The Cape Bounty Arctic Watershed Observatory (CBAWO), located on the south-central coast of the Melville Island (Nunavut, Canada; Fig. 3a and b), was established in 2003 to examine Arctic ecosystem processes that would be impacted by climate warming and permafrost degradation (Lamoureux and Lafrenière, 2018; Treitz et al., 2024). Melville Island is characterized by a polar desert climate (mean annual air temperature: -14.8 ± 1.3 °C) with limited annual precipitation and runoff (< 150 mmyr-1; Beel et al., 2020, 2018; Lamoureux and Lafrenière, 2018) and a short thaw season between June and August (Lamoureux and Lafrenière, 2009). Bedrock in this region is composed of upper Devonian near-shore marine sandstone and siltstone units (Hodgson et al., 1984). Glacial and early Holocene marine sediments later overlaid the region resulting in a landscape characterized by incised low elevation plateaus and gentle hills (Hodgson et al., 1984). Permafrost is continuous and the thickness of the active layer is most often confined to the 0.75–1 m of surface material (Lamoureux and Lafrenière, 2009, 2018; Rudy et al., 2013). Exceptionally warm temperatures and clear skies during July 2007 combined with a major rainfall event (5 year return period), caused extensive thawing of ground ice and degradation of permafrost in Cape Bounty, in the form of hillslope thermokarst structures (Lamoureux and Lafrenière, 2009). This degradation manifested locally as RTSs and ALDs.

The two different types of hillslope thermokarsts were sampled the 7 and 10 August, 2018, i.e., a RTS (Fig. 3c) and an ALD (Fig. 3d), both for so-called “disturbed” and “undisturbed” locations. Samples were taken at 10 cm increments in the active layer until the top of the frozen layer was reached (where only one sample was taken). A location is defined as disturbed (D) when disturbance (ALD or RTS) is present and sampling was conducted within the headwall (Fig. 3e and f). The corresponding undisturbed (UD) site is located at 29 m (ALD) and 234 m (RTS) from the disturbance, in an area where the collapse had not (yet) occurred. The ALD sampled in this study is within the Ptarmigan sub-catchment and formed in 2007 (Lamoureux and Lafrenière, 2009). The catchment hosted other ALDs in 2007, collectively covering 12 % of the total catchment area (21.3 ha; Lafrenière and Lamoureux, 2013). Field observations showed that, while the ALD was not active in summer 2018 at the time of sampling, the RTS was still very active (up to 1 m annual retreat rate) and exposed fresh sediments from that summer.

We used X-ray diffraction (XRD) to characterize the crystalline mineral phases in the samples. The mineralogy of the bulk samples was determined on non-oriented powder finely ground in a mortar (CuKα, Bruker Advance D8 diffractometer, detection limit 5 % by weight) in the sample closest to the surface and in the deepest sample for the four modalities (n = 8).

We measured the total concentrations of Fe, Al, Mn, Ca and K in all samples (n = 33) using a portable X-ray fluorescence (XRF) device (Niton XL3t GOLDD + pXRF; ThermoFisher Scientific, Waltham, the United States). Measurements were carried out in laboratory conditions (ex-situ) on air-dried samples to prevent the introduction of additional variability (e.g. water content, sample heterogeneity). Briefly, the samples were placed on a circular plastic cap (2.5 cm in diameter), its base covered with a thin transparent film (prolene 4 µm). To ensure that detected intensities were not underestimated, the minimum sample thickness in the cap was set at 2 cm (Ravansari et al., 2020), and the total analysis time was set at 90 s to standardize each measurement. Concentrations measured by pXRF were calibrated using a method following Monhonval et al. (2021a). Linear regressions were used to correct pXRF concentrations for trueness on all samples (n = 33). These regressions were obtained from element concentrations measured by pXRF and by inductively coupled plasma optical emission spectrometry (ICP-OES) after alkaline fusion on samples from different permafrost environments (n = 172), including 8 samples from the present study. Those 8 samples were the top and bottom samples for each modality (ALD-D, ALD-UD, RTS-D, RTS-UD). The regressions had good coefficients of determination (robust R2 ≥ 0.9 for Fe, Ca, K; robust R2 ≥ 0.8 for Mn; robust R2 ≥ 0.6 for Al; Fig. A1). In the following, the total element concentration measured by XRF and corrected for trueness will be referred to as Fet, Alt, Mnt, Cat, Kt.

The water-soluble concentration in Ca and K were determined by water extraction on all samples (n = 33). This measurement is used to assess to what extent a soil or sediment has undergone leaching of water-extractable soluble elements. The water extraction protocol follows that in Thomas et al. (2023), consisting of placing 2 g of soil in 20 mL of ultrapure water (resistivity range = 10–18 MΩcm) for 2 h at room temperature and targets the elements released in the soluble phase (Ca, K). The element concentrations are measured in solution after filtration at 0.2 µm by ICP-OES. We also measured the pH (pH Electrode Inlab micro) and conductivity (probe Inlab 715-4 mm) on the water extract (Mettler Toledo SevenCompact DuoS213). In the following, elements extracted with ultrapure water will be referred to as the symbol of the corresponding element followed by a subscript letter “w” (Caw, Kw).

The total organic carbon (TOC) content was determined on finely ground sediments by substracting the total inorganic carbon from the total carbon on all samples (n = 33) and is reported as the mass of OC as a percentage of the total mass of dry soil (weight percent; %wt). The total carbon was measured after dry combustion via a Vario El Cube Elemental Analyser in accordance with the NF ISO 10694 standard (LOD = 0.1 %wt). The measurement of total inorganic carbon was carried out by measuring the release of carbon dioxide (CO2) following acid attack of 1 g of dry soil with hydrochloric acid, following the XP CEN/TS 16375 standard. Briefly, the CO2 released generates a pressure related to the inorganic carbon content and the relationship is established from a series of standards of 10, 20, 50, 100 and 200 mg CaCO3 (see also Baize, 2018). For the samples analyzed (n = 33), inorganic carbon averaged 0.04 ± 0.02 %wt. Inorganic carbon concentration is therefore not significant.

Two approaches of selective extraction from soil were used as indicators of complexed and poorly crystalline oxide phases (Rennert, 2019). Extraction of Fe, Al, Mn and Ca by sodium pyrophosphate targets organometallic complexes (Bascomb, 1968; Parfitt and Childs, 1988). We recognize a possible contribution from oxide nanoparticles in addition to organically bound metals (Courchesne and Turmel, 2008; Jeanroy and Guillet, 1981; Kaiser and Guggenberger, 2007), but limited by centrifugation as well as filtration of the extract. Extraction of Fe, Al and Mn by ammonium oxalate in the dark targets poorly crystalline oxides and organometallic complexes (Blakemore et al., 1981). The pool of mineral elements that form organometallic complexes or associations with OC is sometimes referred to as “reactive”. This reactive pool combines all poorly crystalline, amorphous and complexed forms of Fe, Mn, Al (and Ca), and corresponds here to extraction with ammonium oxalate (Fig. 2b). The selective extractions were carried out on all samples (n = 33). Concentrations in Fe, Al, Mn and Ca were measured in solution by ICP-OES after each extraction. In the following, elements extracted by the pyrophosphate and oxalate methods will be referred to as the symbol of the corresponding element followed by a subscript letter indicating the type of extraction, i.e. “p” for pyrophosphate extraction (Fep, Alp, Mnp, Cap) and “o” for oxalate extraction (Feo, Alo, Mno). The concentration of poorly crystalline oxides is calculated as the difference between the concentration in the oxalate extract and the pyrophosphate extract (Feo-Fep; Alo-Alp; Mno-Mnp).

The pool of OC selectively extracted with sodium pyrophosphate (Bascomb, 1968; Jeanroy and Guillet, 1981; Parfitt and Childs, 1988) was measured on the same solutions as those used for the selective extraction of metals (n = 33). Briefly, we measured dissolved OC released after dispersion by pyrophosphate using a Shimadzu TOC-L analyzer (measuring non-purgeable OC; LOD = 50 µgL-1). In the following, this carbon extracted by pyrophosphate will be referred to as Cp.

For oxalate extracted carbon, direct measurement is impractical as ammonium oxalate ((NH4)2C2O4) and oxalic acid (H2C2O4) are organic reagents. As a proxy, we measured the absorbance at 430 nm in the oxalate extract (via a Genesys 10 S VIS spectrophotometer, with the extractant solution as a blank) to evaluate the organic acid concentration. The optical density of the oxalate extract (ODOE) is mainly influenced by the extracted fulvic acids present in the oxalate extract (Daly, 1982).

The selective extractions presented here target organic carbon in the form of small fragments of biopolymers stabilized by chemical bonds. Some larger fragments of biopolymers – also stabilized by chemical bonds – will not be completely dissolved by selective extractions and will therefore not be included in this pool. These must therefore be quantified using density fractionation (see Sect. 2.8).

The samples followed a two-stage fractionation protocol, based on aggregate size and density (Fig. 4) adapted from and extensively described in Burgeon et al. (2021). This protocol was applied to the sample closest to the surface and to the deepest sample for the four modalities (n = 8) to separate free particulate organic matter from mineral-associated organic matter, representing organic matter occluded in aggregates and sharing chemical bonds with minerals. In the first stage, we performed a series of wet sieving (in a bath of 2 cm of distilled water over the mesh, we rotated and tilted the sieve for 50 rotations during 2 min) to separate the soil into 3 aggregate size classes: the coarse sand size fraction (2000–250 µm), the fine sand size fraction (250–50 µm) and the silt and clay size (S&C) fraction (< 50 µm). Particles larger than 2 mm were eliminated. The second stage used sodium polytungstate (SPT – 1.85 gcm-3) to differentiate light fractions non occluded in aggregates considered as particulate organic matter (< 1.85 gcm-3) from heavy fractions (> 1.85 gcm-3): aggregate of coarse sand size, aggregate of fine sand size and heavy fractions of free silt and clay size, aggregated or not. Briefly, we submerged 5 g of 60 °C oven dried fractions obtained from the first stage in 50 mL of SPT. These were then gently shaken head over head 10 times to soak the sample and remove air bubbles and placed in a vacuum chamber (140 kPa) for 10 min then left to rest under atmospheric pressure for 20 min. All samples were then centrifuged at 4700 g for 10 min and the supernatant, considered as the light fraction, was collected on glass fiber filters (performed thrice). The light fraction was thoroughly rinsed with distilled water and dried at 60 °C. The pellet was shaken into suspension using distilled water, then centrifuged. The distilled water was discarded in order to remove the SPT. This step was performed multiple times to recover heavy fraction dominated by mineral associated organic matter. On each fraction, the total carbon was measured by dry combustion via a Vario El Cube Elemental Analyser (LOD = 0.1 %wt). This carbon content is reported as the mass of OC as a percentage of the total mass of the corresponding fraction (% wt, fraction).

We performed computations for statistical analysis using R software version R.4.5.1 (R Core Team, 2025). Robust mixed-effect models presented in this study were fitted using the rlmer function from the robustlmm package v3.3-3 (Koller, 2016). The variable “profile” (RTS-D, RTS-UD, ALD-D, ALD-UD) has been included as a random factor, with random intercept and slope. We report the marginal R2 (Rm2), representing the population-level trend (variance explained by fixed effects) in the main text. The conditional R2 (Rc2), which includes both fixed and random effects and reflects total model performance, is additionally shown on the plots for reference. When numerical statistics are presented in the text for dataset descriptions, the mean ± standard deviation of the distribution is presented. For comparing two datasets, we performed nonparametric statistical Wilcoxon test.

The diffractograms of the samples characterized indicate the presence in all samples of primary silicate minerals (quartz, feldspars and micas) and secondary silicate minerals (kaolinite, illite, vermiculite, chlorite) (Figs. C1–C4) with similar diffractogram patterns in all samples.

Total concentrations in major elements (Ca, K, Fe, Al, Mn) do not vary significantly between the RTS and the ALD for the disturbed and undisturbed modalities (Fig. B1). Total iron concentration is 36 ± 4 gkg-1, total aluminum concentration is 71 ± 11 gkg-1 and total manganese concentration is 0.53 ± 0.09 gkg-1. The concentrations are homogeneous along the depths. However, total calcium (Cat; 2.3 ± 0.4 gkg-1) and potassium (Kt; 15 ± 2 gkg-1) concentrations increase slightly with depth for the sediments exposed by the retrogressive thaw slump (RTS-D) compared to the undisturbed (RTS-UD) location (Fig. B1d and e).

The proportion of highly soluble water extractable Ca and K shows a clear increase with depth for sediments exposed by the retrogressive thaw slump (RTS-D; Caw/Cat; Kw/Kt; Fig. 5a and b), likewise for electrical conductivity of the solution (Fig. 5c) and pH (Fig. 5d). In contrast, there is no trend with depth for active layer detachment samples (Fig. 5e–h).

Four cations were initially examined for the formation of organo–metallic complexes, i.e., extracted by pyrophosphate: iron, aluminum, manganese and calcium. Calcium extracted by pyrophosphate (Cap) was not further investigated, as it showed no correlation with pyrophosphate-extracted carbon (robust Rm2 = 0.01). We acknowledge that analytically, calcium extracted with pyrophosphate or other extractants reflects both calcium bound to OC and calcium released by dissolution of highly soluble mineral phases, and the pool of organically bound calcium cannot be isolated. Yet, calcium can sometimes play an important role in OC stabilization (e.g., Rowley et al., 2018), by forming cationic cation bridges or beyond, by associating with specific soil OC decomposition products (Rowley et al., 2025 and references therein). Among the three other metals involved in complexation, aluminum is dominant (Alp/(Fep+Alp+Mnp) = 55 ± 6 % when concentrations are expressed on a molar basis: mmolkg-1), followed by iron (Fep/(Fep+Alp+Mnp) = 40 ± 7 %) and manganese (Mnp/(Fep+Alp+Mnp) = 5 ± 3 %). We observe a distinct decrease in the concentration of metals forming complexes in the deepest samples of the profile exposed by the thaw slump (RTS-D; Fig. 6a), while the concentration of all metal complexes are homogeneous along the depth for the three other profiles (RTS-UD; ALD-D; ALD-UD; Fig. 6b–d). Aluminum, besides being the dominant metal in terms of molar concentration, is also the metal that best explains the distribution of complexed OC (mixed-effect model between Alp and Cp; robust Rm2 = 0.89; Fig. D1b), followed by iron (Fep; robust Rm2 = 0.58; Fig. D1a) and manganese (Mnp; robust Rm2 ∼ 0; Fig. D1c). The proportions of these metals in complex form in relation to total concentrations are low, with 0.3 ± 0.2 % for aluminum (Alp/Alt), 0.9 ± 0.6 % for iron (Fep/Fet) and 6 ± 4 % for manganese (Mnp/Mnt).

For oxalate extraction, the proportion extracted relative to total concentration is 5 to 20 times higher than for pyrophosphate (Figs. D2–D4), with 1.4 ± 0.2 % for aluminum (Alo/Alt), 10 ± 2 % for iron (Feo/Fet) and 63 ± 14 % for manganese (Mno/Mnt). As for the proportion of poorly crystalline oxides, it stands at 1.1 ± 0.2 % for aluminum (Alo-Alp)/Alt), 9 ± 2 % for iron (Feo-Fep)/Fet) and 56 ± 14 % for manganese (Mno-Mnp)/Mnt). Concentrations and proportion of poorly crystalline oxides are homogeneous along depth, except in the headwall exposed by the retrogressive thaw slump (RTS-D), where there is a clear increase with depth in the proportion of poorly crystalline iron oxides (Fig. D2a) and poorly crystalline aluminum oxides (Fig. D3a), relative to the total.

Total organic carbon concentrations are low along the profiles, i.e., at 1.1 ± 0.2 %wt. For the profile exposed by the thaw slump (RTS-D), we see an increase to 2 %wt at 20 cm depth, before dropping to 0.9 %wt for the deepest samples (Fig. 7a). The concentration of carbon forming complexes with metals is 0.14 ± 0.07 %wt on average (Cp; Fig. 7). Relative to the TOC concentration, this means that 13 ± 5 % of TOC is in the form of complexes with metals (Cp/TOC; Fig. D6), and trends along the depth follow those of metal complexes (see Fig. 6).

Assuming a maximal sorption capacity of 0.22 g_OC gFe-1 as ferrihydrite (Wagai and Mayer, 2007), we can calculate a concentration of up to 0.08 ± 0.02 %wt of OC bound to poorly crystalline iron oxides (Feo-Fep; Fig. 7). We acknowledge that the samples could potentially also contain Al-bearing amorphous silica (e.g., allophanes, imogolite; Rennert, 2019) extracted with oxalate, yet these cannot be quantified and singled out with confidence. Besides, the concentration in Al-bearing amorphous silica or oxides (Alo-Alp) is on average 4 times lower than those containing iron (Feo-Fep; Fig. D5), as are concentrations in poorly crystalline manganese oxides (Mno-Mnp; 12 times lower; Fig. D5). These Al- or Mn-bearing amorphous silica and poorly crystalline oxides have therefore not been considered for associations with OC. In the following, the calculation of OC bound to poorly crystalline iron oxides will be referred to as Camorph and refer to a maximal sorption capacity of OC to poorly crystalline iron oxides only. Relative to TOC, this means that up to 6 ± 2 % of the TOC is bound to poorly crystalline iron oxides in these profiles (Fig. D6). We estimate accordingly that the combination of the mechanisms of complexation and adsorption to poorly crystalline iron oxides results in 20 ± 4 % of the TOC pool being chemically bound with mineral surfaces or metallic cations (Cp+Camorph)/TOC).

The optical density of the oxalate extract (ODOE) averages 0.039 ± 0.012. This value correlates well with the concentration of OC forming complexes with metals (linear regression between ODOE and Cp; robust Rm2 = 0.84; Fig. D7a) and TOC (robust Rm2 = 0.59; Fig. D7e), but shows poor to no correlation with oxalate extracted metals (robust Rm2 = 0.05 with Feo; Fig. D7b; robust Rm2 = 0.23 with Alo; Fig. D7c and robust Rm2 = 0.28 with Mno; Fig. D7d) or with the proportions in poorly crystalline oxides (Fig. D7f–h). On the other hand, ODOE correlates very well with the sum Cp + Camorph (robust Rm2 = 0.92), supporting our approach to calculate Camorph.

Sediments exposed by the retrogressive thaw slump (RTS-D) exhibit a significant decrease in OC forming complexes with metals (Cp) along depth (Wilcoxon tests comparing the 5 surface samples with the 4 deepest samples; p-value < 0.05; see also Fig. 7a). In contrast, no clear trend is observed for the other three modalities (p-value > 0.3 for RTS-UD, ALD-D and ALD-UD; Fig. 7b–d). The OC associated with poorly crystalline Fe oxides (Camorph) in RTS-D sediments shows an opposite pattern, with concentrations increasing slightly with depth (p-value = 0.2), although Camorph concentrations are, on average, half of those for Cp (Fig. 7). When both mechanisms (Cp and Camorph) are combined, there is still a significant decrease in overall stabilized OC concentration with depth in RTS-D sediments (p-value < 0.05; Fig. 7a).

The mass fractions of 2000–250, 250–50 and < 50 µm aggregates are 15 ± 4 %, 37 ± 9 % and 48 ± 10 %, respectively, with no trend with depth or between RTS and ALD. The particulate organic matter resulting from density fractionation represents 1.1 ± 0.2 % on average of the total soil mass. It is divided into 0.24 ± 0.06 % for 2000–250 µm size, 0.4 ± 0.1 % for 250–50 µm size and 0.5 ± 0.1 % for < 50 µm size. The mineral-associated fraction, which represents 98.3 ± 0.2 % of the total mass, is divided into 15 ± 4 % for 2000–250 µm, 36 ± 8 % for 250–50 µm size and 47 ± 10 % for < 50 µm (Table E1). The OC concentration is much higher for the particulate organic matter fractions (33 ± 10 %_wt,fraction) than for the mineral associated organic matter fractions (0.7 ± 0.3 %_wt,fraction; Table E1). When we recalculate these OC contents taking into account the mass proportion of each fraction in the soil, we obtain that the free particulate organic matter fraction represents a content of 0.4 ± 0.1 %wt on average (Table E1; Fig. 8); divided in 0.06 ± 0.03 %wt, 0.12 ± 0.03 %wt and 0.20 ± 0,07 %wt for 2000–250, 250–50 and < 50 µm sizes, respectively (Fig. E1). The mineral associated organic matter fraction represents a content of 0.7 ± 0.1 %wt (Table E1; Fig. 8), divided in 0.08 ± 0.04 %wt, 0.20 ± 0.09 %wt and 0.4 ± 0.1 %wt for 2000–250, 250–50 and < 50 µm sizes, respectively (Fig. E1). The free particulate organic matter represents on average 36 ± 10 % of the bulk soil TOC content, and the mineral-associated organic matter fraction represents 64 ± 10 % of TOC. For sediment exposed by the RTS, the content of mineral-associated OC decreases with depth, i.e., from 0.84 %wt mineral associated organic C for the shallowest sample to 0.41 %wt at depth (Fig. 8a), as it is the case for the chemical extractions (Figs. 8a and 7a).

Our estimate is that 20 ± 4 % of the TOC (Cp+Camorph)/TOC) in the soil and sediments from Cape Bounty is chemically bound with mineral surfaces or metallic cations, which is in the low range of values reported elsewhere. Indeed, it compares with 43 ± 20 % from other studies on hillslope thermokarst or Yedoma sediments in Siberia and Canada (Monhonval et al., 2021a, 2022; Thomas et al., 2023, 2024), all of which report fairly low TOC content, i.e. on average 1.6 ± 1.5 %wt, which is in the same range as the samples from this study (1.1 ± 0.2 %wt; Sect. 3.3). This confirms that – for sites with comparable TOC contents – the level of stabilization through chemical interactions appears to be about half as high at Cape Bounty in comparison to other studies in the Arctic. This could be partially explained by the particular location of Cape Bounty, which is at the highest latitude (74°55^′ N) compared to the study sites referenced here above (66°45^′–67°34^′ N). In addition, Cape Bounty landscapes consists of Holocene syngenetic permafrost developed in marine sediments (England et al., 2009; Hodgson et al., 1984; Lajeunesse and Hanson, 2008; Paquette et al., 2023) compared to the Pleistocene age of sites referenced above. This means that Cape Bounty is roughly 800–900 km farther north than the comparative sites, with a polar desert climate (mean annual air temperature: -14.8 ± 1.3 °C and low precipitations: < 150mmyr-1; Beel et al., 2018; Lamoureux and Lafrenière, 2018). Such past permafrost history and climatic conditions (i.e., negative temperature and low precipitation) allow only limited pedological development and constitute no favorable conditions for forming mineral–OC bounds. This could partially explain why the level of OC chemical stabilization is generally low at Cape Bounty. Similar low proportions have also been reported in sediments from the Eurasian Arctic shelf, with mineral-interacting proportion of TOC of 0.5 %–22 % (Salvadó et al., 2015; TOC content = 1.4 ± 0.4 %wt) with samples collected between 70 and 80° N. García-Palacios et al. (2024) have shown conversely that higher concentrations of mineral-associated OC are found at lower temperatures, arguing a role of temperature limitation for persistence of the mineral-protected fraction of OC. This suggests that, once the mineral–OC interactions have been formed, cold can help to maintain them over time.

Permafrost peatland environments also show fairly low levels of mineral-interacting proportion of TOC, namely in 20 ± 10 % in lowland thermokarst degradations in Eight Mile Lake, Alaska (Monhonval et al., 2023) and 10 %–15 % in a palsa mire in Sweden (Patzner et al., 2020). In peatland ecosystems, however, the concentration of TOC is much higher, (e.g. 21 ± 19 %wt from Monhonval et al., 2023) and it can be assumed that the surfaces and metallic cations available for mineral–OC interactions are saturated and unable to establish further bonds (Thomas et al., 2024).

The results of aggregate and density fractionation at Cape Bounty, i.e., 64 ± 10 % of the TOC being both physically and chemically stabilized, show on the other hand proportions that are in the same range compared with the literature. By comparison, the same method applied to Siberian permafrost sediments showed a stabilization 73 ± 17 % of the TOC (Dutta et al., 2006; Gentsch et al., 2015; Martens et al., 2023). These proportions are extremely variable across the Arctic and show no latitudinal trend. Future work needs to address a systematic analysis of past and present environmental controls (e.g., soil and air temperature, precipitation, soil moisture) and edaphic controls (e.g., lithology, pH, pyrophosphate- and oxalate-extractable Fe and Al, silt and clay content) of the concentration and proportion of mineral-interacting OC. Total precipitation and effective soil moisture appear promising since they have been shown to be significant positive drivers of soil carbon accumulation in dry tundra (Klaminder et al., 2009) and mineral-interacting OC at the global scale (Kramer and Chadwick, 2018), as are silt and clay content (e.g., Georgiou et al., 2022).

The retrogressive thaw slump at Cape Bounty exposes sediments at depth whose OC is less protected by organo–mineral interactions (Figs. 7a and 8a). In particular, organometallic complexation, an efficient mechanism for OC protection (Kleber et al., 2015; Mikutta et al., 2006), are virtually non-existent in deep sediments exposed by the RTS (RTS-D). As it happens, deep sediments in the RTS-D are also those that are rich in soluble mineral elements (Fig. 4a–c; see also Lamhonwah et al., 2017) with higher pH (Fig. 4d). For the other profiles studied (i.e., RTS-UD, ALD-D and ALD-UD), however, there was no change with depth, either for stabilized carbon (Fig. 7b–d) or for soluble elements, conductivity and pH (Fig. 4). Yet all four profiles have a similar mineralogy (Figs. C1–C4) supporting a similar lithology of the parent material. We attribute the observed trends in the RTS-D to the fact that it probably exposes sediments that were previously perennially frozen at lower depths (as they are rich in solutes) and result from a syngenetic permafrost aggradation (French, 2007; Paquette et al., 2023), thereby limiting pedological development and the potential to form organo–mineral interactions. The retrogressive nature of a thaw slump guarantees exposure to this type of fresh sediment every summer until the slump is no longer active (Fig. 9a).

This is in agreement with the conclusion drawn at the Peel Plateau in Canada (Thomas et al., 2023) and in the Batagay thaw slump in Siberia (Thomas et al., 2024), where shallowest soils and sediments show significantly higher Cp concentrations than deep sediments, for which pedological development has been limited. Historical permafrost thaw dynamics therefore appear to be a first-order factor dictating the concentration of stabilized OC in a given sediment type. Data gathered in this study therefore confirm that thaw slumps, which are likely to develop more intensively in the future (Kokelj et al., 2015, 2021), will potentially further expose masses of deep OC that is less well protected by organo–mineral interactions. We can point out that this limited protection in deep horizons is driven by Cp and, at Cape Bounty, is partially offset by a higher concentration of Camorph (Sect. 3.3). As Camorph constitutes only a minor fraction of the TOC pool however, its offsetting effect is negligible (Figs. 7a and 8a).

We would argue that materials exposed by the ALD are sediments that have undergone multiple freeze-thaw cycles since the ALD development in 2007 up to the sampling in 2018. Therefore, the exposed materials in the ALD headwall have thawed for several summers in a row. In addition, in 2018, the active layer thickness at the time of the sampling was lower (i.e., ∼ 0.6 m on average) due to low summer temperatures (Beel et al., 2021) than average active layer thickness (i.e., 0.75–1.0 m) from previous years (Lamoureux and Lafrenière, 2018) (Fig. 9b and c).

Using a mass balance approach, we can derive a first order estimate of the OC stock exposed by the ALD and the RTS per square meter of exposed horizontal surface area and for the entire headwall. Assuming an excess ice content of 50 % for depths ≥ 60 cm (Lamhonwah et al., 2016) and dry bulk densities of 1.53 gcm-3 for depths of 0–7 cm, 1.55 gcm-3 for depths of 7–26 cm, 1.67 gcm-3 for depths of 26–46 cm, and 1.71 gcm-3 for depths greater than 46 cm (Stanton, 2023), it appears that the RTS exposes a greater stock of OC than the ALD (∼ 13.5 kgm-2 for RTS-D versus ∼ 9.8 kgm-2 for the ALD-D; Fig. D8). This besides supports evidence that the proportion of OC stabilized in sediments exposed by the RTS is lower than that of the ALD (∼ 17.2 % for the RTS-D versus ∼ 18.4 % for the ALD-D; Fig. D8). It is worth pointing out that OC stocks exposed at Cape Bounty are quite low compared to OC stocks exported by thaw slumps such as at the Peel Plateau (∼ 150 to ∼ 400 kgm-2; Thomas et al., 2023) or Batagay (∼ 460 kgm-2; Thomas et al., 2024), which is mainly due to the shallow headwall heights in Cape Bounty.

Measurements of the proportion of OC stabilized through mineral–OC interactions greatly differ between the selective extractions protocol (20 ± 4 %; Fig. 7) and the aggregate and density fractionation protocol (64 ± 10 %; Table E1; Fig. 8). One could therefore argue that chemical extractions (pyrophosphate and oxalate) greatly underestimate the proportion of OC that is stabilized in the sediments at the Cape Bounty. Yet, these two protocols do not target the same pools of OC. Chemical extractions target (i) organometallic complexes via pyrophosphate and (ii) organometallic complexes and OC bound through adsorption onto mineral surfaces such as poorly crystalline iron oxides via oxalate (Fig. 2). These two chemical extractions only include small fragments of biopolymers, which are fully dissolved during extraction. Large fragments of biopolymers can also be stabilized by complexation or adsorption, but will not be soluble and therefore not measured in the oxalate or pyrophosphate extract, due to centrifugation. Density fractionation, on the other hand, targets both (i) the same pool as chemical extractions, but also (ii) the OC occluded within the aggregates of various sizes as well as all organic molecules sharing chemical bounds with minerals (even the large fragments of biopolymers that are not completely dissolved by the extractants). The different pools of OC at Cape Bounty and the associated stabilization mechanisms can be summarized as follows: i.

up to 6 ± 2 % of the TOC is stabilized through adsorption of organic molecules onto poorly crystalline iron oxides (Camorph; Fig. 10). This is calculated based on a maximum sorption capacity of 0.22 g_OC gFe-1as ferrihydrite (Wagai and Mayer, 2007; Feo-Fep) and therefore reflects the difference between oxalate and pyrophosphate extractions (Fig. 2b). The mechanism involves the attachment – by ligand exchange – of organic molecules to metal atoms (Al, Fe) structurally present in the minerals (Fig. 2a). The chemical links are coordinate covalent bonds, which provide the longest residence times in soils (Cui et al., 2014; von Lützow et al., 2006; Mikutta et al., 2006), i.e., more than 100 years, or even millennia (Kleber et al., 2015) under stable physico-chemical conditions (pH, redox), as poorly-crystalline surfaces are among the most effective sorbents for dissolved OC (e.g., Kaiser et al., 1997).

Chemical stabilization mechanisms targeting small fragments of biopolymers therefore account for a relatively small proportion of the TOC (20 ± 4 %) but likely support residence times spanning hundreds to thousands of years with no drastic environmental changes (e.g., pH, redox, organic matter inputs, etc.). This proportion may be underestimated, as it does not take into account associations with calcium (Rowley et al., 2025). In contrast, stabilization of larger molecules via chemical bonds or physical protection within aggregates involve a comparatively larger fraction (45 ± 8 %) of TOC, though the residence times of aggregates could be expected to be much shorter, typically on the scale of decades.

The average figures presented above do not reflect the signature of the deep horizon exposed by the thaw slump (RTS-D). For this horizon, the OC bound to poorly crystalline oxides contributes twice as much as the average (12 % of TOC; Fig. 10), but this is offset by the virtual absence of OC forming complexes with metals (1 % of TOC; Fig. 10) and by the low contribution of OC occluded in aggregates or in stable large fragments of biopolymers (32 % of TOC; Fig. 10). Consequently, more than half of the TOC is in the form of free particulate organic carbon in this horizon, meaning that the thaw slump exposes potentially more labile OC (Fig. 10). We attribute this to the limited pedological development in this horizon (Sect. 4.2), yet the actual fate of this OC after material export from the thaw slump headwall remains uncertain. Thomas et al. (2023) showed that the pools of Cp + Camorph were well preserved in materials exported by thaw slumps in the Peel Plateau in Canada. This supports that, although limited, this mineral-interacting OC can be expected to be well protected over the long term if the physico-chemical conditions do not change. It should be noted, however, that pH or redox changes – e.g. caused by water saturation – would be likely to cause a release of mineral-bound OC (Monhonval et al., 2023; Patzner et al., 2020). Yet, hillslope thermokarst landforms are generally located on well-drained terrains leading to redox conditions considered, arguably, as oxic and relatively stable (Abbott and Jones, 2015). Since thaw slumps expose deep poorly weathered material at depth, a substantial decrease in pH, which would induce dissolution of Al-OC bonds, is unlikely at Cape Bounty, where pH levels are close to neutral (Fig. 5d and h). The fate of the sediments carried downstream from the headwall and into the debris tongues remains yet unclear, and further research needs to evaluate mineral–OC interactions along the sediment cascade of materials exported by thaw slumps.