Introduction

Although wetlands cover only 5%–8% of the global terrestrial area, they contribute 20%–30% of the global organic carbon (OC) stock in soils1,2,3. As OC in wetlands generally accumulates in the surface layer from plant and aquatic inputs, most research focuses on surface sediment layers2,4. The OC stock in deeper sediment layers ( > 1 m) remains highly controversial. Some studies suggest that it is typically low5,6, while other studies have shown that under specific geological historical sedimentation conditions (such as the Holocene thermal maximum), the amount of OC inputs into sediments may exceed that in the modern surface sediments, resulting in higher OC stocks in deep sediments3,7,8.

The primary reason that deep OC is often overlooked may lie in its “recalcitrance”. Generally, OC stability increases with burial depth due to a higher proportion of recalcitrant compounds and aromatic structures that resist decomposition9. Moreover, the anoxic (or anaerobic) and relatively cold conditions limit microbial metabolic activity and reduce the carbon decomposition rate. However, significant uncertainties persist regarding the OC stability of deep wetland sediments. For instance, in alpine wetlands with high anaerobic conditions, the aliphatic content increases with depth in catotelm layers, indicating a relative decrease in OC stability in deeper layers10,11. These uncertainties may stem from environmental factors such as climate and burial conditions. Climate modulates sediment OC stability by altering hydrological conditions, plant communities, and associated ecological processes through changes in temperature and precipitation3,10,12,13,14. Burial conditions, shaped by topography and long-term sedimentation, determine the storage environment of OC, thereby regulating its decomposition dynamics15,16. The effects of this factor manifest over centennial timescales and have been largely overlooked in ecological research. Once deeply buried (i.e., during flooding), OC decomposition rates may decrease, leading to an increased organic carbon accumulation rate (OCAR)16,17. For example, fine grained sediments, such as clay and silt, provide better adsorption and aggregation sites for labile OC, thereby reducing microbial decomposition and preserving OC lability18. Nevertheless, the specific mechanisms through which historical climate fluctuations and hydrological regimes have collectively influenced sedimentary carbon sources and burial dynamics, thereby mediating OC accumulation and stability in deep sediments, remain rarely investigated.

Hailed as the “Asian Water Tower”, the Qinghai-Xizang Plateau (QXP, also known as the Tibetan Plateau) encompasses over 100,000 km2 of wetlands and serves as one of the largest carbon reservoirs in China19. Most studies regarding these critical alpine wetlands have prioritized the surface soil carbon pool. Limited studies on deep sediment OC have primarily focused on carbon accumulation rates and timelines, suggesting that sediment OC formation in most wetlands dates back to the early Holocene20,21,22. However, the rapidly rising temperatures23,24, intensifying climatic extremes25, and earthquake-induced fractures26 have significantly elevated the risk of decomposition and release of deep OC, which was previously buried in cold, anoxic environments and is generally considered highly recalcitrant. Such processes are likely to generate positive feedback on global warming and urgently require in-depth investigation.

To address these knowledge gaps, we performed 42 sediment cores (each ranging from 5.7 m to 23.5 m in depth) across seven typical wetlands in central and southern QXP to assess the vertical distribution of sediment OC content and the OC stocks in deep sediments (Fig. 1). Moreover, we selected the sediment core Naqu4 (NQ4) with the most continuous sedimentary sequence to investigate sediment OC stability and historical drivers by integrating Fourier-transform infrared spectroscopy (FTIR), X-ray fluorescence (XRF), and grain size analysis. Our findings uncover that deep sediments (over 1 m and above the first aquiclude; see Methods) store ~70% of the total OC in alpine wetlands, a substantial portion of which is labile. Both historical warming and an intercalated sedimentary structure, formed by silt and fine-grained clay, are essential drivers of these deep-buried labile OC pools.

Fig. 1: Drilling location map. The selected seven wetlands are marked by red stars.
Fig. 1: Drilling location map. The selected seven wetlands are marked by red stars.
Full size image

The distribution of alpine wetlands is highlighted in green65, while Quaternary swamp and swamp-lacustrine sedimentary facies are represented in blue66. The wetlands include Naqu (NQ), Niyaqu (NYQ), Shenzha (SZ), Dangqu (DQ), Yadong (YD), Jiali (JL), and Zhongba (ZB). The inset map on the upper right highlights the location of Xizang Autonomous Region within China (green shading).

Results and discussion

Sediment OC content and stocks

Across seven wetlands, OC content generally exhibited an overall increasing trend since the Holocene (Fig. 2). Surface sediments ( < 1 m) contain the highest OC content, reaching up to 142.69 g kg−1, which aligns with the conventional understanding that the uppermost layers serve as primary OC reservoirs due to continuous organic matter deposition and constraints on microbial decomposition2,3,4. However, our results challenged the previous understanding that OC content declines monotonically with depth6,27. In our study, many deep-sediment layers exhibited OC content comparable to or exceeding surface sediments. For instance, at depths of 3.2 m in NQ4 and 2.4 m in Dangqu (DQ), OC content reached 144.64 ± 5.04 g kg−1, surpassing surface layer content of 133.65 ± 25.63 g kg−1 (p = 0.04). Similar results were also observed at depths of 1.2 m, 2.6 m, and 4 m in Shenzha (SZ), as well as at ~15.5 m in ZB, where the average deep-sediment OC content accounts for 72.63% to 146.80% of those in the surface layer. These findings are consistent with observations by Chen et al. 28, who reported elevated OC concentrations in deep-sediment layers (2–5 m) of the Zoige peatland on the QXP. The long-term preservation of high OC content in deeper sediments likely results from substantial historical organic matter deposition, coupled with burial conditions that suppress decomposition. Low temperatures and oxygen levels create an environment that constrains microbial activity, thereby reducing OC turnover and promoting long-term carbon sequestration in wetland sediments28,29,30.

Fig. 2: Vertical distribution of organic carbon (OC) content in the seven typical wetlands on the Qinghai-Xizang Plateau.
Fig. 2: Vertical distribution of organic carbon (OC) content in the seven typical wetlands on the Qinghai-Xizang Plateau.
Full size image

Naqu (NQ) (a), Niyaqu (NYQ) (b), Shenzha (SZ) (c), Dangqu (DQ) (d), Yadong (YD) (e), Jiali (JL) (f), and Zhongba (ZB) (g). The solid blue line indicates the sediment depth corresponding to the onset of the Holocene. The solid red line indicates the sediment depth corresponding to the bottom boundary of the unconfined aquifer.

Moreover, deep sediments stored a disproportionately large fraction of total OC stocks, contributing ~70% of the total active OC pool, extending from the surface to the lower boundary of unconfined aquifers (Table 1). Within this depth range, deep-buried OC may be released into surrounding surface water bodies and subsequently enter the atmosphere due to intense hydrological processes in the shallow aquifers of wetlands. These results align with many previous studies, which reported that deep sediment OC stocks could account for more than ~60% of the total OC storage, underscoring their substantial yet often overlooked role in carbon sequestration27,31. Such findings emphasize that deep sediments play a far more significant role in long-term carbon sequestration than previously recognized on the QXP. Current carbon stock assessments and carbon cycle models predominantly focus on surface sediments, potentially leading to an underestimation of wetland carbon storage capacity3,5,6. Given this gap, understanding the accumulation mechanisms and stability of deep-sediment OC is critical for assessing future climate change impacts. Hence, we analyzed these aspects using the most continuous sedimentary sequence (NQ4; see “Methods”).

Table 1 Organic carbon (OC) stocks and their relative proportions at different sediment depths (mean ± 95% CI) in typical wetlands
Full size table

The OC accumulation and its associations with historical climate changes

The NQ4 sediment core, located near the Naqu River, is characterized by extensive Quaternary swamp and swamp-lacustrine sedimentary facies beneath the modern wetlands. Chronological analysis indicated a maximum age of 35.716 ± 1.04 cal ka BP (where 0 cal ka BP corresponds to AD 1950) (Table S2 and Figure S1a). The Holocene sediment accumulation rate (SAR) averaged approximately ~0.745 mm a−1 (Figure S1b).

The OCAR, derived from sediment OC content and age, fluctuated since the Holocene (Fig. 3a). This trend closely paralleled variations in Ca/Si and Mo inc/Al ratios (r = 0.38 and p < 0.01, r = 0.41 and p < 0.05, respectively; Figure S3). In contrast, OCAR exhibited an inverse relationship with Rb/Sr and Fe/Ca ratios (r = −0.46 and p < 0.01, r = −0.39 and p < 0.05, respectively; Fig. 2a and S3). During the early Holocene (8.2–11.7 cal ka BP), OCAR remained relatively low ( ~ 4.87 g m−2 a−1), coinciding with reduced Ca/Si and Mo inc/Al values and elevated Rb/Sr and Fe/Ca ratios. These geochemical patterns suggest that arid conditions prevailed in Xizang, leading to a slower sediment accumulation rate within oxidative depositional environments32. Consistent with paleoclimate records from other wetlands on the QXP during the same period, the mean annual temperature (MAT) at Ngamring Co33 was 5.96 ± 0.76 °C lower than modern levels, and the δ18O value at Selin Co34 was positive (average of −3.41 ‰) (Fig. 4). This regional pattern confirms that the early Holocene was dominated by a cold and dry climate, which suppressed wetland productivity and organic carbon accumulation. In contrast, from the middle to late Holocene (after 8.2 cal ka BP), OCAR exhibited a multimodal trend, with pronounced peaks at ~5.2 cal ka BP (61.32 g m−2 a−1), ~2.9 cal ka BP (66.37 g m−2 a−1), ~1.9 cal ka BP (52.31 g m−2 a−1), and ~1.1 cal ka BP (40.35 g m−2 a−1). These peaks coincided with the highest Ca/Si and Mo inc/Al values and relatively low Rb/Sr and Fe/Ca ratios, indicating a climatic shift toward warmer and wetter conditions. These warm-wet phases enhanced wetland productivity, increasing the input of organic matter, while the generally cold conditions of the QXP further suppressed microbial decomposition, collectively fostering significant OC accumulation35. This transition is further corroborated by lacustrine sedimentary records from Dagze Co36 and Ngamring Co33. For example, during the Holocene Thermal Maximum phase at ~5.2 cal ka BP, the vegetation cover reconstructed from Dagze Co pollen data reached a peak of 62.03%36, while the δD record from Ngamring Co sediments showed a relative enrichment value of −220‰33. Additionally, the OCAR peak during this period was temporally coupled with the drift ice record from the North Atlantic (a low value of 3.62%, phase difference < ± 0.1 ka)37, further suggesting a possible remote correlation between high-latitude climate forcing and carbon dynamics on the QXP. These findings underscore the essential role of historical climate in forming wetland carbon pools. Accordingly, our analyses further investigated its influences on carbon stability.

Fig. 3: Vertical distribution of organic carbon accumulation and associated sedimentary properties in core Naqu4.
Fig. 3: Vertical distribution of organic carbon accumulation and associated sedimentary properties in core Naqu4.
Full size image

Organic carbon accumulation rate (OCAR) and XRF-derived geochemical proxies (K, Ti, Rb/Sr, Fe/Ca, Ca/Si, Mo inc/Al) (a); FTIR-derived functional groups (aromatic, alkane, and aliphatic) and the degree of aliphatization (DA) in organic matter (b); and Sediment texture parameters, including clay, silt, and sand contents, mean grain size, and sorting (c). The red lines separate the surface ( < 1 m) from deep sediments.

Fig. 4: Climate records from the central Qinghai-Xizang Plateau and monsoon-dominated regions since the Holocene.
Fig. 4: Climate records from the central Qinghai-Xizang Plateau and monsoon-dominated regions since the Holocene.
Full size image

The dataset includes the organic carbon accumulation rate (OCAR) from the NQ4 core (this study) (a), mean annual temperature (MAT) from Ngamring Co33 (b), δ18O from Selin Co34 (c), and δD from Ngamring Co33 (d). Reconstructed vegetation cover based on pollen data from Dagze Co36 (e) and drift ice record from the North Atlantic37 (f) are also shown.

The OC stability in deep sediments

The aliphatic content and its ratio to aromatic content are widely recognized indicators for characterizing OC stability9,38. Our results indicated that the relative content of aliphatic functional groups (1420/1040) and the degree of aliphatization [(2850 + 2919)/790] closely tracked variations in deep-sediment OC content since the Holocene (Fig. 3b). Both parameters exhibited a significant positive correlation with OC content, with correlation coefficients of 0.94 and 0.93 (p < 0.001 and p < 0.001, respectively; Figure S3). These findings suggest that OC-rich layers, particularly in deep sediments, contain higher levels of labile OC. However, in contrast to the observations of Routh et al. 9 and Kalisz et al. 10,11, our results did not show a continuous increase in labile OC with depth. Instead, labile OC fluctuated, with markedly higher degrees of aliphatization observed at specific depths, particularly at ~1.3 m and ~3.2 m. The degrees of aliphatization at these depths reached 0.49 and 0.43, respectively, substantially exceeding the peak value of 0.23 observed in surface sediments. These results highlight a more complex relationship between deep-sediment OC stability and depth, suggesting the existence of other drivers, such as historical climate.

Furthermore, we identified sedimentary structure as a crucial driver of OC stability in deep sediments. Variations in sediment density (BD), validated using X-ray tomography (XRT) intensity as an independent physical proxy (linear negative correlation; R2 = 0.29, p < 0.001), reflect temporal shifts in sediment composition and packing (Figures S6, S7). Generally, clay content increased with depth within the 0–1 m range and below 3 m, while fluctuating between 1–3 m. In contrast, silt and sand fractions peaked at depths of ~1.3 m (~1.9 cal ka BP) and ~3.2 m (~5.2 cal ka BP), and exhibited a clear trend toward finer grain sizes (average grain size increasing from ~5.5 φ to ~6.7 φ) above these depths, accompanied by moderate sorting (1.9–2.4 φ), with both depths covered by high-clay layers (Fig. 3c). This distinct stratigraphic pattern suggests alternating depositional regimes linked to climatic variability10,28. During warm-wet periods (e.g., ~5.2 and ~2.9 cal ka BP), enhanced runoff led to coarser silt deposition with higher OC input, forming low-density layers (BD < 0.9 g cm−³). During dry-cold phases (e.g., ~4.2 and ~1.3 cal ka BP), fine-grained clay deposition dominated, increasing BD and limiting oxygen diffusion. The overlying clay layers formed a protective intercalated structure that limited oxygen exposure and physical disturbance, thereby enhancing OC preservation. Moreover, a strong correlation was observed between the degree of aliphatization and clay or silt content (r = −0.59 and p < 0.01, r = 0.54 and p < 0.01, respectively; Figure S3), suggesting that the intercalation of clay and silt layers played a critical role in facilitating the accumulation and preservation of labile OC in deep sediments.

Drivers of the OC stability in deep sediments

To integrate the abovementioned findings, the degree of aliphatization showed significant correlations with OCAR (R2 = 0.56, p < 0.001), clay (R2 = 0.33, p < 0.001), and silt (R2 = 0.21, p = 0.002) (Fig. 5a). The OCAR was positively correlated with MAT (R2 = 0.26, p < 0.001) and silt (R2 = 0.29, p < 0.001), and negatively correlated with sediment age (R2 = 0.33, p < 0.001), depth (R2 = 0.21, p = 0.002), and clay (R2 = 0.22, p = 0.002) (Fig. 5b). A partial least squares path modeling (PLS-PM) was performed to identify the key drivers of OC stability in deep sediments. The analysis revealed OCAR, MAT, and silt content as key factors influencing sediment OC stability (Fig. 6).

Fig. 5: Relationships of organic carbon properties with environmental and sedimentary variables.
Fig. 5: Relationships of organic carbon properties with environmental and sedimentary variables.
Full size image

Relationships between the degree of aliphatization (DA) and its potential influencing variables (a); and Relationships between organic carbon accumulation rate (OCAR) and its potential influencing variables (b). MAT: mean annual temperature; MAP: mean annual precipitation; Vegetation: vegetation cover.

Fig. 6: Partial least squares path modeling (PLS-PM) illustrates the effects of climatic variables and burial conditions on organic carbon stability since the Holocene.
Fig. 6: Partial least squares path modeling (PLS-PM) illustrates the effects of climatic variables and burial conditions on organic carbon stability since the Holocene.
Full size image

MAP mean annual precipitation, MAT mean annual temperature, OCAR organic carbon accumulation rate. Significance levels are marked as follows: *p < 0.05, **p < 0.01, and ***p < 0.001. Orange solid lines represent positive effects, while blue dashed lines indicate negative effects. Line thickness reflects the interaction strength between variables.

The high path coefficient from OCAR to sediment OC stability (0.99, p < 0.001) suggests that variations in carbon input predominantly control the preservation of labile OC pools. The MAT indirectly enhanced sediment OC stability by increasing OCAR, with a statistically significant path coefficient of 0.99 (p < 0.001). This result is consistent with previous studies, which reported that elevated OCAR in northern peatlands was closely linked to the warm Holocene climate13,20,21. During that period, rapid OC accumulation occurred alongside suppressed decomposition under anaerobic conditions (often indicated by high Mo inc/Al ratios; Fig. 3a), facilitating the burial of a substantial proportion of labile carbon in these sediment layers39,40.

Sediment structure exerts a key physical control over the long-term stability of buried OC. Silt content enhances stability primarily by increasing OCAR (path coefficient = 0.51, p < 0.001), likely due to its association with high sedimentation rates and rapid burial that limit oxygen exposure. In contrast, although negatively correlated with OCAR (Fig. 5b), clay content supports OC stability by forming compact, low-porosity layers that reduce oxygen diffusion and protect the underlying higher proportion of labile OC. Additionally, these clay layers may promote the development of stable clay-humus complexes through interactions with organic matter41,42. Over time, ongoing deposition, burial, and compaction create increasingly oxygen-limited conditions above the rapidly accumulated silt layers, thereby suppressing decomposition and enhancing labile OC preservation. Such an intercalated sedimentary structure is particularly pronounced in wetlands and coastal tidal flats with high sedimentation rates43,44, representing an overlooked geological mechanism mediating wetland carbon stability.

Unlike many existing studies35,45, we found that mean annual precipitation (MAP) negatively affected sediment OC stability by influencing OCAR, with a path coefficient of −0.46 (p < 0.05). This unusual finding is likely because the study site is a riverine wetland. High precipitation may induce strong hydrological disturbances (e.g., floods), undermining vegetation growth and reducing litter input25. Besides, coarsening and poor sorting of sediment grains may promote oxygen and pore water exchange, leading to increased microbial aerobic respiration and lateral output of dissolved OC via surface runoff, consequently exacerbating OC losses and constraining sediment OC retention46,47.

Among the factors examined, variables such as sediment age-depth and vegetation exhibited relatively minor effects (p > 0.05). Overall, sediment OC stability was primarily governed by sediment grain characteristics, mineral interactions, and the accessibility of OC to microbial degradation rather than by burial age or depth. Under the anoxic conditions prevalent in wetlands, the influence of age and depth on sediment OC oxidation was significantly attenuated10,11,48. Notably, our study did not reveal a strong vegetation effect on OC stability, which contrasts with findings from many ecological studies49,50. This is probably because the long-term cumulative effects of historical vegetation dynamics may have been weakened by surface sediment disturbances driven by climate variability or shifts in burial conditions. As a result, OC stability in deep sediments was more strongly dictated by mineral-associated protection mechanisms rather than by the residual influence of past vegetation51,52. Additionally, the vegetation data used in this study were palynologically reconstructed, a method limited by low species resolution and potential biases from windborne pollen and spores. Future studies could involve sedimentary ancient DNA (sedaDNA) analysis to obtain a more accurate assessment of historical vegetation effects on OC accumulation53,54.

Taken together, our findings highlight the complex interplay between the warm climate and sedimentation processes in shaping the long-term stability of sediment carbon pools in alpine wetlands.

Conclusions

This study on the Qinghai-Xizang Plateau reveals that, unlike well-studied alpine grasslands, OC in wetlands does not decline linearly with depth. Deep sediments in the unconfined shallow aquifer store ~70% of the total OC, much of which is labile, challenging the prevailing assumption that OC in deep soils is both scarce and highly recalcitrant. While temperature remains a crucial driver of deep-sediment OC accumulation and chemical stability, our work introduces an overlooked geological mechanism in which the “clay-silt-clay” sedimentary structure mediates OC stability. These findings underscore the need to include these long-neglected labile carbon pools in climate change research and assessments, particularly as rising temperatures24, intensifying extreme precipitation events55, and earthquake-induced fractures26 on the QXP may trigger deep OC decomposition, thereby exacerbating global warming. Recognizing the susceptibility of these deep-sediment OC pools to environmental perturbations is crucial for refining predictions of carbon cycling and their potential influence on future climate trajectories. This requires interdisciplinary research across ecology, hydrology, geochemistry, and paleoclimatology, as well as new methods such as sedimentary ancient DNA analysis.

Methods

Study area and sampling methods

The study was conducted in the Xizang Autonomous Region of China, located between 78°25’–99°06’E and 26°50’–36°53’N, with an average altitude of ~4500 m. The region experiences a highly variable climate, characterized by cold and dry conditions in the northwest, and warmer and more humid conditions in the southeast. The annual temperature ranges from −2.8 to 11.9 °C, while precipitation varies from 74.8 mm to 901.5 mm. Dominant vegetation species in the region include Kobresiatibetensis, Carex parvula O. Yano, and Bistorta vivipara (L.) Gray.

Fieldwork was conducted across seven typical alpine wetlands located in Naqu (NQ), Niyaqu (NYQ), Shenzha (SZ), Dangqu (DQ), Yadong (YD), Jiali (JL), and Zhongba (ZB) (Fig. 1, Table S1), most of which were situated within national parks or nature reserves. No signs of land cultivation, drainage, or other disturbances were observed in the wetland areas of the drilling sites. The drilling depths ranged from 5.7 m to 23.5 m, penetrating the unconfined groundwater layer and exposing one or more confined groundwater layers. At each site, a total of six duplicate cores (90 mm in diameter) were collected at ~100 m intervals. The collected cores were cryo-sealed for transport to the laboratory, where they were longitudinally split and photographed. Subsamples were then dried, weighed, ground, and sieved through a 2 mm sieve to remove visible stones, while retaining plant debris. The physicochemical properties of the samples, including organic carbon (OC) content and grain size, were analyzed. The OC content was determined using the dichromate oxidation-colorimetric method56. To minimize potential interference from Fe and Ca during OC determination by dichromate oxidation, phosphoric and sulfuric acids were added to chelate Fe3+ and prevent CaSO4 precipitation, while blank corrections were applied to account for residual inorganic reducing agents. The grain size distribution was analyzed using a laser grain size analyzer (Microtrac S3500, USA), which provided data on the content of clay, silt, and sand. The mean grain size and sorting were calculated according to the method of Folk and Ward (1957)57. To determine bulk density, 157 undisturbed sediment samples were collected from the vertical profiles of 21 sediment cores using a standard 100 cm3 container, accounting for over 10% of all core samples (n = 1537). Each sample was dried at 105 °C to a constant weight. Bulk density (BD) was then calculated as the ratio of dry mass to the volume of the container. Following previous research dealing with a large amount of drilling samples6,58, an empirical model between bulk density and sediment OC content was established using these 157 samples (R2 = 0.54; Figure S2). To assess model robustness, we recalibrated the BD-OC regression using 10-fold cross-validation (R2 = 0.54; RMSE = 0.24 g cm−3) and performed residual diagnostics, which showed no systematic bias (Durbin-Watson test, p = 0.36). Notably, although cross-validation and Monte Carlo propagation confirmed its robustness, high- and low-OC extremes may still be underestimated. The model was then used to assess the BD of other sediment samples collected from the wetland cores.

Sediment dating

A total of 33 bulk sediment samples were selected from the seven wetlands for radiocarbon dating, with 3–10 samples per site according to the sediment depth and number of layers (Table S2). The samples were freeze-dried, sealed, and subsequently analyzed for radiocarbon content using accelerator mass spectrometry (AMS, USA) at the Miami Beta Analysis Company. The obtained radiocarbon dates were calibrated using the IntCal20 calibration dataset59. To construct an age-depth model for the entire depositional sequence, the calibrated dates were interpolated through a Bayesian approach. The depth marking the start of the Holocene was identified through this interpolation method. All analyses were performed using the Bacon 2.2 package in R 3.1.060.

Fourier-transform infrared spectroscopy (FTIR) for OC stability

The FTIR spectroscopy was conducted using a Cary 660 FTIR spectrometer (Agilent, USA) to analyze the molecular composition of natural organic matter in sediment samples. For analysis, each sample was prepared by mixing 2 mg of powdered sediment with 200 mg of FTIR-grade KBr, which was then pressed into 13 mm pellets. Spectra were obtained by averaging 32 scans per sample, with baseline corrections applied after subtracting the KBr background61. To precisely determine the wavenumber positions of characteristic peaks, we employed the baseline correction method and R script developed by Hodgkins et al.61. Peak intensities were then converted into relative abundances (Figure S5).

Four functional groups—aromatic, carbohydrate, alkane, and aliphatic—were analyzed to quantify the chemical stability of OC in the sediment samples. The relative content of each functional group was determined using specific peak ratios: aromatic (790/1040), alkane (1420/1040), and aliphatic (2850/1040 and 2919/1040). Additionally, the degree of aliphatization was defined as the ratio of aliphatic to aromatic content, calculated as (2850/790 + 2919/790).

X-ray fluorescence (XRF) scanning for palaeoenvironmental and palaeoclimatic analyses

The sealed, segmented core samples were transported under cold-chain conditions to the Tibetan Plateau Key Laboratory of Earth System and Environment at the Institute of Tibetan Plateau Research, Chinese Academy of Sciences (ITP-CAS). Elemental composition, optical images, and radiographic images were analyzed using an Itrax XRF core scanner (Cox Itrax, Sweden). XRF scanning was performed using a 3 kW molybdenum target tube, operated at 30 kV and 50 mA, with a 1 mm resolution and a 15-second exposure time. X-ray imaging was captured at 30 kV and 10 mA, also at 1 mm resolution, with a 100 ms exposure time. The raw XRF spectra were processed using Q-Spec software to derive elemental peak area counts and X-ray transmission (XRT) intensity.

To analyze palaeoenvironmental and palaeoclimatic conditions, specific elemental ratios were selected, namely, potassium (K), titanium (Ti), rubidium to strontium (Rb/Sr), iron to calcium (Fe/Ca), calcium to silicon (Ca/Si), and molybdenum to aluminium (Mo inc/Al). These ratios serve as proxies for various environmental and climatic conditions: K and Ti indicate dust input and debris transport during dry periods; Rb/Sr and Fe/Ca reflect the dominance of chemical weathering over physical weathering, while Ca/Si indicates more substantial physical weathering; Mo inc/Al serves as a proxy for paleo-redox conditions62,63.

Statistical analyses

In order to quantify the distribution and proportion of OC stocks in the sediments of alpine wetlands at various depths, we selected six duplicate core samples from seven representative wetlands in Xizang. These samples were analyzed for OC content and dated to determine the onset of the Holocene. Additionally, to explore the mechanisms underlying the formation of deep sediment OC, we selected core #4 from Naqu, an area characterized by extensive swamp sedimentation, swamp-lacustrine sedimentation, and a widespread distribution of modern wetlands. We analyzed the grain size, FTIR spectra, and XRF scans to assess OC stability, while incorporating paleoenvironmental and paleoclimatic data via partial least squares path modeling (PLS-PM) to identify the primary drivers. To minimize random errors, grain size and FTIR data were smoothed using a 5-point moving average, and XRF scan data were smoothed using a 20-point moving average.

Additional palaeoenvironmental and palaeoclimatic data were collected from the vicinity of NQ4 to serve as a reference. These data were used to examine relationships between mean annual temperature (MAT), mean annual precipitation (MAP), vegetation cover, sediment age and depth, clay and silt content, organic carbon accumulation rate (OCAR), and the degree of aliphatization. Prior to multivariate modeling, we first performed simple linear regression analyses to evaluate the individual effects of each predictor variable on OCAR and the degree of aliphatization. These regressions provided a quantitative basis for identifying potential key drivers, and the results are presented in Fig. 5. To reduce dimensionality and avoid collinearity, age and depth were subsequently merged via principal component analysis (PCA), with PC1, explaining 97.6% of the variance. Variance inflation factor (VIF) values for all retained predictors were below five. Building on these preparatory steps, PLS-PM was employed to identify the pathways influencing sediment OC stability. This method is particularly useful for demonstrating cause-and-effect relationships between observed and latent variables64. The MAT data were obtained from Sun et al. 33, while the MAP data were derived from the δ18O results presented by Gu et al. 34. Both climate variables were aligned with sample ages based on the established chronology. Vegetation data for the region were sourced from the reconstructed vegetation cover by Liu et al. 36 to ensure both accuracy and geographic relevance. Other parameters, including sediment age and depth, clay and silt content, OCAR, and degree of aliphatization, were obtained through measurements conducted in this study. This statistical model was constructed using the “innerplot” function in the “plspm” package64. All statistical analyses were conducted in R version 4.1.3.

Organic carbon density (OCD) and OC stock for a given wetland sediment depth were calculated using Eqs. (1) and (2), respectively:

$${OCD}=\frac{{\sum }_{i=1}^{n}O{C}_{i}\times B{D}_{i}\times {H}_{i}}{{\sum }_{i=1}^{n}{H}_{i}}$$
(1)
$${OC}\, {\mbox{stock}}=\frac{A\times {\sum }_{i=1}^{n}O{C}_{i}\times B{D}_{i}\times {H}_{i}}{1000}$$
(2)

where OCi, OCD, BDi, and Hi represent the average organic carbon content (g kg-1), organic carbon density (kg m-3), bulk density (g cm-3), and equal interval depths (cm) of six cores in per wetland, respectively. The OC stock represents the wetland sediment organic carbon stock (Tg), while A represents the wetland area (km2).

Sediment accumulation rate (SAR) and OCAR were calculated using Eqs. (3) and (4), respectively:

$${SAR}=\frac{{H}_{i}\times 10}{{T}_{i}}$$
(3)
$${OCAR}=\frac{{H}_{i}\times B{D}_{i}\times O{C}_{i}\times 1000}{{T}_{i}}$$
(4)

where SAR, OCAR, and Ti represent the sediment accumulation rate (mm a-1), OCAR (g m-2 a-1), and the age at equal interval depths Hi (cm), respectively (Figure S1).

To propagate the model uncertainty to OC stock and OCAR, we ran 10,000 Monte Carlo iterations. In each iteration, we randomly sampled BD from its 95 % prediction interval (conditional on OC content) and recalculated OC stock and OCAR. The 2.5% and 97.5% quantiles of the resulting BD, OC stock, and OCAR sets provided 95% confidence intervals (Table 1 and Figure S7), and the resulting 95% confidence intervals for OC stock estimates ( ± 1.56% to ±4.87%) were considerably lower than the spatial variability observed across sites (coefficient of variation = 44.14%), suggesting that the model-induced uncertainty is substantially smaller than natural heterogeneity.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.