocarbon ages signalize sediment mixing as plant macrofossils have an age of 5.5 cal ka BP at 2 cm depth and 5.3 cal ka BP at 37 cm depth. Along the core, TOC contents are on average 4.5 wt % ± 2.2 wt % and show a decreasing trend. Until 87 cm depth, TN averages 0.3 wt % ± 0.1 wt %, and contents are <0.1 wt % below this depth. The C:N ratio is relatively homogenous with an average of 18.4 ± 0.9. δ13C lies on average at -27.7 ‰ ± 0.1 ‰ until 76 cm depth (Unit I). From 81 to 109 cm depth (Unit II) δ13C precipitously increases to -25.4 ‰, which is the maximum value compared to all cores. No correlation is found between C:N and δ13C (r=-0.26, p=0.32; Fig. S1).

The sediments of Elson Lagoon reveal a decreasing trend of TAC with depth (average 32.1 ± 28.5 µgg-1Sed; 483.9 ± 432.2 µgg-1TOC; Fig. 3). n-Alkane proxies compare generally well to the thermokarst lagoon with the exception of the CPI. The degradation proxy indicates less degraded material (average 11.4 ± 2.4) in the non-thermokarst lagoon compared to the thermokarst lagoon. Otherwise, the average ACL (26.6 ± 0.2), terrestrial n-alkane ratio (0.3 ± 0.0), and Paq (0.6 ± 0.0) are similar as in the thermokarst lagoon.

Our five thermal sediment stages comprise seasonally frozen including the max. active layer depths of the upland and the DLB after Nyland et al. (2021), perennially frozen referring to the permafrost domain of the upland, unfrozen (talik/cryopeg of West and East Twin Lake), refrozen corresponding to DLB permafrost and lagoon (thermokarst lagoon and Elson Lagoon) (for details see Table S9 in the Supplement).

Kruskal–Wallis tests show that significant differences exist among the thermal regimes for TOC (p<0.001), C:N (p<0.001), and δ13C (p=0.02) (Table S9). Significance levels of pairwise Mann–Whitney test results are visualized in Fig. 4. The PERMANOVA test considering the combination of TOC, C:N and δ13C indicated significant differences between thermal regimes (r=0.40, p=0.001). Post-hoc testing revealed that unfrozen sediments are significantly different from frozen (r=0.40, p=0.05), refrozen (r=0.55, p=0.02), and lagoon sediments (r=0.40, p=0.04). Frozen sediments further differ from refrozen (r=0.40, p=0.03) and lagoon (r=0.55, p=0.01) sediments. Also, significant differences exist between refrozen and lagoon sediments (r=0.40, p=0.03).

Grouping of sediment samples was conducted after previously defined salinity stages (Sect. 2.2.2, for details see Table S10 in the Supplement).

Significant differences in TOC (p<0.001), C:N (p<0.001), and δ13C (p<0.001) exist across at least some of the defined salinity stages according to Kruskal–Wallis (Table S10). Pairwise Mann–Whitney test significant levels are plotted in Fig. 5. PERMANOVA testing indicated that significant differences in salinity categories exist (r=0.55, p=0.001). Post-hoc testing revealed weak but significant differences between brackish and hypersaline samples (r=0.07, p=0.05).

Since thermal and salinity stages were considered individually concerning their influence on organic matter characteristics, their combined impact was also investigated. Combining the thermal and salinity stages into unique groups resulted in eleven groups (Table 1). In PERMANOVA tests specific groups were excluded from the analysis due to low sample counts (n<5 in G5, G7, G8, G10, Table 1). The PERMANOVA indicated that groups differ significantly from each other concerning TOC, C:N and δ13C (r=0.58, p=0.001). Post-hoc pairwise PERMANOVA tests revealed that Group 2 and Group 6 were significantly different (r=0.87, p=0.03). Also, Group 3 differed significantly from Group 9 (r=0.87, p<0.01).

The recent active layer depth in the tundra ranges between 29 and 47 cm (Nyland et al., 2021). This aligns well with the increased TOC and TN contents of Unit I (Fig. 2a). Measured electrical conductivity in the sediments reflects freshwater conditions, consistent with observations from the same upland (Jones et al., 2023) and in nearby streams (Lougheed et al., 2020). Unit II represents the range of the late-season thaw depth and the transient layer, which can be recognized by increased water/ice contents (Shur et al., 2005). The permafrost begins below 47 cm (Unit III). Bockheim et al. (1999) found that >75 % of soils in the region are affected by cryoturbation; signs for this are found between 70 and 100 cm depths (Unit IV, Table S2). TOC and TN contents are increased compared to the units above and below, and the radiocarbon dates reveal an age-depth inversion. Cryoturbation at this depth could point towards a paleo active layer from the mid Holocene. Relatively homogenous conditions in the second meter (Unit V, VI) indicate fairly stable conditions from the Late Glacial into the early Holocene period.

Through the investigation of pollen records, Meyer et al. (2010) describe a shift from a purely grass dominated tundra in the Late Glacial towards a grass, sedge, and dwarf shrub mix in the early Holocene. Considering the terrestrial n-alkane proxies ACL and the n-alkane ratio, it could be inferred that this slowly occurring shrubification throughout the Holocene can be confirmed by our n-alkane analysis due to the tendency towards lower ACL and n-alkane ratio values in Holocene samples compared to the deepest (Late Glacial) biomarker sample (Fig. 3). Furthermore, an aquatic influence indicated by the Paq may be explained by the presence of Sphagnum moss in ice-wedge troughs (Lara et al., 2015). However, Paq could also be affected by mid-chain-length dominated Betula shrubs, whose n-alkane distributions can overlap with those of aquatic vegetation, limiting its source specificity (Ficken et al., 2000; Weber and Schwark, 2020).

Different from the gradual top-down thaw processes in the upland, thermokarst lakes are affected by abrupt thaw processes which can reach deeper deposits over short time scales (Grosse et al., 2011; Webb et al., 2025). While uplands and thermokarst lakes are sedimentologically connected due to the same parent material, the biogeochemical fingerprints of lakes may have changed due to recent thaw processes and lacustrine deposition. Comparing our two lake sites is challenging, as the West Twin Lake core is short, thus, a full comparison to the longer East Twin Lake core is somewhat limited. The main difference lies in the salinity between the lake sediments. While brackish conditions occur beneath West Twin Lake, the sediments of East Twin Lake have predominantly saline porewater (Fig. 2b and c). In East Twin Lake, electrical conductivity was observed to have almost tripled between 2016 and 2022 (Jones et al., 2023). As negative temperatures were measured in these sediments (Jones et al., 2023), we defined the unfrozen sediments as a cryopeg. Since brackish talik sediments were determined in the freshwater West Twin Lake, a near-future shift towards brackish lake water is likely. East Twin Lake's sediments (ca. 123 cm, Unit I) can be distinguished from the underlying late Pleistocene marine influenced sediments (Unit III) by considering the TOC contents (Eisner et al., 2005). These are in the range of values observed in the Teshepuk Lake area, ca. 120 km east of the Utqiaġvik Peninsula (Lenz et al., 2016). Paq indicates an increased aquatic influence in the upper sediments. Yet ACL and the n-alkane ratio also indicate terrestrial plant input into the lake which can be explained by enhanced lateral expansion (shoreline erosion) in recent years (Jones et al., 2023). It needs to be stressed that in our study the ACL decreases with decreasing CPI values (r=0.79, p<0.001), meaning that the vegetation signal is influenced by organic matter degradation (strongest in East Twin Lake).

The transition from Unit II to III in the East Twin Lake core is characterized by a sharp decrease of the mean grain size from fine sand to silt. The sand might be aeolian which deposited in the surface depression (Carter, 1981; Eisner et al., 2005). In both units, the C:N and δ13C ratios point towards a marine/lacustrine algae production (Fig. 4b and c; Meyers, 1994), which aligns with the high salinity (Fig. 2c). This pattern matches with the investigation of Meyer et al. (2010), who observed more enriched δ13C values in the salty late Pleistocene sediments at a permafrost tunnel on the Utqiaġvik Peninsula.

In view of the landscape alteration, the next step is the drainage of thermokarst lakes where permafrost is re-forming. The DLB core revealed mixed signals. Electrical conductivity is similar to the upland, which is linked to Holocene ages, whereas the sediments are sandier, comparable to the East Twin Lake sediments. Brown et al. (2003) estimated that the lake drained about 100–150 years ago; however, our radiocarbon dating revealed that post-drainage peat formation was initiated ca. 700 cal a BP (Fig. 2d). DLBs sequester organic carbon in the form of peat (Bockheim et al., 2004; Hinkel et al., 2003; Jones et al., 2012), which can be recognized by the highest TOC contents and C:N ratios in the upper 15 cm (Unit I). We argue that the most recent lake phase in the DLB started 1.3 cal ka BP and therefore lasted only approximately 600 years when considering the TOC contents (Unit II). Between 21 and 31 cm depth, TOC contents are in the range of contents found in the East Twin Lake sediments, and Paq signalizes aquatic conditions. Our observation of the former lake phase shows that these periods are in this case shorter than previously estimated by Fuchs et al. (2019), who found that lakes persist a minimum of 1000 years in northern Alaska. The relatively carbon-poor and alkane-depleted sediments beneath the lake phase might be a refrozen talik starting beneath the modern active layer (Unit III). The brackish and refrozen sediments differed statistically from frozen upland deposits affected by freshwater conditions (Sect. 3.2.3). In Unit IV, increased TOC contents and a high n-alkane ratio – indicating the presence of herbaceous plants (Schäfer et al., 2016) – point toward one or even two wetland periods with organic matter accumulation, which likely occurred during the Holocene Thermal Maximum (Jones and Yu, 2010; Kaufman et al., 2004). The low TOC and TN content at 93 cm depth make it difficult to clarify whether a first wetland period was followed by another, as this impression could also result from cryoturbation. The age-depth inversion is an indication of both a refrozen talik and intensified freeze-thaw dynamics during the early Holocene. Unit V represents the late Pleistocene to early Holocene transition as indicated by the trend towards more negative δ13C values (Fig. 2d).

If thermokarst lakes or their drained basins are connected and inundated by sea water, they become thermokarst lagoons. With a size of 80 ha, the semi-open thermokarst lagoon is a relatively small lagoon based on the circum-Arctic thermokarst lagoon assessment published by Jenrich et al. (2025b). TOC contents are higher compared to other lagoons, such as on the Bykovsky Peninsula in Siberia (Jenrich et al., 2021; Schirrmeister et al., 2018; Ulyantsev et al., 2017; Yang et al., 2023), in northwestern Canada (Jenrich et al., 2025a), and the Teshepuk Lake area (Giest et al., 2025). Shoreline erosion and fluvial organic matter input from streams (Fig. 1) are likely driving factors for this pattern.

Elson Lagoon, which is not a thermokarst lagoon, receives terrestrial sediment input through coastal erosion, and deposition takes place as the barrier islands protect the waters of the lagoon (Brown et al., 2003; Ping et al., 2011; Zimmermann et al., 2022). Our investigations indicate that the upper 80 cm of the sediments are terrestrial deposits, since the biogeochemical parameters are overall very similar to the values found in the sediments of the thermokarst lagoon (Unit I; Figs. 2f and 3). These are underlain by likely late Pleistocene to early Holocene sandy sediments (Unit II).

The age-depth inversions signalize sediment mixing in both lagoons, which is favored through the semi- to unfrozen conditions (Table S1). The hypersalinity in the sediments might be the result from concentrating salt through bedfasting ice in the shallow lagoons (Jenrich et al., 2021). Hypersaline conditions in Elson Lagoon were quantified earlier by Osterkamp and Harrison (1985) as well as Overduin et al. (2012).