The RTS sites studied here (Fig. 1a) have been described in detail previously (Bröder et al., 2021; Keskitalo et al., 2021) and have featured in related work on sediment dynamics and carbon fluxes (Littlefair et al., 2017; Shakil et al., 2020; Thomas et al., 2023; Zolkos et al., 2019). Sites CB and SF are smaller, more recently initiated RTS systems at higher elevation within tundra-like vegetation, whereas sites FM2 and FM3 are older, much larger and at lower elevation in more forested settings (Table 1). Samples were collected from the four key geomorphological features common to each slump: seasonally thawed active layer (AL), Holocene (HO) (and deeper Pleistocene (PL) where exposed) permafrost (PF) layers, freshly thawed slump debris (DB), and suspended sediments in runoff (RU). These zones are illustrated in the images of the FM3 headwall in Fig. 1b.

Site parameters were measured during the sampling campaign in 2017 and are described in more detail by Bröder et al. (2021) and Zolkos et al. (2019) (Table 1). In short, active-layer material was sampled from the headwall, permafrost blocks were cut directly from exposed headwalls, and debris and runoff sediments were collected with stainless-steel scoops following the procedures outlined in Bröder et al. (2021). Samples were placed in pre-cleaned containers, stored frozen until return to the laboratory, and subsequently freeze-dried, ground, and homogenised prior to analysis. Inorganic carbon was removed by acid fumigation with concentrated hydrochloric acid (HCl) for 72 h at 60 °C, following standard procedures for solid-phase organic-matter analysis (Harris et al., 2001; Komada et al., 2008). Pre-treated samples from all four sites (CB, SF, FM2, FM3) were first analysed using solid total organic carbon analysis (SoliTOC) to provide a bulk assessment of OM content and thermal lability (Mittelbach et al., 2025). A subset of samples from the larger slumps (FM2 and FM3) was subsequently investigated by online ramped oxidation-accelerator mass spectrometry (ORO-AMS) and thermally sliced-pyrolysis-gas chromatography mass spectrometry (Ts-Py-GCMS) to determine the radiocarbon age, energy distribution, and molecular composition of thermally resolved fractions.

SoliTOC analyses were carried out to obtain an overall characterisation of OM thermal stability across the geomorphological features of each thaw slump. For each sample, ∼50 mg of pre-treated material was loaded into a ceramic crucible and analysed using a SoliTOC Cube analyser (Elementar GmbH) following the German industrial standard DIN 19539 (Deutsches Institut für Normung, 2016). Instrument calibration employed a calcium-carbonate (CaCO3) standard containing 12 % total inorganic carbon (minimum p.A. quality) mixed with aluminium oxide (Al2O3). Analytical accuracy was verified using two certified reference materials: a high-organic-carbon sediment (Säntis SA33802151, 7.45 % C) and a low-organic-carbon soil (Säntis SA33802152, 1.54 % C). Precision within the analytical sequence was assessed using repeated analyses (n=5 for each standard), yielded standard deviations of 0.05 % C and 0.02 % C, corresponding to relative standard deviations of 0.7 % and 1.3 %, respectively. Mean recoveries were 97.35 % (SA33802151) and 99.51 % (SA33802152), confirming stable and accurate instrument performance.

Organic carbon was quantified using three operationally defined temperature steps following DIN 19539: total organic carbon released during the 400 °C isothermal step (TOC₄₀₀), representing a thermally labile pool; residual oxidisable carbon (ROC), released during the subsequent 600 °C step; and total inorganic carbon (TIC), released during the final 900 °C step. Rapid heating between steps minimises CO2 release during temperature ramps, such that carbon is operationally assigned to the target isothermal intervals rather than to a continuous temperature ramp (Mittelbach et al., 2025).

The sum of TOC₄₀₀ and ROC was defined as total organic carbon (TOC). Because a small fraction of thermally recalcitrant OC appears to combust within the nominal TIC window, the operational fractions TOC₄₀₀ and ROC should be interpreted strictly as method-defined thermal lability pools. The ROC/TOC ratio, first introduced as an operational index of recalcitrance by Mittelbach et al. (2025), may therefore serve as a proxy for intrinsic oxidation resistance but is not necessarily equivalent to biological lability (see Results and Discussion for an assessment of methodological limitations). For completeness, total carbon (TC), defined as the sum of TOC and TIC, was also reported to verify bulk carbon consistency across analytical methods (Table S2 in the Supplement).

To resolve the age structure of OM thermal-lability fractions, ORO-AMS analyses were conducted using the setup described by Bolandini et al. (2025). This method captures and measures radiocarbon activity of CO2 that is released across a series of temperature windows. Briefly, the system features a dual-oven configuration: the first oven (where the sample is loaded) applies a linear temperature ramp to progressively oxidise OM from the sample, while the second oven is maintained at constant temperature and contains catalytic material to ensure complete oxidation and removal of non-carbon species. Released CO2 within each temperature window is then purified, trapped using a dual trap molecular zeolite interface (De Maria et al., 2021), it is first purified, and transferred to an accelerator mass spectrometer (Low Energy Accelerator, LEA, IonPlus) for radiocarbon measurement (Ramsperger et al., 2024; Synal et al., 2007).

In this study, approximately 40–50 mg of acid-fumigated, homogenised material from each geomorphological feature of the larger thaw slumps (FM2 and FM3) was analysed individually, including replicate combustions of selected samples (10 primary analyses plus 3 replicates). Additional combustions from FM2, FM3, CB, and SF were conducted for pre-screening and bulk characterisation. In total, more than 30 ORO combustions were performed across all sites. Samples were loaded into pre-combusted quartz tubes placed within the oxidation reactor. The ramping furnace was programmed to heat linearly at 5 °Cmin-1 from 150 to 900 °C under a continuous flow of 18 % O2 in He (90 mLmin-1). This oxygen-rich carrier gas was used to promote complete oxidative decomposition of the sample and minimise charring during ORO–AMS analysis. The carrier-gas flow of 90 mLmin-1 was selected based on previous optimisation of the ORO–AMS setup, where this setting provided stable gas transport, limited reflux or back-mixing, and ensured reproducible CO2 transfer to the trapping interface (Bolandini et al., 2025). A heating rate of 5 °Cmin-1 was used to provide sufficient thermal resolution while maintaining adequate CO2 yield per temperature interval for AMS analysis. The influence of ramp rate on thermogram shape and activation-energy estimates is explicitly treated in the kinetic framework used here (Hemingway et al., 2017). Nevertheless, because ramp rate, gas composition, flow rate, and system configuration can influence thermogram shape and apparent thermal metrics, comparisons among RPO/ORO studies should consider operational differences, as also noted for other ramped oxidation and thermal–radiocarbon approaches (Dasari and Widory, 2022; Garnett et al., 2023; Stoner et al., 2023).

Radiocarbon analyses were performed on predefined temperature windows (150–240, 240–300, 300–350, 350–400, 400–455, 455–510, and 510–600 °C), selected to provide a consistent temperature framework for comparison with the SoliTOC decomposition scheme. While this alignment facilitates cross-method interpretation, it does not imply direct equivalence of OM fractions, given the differing analytical conditions and reaction pathways involved. For consistency, we also calculated an ORO-based ROC/TOC ratio by integrating CO2 released between 400 and 600 °C relative to the total CO2 released below 600 °C. Temperatures above 600 °C were excluded from radiocarbon analysis because CO2 yields were insufficient to sustain a stable AMS ion current and because our focus was on the sub-600 °C domain that corresponds to the operational ROC threshold used in SoliTOC. However, small amounts of refractory OC may combust above 600 °C in some samples (see Results and Discussion for an assessment of methodological limitations). Samples were ramped to 900 °C, but thermograms are displayed only up to 800 °C because only minimal additional CO2 was released above this temperature.

Evolved CO2 was continuously monitored and recorded as thermograms (i.e., plots of CO2 yield as a function of temperature) providing real-time information on OM oxidation behaviour and decomposition kinetics. Gaussian and Savitzky–Golay smoothing were applied to minimise CO2 and temperature instrumental noise. Radiocarbon results were normalised to Ox-II reference material and corrected for machine and procedural blanks following standard ETH protocols (Synal et al., 2007) and are reported as fraction modern (F¹⁴C), following the guidelines detailed by Reimer et al. (2004, 2020). Typical analytical uncertainties ranged from ±0.003 to ±0.010 F¹⁴C depending on CO2 yield (Stuiver and Polach, 1977). For quality control, bulk-equivalent F¹⁴C was reconstructed as the CO2-weighted mean of F¹⁴C results for each ORO-AMS thermal window and compared to previously published bulk radiocarbon measurements (Supplementary Discussion Table S1, Fig. S1, Table S3 from Bröder et al., 2021).

Subsequent data analysis integrating CO2 thermograms with F¹⁴C results was performed using the open-source Python package “rampedpyrox” (Hemingway, 2016). Interpretation followed the mechanistic framework of Hemingway et al. (2017, 2019), which links decomposition profiles to underlying distributions of OM activation energy, E. These distributions, termed p(0,E), represent the resistance of OM to oxidative decomposition and thus provide an integrated measure of OM reactivity governed by molecular composition and stabilisation mechanisms. To further compare p(0,E) distributions across samples, three metrics were extracted: the mean activation energy, μE, the standard deviation of activation energy, σE, and the activation energy at which CO2 release reaches its peak, Emax.

Ts-Py-GCMS generates molecular fingerprints by thermally decomposing (in the absence of oxygen) non-volatile OM into volatile compounds, which are then separated by gas chromatography and identified via mass spectrometry based on molecular weight and fragmentation patterns (Derenne and Quéné, 2015; De Leeuw and Largeau, 1993; Lewis, 1993). Unlike more conventional flash pyrolysis methods, which generate a molecular fingerprint for one specific temperature (e.g., Kaal et al., 2009; Tolu et al., 2015), the Ts-Py-GCMS approach used here applies the same step-wise temperature windows as the ORO–AMS method. As for ORO-AMS, seven windows between 150 and 600 °C were analysed (150–240, 240–300, 300–350, 350–400, 400–455, 455–510, and 510–600 °C), together with an additional high-temperature window (600–850 °C) to capture the most thermally resistant components. No measurable carbon was detected below 150 °C. While the underlying processes differ (pyrolytic decomposition for Ts-Py-GCMS versus oxidative combustion for ORO–AMS and SoliTOC), the use of a common temperature framework provides a reference for comparison of trends across thermal windows, without implying direct equivalence of OM fractions.

Analyses were performed using an Agilent 7890A gas chromatograph (GC) coupled via a heated transfer line to a time-of-flight mass spectrometer (BenchTOF, Markes International). The GC was equipped with a Gerstel thermal desorption unit (TDU) pyrolysis system (Gerstel) connected to a cooled injection system (CIS), with a liquid nitrogen cryotrap for compound focusing (Gerstel). The pyrolysis unit operated in evolved gas analysis (EGA) mode with a heating rate of 1 °Cs-1 and a maximum temperature of 850 °C. The CIS was initially cooled to -150 °C to trap the volatiles, followed by a fast-heating ramp to 320 °C with a 1 min equilibration time, while the GC inlet temperature was maintained at 300 °C. Chromatographic separation was carried out using a DB5-ms column (30m×0.25mm×0.25µm). The GC oven programme consisted of a 5 min isothermal hold at 40 °C, followed by a ramp of 5 °Cmin-1 to 270 °C, then 10 °Cmin-1 to 320 °C with a final 10 min hold. Each run lasted a total of 66 min.

Mass spectra were acquired using the BenchTOF instrument operating with an ionisation energy of 70 eV. The transfer line was maintained at 310 °C, and the ion source at 300 °C. The instrument scanned over an m/z range of 50 to 700 with a time-of-flight resolution better than 7000 (full width at half maximum, FWHM) and mass accuracy within ±0.1 Da.

Mass spectrometry data were processed using the OpenChrom software (Wenig and Odermatt, 2010), with compound identification aided by matching against the NIST23 Mass Spectral Library. Spectral matches were only considered valid when their match factor exceeded a reliability threshold of 70 %, a commonly accepted cutoff for tentative identification (Bravo et al., 2017; Tolu et al., 2015). Compound identification and classification followed the approach described by Bravo et al. (2017) and Tolu et al. (2015), integrating match quality, literature-based retention-time patterns, and biomarker grouping to assign peaks to major OM compound classes. Because Ts-Py-GCMS involves thermal decomposition under oxygen-free conditions, charring and secondary pyrolysis reactions may occur, particularly at higher temperatures. Therefore, identified compounds are interpreted as operational pyrolysis products rather than direct molecular inventories of the original OM.

Compound classes identified in this study include branched/cyclic lipids (as markers of microbial origin or thermally altered OM) and n-alkyl lipids (straight-chain alkanes and alkenes derived from aliphatic biopolymers or thermally transformed OM). Lignin derivatives were used as biomarkers of vascular plant-derived OM (Kaal et al., 2016; Tolu et al., 2015). Pyrolysis products of carbohydrates and carbohydrate–cellulose derivatives (e.g., levoglucosan and furfural) were used to represent fresh biological inputs from plants or microbial exudates (Derenne and Quéné, 2015; Schnitzer and Monreal, 2011). Aromatic compounds, including phenols and polycyclic aromatic hydrocarbons (PAHs), derive from thermally stable precursor molecules that can form through microbial degradation, combustion, or advanced diagenesis and may also indicate mineral stabilisation of OM (Bravo et al., 2017; Kaal et al., 2009). Finally, N-containing compounds represent nitrogen-bearing molecules originating from microbial biomass, degraded proteinaceous material, or nitrogen-rich polymers such as chitin or peptidoglycans (Derenne and Quéné, 2015; Schnitzer and Monreal, 2011).

Across all four slumps, the DB, RU, and PF features exhibit relatively consistent TOC₄₀₀ and ROC contents, with TOC₄₀₀ generally between 1.1 % and 1.3 %, and ROC between 0.5 % and 0.8 % (Fig. 2). By contrast, the AL shows much greater variability across sites. In the smaller slumps (CB and SF), AL samples display TOC₄₀₀ values around 1.5 % and ROC near 1.2 %, slightly higher than in the other features of those RTS. In the larger slumps (FM2 and FM3), however, AL TOC₄₀₀ concentrations are markedly higher, ranging from 5 % to over 16 %, while ROC remains between 0.9 % and 1.4 %.

The two largest slumps, FM2 and FM3, were selected for thermally sliced radiocarbon and chemical-fingerprint analysis because their well-developed geomorphic features (distinct AL, PF, DB, and RU zones) provide the most representative and internally consistent record of thaw-slump evolution across the Peel Plateau. Normalised CO2 thermograms reveal distinct differences in thermal behaviour across geomorphic features and sites (Fig. 3a and c). Again, AL samples show the most pronounced contrasts. In both FM2 and FM3, CO2 release begins early (between 170 and 200 °C) and peaks sharply at relatively low temperatures. The FM2 AL sample exhibits a dominant peak near 370 °C with a shoulder plateauing at ∼400 °C, while FM3 AL2 and AL3 display bimodal structures: AL2 peaks at ∼300 and ∼450 °C, and AL3 at ∼300 °C with a smaller secondary peak near 450 °C. These patterns correspond to comparatively low activation energies (FM2 AL: μE=152 kJmol-1, Emax=159 kJmol-1; FM3 AL2/AL3: μE=152–156 kJmol-1, Emax=140–173 kJmol-1). Activation-energy distributions are broader in AL samples (μE≈21–24 kJmol-1) than in PF-derived units, which show similarly elevated but more consolidated distributions (σE≈17–24 kJmol-1).

In both slumps, PF samples (FM2 HO1, HO2, PL; FM3 HO) display broad, asymmetric peaks. CO2 release begins gradually near 250 °C, peaks between 370 and 400 °C, and often shows a secondary shoulder near 470 °C with extended high-temperature tails. These features align with consistently higher activation energies than in AL samples (μE≈162–171 kJmol-1; Emax≈155–161 kJmol-1). DB and RU samples follow similar trends to their corresponding PF layers. In FM3, both DB and RU thermograms closely resemble the HO profile and show similarly elevated activation-energy metrics (μE≈163–169 kJmol-1; Emax≈155–161 kJmol-1). The FM3 RU sample exhibits an earlier onset of CO2 release (∼230 °C), with peaks between ∼330 and 380 °C and a secondary shoulder near 450 °C. The FM2 RU sample departs slightly from this pattern, with a peak centred near 370 °C and a pronounced shoulder around 450 °C, accompanied by activation-energy values comparable to PF (μE≈167 kJmol-1; Emax≈157 kJmol-1).

SoliTOC- and ORO–AMS–derived ROC/TOC ratios show consistent patterns across sample types, with higher values in PF, DB, and RU (ROC/TOC≈0.25–0.45) and substantially lower values in AL (typically <0.15). However, ORO–AMS systematically yields higher absolute ROC/TOC values compared to SoliTOC (Fig. S2 in the Supplement). Method-comparison analyses show that ROC/TOC patterns are robust across SoliTOC and ORO–AMS, whereas TIC-related metrics are method-dependent and therefore not used for interpretation; full details are provided in the Supplement (Figs. S2–S4; Tables S2 and S3).

Radiocarbon age distributions across the thermal lability spectrum were analysed for a subset of samples from the two largest slumps, FM2 and FM3 (Fig. 3b and d; Figs. S5–S14 and Table S3 in the Supplement). The FM2 AL exhibits a modest decline in F¹⁴C values with increasing temperature from 0.725 ± 0.008 at 150–240 °C to 0.625 ± 0.007 at 510–600 °C. (Fig. 3b). A minor decrease also occurs at 300–350 °C (0.672 ± 0.008), followed by the highest F¹⁴C at 400–455 °C (0.729 ± 0.008) and a gradual decline at higher temperatures. By contrast, AL samples from FM3 (Fig. 3d) show consistently lower F¹⁴C values. FM3 AL2 remains relatively stable, ranging from 0.337 ± 0.005 at 150–240 °C to 0.327 ± 0.006 at 510–600 °C, while AL3 is even more 14C-depleted, ranging from 0.309 ± 0.006 to 0.267 ± 0.007, with no significant trend over the thermal range.

Across PF, DB, and RU, F¹⁴C generally decreases with increasing temperature (Fig. 3b and d). The trend is clearest in PF samples; e.g., in FM2 (Fig. 3b), HO1 F¹⁴C values decrease from 0.262 ± 0.005 to 0.102 ± 0.004, HO2 from 0.209 ± 0.005 to 0.123 ± 0.004, and the PL layer reaches 0.015–0.021 in the highest temperature windows, indicating near-radiocarbon-free material. Interestingly, the FM2 HO2 profile shows a partial reversal, with F¹⁴C reaching a minimum at 350–400 °C (0.042 ± 0.002) before rising to 0.123 ± 0.004 at 510–600 °C. FM3 HO (Fig. 3d) exhibits a similar depletion trend, decreasing from 0.106 ± 0.004 to 0.023 ± 0.002, with a signal near radiocarbon-dead for the highest temperature window.

In FM2, the RU sample exhibits a pronounced decline in F¹⁴C from 0.239 ± 0.006 to 0.097 ± 0.003 across the thermal windows, a pattern closely resembling the behaviour of the HO1 layer rather than the deeper PL PF. In FM3, both DB and RU decrease from initial values of ∼0.13 and ∼0.12 to 0.025 and 0.041, respectively, at 510–600 °C, which is almost identical to HO PF. Together, these patterns show that F¹⁴C generally decreases with increasing thermal resistance across most features, with the exception of the AL and a partial rebound observed in the HO2 profile. Finally, weighted-average bulk F¹⁴C reconstructed from all ORO–AMS thermal windows closely match independent EA–AMS bulk measurements (Fig. S1 in the Supplement; R2=0.99, RMSD≈0.03), demonstrating that the thermal-integration approach reproduces bulk F¹⁴C within analytical uncertainty.

Interpretable Ts-Py-GCMS chromatograms were obtained for (almost) all thermal windows of AL, HO2 and RU from FM2 and AL2, AL3, HO and DB from FM3 (Fig. 4), whereas other samples (FM2 HO1 and PL, FM3 RU) did not yield usable data due to low signal intensity or excessive noise in most thermal windows (Figs. S15 and S16 in the Supplement). Compound-class distributions generally reflected a progressive shift from labile, oxygen-rich OM at low temperatures to more compositionally altered, thermally stable material at higher temperatures across all thaw-slump components (peak lists for all analysed samples are provided in Table S4 in the Supplement). To facilitate interpretation across methods, molecular compound-class distributions are presented alongside F¹⁴C values derived from ORO–AMS for corresponding temperature intervals, providing a combined view of OM composition, thermal stability, and radiocarbon age.

In FM2 (Fig. 4a–c), AL, HO2, and RU samples display temperature-dependent shifts in the proportions of specific compound classes. Groups indicating fresh (and potentially bioavailable) OM such as carbohydrates, including cellulose-derived compounds, dominate the 240–300 °C window but are largely absent above 510 °C. Phenols occur across the full thermal range, with variable intensities among geomorphic features, whereas lignin derivatives are detected only in the AL sample. In contrast, proportions of n-alkyl lipids and aromatic hydrocarbons increase progressively with temperature, becoming most abundant between 510 and 600 °C. N-containing compounds appear mainly at mid to high temperatures (300–510 °C) and re-emerge in AL and RU in the highest temperature window (600–850 °C). More specifically, the AL sample contains abundant hemicellulose- and cellulose-derived carbohydrates within the 240–300 °C window (e.g., xylose, arabinose, glucose, furfural), followed by phenols 300–350 °C and a distinct shift toward aliphatic lipids and aromatic hydrocarbons between 400 and 510 °C. HO2 contains furfural and branched–cyclic lipids 300–350 °C, transitioning to n-alkyl lipids and aromatics in the 510–850 °C range. RU displays methylstyrene compounds at 240–300 °C, cellulose pyrolysis products between 300–350 °C, and increasingly aromatic profiles above 455 °C, with mainly condensed PAHs at 850 °C.

In FM3 (Fig. 4d–f), AL2 (Fig. 4d) and AL3 (Fig. S16) exhibited compound-class distributions broadly consistent with those observed for FM2 AL, with a clear progression from carbohydrates and phenols at lower temperatures to increasing contributions of aliphatic lipids and aromatic compounds at later thermal intervals. Between 350–400 °C, both AL2 and AL3 transitioned toward more thermally stable compounds, with rising proportions of phenols, n-alkyl lipids, and aromatics. Above 455 °C, PAHs and N-containing compounds became dominant, particularly in AL3. In contrast, the HO and DB samples (Fig. 4e and f) exhibited an earlier release of hydrocarbons and n-alkyl lipids already within the 150–240 °C window. Carbohydrates were largely absent from HO but appeared in DB above 350 °C, while phenols and N-containing compounds were mainly detected at higher temperatures. Chromatograms from the 455–510 °C interval in HO did not yield reliable compound matches. At lower temperatures (240–300 °C), cellulose pyrolysis products (e.g., furfural), methylstyrene compounds, and short- to mid-chain n-alkyl lipids occurred across AL2, AL3, HO, and DB. AL2 and AL3 also contained phenols consistent with lignocellulose decomposition. Both HO and DB yielded n-aldehydes and levoglucosan, while DB additionally released nonanal and decanal already at 240–300 °C. At the highest temperatures (600–850 °C), persistent compounds included long-chain n-alkyl lipids, condensed aromatics (e.g., naphthalene derivatives), and N-containing compounds, representing the residual products of OM decomposition.

Thermal behaviour, radiocarbon patterns, and molecular compositions together show that fundamentally different OM pools are present within the different geomorphic features of the four slumps (Figs. 2–4). Thermograms from FM2 and FM3 demonstrate that AL material begins oxidising at much lower temperatures than PF, DB and RU and exhibits a distinct structure, expressed either as a sharp early peak or as a bimodal profile with comparable low- and mid-temperature contributions (Fig. 3a and c). These patterns are consistent with lower mean activation energies in AL samples (μE≈152–156 kJmol-1) and, particularly in FM3, broader activation-energy distributions (σE≈21–24 kJmol-1), reflecting energetically heterogeneous and comparatively reactive OM pools. In contrast, PF, DB, and RU samples show later onsets of CO2 release, unimodal peaks centred near ∼370–400 °C and pronounced high-temperature tails characteristic of thermally stable carbon. These features coincide with higher μE values (≈162–171 kJmol-1) and consistently elevated but more uniform σE values (≈17–24 kJmol-1) across deeper units, indicating a dominance of energetically resistant OM with less variability in stabilisation mechanisms than observed in the AL.

Py-GCMS molecular fingerprints provide a compositional context for these contrasting thermogram shapes when evaluated within the common temperature framework (Fig. 4). In low-temperature windows (240–350 °C), AL samples are relatively enriched in carbohydrate- and cellulose-derived pyrolysates compared to PF, DB, and RU, consistent with the dominance of recently produced, oxygen-rich OM that oxidises early. At higher temperatures (>455 °C), all features show increasing contributions from aromatic hydrocarbons and long-chain n-alkyl lipids; however, these compounds dominate the high-temperature fractions of PF, DB, and RU, whereas AL retains a more mixed molecular signature. This enrichment of condensed and lipid-rich structures in PF-derived material aligns with their broad thermogram peaks and extended high-temperature tails. Similar temperature-dependent compositional shifts have been reported for permafrost OM elsewhere, where aromatic and lipid-rich components control high-temperature reactivity (Tolu et al., 2015; Zaccone et al., 2011). Together, these observations identify the primary molecular divide between biologically active surface horizons and deeper, cryogenically preserved permafrost-derived pools.

Radiocarbon profiles across thermal windows reinforce this interpretation (Fig. 3b and d). AL samples in FM2 maintain high F¹⁴C values (∼0.73–0.63), whereas the FM3 AL samples show lower values (∼0.33–0.27), reflecting that AL2 and AL3 were collected from deeper positions within the active layer than the surface AL sampled at FM2, leading to smaller contributions of recent vegetation and instead greater contributions of older, legacy OM. Although AL samples from both RTS do not display strictly invariant F¹⁴C values across thermal windows, they show only modest variation relative to PF, DB and RU samples, and several AL fractions as well as the FM2 HO2 sample even show local increases in F¹⁴C with increasing temperature, indicating that a small fraction of comparatively young OM persists into higher-temperature (higher-energy) windows. Such complexity suggests that much of the carbon oxidised across low- to mid-temperature windows derives from surface-influenced or recently cycled sources with overlapping activation-energy domains rather than a simple “young = low-T / old = high-T” structure. Also, F¹⁴C values are very different between AL samples from FM2 and FM3; this aligns with cryoturbated or compositionally heterogeneous soil horizons within this seasonally thawed layer, where young and older OM can co-occur within similar energetic ranges. In contrast to AL samples, PF, DB and RU display systematic and often steep declines in F¹⁴C with increasing temperature, consistent with sequential oxidation of progressively older and more refractory pools.

Activation energy-resolved F¹⁴C spectra further support these trends (Figs. S17 and S18 in the Supplement). AL material retains high to intermediate F¹⁴C across low-to-mid E, with only a minor old fraction emerging in the high-E tail. In contrast, PF, DB and RU are uniformly depleted across nearly the entire p(0,E) spectrum, with the strongest depletion at highest E values, indicating that the most oxidation-resistant fractions are also the oldest. FM3 AL horizons show large σE values (≈21–24 kJmol-1), reflecting substantial internal heterogeneity due to mixed plant inputs, cryoturbation, and variable degrees of protection. These energetic patterns highlight that OM stability arises from interactions between molecular composition, cryogenic preservation, and physical or mineral protection (Grant et al., 2019; Hemingway et al., 2017, 2019).

Similarly, μE vs. bulk F¹⁴C relationships reveal clear clustering (Figs. S17 and S18). AL samples from both slumps occupy a low μE, high F¹⁴C domain, whereas PF, DB and RU plot consistently at higher μE and lower F¹⁴C. As expected, these patterns mirror the thermogram shapes and activation-energy spectra: low μE values reflect lower kinetic barriers associated with labile or less-protected material, whereas high μE values reflect the stronger stabilisation of permafrost-derived pools. σE patterns follow the same structure: broader distributions in AL, narrower ones in PF-derived material. Exceptions – including the HO₂ high-E young fraction and several AL windows containing young OM at higher temperatures – likely reflect the presence of protected or mineral-associated OM within active-layer and upper-permafrost horizons.

Across both slumps, F¹⁴C decreases systematically with increasing μE (Fig. 5a). The corresponding μE–ROC/TOC relationship (Fig. S19 in the Supplement) shows that these energetic differences are reflected in operational thermal recalcitrance: AL consistently exhibits low ROC/TOC values at low μE, whereas PF and mobilised units DB and RU occupy a high-ROC/TOC, high-μE field. Importantly, these relationships describe feature-level end-member behaviour rather than within-feature structure. As shown above, individual AL samples can host young and old carbon across overlapping activation-energy ranges and thus do not follow a strict “young = low-T / old = high-T” rule. Instead, these broader μE–F¹⁴C and μE–ROC/TOC trends emerge when contrasting surface versus permafrost-derived pools at the scale of geomorphic units. Together, these trends define two internally coherent stability domains – “young and labile” versus “old and recalcitrant” – within which OM from FM2 and FM3 can be consistently interpreted. Having established these domains for the two fully characterised slumps, we now extend the comparison to all four RTS using the bulk metrics (ROC/TOC and bulk F¹⁴C) available across sites.

Across the Peel Plateau sites, ROC/TOC ratios (from SoliTOC analyses) and bulk F¹⁴C patterns reveal that slump morphology, vegetation, and active-layer thickness regulate only the surface OM pools, whereas deeper permafrost-derived material remains compositionally and thermally uniform (Fig. 5b). At the forested slumps FM2 and FM3, AL horizons show higher TOC₄₀₀ and distinct radiocarbon signatures that reflect greater biological inputs and deeper seasonal thaw. FM2 AL is the youngest and most labile, consistent with substantial modern vegetation input. FM3 AL is more heterogeneous because AL2/AL3 samples were collected at greater depths within this cryoturbated layer containing a mix of young and older OM. In contrast, CB and SF – both tundra sites with thin active layers – show lower TOC₄₀₀ but still young bulk F¹⁴C, indicating low OM input rather than rapid turnover. Despite this ecological variability at the surface, PF, DB, and RU units from all four RTS consistently exhibit higher ROC/TOC and lower bulk F¹⁴C, showing that the deep, cryogenically preserved, low-OC permafrost substrate mobilised by abrupt thaw and erosion is effectively invariant across the region.

These across-slump differences are most clearly expressed when bulk age and thermal partitioning are combined in ROC/TOC–F¹⁴C space (Fig. 5b). AL samples from all RTS plot within a “young, thermally labile” domain, but their position reflects ecological setting: FM2 AL is the youngest and most labile; FM3 AL is moderately depleted and intermediate in stability; and tundra AL (CB/SF) is more thermally resistant yet still young in radiocarbon age owing to limited biological input. In contrast, PF, DB and RU from every slump occupy a compact, old and recalcitrant cluster. This demonstrates that the mechanistic distinctions identified previously are not site-specific: the PF–DB–RU continuum forms a consistent, regionally coherent stability field regardless of vegetation, or slump size. This pattern aligns with prior studies showing that RTS in the Peel Plateau predominantly mobilise radiocarbon-depleted particulate OM from deeper permafrost with limited compositional alteration during initial transport (Bröder et al., 2021; Keskitalo et al., 2021; Shakil et al., 2020; Zolkos et al., 2019).

Overall, RTS-scale contrasts indicate that ecological setting and slump morphology primarily influence the composition and stability of active-layer OM, whereas deeper permafrost-derived pools exhibit consistent thermal and radiocarbon behaviour across all slumps. Importantly, all four RTS display uniformly low TOC in PF (HO and PL), DB and RU units, with no evidence for peat-rich or historically high-productivity ecosystems. This pattern aligns with regional mapping and previous work showing that the Peel Plateau is largely underlain by glacial/moraine-derived, ice-rich sediments rather than organic-rich deposits (Kokelj et al., 2017; Zolkos et al., 2018). The uniformity of PF-derived material across slumps therefore reflects mobilisation of a broadly homogeneous, low-OC, cryogenically preserved substrate, setting the stage for evaluating its regional carbon-cycle significance in the next section.

Distilling the feature-level and RTS-level patterns into a regional perspective shows that slumps on the Peel Plateau primarily mobilise old, thermally stable permafrost-derived carbon, reflecting both their geomorphic configuration and the moraine–till substrate underlying soils of the region. Across FM2, FM3, CB and SF, deeper PF – as well as mobilised components DB and RU – consistently share low bulk F¹⁴C, high ROC/TOC and high μE (Figs. 5 and S17–S19), indicating mobilisation and export of a broadly uniform pool of previously cryogenically preserved, relatively oxidation-resistant OM. These signatures match regional observations that Peel Plateau slumps export particulate OM largely originating from permafrost layers exposed at the headwalls rather than recently produced vegetation or active-layer material (Bröder et al., 2021; Keskitalo et al., 2021; Shakil et al., 2020; Zolkos et al., 2019).

The RU samples therefore represent an important transitional pool linking mobilisation of terrestrial material to riverine export and potential in-stream processing. Their similarity to PF and DB material indicates that runoff exports old, thermally stable particulate OM with limited alteration during early mobilisation. This material may be transported downstream and deposited in riverine, deltaic, or coastal sediments, acting as a transient or longer-term particulate carbon sink. However, river systems are not passive conduits: hydrodynamic sorting, oxygen exposure, changes in mineral association, and microbial processing may alter OM reactivity during transport. Under such variable conditions, parts of this thermally stable, aged carbon could still be transformed into dissolved or gaseous forms and contribute to a translocated, delayed greenhouse-gas release, as recently suggested for aged carbon leakage from the Mackenzie River system (Dasari et al., 2024).

The stability of the exported OM is further supported by incubation studies from the Peel Plateau and comparable moraine–till permafrost settings elsewhere in the Arctic. Laboratory and field incubations on mineral-rich permafrost soils from north-west Canada, Alaska, and other glaciated terrains demonstrate that respiration-resistant carbon pools are dominated by mineral-associated and physically protected fractions, while only a small labile component is rapidly decomposed following thaw (Estop-Aragonés et al., 2020; Littlefair et al., 2017; Schädel et al., 2014; Vaughn and Torn, 2019). These systems are characterised by relatively low TOC and strong mineral control on OM stabilisation, closely matching the geomorphic and substrate conditions of the Peel Plateau. The persistence of old, high- μE, high-ROC/TOC carbon in PF, DB and RU observed here mirrors these results, indicating that slump processes in this region primarily redistribute protected permafrost material downslope with minimal compositional alteration, rather than triggering substantial early-stage degradation. These findings contrast with observations made for so-called Yedoma permafrost. This late Pleistocene, syngenetic, silt-dominated, ice-rich permafrost contains some of the highest TOC and ground-ice contents in the Arctic, storing ∼327–466 Gt C globally and representing up to one-third of the deep-frozen carbon pool (Martens et al., 2023; Strauss et al., 2017, 2025). This substrate contrast has direct implications for carbon dynamics: erosion of Yedoma deposits can mobilise extremely carbon-rich and comparatively microbially labile OM, as shown by incubation and field studies reporting high respiration rates and rapid carbon losses following thaw (Knoblauch et al., 2013; Strauss et al., 2017; Vonk et al., 2013). In contrast, thaw slumps developed in moraine–till terrains mobilise permafrost carbon characterised by lower TOC concentrations and strong mineral association, resulting in compositionally uniform, physically protected OM that resists rapid oxidation. Substrate type therefore plays a central role in determining the fate of eroded permafrost OM and must be explicitly considered when extrapolating thaw-slump impacts to the pan-Arctic scale.