Catchment lithology controls net carbon balance along the Antarctic Peninsula Authors: Songfan He, Xiaowen Zhang, Huanting Hu, Maojun Yan, Huiyuan Yang, Qiyue Gu, Thomas S. Bianchi, Xiaomei Xu, Xingqian Cui Published: 2026 ABSTRACT Climate-driven ecosystem change in Antarctica is reshaping carbon dynamics and influencing global climate. Under sustained warming, mobilization and reburial of organic carbon (OC) in coastal Antarctica may represent an important carbon sink, yet regional OC burial and oxidation remain poorly quantified. Here, we apply ramped pyrolysis/oxidation analysis and isotopic end-member mixing models to investigate sedimentary OC along two Western Antarctic Peninsula (WAP) fjords underlain by contrasting bedrock types. We show that the fjord underlain by sedimentary rocks exhibits ten-fold higher burial flux of petrogenic OC (OCpetro) than that dominated by volcanic rocks. We estimate burial fluxes of 6.1–8.8 Gg C yr−1 for biospheric OC (OCbio) and 0.3–3.1 Gg C yr−1 for OCpetro in WAP fjords, respectively, with regional OCbio burial likely exceeding OCpetro oxidation. Model simulations indicate that future ecosystem change may increase regional OC burial flux by 2–4 times, which is critical for predicting the long-term carbon-climate feedback. Introduction Antarctica has undergone protracted warming and extensive glacier retreat for decades1–4. Ice shelf thinning, ice front retreat, and grounding line discharge have profoundly reshaped regional geomorphology and hydrological dynamics5,6. Concurrently, biogeochemical cycles have been restructured by elevated glacial runoff to coastal regions7. Increased nutrient discharge by meltwater may boost coastal primary productivity and promote carbon dioxide (CO2) sequestration in the euphotic zone8–10. Meanwhile, remobilization and subsequent decomposition of subglacial carbon represents a potential carbon source driven by glacial retreat11. Elevated primary productivity and potential carbon efflux from the subglacial carbon reservoir highlight the dual role of Antarctica as either a substantial carbon sink or source under a changing climate10,11. Therefore, the responses of Antarctic carbon cycle to persistent warming, as reflected by the overall balance between carbon sequestration and decomposition, are critical for constraining future climate trajectories. Regional net carbon balance is directly governed by both inorganic and organic carbon (OC) cycling12. Among inorganic processes, silicate weathering is the largest carbon sink, while sulfide oxidation and coupled carbonate dissolution is an important carbon source13,14. Nonetheless, silicate weathering in Antarctica is restricted by relatively low temperature and the arid environment15. Direct CO2 emission following sulfide oxidation is likely limited by carbonate availability in Antarctica16, although the resulting proton equivalents can regulate the ocean alkalinity and carbon emissions over intermediate timescales (104–105 yrs)14. Compared to the inorganic carbon cycling, the balance between biospheric OC (OCbio) burial and petrogenic OC (OCpetro) oxidation (i.e., OC cycling) predominantly controls regional carbon balance in Antarctica. Glacial excavation sustains the oxidation of old carbon (e.g., bedrock-derived OCpetro) that acts as a potential carbon source17. Meanwhile, glacial retreat will expand ice-free areas, enhance hydrological connectivity, and create more habitable niches across terrestrial and aquatic ecosystems7,9,18. These ecosystem changes will promote atmospheric CO2 drawdown by increasing primary productivity and enhancing coastal erosion and reburial of terrestrially-derived OCbio7,18–20. However, quantitative constraints on the two counteracting processes, and thus the net carbon balance in Antarctica, remain limited, due to extensive ice cover21 and the lack of systematic studies. Given their disproportionate role in OC burial22–25, estuaries are important locations to investigate OC cycling. Serving as glacially carved estuaries, fjords are pervasively distributed along the Antarctic Peninsula26,27. Sedimentary OC buried in these fjords is composed of biospheric and petrogenic provenances, with the former further divided into marine (OCmar) and terrestrial (OCterr). Recent studies identify Antarctic fjords and other coastal regions as centers of glacial retreat, ecosystem changes, and phytoplankton accumulation3,20,28,29. They are also expected to be emerging hotspots for OCbio burial and OCpetro oxidation under future climate scenarios11,18,19,24. Therefore, a mechanistic understanding of OC cycling is needed to better constrain regional carbon fluxes. Sedimentary and volcanic rocks, as primary sources of sediment along the land-ocean continuum, may impact the carbon cycle in different ways30,31. Particularly, the weathering of sedimentary rocks represents a potential carbon source, due to relatively high rates of OCpetro oxidation and sulfide oxidation31,32. In comparison, CO2 release during volcanic rock weathering is limited by insufficient OCpetro and sulfides31. Although lithological controls on the net carbon balance have been revealed by global datasets30,33, their quantitative effects are often obscured by other factors (e.g., climate) at regional scales. Notably, catchments in the Antarctic Peninsula are characterized by similar environmental settings but different bedrock lithologies34, making them ideal for uncovering lithological controls on OC cycling19. Here, we systematically examine the composition of sedimentary OC (OCmar, OCterr, and OCpetro) in two Antarctic fjords (Flandres Bay and Lapeyrère Bay) underlain by contrasting bedrock types (Fig. 1). Flandres Bay, located on the Antarctic Peninsula, is denoted as AP; Lapeyrère Bay is abbreviated as AI, given its geographic location on the adjacent Anvers Island. The catchment of Flandres Bay (AP) is primarily underlain by Paleozoic to Mesozoic sedimentary rocks, while the catchment bedrock in Lapeyrère Bay (AI) is predominantly composed of Cenozoic volcanic rocks34 (Fig. 1). Twelve core-top sediments (six from each fjord) were collected to constitute complete land-to-ocean transects. Bulk OC content, dual carbon isotopes (δ13C and F14C), ramped pyrolysis/oxidation (RPO), and mineral surface area (MSA) were measured for all samples, while four representative samples were selected for the RPO-13C/14C analysis (Methods). By compiling regional datasets, we further assessed the burial flux of OCbio, the burial and oxidation fluxes of OCpetro, and the overall net carbon balance in the Antarctic Peninsula. By comparing with global compilations, we proposed and simulated a two-stage variation of OC burial flux in WAP fjords under future climate change. Our study highlights the unexplored carbon burial potential in western Antarctic Peninsula (WAP) fjords, which is projected to increase with future climate change.Fig. 1Maps of the Western Antarctic Peninsula indicating sampling sites.A Satellite view of the Western Antarctic Peninsula, with zoomed-in regions (orange rectangles) and lithology of rock outcrops annotated53. B Spatial distributions of sampling sites in Lapeyrère Bay (AI). C Spatial distributions of sampling sites in Flandres Bay (AP). Orange dots in (B, C) are sampling sites. Maps were created with QGIS (version 3.40.10). Maps of the Western Antarctic Peninsula indicating sampling sites. A Satellite view of the Western Antarctic Peninsula, with zoomed-in regions (orange rectangles) and lithology of rock outcrops annotated53. B Spatial distributions of sampling sites in Lapeyrère Bay (AI). C Spatial distributions of sampling sites in Flandres Bay (AP). Orange dots in (B, C) are sampling sites. Maps were created with QGIS (version 3.40.10). Results and discussion Heterogeneous sedimentary OC distributions controlled by catchment lithology The spatial distributions of sedimentary OC content (i.e., OC%) and δ13C of OC are broadly identical along two fjords with an apparent land-to-ocean gradient. The significantly lower mineral surface area (MSA) in the inner fjords (AP-1 to AP-3 and AI-1 to AI-3) compared to the outer basins (p = 0.01) indicates a dominant contribution of coarse particles in areas proximal to glacier fronts (Fig. 2A). This is further supported by the extremely low inner-fjord OC% (on average, 0.18% and 0.10%) that indicates reburial of organic-poor debris from glacial erosion35,36. In comparison, consistent increases of OC% and MSA toward outer fjords suggest the association of OC with the fine-grained particles during lateral transport (Fig. 2A). A broadly negative shift in δ¹³Cₒᵣg is observed along both transects, with the most depleted value (−24.7‰) occurring in the outer basin (Fig. 2A), which is in marked contrast to the inner-to-outer gradients commonly reported for other fjords37,38. Nonetheless, this negative excursion in δ¹³C is consistent with a recent study from an Arctic fjord39, suggesting that depleted δ¹³C values of marine OC may represent a distinct and widespread signature in polar regions.Fig. 2Geochemical and thermochemical properties of sedimentary organic carbon (OC) along Flandres Bay (AP) and Lapeyrère Bay (AI).A Distributions of total organic carbon content (OC%), stable carbon isotope of organic carbon (δ13C) relative to Vienna Pee Dee Belemnite (VPDB), mineral surface area (MSA), and fraction modern carbon (F14C) values along two fjords. Purple dots and bars represent samples along Flandres Bay (AP), while orange dots and bars are samples along Lapeyrère Bay (AI). Gray dots and dashed lines represent lateral gradients of δ13C values. “Inner” and “Outer” represent the inner (close to the glacial front) to the outer fjord transect. B Linear regression plot of OC%×F14C against OC% indicates distinct petrogenic OC% in the two fjords. C Activation energy (E) distributions of sedimentary OC along AP and AI. Purple colored dashed and solid lines in (B) represent linear fitting curves of all AP samples and after removing one outlier (AP-6), respectively. Gray dashed lines in (C) represent the determined threshold of 215 kJ mol−1. Geochemical and thermochemical properties of sedimentary organic carbon (OC) along Flandres Bay (AP) and Lapeyrère Bay (AI). A Distributions of total organic carbon content (OC%), stable carbon isotope of organic carbon (δ13C) relative to Vienna Pee Dee Belemnite (VPDB), mineral surface area (MSA), and fraction modern carbon (F14C) values along two fjords. Purple dots and bars represent samples along Flandres Bay (AP), while orange dots and bars are samples along Lapeyrère Bay (AI). Gray dots and dashed lines represent lateral gradients of δ13C values. “Inner” and “Outer” represent the inner (close to the glacial front) to the outer fjord transect. B Linear regression plot of OC%×F14C against OC% indicates distinct petrogenic OC% in the two fjords. C Activation energy (E) distributions of sedimentary OC along AP and AI. Purple colored dashed and solid lines in (B) represent linear fitting curves of all AP samples and after removing one outlier (AP-6), respectively. Gray dashed lines in (C) represent the determined threshold of 215 kJ mol−1. Despite similar distributions of sedimentary OC% and δ13C, bulk-OC 14C activity (expressed in F14C values; fraction modern carbon) exhibits contrasting spatial trends along the two fjords (Fig. 2A), which is likely associated with distinct inputs of OCpetro (F14C = 0). The F14C values of AP samples range from 0.29 to 0.71 and are significantly lower than those of AI (Table S1; p < 0.01), potentially indicating higher OCpetro inputs along AP. A linear regression approach was applied to approximate fjord-scale petrogenic OC% (Methods), while acknowledging strict prerequisites for its valid application35,40. The average petrogenic OC% along AP and AI is estimated to be 0.087% and 0.008% (absolute weight), respectively (Fig. 2B). At the head of AP, where sedimentary OC% ranges from 0.12 to 0.24% (Table S1), OCpetro on average accounts for approximately 50% of sedimentary OC, suggestive of predominant inputs of bedrock-derived carbon from glacial erosion. In comparison, the average F14C values of OCbio are almost identical between AP (0.80) and AI (0.84), as demonstrated by the two parallel lines in Fig. 2B. RPO technique was further used to confirm OCpetro contribution within samples (Methods). This technique sequentially decomposes sedimentary OC within a gradually ramping process and quantitatively distinguishes OC with different bond strength by thermal activation energy (E)41–43. Compared to OCbio, OCpetro is typically more resistant to decomposition and is characterized by stronger bond strength32,42,43 (as expressed by higher E value). The RPO results indicate a bimodal activation energy distribution along AP (Figs. 2C and S1), where the prevailing high-E peaks in AP-1 to AP-3 are within the window of OCpetro32,44, demonstrating the latter as an important component of sedimentary OC along AP. The intensity of high-E peak abruptly diminishes toward outer part of AP, which coincides with the transition of the major OC provenance from glacially eroded debris to relatively labile OCbio. In comparison, all E distributions of AI samples are highly overlapped within comparably narrow E ranges (Fig. 2C), indicative of homogeneous OC properties devoid of petrogenic inputs. CO2 fractions of different temperature intervals during the RPO analysis (i.e., RPO fractions) were collected for representative samples and measured for δ13C and F14C values (“Methods”; Table S3). Fraction-specific F14C values exhibit a descending trend as E increases (Fig. 3), which is consistent with previous studies43,45. Interestingly, the fractional F14C values of high-E fractions are significantly lower in AP than those of AI (p < 0.01), with the lowest value (0.0132 ± 0.0007) being diagnostic of the petrogenic provenance (Fig. 3 and Table S3). By comparing E distributions of AP and AI samples, we used fE ≥ 215 (i.e., proportion of OC with E ≥ 215 kJ mol−1) to approximate the proportion of OCpetro that has not been chemically altered (Supplementary Information). This E threshold basically excludes OCbio and weathered OCpetro that are typically characterized by relatively low E values32. The average fE ≥ 215 value of the inner AP (0.48 ± 0.03) is about an order of magnitude higher than that of the inner AI (0.05 ± 0.01), which is in great consensus with F14C results and corroborates a significantly greater export of OCpetro to AP (Table S2).Fig. 3Fraction-specific activation energy (E) distributions and isotopic compositions of representative samples along two fjords using ramped pyrolysis/oxidation-carbon isotope (RPO-13C/14C) analysis.A Stable carbon isotope (δ13C) and fraction modern carbon (F14C) values of ramped pyrolysis/oxidation (RPO) fractions from inner fjord (AP-2) to outer fjord (AP-5). B δ13C and F14C values of RPO fractions from AI-2 and AI-5. Open circles are δ13C values, while filled circles are F14C values. The x values of circles are fractional μE values (colored dots) with fractional σE (horizontal deviations) labeled. Gray dashed lines are the E thresholds of 150 and 215 kJ mol−1, respectively. Gray regions are E distributions of corresponding RPO fractions, and solid curves are overall E distributions obtained by summing up all fractions. The vertical deviations of isotopic values are smaller than the data points. RI: relative intensity; AP-2 and AP-5: individual samples from Flandres Bay (AP); AI-2 and AI-5: individual samples from Lapeyrère Bay (AI); μE: the mean value of E; σE: the standard deviation of E; VPDB: Vienna Pee Dee Belemnite. Fraction-specific activation energy (E) distributions and isotopic compositions of representative samples along two fjords using ramped pyrolysis/oxidation-carbon isotope (RPO-13C/14C) analysis. A Stable carbon isotope (δ13C) and fraction modern carbon (F14C) values of ramped pyrolysis/oxidation (RPO) fractions from inner fjord (AP-2) to outer fjord (AP-5). B δ13C and F14C values of RPO fractions from AI-2 and AI-5. Open circles are δ13C values, while filled circles are F14C values. The x values of circles are fractional μE values (colored dots) with fractional σE (horizontal deviations) labeled. Gray dashed lines are the E thresholds of 150 and 215 kJ mol−1, respectively. Gray regions are E distributions of corresponding RPO fractions, and solid curves are overall E distributions obtained by summing up all fractions. The vertical deviations of isotopic values are smaller than the data points. RI: relative intensity; AP-2 and AP-5: individual samples from Flandres Bay (AP); AI-2 and AI-5: individual samples from Lapeyrère Bay (AI); μE: the mean value of E; σE: the standard deviation of E; VPDB: Vienna Pee Dee Belemnite. Generally, the yield of OCpetro in a catchment is regulated by multiple factors, including latitude (climate), denudation rate, catchment area, and bedrock lithology. Latitudinal controls on OCpetro erosion are likely similar for the two fjords, as they are situated at the same latitude (64–65°S; Table S1) with comparable climatic settings46. Previous studies demonstrate that the intensity of physical erosion is a dominant control on OCpetro yield47. However, it cannot explain the tenfold discrepancy given the generally low erosion rates (<0.1 mm yr−1)48,49 and comparable sedimentation rates50 in the two fjords. Enhanced sediment delivery is also unlikely to be the main cause, as the catchment areas of the two fjords are broadly of the same order of magnitude27. Furthermore, elevated sedimentary delivery does not necessarily result in higher petrogenic OC%, owing to the dilution effect of inorganic material35. Other potential fossil OC sources that lead to depleted F14C signatures, such as black carbon, aerosol deposition, and glacial meltwater, are quantitatively limited51,52 and cannot fully explain the spatial heterogeneity. Therefore, we propose that the delivery of OCpetro to the sedimentary basin is primarily dictated by the intrinsic OC% of bedrock. The catchment of AP is predominantly underlain by Paleozoic to Mesozoic sedimentary rocks, whereas AI is dominated by Cenozoic volcanic rocks34,53. Contrasting OCpetro yields in the two fjords are consistent with recent studies that reveal relatively higher contents of OCpetro in sedimentary rocks30,54. Erosion of sedimentary rock redistributes rock-derived OCpetro to the broader sedimentary basin47,55. In contrast, erosion of volcanic rock primarily results in the inorganic dilution of sedimentary OC in the basin, with minimal OCpetro inputs31. This lithological control on OCpetro contents in sediments is observed globally. Catchments dominated by sedimentary rocks are characteristic of significant and variable amounts of OCpetro transported downstream to fjords, exemplified by those in southeast Alaska35, Svalbard56, and South Georgia57. On the contrary, limited OCpetro burial is found in fjords adjacent to volcanic bedrock58–60. Overall, catchment lithology, which determines the richness of rock-bound OC, exerts a first-order control on the budget of OCpetro exported to sedimentary basins. Quantitative estimate of net carbon balance in the Antarctic Peninsula The contrasting OC provenances along the two fjords highlight the need for quantitative estimates of OCbio (OCmar and OCterr) and OCpetro to accurately assess the net carbon balance47. Carbon isotopic compositions of various OC sources are constrained using RPO fractions for source apportionment. By integrating fractional isotopic signatures, δ13C values of OCmar and OCpetro are clearly revealed as −25‰ to −23‰ and about −21‰, respectively (Fig. S2). Fraction-specific F14C values span a broad range with the peak at ~0.85 (Fig. 3; Table S3), systematically lower than those of vegetated fjords in southeast Alaska and New Zealand43. This peak is consistent with the “reservoir age effect” in the WAP of around 1300 yr BP (F14C ≈ 0.85)41, as a result of surface sea ice barriers and upwelling of deep seawater. Its prevalence along both fjords suggests that this moderately aged signature is well integrated into modern OCmar. Notably, beyond the reservoir age effect, more pronounced pre-aged F14C signatures (F14C ≈ 0.72–0.75) are prevalent along the two fjords (Figs. 3 and S2). The relatively 14C-depleted signatures may result from the inputs of pre-aged OCterr or from the mixing of OCmar and OCpetro. However, the theoretical fractional F14C value from simple mixing of OCmar and OCpetro is statistically distinct from the measured value (Supplementary Information), which indicates that the pre-aged F14C results from a distinct OC end-member (i.e., OCterr). In the WAP, poorly developed soils and moss peats are likely dominant sources of pre-aged OCterr61–63. Indeed, previous work reported an average δ13C value of −25.1‰ and an F14C of ~0.71 for peat ecosystems62, which are statistically identical to RPO fractions in this study, further supporting moss peats as the primary source of OCterr in this region. Given isotopic constraints on end-members above, a three-end-member mixing model (OCmar, OCterr, and OCpetro) was employed to accurately quantify the relative contributions of different OC sources (Fig. 4A). Using Monte Carlo resampling (Methods), the regional average fractions of OCmar, OCterr, and OCpetro were estimated as 50 ± 21, 16 ± 15, and 33 ± 13% for AP, and 69 ± 13, 28 ± 14, and 3 ± 3% for AI, respectively (Fig. 4C; Table S4), supporting distinct contributions of OCpetro in the two fjords (t-test; p < 0.001). A complementary end-member mixing model, based on F14C and fE ≥ 215 values (Figs. 4B and S3), further confirms the robustness of the three-end-member mixing model (Supplementary Information). Meanwhile, proportions of end-members were also resolved for all RPO fractions, which reveal general gradients of increasing OCpetro contributions toward high-E space (Fig. 4D). A minor proportion of OCpetro with relatively low E may also reflect input from chemically weathered bedrock32. On average, continentally-derived OC (i.e., OCpetro and OCterr) accounts for one-third to a half of sedimentary OC reservoirs, indicating considerable OC mobilization via glacial erosion.Fig. 4Three-end-member mixing model used to constrain proportions of marine, terrestrial, and petrogenic organic carbon (OC) along Flandres Bay (AP) and Lapeyrère Bay (AI).A Three-end-member mixing model using fraction modern carbon (F14C) and stable carbon isotope (δ13C) values. B Three-end-member mixing model using F14C and fE≥215 (the proportion of OC with activation energy ≥215 kJ mol−1) values. C Proportions of different OC provenances for bulk samples along two fjords. D Proportions of different OC provenances for ramped pyrolysis/oxidation (RPO) fractions. Circles and smaller diamonds (with standard deviations labeled) in (A) represent bulk samples and RPO fractions, respectively. The fE ≥ 215 values in (B) are calculated from activation energy (E) distributions of samples and RPO fractions. Dashed lines in (A, B) represent theoretical end-member mixing lines. End-member proportions shown in (C, D) are model results of (A) and are labeled with uncertainties estimated by Monte Carlo resampling. “Inner” and “Outer” in (C) represent the gradient from the inner fjord to the outer fjord. “F1” and “F4” in (D) represent the gradient from the first RPO fraction to the fourth (last) RPO fraction of individual samples (AP-2, AP-5, AI-2, and AI-5). VPDB: Vienna Pee Dee Belemnite. Three-end-member mixing model used to constrain proportions of marine, terrestrial, and petrogenic organic carbon (OC) along Flandres Bay (AP) and Lapeyrère Bay (AI). A Three-end-member mixing model using fraction modern carbon (F14C) and stable carbon isotope (δ13C) values. B Three-end-member mixing model using F14C and fE≥215 (the proportion of OC with activation energy ≥215 kJ mol−1) values. C Proportions of different OC provenances for bulk samples along two fjords. D Proportions of different OC provenances for ramped pyrolysis/oxidation (RPO) fractions. Circles and smaller diamonds (with standard deviations labeled) in (A) represent bulk samples and RPO fractions, respectively. The fE ≥ 215 values in (B) are calculated from activation energy (E) distributions of samples and RPO fractions. Dashed lines in (A, B) represent theoretical end-member mixing lines. End-member proportions shown in (C, D) are model results of (A) and are labeled with uncertainties estimated by Monte Carlo resampling. “Inner” and “Outer” in (C) represent the gradient from the inner fjord to the outer fjord. “F1” and “F4” in (D) represent the gradient from the first RPO fraction to the fourth (last) RPO fraction of individual samples (AP-2, AP-5, AI-2, and AI-5). VPDB: Vienna Pee Dee Belemnite. End-member model results enable a further assessment of regional OC fluxes. By compiling sedimentation rates and basin areas of major WAP fjords, regional burial flux of sedimentary OC is estimated to be 9.2 ± 2.0 Gg C yr−1 (Methods). The estimated OC burial flux falls at the lower end of the range reported for global fjord systems35,64, indicating WAP fjords currently play a secondary role in sedimentary OC sequestration compared to their mid-latitude and subpolar counterparts. Regarding regional OCpetro burial flux, the two fjords in this study likely provide the upper and lower bounds for the estimate, representing end-member scenarios of catchments predominantly underlain by sedimentary rocks versus volcanic rocks, respectively. Accordingly, OCpetro flux along WAP fjords is estimated to be 0.3–3.1 Gg C yr−1 using the estimated OCpetro proportion range of 3–33% (“Methods”; Table S4). After isolating OCpetro, the burial flux of OCbio is 6.1–8.8 Gg C yr−1 (Methods). Overall, our results indicate that OCbio dominates OC burial in WAP fjords, while OCpetro burial flux is highly variable, modulated by the relative contributions of sedimentary and volcanic rocks. Beyond total OC burial flux, the regional net carbon balance and its long-term change are governed by two key factors: the burial of OCbio and the ultimate fate (i.e., preservation or degradation) of OCpetro during erosion-transport processes. Accordingly, we assessed the current net carbon balance in the Antarctic Peninsula by two independent approaches (Methods). OCpetro oxidation occurs primarily during fluvial transport, in subaerial regions, and through subglacial grinding. The variation in OCpetro concentrations during lateral transport is assessed using OC% and fE ≥ 215 (Supplementary Information). Nearly constant OC% × fE ≥ 215 values along AP suggest minimal degradation of OCpetro in fjords (Fig. 5A), which is likely due to its high maturity and refractory nature. By applying a forward approach with a regional dataset54, the subaerial oxidation flux of OCpetro in the Antarctic Peninsula is estimated to be 4.0–7.8 Gg C yr−1 (Methods). However, the oxidation flux is likely overestimated, given that it integrates the entire ice-free regions (Methods), and relies on global oxidation rates that are generally higher than those in Antarctica30,54. Therefore, OCpetro oxidation flux of the entire Antarctic Peninsula (including catchments and subaqueous fjords) is almost completely offset by OCbio burial (6.1–8.8 Gg C yr−1; Methods) in WAP fjords alone, suggesting net carbon sequestration in the Antarctic Peninsula at present.Fig. 5Organo-mineral constraints on the ultimate fate of petrogenic organic carbon (OC) and the future OC burial flux in Western Antarctic Peninsula (WAP) fjords.A Distribution of OC% × fE≥215 (total organic carbon content × the proportion of OC with activation energy ≥ 215 kJ mol−1) of samples along Flandres Bay (AP-1 to AP-6). B Correlation between fraction modern carbon (F14C) and mineral surface area (MSA) values along two fjords. C OC/MSA (total organic carbon content/mineral surface area) of WAP fjords and from global compilations43. WAP in this study (purple box), deglaciating (gray box), and vegetated fjords (blue box) are shown, respectively. The whiskers in (C) denote data within 1.5 × IQR (interquartile range), and the arrow indicates the anticipated increase in OC/MSA for WAP fjords after glacial retreat. D The predicted two-stage variation of sedimentary OC burial flux following deglaciation and vegetation expansion. The x-axis in (D) represents two successive stages (as separated by the dashed line) with their possible timescales labeled in the figure. The y-axis in (D) represents the dimensionless ratio of future to current OC burial flux in WAP fjords. The purple dot with standard deviation in (D) represents the current status of OC burial in WAP fjords, while the dashed area and solid line represent the model-simulated 25th–75th percentiles and 50th percentile of normalized future regional OC burial flux, respectively. Organo-mineral constraints on the ultimate fate of petrogenic organic carbon (OC) and the future OC burial flux in Western Antarctic Peninsula (WAP) fjords. A Distribution of OC% × fE≥215 (total organic carbon content × the proportion of OC with activation energy ≥ 215 kJ mol−1) of samples along Flandres Bay (AP-1 to AP-6). B Correlation between fraction modern carbon (F14C) and mineral surface area (MSA) values along two fjords. C OC/MSA (total organic carbon content/mineral surface area) of WAP fjords and from global compilations43. WAP in this study (purple box), deglaciating (gray box), and vegetated fjords (blue box) are shown, respectively. The whiskers in (C) denote data within 1.5 × IQR (interquartile range), and the arrow indicates the anticipated increase in OC/MSA for WAP fjords after glacial retreat. D The predicted two-stage variation of sedimentary OC burial flux following deglaciation and vegetation expansion. The x-axis in (D) represents two successive stages (as separated by the dashed line) with their possible timescales labeled in the figure. The y-axis in (D) represents the dimensionless ratio of future to current OC burial flux in WAP fjords. The purple dot with standard deviation in (D) represents the current status of OC burial in WAP fjords, while the dashed area and solid line represent the model-simulated 25th–75th percentiles and 50th percentile of normalized future regional OC burial flux, respectively. Uncertainties may arise from other potential CO2 sources (e.g., subglacial OCpetro decomposition and carbonate dissolution following sulfide oxidation), which are poorly constrained due to limited data54. However, CO2 efflux directly driven by sulfide oxidation is likely insignificant at present due to limited exposure, insufficient carbonate supply, and low reaction rates16,54. To account for subglacial OCpetro oxidation, we applied an inverse approach to assess the proportion of OCpetro oxidation required to shift the Antarctic Peninsula to a carbon source (Methods). Our quantitative results show that 67%–97% of OCpetro would need to be oxidized to fully offset the CO2 drawdown driven by OCbio burial. This proposed range is comparable to tropical regions (67%–85%)32,65, and is significantly higher than that estimated in Arctic fjords (25 ± 16%)66. However, extensive oxidation of bedrock OCpetro is unlikely to occur in the Antarctic Peninsula given the lithology, topography, and climate at present. By combining all the evidence, we propose that, although with uncertainties, regional OCbio burial flux exceeds OCpetro oxidation flux, indicating the Antarctic Peninsula is a net carbon sink at present. Carbon burial potential of WAP fjords modulated by organo-mineral interactions The interactions between OC and minerals provide constraints on the long-term stability of sedimentary OC22,45,67,68, which are critical for projecting regional net carbon balance with future climate change. Our dataset reveals a strong, positive correlation between mineral surface area (MSA) and F14C (Fig. 5B), indicating that sedimentary OC composition is regulated by minerals43. The general mineral protection mechanism and the specific organo-mineral associations in WAP fjords provide a basis for predicting future OC burial fluxes. Currently, sediment OC loadings (i.e., the ratios of OC normalized to MSA) in WAP fjords are generally lower than the 0.4 mg m−2 threshold reported for typical estuaries (Table S1), with the average value (0.26 ± 0.10 mg m−2; Table S1) being one-fifth of global fjords (1.35 ± 1.67 mg m−2) and about one-tenth of vegetated fjords (Fig. 5C; 2.32 ± 1.93 mg m−2)43. Notably, low OC loadings indicate unsaturated mineral surfaces for OC adsorption and great potential for carbon sequestration67. Despite a possible decrease in sediment supply (i.e., lower mass accumulation rates) to fjords after extensive glacial retreat (Fig. S4), the subsequent vegetation expansion in WAP fjords can substantially promote OC burial by increasing sediment OC loadings (Fig. 5). After systematic comparisons with deglaciating and vegetated fjord systems, we proposed a two-stage framework for future change in WAP fjords (i.e., deglaciation followed by vegetation expansion; Supplementary Information). Our model simulations, based on global datasets of sediment OC loadings and mass accumulation rates, project that OC burial flux will increase by one-third after deglaciation, and reach 2–4 times the present value after vegetation expansion (Fig. 5D; Supplementary Information). This prospective pattern is jointly regulated by potentially higher marine primary productivity8,10, more terrestrial vegetation18,69, and expanding benthic ecosystems19 after deglaciation, and is corroborated by increasing carbon sequestration in Antarctic coastal polynyas20 and terrestrial ecosystems69. Overall, these findings highlight the growing significance of WAP fjords in sedimentary OC burial, which enhances CO2 drawdown under climate change. Concurrently, climate change and land cover transition likely shift the Antarctic Peninsula from predominantly physical erosion to chemical weathering18, enhancing sulfide oxidation and microbial respiration of OCpetro32,66. However, the overall oxidation flux in the Antarctic Peninsula may be constrained by the restricted distribution of sedimentary rocks, low contents of bedrock OCpetro, and the reduced fluvial erosion after glacial retreat54,70, and is further compensated by enhanced silicate weathering15,30,54. In comparison, the production of high-MSA secondary minerals through chemical weathering can further increase carbon sequestration potential by enhancing preservation of OCbio (Fig. 5B). Therefore, the Antarctic Peninsula is suggested to be a persistent carbon sink under future climate change, while the accurate magnitude of net carbon balance is largely dependent on multiple factors, including catchment lithology, organo-mineral interactions, and microbial activity31,32,43. Overall, our study reveals that both the current and long-term net carbon balance in Antarctica are primarily modulated by catchment lithology and organo-mineral interactions, which is a modern analog to glacial–interglacial transitions in polar regions. Methods Samples Twelve surface sediments (0–5 cm) in the Antarctic Peninsula fjords were collected from six cruises since 1982. Six core-top sediments were sampled from Flandres Bay in the Northwest Antarctic Peninsula and were denoted as AP. Sediments were sequentially numbered from the innermost (i.e., proximal to glacier front) AP-1 to the outermost AP-6 (Fig. 1). Another six core-top sediments were sampled from Lapeyrère Bay on the adjacent Anvers Island (i.e., AI-1 to AI-6) following the same proximal-to-distal gradient. All sediments were freeze-dried before analysis. A small proportion of each sample was saved for mineral surface area (MSA) measurement, while the rest was ground, homogenized, and acidified before the analysis of organic carbon (OC). Mineral surface area measurement The surface area of each sediment was measured following the 10-point Brunauer-Emmett-Teller (BET) method. About 0.5 g unacidified sediment was combusted at 350 °C for 12 h to remove most OC and degassed at 200 °C prior to analysis with a Micromeritics TriStar II 3020 surface area analyzer71. The relative standard deviation of measurement is, on average, less than 2% based on duplicates. Total organic carbon and stable carbon isotope measurement Total organic carbon content (OC%) and δ13C of OC were measured following acid fumigation72. An aliquot of 20–30 mg sediment was weighed in a silver capsule with the addition of Milli-Q water (18.2 MΩ). A 96-well cell culture plate containing the weighed sediments was placed in an enclosed desiccator, with a beaker containing about 50 mL of 12 N aqueous HCl, for 12 h to remove carbonates. After acid fumigation, sediments were oven-dried at 50 °C for 48 h and then wrapped into tin capsules. All acidified sediments were combusted at 950 °C in a Flash elemental analyzer (EA) coupled to a Thermo Scientific DELTA V Advantage isotope ratio mass spectrometer (IRMS). The OC% and δ13C (‰, VPDB) of samples were calibrated against replicates of three standard samples (i.e., USGS-40, USGS-62, and USGS-64). The standard deviations are <2% (relative) for OC% and <0.1‰ (absolute) for δ13C, respectively, based on USGS-40 (n = 5) and duplicates of samples. Radiocarbon analysis Radiocarbon analysis was carried out using the zinc reduction method73. Sediments were immersed in 1 N aqueous HCl for 12 h, washed with Milli-Q water, and freeze-dried to successively remove carbonates, residual HCl, and residual water. Each decarbonated sediment with 50–60 mg copper oxide was introduced by a glass conduit to the bottom of a quartz tube and flame sealed on a vacuum line in the Sedimentary Organic Geochemistry Lab at Shanghai Jiao Tong University. The sealed quartz tube was combusted at 900 °C for 6 h to completely convert OC to CO2. The CO2 gas was further cryogenically purified on the vacuum line and reduced by zinc and titanium hydride to graphite. After graphitization, radiocarbon isotopic compositions (reported as F14C) were measured by accelerator mass spectrometry (AMS) at the Keck Carbon Cycle Accelerator Facility, University of California, Irvine. Based on the primary standard (OX-II; F14C = 1.3407) and an in-house blank (Irati-T2; F14C = 0), modern carbon contamination and dead carbon contamination are estimated to be 0.9 ± 0.45 μg and 0.001 ± 0.000 μg, respectively74. The precision of F14C is better than 0.002. RPO−13C/14C analyses The system of ramped pyrolysis/oxidation (RPO) utilized here has been detailed in previous studies75,76. Following a recent study, sediment decarbonated by 1 N aqueous HCl was analyzed using an in-house assembled RPO system76. An aliquot of acidified sediment containing approximately 0.5 mg OC was introduced to the bottom of a quartz reactor and underwent 5 °C min−1 heating from 50 °C to >1050 °C in a pyrolytic furnace. Carrier gas compositions are 27 mL min−1 helium + 3 mL min−1 5% oxygen (inner tube) and 5 mL min−1 pure oxygen (outer tube). Throughout the analyses, carrier gas flows, furnace temperatures, concentrations of evolved CO2 gas, and the cumulative amount of CO2 gas were recorded. The concentration of evolved CO2 from OC decomposition was calibrated against a diluted CO2 (2000 ppm) reference gas. Four representative sediments along two fjords (i.e., AP-2, AP-5, AI-2, AI-5) were selected for RPO−13C/14C analyses to further characterize OC provenances and spatial dynamics. Throughout RPO analysis, evolved CO2 gas was sequentially collected as four temperature-based fraction splits (i.e., RPO fractions; Table S2). Each fraction was cryogenically trapped in a dual-inlet glass toggling trap and purified on the vacuum line in the Sedimentary Organic Geochemistry Lab at Shanghai Jiao Tong University. After purification, fractional stable carbon isotopes (reported as δ13C) were measured on a Thermo Fisher Gasbench II interfaced with a continuous-flow Delta V Plus IRMS in the Gas Stable Isotope Lab at Shanghai Jiao Tong University; fractional radiocarbon isotopic compositions (reported as F14C) were analyzed at the Keck Carbon Cycle Accelerator Facility, University of California, Irvine. All fractional isotopic values were calibrated against standards and blanks. Modern and dead carbon contaminations were estimated to be 0.8 ± 0.4 and 0 ± 0 μg, respectively, for each fraction (Supplementary Information). The estimated RPO blank accounts for less than 1% (mostly less than 0.5%) of the carbon mass for each RPO fraction, suggesting the RPO-13C/14C results are minimally affected by contamination from experiment procedures77. The precision of measurements of δ13C and F14C is better than 0.1‰ and 0.002, respectively. Data processing The apparent radiocarbon age of each sample was derived from F14C value78, as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\,{Radiocarbon}\,{age}=-8033\times {{\mathrm{ln}}}\left({F}^{14}C\right)$$\end{document}Radiocarbonage=−8033×lnF14C The average petrogenic OC% and the F14C value of biospheric OC (OCbio) were estimated by a linear regression approach40. The linear regression between OC%× F14C (y) and OC% (x) was performed following the least square method, as:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y={\beta }_{0}+{\beta }_{1}x$$\end{document}y=β0+β1x The average petrogenic OC% in a fjord was given by the intercept of Eq. (2) on the x-axis, while biospheric F14C value was approximated as β1. The CO2-temperature array generated by RPO was further converted to the density distribution of activation energy (p[E]) by the inverse model42. The parameters for inversion processes (i.e., the pre-exponential factor and Tikhonov regularization factor) were adopted after Hemingway et al.42. Three RPO parameters were calculated to characterize E distributions43,45 (Table S2). The mean of E distribution was calculated as:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{E}=\int^{+\infty }_{0}{Ep}\left[E\right]{dE}$$\end{document}μE=∫0+∞EpEdE \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\#\end{array}$$\end{document}#The standard deviation of E distribution was calculated as:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{E}=\sqrt{{\mu }_{{E}^{2}}-({{\mu }_{E}})^{2}}$$\end{document}σE=μE2−(μE)2 The proportion of OC within a given E range from a to b kJ mol−1 was calculated as:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f}_{a\le E < b}=\int^{b}_{a}p\left[E\right]{dE}$$\end{document}fa≤E